@article{
 title = {Co-ECL: Covariant Network with Equivariant Contrastive Learning for Oriented Object Detection in Remote Sensing Images},
 type = {article},
 year = {2024},
 keywords = {contrastive learning,equivariant learning,oriented object detection,remote sensing (RS)},
 pages = {516},
 volume = {16},
 websites = {https://www.mdpi.com/2072-4292/16/3/516/htm,https://www.mdpi.com/2072-4292/16/3/516},
 month = {1},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {29},
 id = {aa13f4de-280d-36f3-a7e4-205a85abce92},
 created = {2024-02-16T10:15:19.071Z},
 accessed = {2024-02-16},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-04T07:45:03.892Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Contrastive learning allows us to learn general features for downstream tasks without the need for labeled data by leveraging intrinsic signals within remote sensing images. Existing contrastive learning methods encourage invariant feature learning by bringing positive samples defined by random transformations in feature spaces closer, where transformed samples of the same image at different intensities are considered equivalent. However, remote sensing images differ from natural images in their top-down perspective results in the arbitrary orientation of objects and in that the images contain rich in-plane rotation information. Maintaining invariance to rotation transformations can lead to the loss of rotation information in features, thereby affecting angle information predictions for differently rotated samples in downstream tasks. Therefore, we believe that contrastive learning should not focus only on strict invariance but encourage features to be equivariant to rotation while maintaining invariance to other transformations. To achieve this goal, we propose an invariant–equivariant covariant network (Co-ECL) based on collaborative and reverse mechanisms. The collaborative mechanism encourages rotation equivariance by predicting the rotation transformations of input images and combines invariant and equivariant learning tasks to jointly supervise the feature learning process to achieve collaborative learning. The reverse mechanism introduces a reverse rotation module in the feature learning stage, applying reverse rotation transformations with equal intensity to features in invariant learning tasks as in the data transformation stage, thereby ensuring their independent realization. In experiments conducted on three publicly available oriented object detection datasets of remote sensing images, our method consistently demonstrated the best performance. Additionally, these experiments on multi-angle datasets demonstrated that our method has good robustness on rotation-related tasks.},
 bibtype = {article},
 author = {Zhang, Yunsheng and Ren, Zijing and Ding, Zichen and Qian, Hong and Li, Haiqiang and Tao, Chao},
 doi = {10.3390/RS16030516},
 journal = {Remote Sensing 2024, Vol. 16, Page 516},
 number = {3}
}

Contrastive learning allows us to learn general features for downstream tasks without the need for labeled data by leveraging intrinsic signals within remote sensing images. Existing contrastive learning methods encourage invariant feature learning by bringing positive samples defined by random transformations in feature spaces closer, where transformed samples of the same image at different intensities are considered equivalent. However, remote sensing images differ from natural images in their top-down perspective results in the arbitrary orientation of objects and in that the images contain rich in-plane rotation information. Maintaining invariance to rotation transformations can lead to the loss of rotation information in features, thereby affecting angle information predictions for differently rotated samples in downstream tasks. Therefore, we believe that contrastive learning should not focus only on strict invariance but encourage features to be equivariant to rotation while maintaining invariance to other transformations. To achieve this goal, we propose an invariant–equivariant covariant network (Co-ECL) based on collaborative and reverse mechanisms. The collaborative mechanism encourages rotation equivariance by predicting the rotation transformations of input images and combines invariant and equivariant learning tasks to jointly supervise the feature learning process to achieve collaborative learning. The reverse mechanism introduces a reverse rotation module in the feature learning stage, applying reverse rotation transformations with equal intensity to features in invariant learning tasks as in the data transformation stage, thereby ensuring their independent realization. In experiments conducted on three publicly available oriented object detection datasets of remote sensing images, our method consistently demonstrated the best performance. Additionally, these experiments on multi-angle datasets demonstrated that our method has good robustness on rotation-related tasks.

@article{
 title = {Co-ECL: Covariant Network with Equivariant Contrastive Learning for Oriented Object Detection in Remote Sensing Images},
 type = {article},
 year = {2024},
 keywords = {contrastive learning,equivariant learning,oriented object detection,remote sensing (RS)},
 pages = {516},
 volume = {16},
 websites = {https://www.mdpi.com/2072-4292/16/3/516/htm,https://www.mdpi.com/2072-4292/16/3/516},
 month = {1},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {29},
 id = {b14b84d0-cff3-3bfc-abbd-afc1b5ee47e0},
 created = {2024-03-04T07:27:33.611Z},
 accessed = {2024-03-04},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-04T07:45:04.134Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0,5a010301-acb6-4642-a6b2-8afaee1b741c},
 private_publication = {false},
 abstract = {Contrastive learning allows us to learn general features for downstream tasks without the need for labeled data by leveraging intrinsic signals within remote sensing images. Existing contrastive learning methods encourage invariant feature learning by bringing positive samples defined by random transformations in feature spaces closer, where transformed samples of the same image at different intensities are considered equivalent. However, remote sensing images differ from natural images in their top-down perspective results in the arbitrary orientation of objects and in that the images contain rich in-plane rotation information. Maintaining invariance to rotation transformations can lead to the loss of rotation information in features, thereby affecting angle information predictions for differently rotated samples in downstream tasks. Therefore, we believe that contrastive learning should not focus only on strict invariance but encourage features to be equivariant to rotation while maintaining invariance to other transformations. To achieve this goal, we propose an invariant–equivariant covariant network (Co-ECL) based on collaborative and reverse mechanisms. The collaborative mechanism encourages rotation equivariance by predicting the rotation transformations of input images and combines invariant and equivariant learning tasks to jointly supervise the feature learning process to achieve collaborative learning. The reverse mechanism introduces a reverse rotation module in the feature learning stage, applying reverse rotation transformations with equal intensity to features in invariant learning tasks as in the data transformation stage, thereby ensuring their independent realization. In experiments conducted on three publicly available oriented object detection datasets of remote sensing images, our method consistently demonstrated the best performance. Additionally, these experiments on multi-angle datasets demonstrated that our method has good robustness on rotation-related tasks.},
 bibtype = {article},
 author = {Zhang, Yunsheng and Ren, Zijing and Ding, Zichen and Qian, Hong and Li, Haiqiang and Tao, Chao},
 doi = {10.3390/RS16030516},
 journal = {Remote Sensing 2024, Vol. 16, Page 516},
 number = {3}
}

Contrastive learning allows us to learn general features for downstream tasks without the need for labeled data by leveraging intrinsic signals within remote sensing images. Existing contrastive learning methods encourage invariant feature learning by bringing positive samples defined by random transformations in feature spaces closer, where transformed samples of the same image at different intensities are considered equivalent. However, remote sensing images differ from natural images in their top-down perspective results in the arbitrary orientation of objects and in that the images contain rich in-plane rotation information. Maintaining invariance to rotation transformations can lead to the loss of rotation information in features, thereby affecting angle information predictions for differently rotated samples in downstream tasks. Therefore, we believe that contrastive learning should not focus only on strict invariance but encourage features to be equivariant to rotation while maintaining invariance to other transformations. To achieve this goal, we propose an invariant–equivariant covariant network (Co-ECL) based on collaborative and reverse mechanisms. The collaborative mechanism encourages rotation equivariance by predicting the rotation transformations of input images and combines invariant and equivariant learning tasks to jointly supervise the feature learning process to achieve collaborative learning. The reverse mechanism introduces a reverse rotation module in the feature learning stage, applying reverse rotation transformations with equal intensity to features in invariant learning tasks as in the data transformation stage, thereby ensuring their independent realization. In experiments conducted on three publicly available oriented object detection datasets of remote sensing images, our method consistently demonstrated the best performance. Additionally, these experiments on multi-angle datasets demonstrated that our method has good robustness on rotation-related tasks.

@article{
 title = {Oriented R-CNN and Beyond},
 type = {article},
 year = {2024},
 keywords = {Instance segmentation,Oriented object detection,Oriented region proposal network},
 pages = {1-23},
 websites = {https://link.springer.com/article/10.1007/s11263-024-01989-w},
 month = {1},
 publisher = {Springer},
 day = {29},
 id = {db176d5a-3286-385c-b68e-429b42ca51d8},
 created = {2024-03-18T14:29:53.949Z},
 accessed = {2024-03-18},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-20T12:30:54.392Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 citation_key = {Xie2024},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Currently, two-stage oriented detectors are superior to single-stage competitors in accuracy, but the step of generating oriented proposals is still time-consuming, thus hindering the inference speed. This paper proposes an Oriented Region Proposal Network (Oriented RPN) to produce high-quality oriented proposals in a nearly cost-free manner. To this end, we present a novel representation manner of oriented objects, named midpoint offset representation, which avoids the complicated design of oriented proposal generation network. Built on Oriented RPN, we develop a simple yet effective oriented object detection framework, called Oriented R-CNN, which could accurately and efficiently detect oriented objects. Moreover, we extend Oriented R-CNN to the task of instance segmentation and realize a new proposal-based instance segmentation method, termed Oriented Mask R-CNN. Without bells and whistles, Oriented R-CNN achieves state-of-the-art accuracy on all seven commonly-used oriented object detection datasets. More importantly, our method has the fastest speed among all detectors. For instance segmentation, Oriented Mask R-CNN also achieves the top results on the large-scale aerial instance segmentation dataset, named iSAID. We hope our methods could serve as solid baselines for oriented object detection and instance segmentation. Code is available at https://github.com/jbwang1997/OBBDetection .},
 bibtype = {article},
 author = {Xie, Xingxing and Cheng, Gong and Wang, Jiabao and Li, Ke and Yao, Xiwen and Han, Junwei},
 doi = {10.1007/S11263-024-01989-W/TABLES/16},
 journal = {International Journal of Computer Vision}
}

Currently, two-stage oriented detectors are superior to single-stage competitors in accuracy, but the step of generating oriented proposals is still time-consuming, thus hindering the inference speed. This paper proposes an Oriented Region Proposal Network (Oriented RPN) to produce high-quality oriented proposals in a nearly cost-free manner. To this end, we present a novel representation manner of oriented objects, named midpoint offset representation, which avoids the complicated design of oriented proposal generation network. Built on Oriented RPN, we develop a simple yet effective oriented object detection framework, called Oriented R-CNN, which could accurately and efficiently detect oriented objects. Moreover, we extend Oriented R-CNN to the task of instance segmentation and realize a new proposal-based instance segmentation method, termed Oriented Mask R-CNN. Without bells and whistles, Oriented R-CNN achieves state-of-the-art accuracy on all seven commonly-used oriented object detection datasets. More importantly, our method has the fastest speed among all detectors. For instance segmentation, Oriented Mask R-CNN also achieves the top results on the large-scale aerial instance segmentation dataset, named iSAID. We hope our methods could serve as solid baselines for oriented object detection and instance segmentation. Code is available at https://github.com/jbwang1997/OBBDetection .

@article{
 title = {YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information},
 type = {article},
 year = {2024},
 websites = {https://arxiv.org/abs/2402.13616v2},
 month = {2},
 day = {21},
 id = {eb63a530-cbe1-3cb0-a0fa-ed76542b0f4f},
 created = {2024-03-22T07:47:43.796Z},
 accessed = {2024-03-22},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-22T07:47:47.597Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Today's deep learning methods focus on how to design the most appropriate objective functions so that the prediction results of the model can be closest to the ground truth. Meanwhile, an appropriate architecture that can facilitate acquisition of enough information for prediction has to be designed. Existing methods ignore a fact that when input data undergoes layer-by-layer feature extraction and spatial transformation, large amount of information will be lost. This paper will delve into the important issues of data loss when data is transmitted through deep networks, namely information bottleneck and reversible functions. We proposed the concept of programmable gradient information (PGI) to cope with the various changes required by deep networks to achieve multiple objectives. PGI can provide complete input information for the target task to calculate objective function, so that reliable gradient information can be obtained to update network weights. In addition, a new lightweight network architecture -- Generalized Efficient Layer Aggregation Network (GELAN), based on gradient path planning is designed. GELAN's architecture confirms that PGI has gained superior results on lightweight models. We verified the proposed GELAN and PGI on MS COCO dataset based object detection. The results show that GELAN only uses conventional convolution operators to achieve better parameter utilization than the state-of-the-art methods developed based on depth-wise convolution. PGI can be used for variety of models from lightweight to large. It can be used to obtain complete information, so that train-from-scratch models can achieve better results than state-of-the-art models pre-trained using large datasets, the comparison results are shown in Figure 1. The source codes are at: https://github.com/WongKinYiu/yolov9.},
 bibtype = {article},
 author = {Wang, Chien-Yao and Yeh, I-Hau and Liao, Hong-Yuan Mark}
}

Today's deep learning methods focus on how to design the most appropriate objective functions so that the prediction results of the model can be closest to the ground truth. Meanwhile, an appropriate architecture that can facilitate acquisition of enough information for prediction has to be designed. Existing methods ignore a fact that when input data undergoes layer-by-layer feature extraction and spatial transformation, large amount of information will be lost. This paper will delve into the important issues of data loss when data is transmitted through deep networks, namely information bottleneck and reversible functions. We proposed the concept of programmable gradient information (PGI) to cope with the various changes required by deep networks to achieve multiple objectives. PGI can provide complete input information for the target task to calculate objective function, so that reliable gradient information can be obtained to update network weights. In addition, a new lightweight network architecture -- Generalized Efficient Layer Aggregation Network (GELAN), based on gradient path planning is designed. GELAN's architecture confirms that PGI has gained superior results on lightweight models. We verified the proposed GELAN and PGI on MS COCO dataset based object detection. The results show that GELAN only uses conventional convolution operators to achieve better parameter utilization than the state-of-the-art methods developed based on depth-wise convolution. PGI can be used for variety of models from lightweight to large. It can be used to obtain complete information, so that train-from-scratch models can achieve better results than state-of-the-art models pre-trained using large datasets, the comparison results are shown in Figure 1. The source codes are at: https://github.com/WongKinYiu/yolov9.

@article{
 title = {YOLO-based Object Detection Models: A Review and its Applications},
 type = {article},
 year = {2024},
 keywords = {Computer Vision,Dataset,Deep Learning,Object detection,YOLO},
 pages = {1-40},
 websites = {https://link.springer.com/article/10.1007/s11042-024-18872-y},
 month = {3},
 publisher = {Springer},
 day = {14},
 id = {cb4d5997-03e0-3d08-907e-0303da540cbb},
 created = {2024-03-27T10:34:15.841Z},
 accessed = {2024-03-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-27T13:51:51.492Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {In computer vision, object detection is the classical and most challenging problem to get accurate results in detecting objects. With the significant advancement of deep learning techniques over the past decades, most researchers work on enhancing object detection, segmentation and classification. Object detection performance is measured in both detection accuracy and inference time. The detection accuracy in two stage detectors is better than single stage detectors. In 2015, the real-time object detection system YOLO was published, and it rapidly grew its iterations, with the newest release, YOLOv8 in January 2023. The YOLO achieves a high detection accuracy and inference time with single stage detector. Many applications easily adopt YOLO versions due to their high inference speed. This paper presents a complete survey of YOLO versions up to YOLOv8. This article begins with explained about the performance metrics used in object detection, post-processing methods, dataset availability and object detection techniques that are used mostly; then discusses the architectural design of each YOLO version. Finally, the diverse range of YOLO versions was discussed by highlighting their contributions to various applications.},
 bibtype = {article},
 author = {Vijayakumar, Ajantha and Vairavasundaram, Subramaniyaswamy},
 doi = {10.1007/S11042-024-18872-Y/TABLES/4},
 journal = {Multimedia Tools and Applications}
}

In computer vision, object detection is the classical and most challenging problem to get accurate results in detecting objects. With the significant advancement of deep learning techniques over the past decades, most researchers work on enhancing object detection, segmentation and classification. Object detection performance is measured in both detection accuracy and inference time. The detection accuracy in two stage detectors is better than single stage detectors. In 2015, the real-time object detection system YOLO was published, and it rapidly grew its iterations, with the newest release, YOLOv8 in January 2023. The YOLO achieves a high detection accuracy and inference time with single stage detector. Many applications easily adopt YOLO versions due to their high inference speed. This paper presents a complete survey of YOLO versions up to YOLOv8. This article begins with explained about the performance metrics used in object detection, post-processing methods, dataset availability and object detection techniques that are used mostly; then discusses the architectural design of each YOLO version. Finally, the diverse range of YOLO versions was discussed by highlighting their contributions to various applications.

@article{
 title = {YOLO-based Object Detection Models: A Review and its Applications},
 type = {article},
 year = {2024},
 keywords = {Computer Vision,Dataset,Deep Learning,Object detection,YOLO},
 pages = {1-40},
 websites = {https://link.springer.com/article/10.1007/s11042-024-18872-y},
 month = {3},
 publisher = {Springer},
 day = {14},
 id = {3badd1a5-bcf7-37b2-8e8e-5c0605fba036},
 created = {2024-03-27T10:36:53.835Z},
 accessed = {2024-03-27},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-04-09T18:54:39.043Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 citation_key = {Vijayakumar2024},
 private_publication = {false},
 abstract = {In computer vision, object detection is the classical and most challenging problem to get accurate results in detecting objects. With the significant advancement of deep learning techniques over the past decades, most researchers work on enhancing object detection, segmentation and classification. Object detection performance is measured in both detection accuracy and inference time. The detection accuracy in two stage detectors is better than single stage detectors. In 2015, the real-time object detection system YOLO was published, and it rapidly grew its iterations, with the newest release, YOLOv8 in January 2023. The YOLO achieves a high detection accuracy and inference time with single stage detector. Many applications easily adopt YOLO versions due to their high inference speed. This paper presents a complete survey of YOLO versions up to YOLOv8. This article begins with explained about the performance metrics used in object detection, post-processing methods, dataset availability and object detection techniques that are used mostly; then discusses the architectural design of each YOLO version. Finally, the diverse range of YOLO versions was discussed by highlighting their contributions to various applications.},
 bibtype = {article},
 author = {Vijayakumar, Ajantha and Vairavasundaram, Subramaniyaswamy},
 doi = {10.1007/S11042-024-18872-Y/TABLES/4},
 journal = {Multimedia Tools and Applications}
}

In computer vision, object detection is the classical and most challenging problem to get accurate results in detecting objects. With the significant advancement of deep learning techniques over the past decades, most researchers work on enhancing object detection, segmentation and classification. Object detection performance is measured in both detection accuracy and inference time. The detection accuracy in two stage detectors is better than single stage detectors. In 2015, the real-time object detection system YOLO was published, and it rapidly grew its iterations, with the newest release, YOLOv8 in January 2023. The YOLO achieves a high detection accuracy and inference time with single stage detector. Many applications easily adopt YOLO versions due to their high inference speed. This paper presents a complete survey of YOLO versions up to YOLOv8. This article begins with explained about the performance metrics used in object detection, post-processing methods, dataset availability and object detection techniques that are used mostly; then discusses the architectural design of each YOLO version. Finally, the diverse range of YOLO versions was discussed by highlighting their contributions to various applications.

@article{
 title = {On Boundary Discontinuity in Angle Regression Based Arbitrary Oriented Object Detection},
 type = {article},
 year = {2024},
 keywords = {Angle regression,Artificial neural networks,Codes,Detectors,Feature extraction,Proposals,Visualization,YOLO,boundary discontinuity,oriented object detection,phase-shifting coder},
 publisher = {IEEE Computer Society},
 id = {71f30f73-2b51-3cb6-ac6c-20f1672b2553},
 created = {2024-06-25T11:18:09.906Z},
 accessed = {2024-06-25},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2025-01-27T13:24:38.846Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {With vigorous development e.g. in autonomous driving and remote sensing, oriented object detection has gradually been featured. The majority of existing methods directly perform regression on the rotation angle, which we argue has fundamental limitations of boundary discontinuity (even if using Gaussian or RotatedIoU-based losses). In this paper, a novel angle coder named phase-shifting coder (PSC) is proposed to address this issue. Different from another well-explored alternative i.e. angle classification, PSC achieves boundary-discontinuity-free in a continuous and differentiable manner and thus can work together with Gaussian or RotatedIoU-based methods to further boost their performance. Moreover, by rethinking the boundary discontinuity of elongated and square-like objects as rotational symmetry of different cycles, a dual-frequency version (PSCD) is proposed to accurately predict the orientation of both types of objects. Visual analysis and extensive experiments on several popular backbone detectors and datasets demonstrate the effectiveness and the potentiality of our approach. When facing scenarios requiring high-quality bounding boxes, the proposed methods are expected to give a competitive performance.},
 bibtype = {article},
 author = {Yu, Yi and Da, Feipeng},
 doi = {10.1109/TPAMI.2024.3378777},
 journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence}
}

With vigorous development e.g. in autonomous driving and remote sensing, oriented object detection has gradually been featured. The majority of existing methods directly perform regression on the rotation angle, which we argue has fundamental limitations of boundary discontinuity (even if using Gaussian or RotatedIoU-based losses). In this paper, a novel angle coder named phase-shifting coder (PSC) is proposed to address this issue. Different from another well-explored alternative i.e. angle classification, PSC achieves boundary-discontinuity-free in a continuous and differentiable manner and thus can work together with Gaussian or RotatedIoU-based methods to further boost their performance. Moreover, by rethinking the boundary discontinuity of elongated and square-like objects as rotational symmetry of different cycles, a dual-frequency version (PSCD) is proposed to accurately predict the orientation of both types of objects. Visual analysis and extensive experiments on several popular backbone detectors and datasets demonstrate the effectiveness and the potentiality of our approach. When facing scenarios requiring high-quality bounding boxes, the proposed methods are expected to give a competitive performance.

@article{
 title = {Explicit and Implicit Box Equivariance Learning for Weakly-Supervised Rotated Object Detection},
 type = {article},
 year = {2024},
 keywords = {Annotations,Box equivariance,Detectors,Feature extraction,Neck,Object detection,Shape,Task analysis,rotated object detection,weakly-supervised learning},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {d71f3d03-9c42-37ca-957f-485bc26ddca8},
 created = {2024-06-26T06:46:30.000Z},
 accessed = {2024-06-26},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-07-03T09:41:20.558Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Rotated object detection has emerged as a preferred alternative to the conventional Horizontal Box (HBox) object detection in various tasks. However, the majority of existing detection datasets are still annotated using HBox. Re-annotating these datasets for rotated object detection is notably inefficient and labor-intensive. Recently, weakly-supervised detectors based directly on HBox have provided a solution to bridge the gap between the rotated bounding box (RBox) prediction and the weak annotations. Nonetheless, they tend to overlook the potential impact of factors such as bounding box position, aspect ratio, and scale on angle prediction. Additionally, inconsistencies arise when evaluating rotation variations, hindering the inference performance. In light of these limitations, we propose an explicit and implicit equivariance detector (EIE-Det) to predict fine-grained RBox based on HBox annotations. Along with a shared backbone and neck structure, our proposed EIE-Det comprises a weakly supervised (WS) branch for predicting coarse RBox from HBox labels, an explicit equivariance (EE) branch for learning rotation consistency, and an implicit equivariance (IE) branch for learning position, aspect ratio, and scale consistency. Furthermore, we introduce the Tanimoto Coefficient for vector-based bounding box regression, enabling a comprehensive assessment of model prediction bias, thereby eliminating interference from variations in box aspect ratio and scale. Experiments demonstrate that our method not only significantly surpasses state-of-the-art models supervised with HBox but also achieves accuracy comparable to fully-supervised detectors (75.74<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and 58.50<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> <inline-formula><tex-math notation="LaTeX">$AP_50$</tex-math></inline-formula> on DOTA and DIOR-R, respectively).},
 bibtype = {article},
 author = {Wang, Linfei and Zhan, Yibing and Lin, Xu and Yu, Baosheng and Ding, Liang and Zhu, Jianqing and Tao, Dapeng},
 doi = {10.1109/TETCI.2024.3398020},
 journal = {IEEE Transactions on Emerging Topics in Computational Intelligence}
}

Rotated object detection has emerged as a preferred alternative to the conventional Horizontal Box (HBox) object detection in various tasks. However, the majority of existing detection datasets are still annotated using HBox. Re-annotating these datasets for rotated object detection is notably inefficient and labor-intensive. Recently, weakly-supervised detectors based directly on HBox have provided a solution to bridge the gap between the rotated bounding box (RBox) prediction and the weak annotations. Nonetheless, they tend to overlook the potential impact of factors such as bounding box position, aspect ratio, and scale on angle prediction. Additionally, inconsistencies arise when evaluating rotation variations, hindering the inference performance. In light of these limitations, we propose an explicit and implicit equivariance detector (EIE-Det) to predict fine-grained RBox based on HBox annotations. Along with a shared backbone and neck structure, our proposed EIE-Det comprises a weakly supervised (WS) branch for predicting coarse RBox from HBox labels, an explicit equivariance (EE) branch for learning rotation consistency, and an implicit equivariance (IE) branch for learning position, aspect ratio, and scale consistency. Furthermore, we introduce the Tanimoto Coefficient for vector-based bounding box regression, enabling a comprehensive assessment of model prediction bias, thereby eliminating interference from variations in box aspect ratio and scale. Experiments demonstrate that our method not only significantly surpasses state-of-the-art models supervised with HBox but also achieves accuracy comparable to fully-supervised detectors (75.74$\%$ and 58.50$\%$ $AP_50$ on DOTA and DIOR-R, respectively).

@misc{
 title = {Point2RBox: Combine Knowledge from Synthetic Visual Patterns for End-to-end Oriented Object Detection with Single Point Supervision},
 type = {misc},
 year = {2024},
 pages = {16783-16793},
 websites = {https://github.com/yuyi1005/point2rbox-mmrotate},
 id = {38fcb447-e7c6-371a-b21e-f4795b28b0fc},
 created = {2024-06-26T06:49:14.474Z},
 accessed = {2024-06-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-06-26T06:49:20.444Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {With the rapidly increasing demand for oriented object detection (OOD), recent research involving weakly-supervised detectors for learning rotated box (RBox) from the horizontal box (HBox) has attracted more and more attention. In this paper , we explore a more challenging yet label-efficient setting, namely single point-supervised OOD, and present our approach called Point2RBox. Specifically, we propose to leverage two principles: 1) Synthetic pattern knowledge combination: By sampling around each labeled point on the image, we spread the object feature to synthetic visual patterns with known boxes to provide the knowledge for box regression. 2) Transform self-supervision: With a transformed input image (e.g. scaled/rotated), the output RBoxes are trained to follow the same transformation so that the network can perceive the relative size/rotation between objects. The detector is further enhanced by a few devised techniques to cope with peripheral issues, e.g. the anchor/layer assignment as the size of the object is not available in our point supervision setting. To our best knowledge, Point2RBox is the first end-to-end solution for point-supervised OOD. In particular, our method uses a lightweight paradigm, yet it achieves a competitive performance among point-supervised alternatives, 41.05%/27.62%/80.01% on DOTA/DIOR/HRSC datasets.},
 bibtype = {misc},
 author = {Yu, Yi and Yang, Xue and Li, Qingyun and Da, Feipeng and Dai, Jifeng and Qiao, Yu and Yan, Junchi}
}

With the rapidly increasing demand for oriented object detection (OOD), recent research involving weakly-supervised detectors for learning rotated box (RBox) from the horizontal box (HBox) has attracted more and more attention. In this paper , we explore a more challenging yet label-efficient setting, namely single point-supervised OOD, and present our approach called Point2RBox. Specifically, we propose to leverage two principles: 1) Synthetic pattern knowledge combination: By sampling around each labeled point on the image, we spread the object feature to synthetic visual patterns with known boxes to provide the knowledge for box regression. 2) Transform self-supervision: With a transformed input image (e.g. scaled/rotated), the output RBoxes are trained to follow the same transformation so that the network can perceive the relative size/rotation between objects. The detector is further enhanced by a few devised techniques to cope with peripheral issues, e.g. the anchor/layer assignment as the size of the object is not available in our point supervision setting. To our best knowledge, Point2RBox is the first end-to-end solution for point-supervised OOD. In particular, our method uses a lightweight paradigm, yet it achieves a competitive performance among point-supervised alternatives, 41.05%/27.62%/80.01% on DOTA/DIOR/HRSC datasets.

@misc{
 title = {PointOBB: Learning Oriented Object Detection via Single Point Supervision},
 type = {misc},
 year = {2024},
 pages = {16730-16740},
 websites = {https://github.com/Luo-Z13/pointobb},
 id = {49af9f1b-2a0f-36e8-aa3e-e75696126f49},
 created = {2024-06-26T06:51:37.241Z},
 accessed = {2024-06-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-06-26T06:51:46.434Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Single point-supervised object detection is gaining attention due to its cost-effectiveness. However, existing approaches focus on generating horizontal bounding boxes (HBBs) while ignoring oriented bounding boxes (OBBs) commonly used for objects in aerial images. This paper proposes PointOBB, the first single Point-based OBB generation method, for oriented object detection. PointOBB operates through the collaborative utilization of three distinctive views: an original view, a resized view, and a ro-tated/flipped (rot/flp) view. Upon the original view, we leverage the resized and rot/flp views to build a scale augmentation module and an angle acquisition module, respectively. In the former module, a Scale-Sensitive Consistency (SSC) loss is designed to enhance the deep network's ability to perceive the object scale. For accurate object angle predictions, the latter module incorporates self-supervised learning to predict angles, which is associated with a scale-guided Dense-to-Sparse (DS) matching strategy for aggre-gating dense angles corresponding to sparse objects. The resized and rot/flp views are switched using a progressive multi-view switching strategy during training to achieve coupled optimization of scale and angle. Experimental results on the DIOR-R and DOTA-v1.0 datasets demonstrate that PointOBB achieves promising performance, and significantly outperforms potential point-supervised baselines.},
 bibtype = {misc},
 author = {Luo, Junwei and Yang, Xue and Yu, Yi and Li, Qingyun and Yan, Junchi and Li, Yansheng}
}

Single point-supervised object detection is gaining attention due to its cost-effectiveness. However, existing approaches focus on generating horizontal bounding boxes (HBBs) while ignoring oriented bounding boxes (OBBs) commonly used for objects in aerial images. This paper proposes PointOBB, the first single Point-based OBB generation method, for oriented object detection. PointOBB operates through the collaborative utilization of three distinctive views: an original view, a resized view, and a ro-tated/flipped (rot/flp) view. Upon the original view, we leverage the resized and rot/flp views to build a scale augmentation module and an angle acquisition module, respectively. In the former module, a Scale-Sensitive Consistency (SSC) loss is designed to enhance the deep network's ability to perceive the object scale. For accurate object angle predictions, the latter module incorporates self-supervised learning to predict angles, which is associated with a scale-guided Dense-to-Sparse (DS) matching strategy for aggre-gating dense angles corresponding to sparse objects. The resized and rot/flp views are switched using a progressive multi-view switching strategy during training to achieve coupled optimization of scale and angle. Experimental results on the DIOR-R and DOTA-v1.0 datasets demonstrate that PointOBB achieves promising performance, and significantly outperforms potential point-supervised baselines.

@article{
 title = {HDDet: A More Common Heading Direction Detector for Remote Sensing and Arbitrary Viewing Angle Images},
 type = {article},
 year = {2024},
 keywords = {Data augmentation,deep learning,intersection over union (IoU) loss,object heading detection (OHD),remote sensing},
 pages = {1-14},
 volume = {62},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {7399e72b-a002-3daf-8670-a92a899f936b},
 created = {2024-06-26T07:11:02.101Z},
 accessed = {2024-06-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2025-01-27T13:20:54.487Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Object heading detection (OHD) offers potential for research in the control and traffic analysis sectors. Contemporary methodologies in OHD grapple with a set of distinct limitations: a constrained range of detectable object types, a discernible drop in accuracy for oriented bounding box (OBB) predictions influenced by heading estimations, and the inherent limitations associated with the exclusive perspective of bird's-eye view imagery. This article introduces heading direction detector (HDDet), an advanced method devised for detecting the OBB of objects with heading direction from various viewpoints. It sequentially delineates the circular annotation method (CAM), multidimensional angle encoding (MDAE), and the CosWeight strategy. CAM initially couples OBB data with heading details for accurate target delineation. MDAE follows, optimizing angle encoding to boost the model's training process. The culmination of this approach is CosWeight, which integrates the Rotated-intersection over union (IoU) and heading information into the horizontal box's loss, thereby enhancing the precision of heading predictions. Rigorous testing across diverse datasets, including the SJTU-L dataset where the heading accuracy increased from 64.41% to 94.82% over OHDet, validates HDDet's enhanced capabilities, surpassing existing methodologies in both OBB detection and heading accuracy, and marking a significant advancement over the state-of-the-art OHDet. Additionally, this article presents an engineering vehicle dataset, which is conducive to multiperspective OHD research.},
 bibtype = {article},
 author = {Ding, Siran and Liu, Jingxian and Yang, Fan and Xu, Mai},
 doi = {10.1109/TGRS.2024.3377643},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

Object heading detection (OHD) offers potential for research in the control and traffic analysis sectors. Contemporary methodologies in OHD grapple with a set of distinct limitations: a constrained range of detectable object types, a discernible drop in accuracy for oriented bounding box (OBB) predictions influenced by heading estimations, and the inherent limitations associated with the exclusive perspective of bird's-eye view imagery. This article introduces heading direction detector (HDDet), an advanced method devised for detecting the OBB of objects with heading direction from various viewpoints. It sequentially delineates the circular annotation method (CAM), multidimensional angle encoding (MDAE), and the CosWeight strategy. CAM initially couples OBB data with heading details for accurate target delineation. MDAE follows, optimizing angle encoding to boost the model's training process. The culmination of this approach is CosWeight, which integrates the Rotated-intersection over union (IoU) and heading information into the horizontal box's loss, thereby enhancing the precision of heading predictions. Rigorous testing across diverse datasets, including the SJTU-L dataset where the heading accuracy increased from 64.41% to 94.82% over OHDet, validates HDDet's enhanced capabilities, surpassing existing methodologies in both OBB detection and heading accuracy, and marking a significant advancement over the state-of-the-art OHDet. Additionally, this article presents an engineering vehicle dataset, which is conducive to multiperspective OHD research.

@article{
 title = {Explicit and Implicit Box Equivariance Learning for Weakly-Supervised Rotated Object Detection},
 type = {article},
 year = {2024},
 keywords = {Annotations,Box equivariance,Detectors,Feature extraction,Neck,Object detection,Shape,Task analysis,rotated object detection,weakly-supervised learning},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {9ead5ef3-c321-3e7c-91c0-8ede06c3d8a3},
 created = {2024-07-03T09:39:40.891Z},
 accessed = {2024-07-03},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-07-03T09:40:04.337Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Rotated object detection has emerged as a preferred alternative to the conventional Horizontal Box (HBox) object detection in various tasks. However, the majority of existing detection datasets are still annotated using HBox. Re-annotating these datasets for rotated object detection is notably inefficient and labor-intensive. Recently, weakly-supervised detectors based directly on HBox have provided a solution to bridge the gap between the rotated bounding box (RBox) prediction and the weak annotations. Nonetheless, they tend to overlook the potential impact of factors such as bounding box position, aspect ratio, and scale on angle prediction. Additionally, inconsistencies arise when evaluating rotation variations, hindering the inference performance. In light of these limitations, we propose an explicit and implicit equivariance detector (EIE-Det) to predict fine-grained RBox based on HBox annotations. Along with a shared backbone and neck structure, our proposed EIE-Det comprises a weakly supervised (WS) branch for predicting coarse RBox from HBox labels, an explicit equivariance (EE) branch for learning rotation consistency, and an implicit equivariance (IE) branch for learning position, aspect ratio, and scale consistency. Furthermore, we introduce the Tanimoto Coefficient for vector-based bounding box regression, enabling a comprehensive assessment of model prediction bias, thereby eliminating interference from variations in box aspect ratio and scale. Experiments demonstrate that our method not only significantly surpasses state-of-the-art models supervised with HBox but also achieves accuracy comparable to fully-supervised detectors (75.74<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and 58.50<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> <inline-formula><tex-math notation="LaTeX">$AP_50$</tex-math></inline-formula> on DOTA and DIOR-R, respectively).},
 bibtype = {article},
 author = {Wang, Linfei and Zhan, Yibing and Lin, Xu and Yu, Baosheng and Ding, Liang and Zhu, Jianqing and Tao, Dapeng},
 doi = {10.1109/TETCI.2024.3398020},
 journal = {IEEE Transactions on Emerging Topics in Computational Intelligence}
}

Rotated object detection has emerged as a preferred alternative to the conventional Horizontal Box (HBox) object detection in various tasks. However, the majority of existing detection datasets are still annotated using HBox. Re-annotating these datasets for rotated object detection is notably inefficient and labor-intensive. Recently, weakly-supervised detectors based directly on HBox have provided a solution to bridge the gap between the rotated bounding box (RBox) prediction and the weak annotations. Nonetheless, they tend to overlook the potential impact of factors such as bounding box position, aspect ratio, and scale on angle prediction. Additionally, inconsistencies arise when evaluating rotation variations, hindering the inference performance. In light of these limitations, we propose an explicit and implicit equivariance detector (EIE-Det) to predict fine-grained RBox based on HBox annotations. Along with a shared backbone and neck structure, our proposed EIE-Det comprises a weakly supervised (WS) branch for predicting coarse RBox from HBox labels, an explicit equivariance (EE) branch for learning rotation consistency, and an implicit equivariance (IE) branch for learning position, aspect ratio, and scale consistency. Furthermore, we introduce the Tanimoto Coefficient for vector-based bounding box regression, enabling a comprehensive assessment of model prediction bias, thereby eliminating interference from variations in box aspect ratio and scale. Experiments demonstrate that our method not only significantly surpasses state-of-the-art models supervised with HBox but also achieves accuracy comparable to fully-supervised detectors (75.74$\%$ and 58.50$\%$ $AP_50$ on DOTA and DIOR-R, respectively).

@article{
 title = {Probabilistic Intersection-Over-Union for Training and Evaluation of Oriented Object Detectors},
 type = {article},
 year = {2024},
 keywords = {Computer vision,object detection,performance evaluation},
 pages = {671-681},
 volume = {33},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {1cb88f70-7a67-346e-8bcf-6acbea806dcc},
 created = {2024-07-03T10:40:10.328Z},
 accessed = {2024-07-03},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2025-01-03T15:57:30.766Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Oriented object detection is a challenging and relatively new problem. Most existing approaches are based on deep learning and explore Oriented Bounding Boxes (OBBs) to represent the objects. They are typically based on adaptations of traditional detectors that work with Horizontal Bounding Boxes (HBBs), which have been exploring IoU-like loss functions to regress the HBBs. However, extending this idea for OBBs is challenging due to complex formulations or requirement for customized backpropagation implementations. Furthermore, using OBBs presents limitations for irregular or roughly circular objects, since the definition of the ideal OBB is an ambiguous and ill-posed problem. In this work, we jointly tackle the problem of training, representing, and evaluating oriented detectors. We explore Gaussian distributions-called Gaussian Bounding Boxes (GBBs)-as fuzzy representations for oriented objects and propose using a similarity metric between two GBBs based on the Hellinger distance. We show that this metric leads to a differentiable closed-form expression that can be directly used as a localization loss term to train OBB object detectors. We also show that GBBs present a natural representation as elliptical regions (called EBBs), which inherently mitigate ambiguity representation for circular objects. Finally, we empirically show that the proposed similarity metric computed between two GBBs strongly correlates with the IoU between the corresponding EBBs, motivating the name Probabilistic Intersection-over-Union (ProbIoU). Our experiments show that results using ProbIoU as a regression loss are competitive with state-of-the-art alternatives without requiring additional hyperparameters or customized implementations, and that ProbIoU is a promising alternative to evaluate oriented object detectors. Our code is available at https://github.com/ProbIOU/.},
 bibtype = {article},
 author = {Murrugarra-Llerena, Jeffri and Kirsten, Lucas N. and Zeni, Luis Felipe and Jung, Claudio R.},
 doi = {10.1109/TIP.2023.3348697},
 journal = {IEEE Transactions on Image Processing}
}

Oriented object detection is a challenging and relatively new problem. Most existing approaches are based on deep learning and explore Oriented Bounding Boxes (OBBs) to represent the objects. They are typically based on adaptations of traditional detectors that work with Horizontal Bounding Boxes (HBBs), which have been exploring IoU-like loss functions to regress the HBBs. However, extending this idea for OBBs is challenging due to complex formulations or requirement for customized backpropagation implementations. Furthermore, using OBBs presents limitations for irregular or roughly circular objects, since the definition of the ideal OBB is an ambiguous and ill-posed problem. In this work, we jointly tackle the problem of training, representing, and evaluating oriented detectors. We explore Gaussian distributions-called Gaussian Bounding Boxes (GBBs)-as fuzzy representations for oriented objects and propose using a similarity metric between two GBBs based on the Hellinger distance. We show that this metric leads to a differentiable closed-form expression that can be directly used as a localization loss term to train OBB object detectors. We also show that GBBs present a natural representation as elliptical regions (called EBBs), which inherently mitigate ambiguity representation for circular objects. Finally, we empirically show that the proposed similarity metric computed between two GBBs strongly correlates with the IoU between the corresponding EBBs, motivating the name Probabilistic Intersection-over-Union (ProbIoU). Our experiments show that results using ProbIoU as a regression loss are competitive with state-of-the-art alternatives without requiring additional hyperparameters or customized implementations, and that ProbIoU is a promising alternative to evaluate oriented object detectors. Our code is available at https://github.com/ProbIOU/.

@article{
 title = {A Stochastic-Geometrical Framework for Object Pose Estimation Based on Mixture Models Avoiding the Correspondence Problem},
 type = {article},
 year = {2024},
 keywords = {Camera system,Computer vision,Lateration system,Object pose estimation,Rigid body,Stochastic modeling},
 pages = {822-838},
 volume = {66},
 websites = {https://link.springer.com/article/10.1007/s10851-024-01200-2},
 month = {10},
 publisher = {Springer},
 day = {1},
 id = {ef4e8ebc-8619-34f0-9464-17b7e961ec4c},
 created = {2024-09-10T06:11:22.533Z},
 accessed = {2024-09-10},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-10-11T15:17:19.992Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Pose estimation of rigid objects is a practical challenge in optical metrology and computer vision. This paper presents a novel stochastic-geometrical modeling framework for object pose estimation based on observing multiple feature points. This framework utilizes mixture models for feature point densities in object space and for interpreting real measurements. Advantages are the avoidance to resolve individual feature correspondences and to incorporate correct stochastic dependencies in multi-view applications. First, the general modeling framework is presented, second, a general algorithm for pose estimation is derived, and third, two example models (camera and lateration setup) are presented. Numerical experiments show the effectiveness of this modeling and general algorithm by presenting four simulation scenarios for three observation systems, including the dependence on measurement resolution, object deformations and measurement noise. Probabilistic modeling utilizing mixture models shows the potential for accurate and robust pose estimations while avoiding the correspondence problem.},
 bibtype = {article},
 author = {Hoegele, Wolfgang},
 doi = {10.1007/S10851-024-01200-2/TABLES/2},
 journal = {Journal of Mathematical Imaging and Vision},
 number = {5}
}

Pose estimation of rigid objects is a practical challenge in optical metrology and computer vision. This paper presents a novel stochastic-geometrical modeling framework for object pose estimation based on observing multiple feature points. This framework utilizes mixture models for feature point densities in object space and for interpreting real measurements. Advantages are the avoidance to resolve individual feature correspondences and to incorporate correct stochastic dependencies in multi-view applications. First, the general modeling framework is presented, second, a general algorithm for pose estimation is derived, and third, two example models (camera and lateration setup) are presented. Numerical experiments show the effectiveness of this modeling and general algorithm by presenting four simulation scenarios for three observation systems, including the dependence on measurement resolution, object deformations and measurement noise. Probabilistic modeling utilizing mixture models shows the potential for accurate and robust pose estimations while avoiding the correspondence problem.

@article{
 title = {Advancements in deep learning for accurate classification of grape leaves and diagnosis of grape diseases},
 type = {article},
 year = {2024},
 keywords = {Convolutional neural network,Grape leaf disease,Image classification,Vision transformer},
 pages = {1061-1080},
 volume = {131},
 websites = {https://link.springer.com/article/10.1007/s41348-024-00896-z},
 month = {6},
 publisher = {Springer Science and Business Media Deutschland GmbH},
 day = {1},
 id = {fa0227c2-0c03-37d5-bfbd-93ebded3ee0a},
 created = {2024-09-10T09:29:09.876Z},
 accessed = {2024-09-10},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-05T13:55:20.258Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e},
 private_publication = {false},
 abstract = {Plant diseases cause significant agricultural losses, demanding accurate detection methods. Traditional approaches relying on expert knowledge may be biased, but advancements in computing, particularly deep learning, offer non-experts effective tools. This study focuses on fine-tuning cutting-edge pre-trained CNN and vision transformer models to classify grape leaves and diagnose grape leaf diseases through digital images. Our research examined a PlantVillage dataset, which comprises 4062 leaf images distributed across four categories. Additionally, we utilized the Grapevine dataset, consisting of 500 leaf images. This dataset is organized into five distinct groups, with each group containing 100 images corresponding to one of the five grape types. The PlantVillage dataset focuses on four classes related to grape diseases, namely Black Rot, Leaf Blight, Healthy, and Esca leaves. On the other hand, the Grapevine dataset includes five classes for leaf recognition, specifically Ak, Alaidris, Buzgulu, Dimnit, and Nazli. In experiments with 14 CNN and 17 vision transformer models, deep learning demonstrated high accuracy in distinguishing grape diseases and recognizing leaves. Notably, four models achieved 100% accuracy on PlantVillage and Grapevine datasets, with Swinv2-Base standing out. This approach holds promise for enhancing crop productivity through early disease detection and providing insights into grape variety characterization in agriculture.},
 bibtype = {article},
 author = {Kunduracioglu, Ismail and Pacal, Ishak},
 doi = {10.1007/S41348-024-00896-Z/TABLES/7},
 journal = {Journal of Plant Diseases and Protection},
 number = {3}
}

Plant diseases cause significant agricultural losses, demanding accurate detection methods. Traditional approaches relying on expert knowledge may be biased, but advancements in computing, particularly deep learning, offer non-experts effective tools. This study focuses on fine-tuning cutting-edge pre-trained CNN and vision transformer models to classify grape leaves and diagnose grape leaf diseases through digital images. Our research examined a PlantVillage dataset, which comprises 4062 leaf images distributed across four categories. Additionally, we utilized the Grapevine dataset, consisting of 500 leaf images. This dataset is organized into five distinct groups, with each group containing 100 images corresponding to one of the five grape types. The PlantVillage dataset focuses on four classes related to grape diseases, namely Black Rot, Leaf Blight, Healthy, and Esca leaves. On the other hand, the Grapevine dataset includes five classes for leaf recognition, specifically Ak, Alaidris, Buzgulu, Dimnit, and Nazli. In experiments with 14 CNN and 17 vision transformer models, deep learning demonstrated high accuracy in distinguishing grape diseases and recognizing leaves. Notably, four models achieved 100% accuracy on PlantVillage and Grapevine datasets, with Swinv2-Base standing out. This approach holds promise for enhancing crop productivity through early disease detection and providing insights into grape variety characterization in agriculture.

@article{
 title = {Artificial Intelligence Techniques in Grapevine Research: A Comparative Study with Extensive Review on Datasets, Diseases, and Techniques Evaluation},
 type = {article},
 year = {2024},
 keywords = {artificial intelligence,diseases,grapevine,machine learning,smart agriculture,smart sensors,vineyards},
 websites = {https://www.preprints.org/manuscript/202407.1623/v1},
 month = {7},
 publisher = {Preprints},
 day = {19},
 id = {3367b9b3-367e-3248-943a-44b63cea986d},
 created = {2024-09-10T09:29:24.997Z},
 accessed = {2024-09-10},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-10T09:29:25.154Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e},
 private_publication = {false},
 bibtype = {article},
 author = {Gatou, Paraskevi and Tsiara, Xanthi and Spitalas, Alexandros and Sioutas, Spyros and Vonitsanos, Gerasimos},
 doi = {10.20944/PREPRINTS202407.1623.V1}
}

@article{
 title = {Vine Diseases Detection Trials in the Carpathian Region with Proximity Aerial Images},
 type = {article},
 year = {2024},
 pages = {29-34},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {9d078c89-596f-31e7-b761-7a24c4dfe461},
 created = {2024-09-16T07:03:17.317Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:03:31.850Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {With the recent advances in the Agro 4.0 domain, the advanced inspection tools from the robotics domain in agriculture got into the focus. The automatic field inspection with autonomous robots started to get accepted by the farmers. However, one of the main challenges of the visual inspection is the geographical variety of the plants. In this work, we focus on the vineyard-specific inspection with low-Altitude flying drones and we analyze the geographical correlation among the typical grape species in the Carpathian Basin using foundation machine learning models.},
 bibtype = {article},
 author = {Tamas, Levente and Gubo, Stefan and Lukic, Tibor},
 doi = {10.1109/SAMI60510.2024.10432919},
 journal = {2024 IEEE 22nd World Symposium on Applied Machine Intelligence and Informatics, SAMI 2024 - Proceedings}
}

With the recent advances in the Agro 4.0 domain, the advanced inspection tools from the robotics domain in agriculture got into the focus. The automatic field inspection with autonomous robots started to get accepted by the farmers. However, one of the main challenges of the visual inspection is the geographical variety of the plants. In this work, we focus on the vineyard-specific inspection with low-Altitude flying drones and we analyze the geographical correlation among the typical grape species in the Carpathian Basin using foundation machine learning models.

@article{
 title = {Evaluating the potential of high-resolution hyperspectral UAV imagery for grapevine viral disease detection in Australian vineyards},
 type = {article},
 year = {2024},
 keywords = {GLRaVs,GVA,Grapevine Leafroll Disease,Low-altitude remote sensing,PLS-DA,Shiraz Disease,Virus detection},
 pages = {103876},
 volume = {130},
 month = {6},
 publisher = {Elsevier},
 day = {1},
 id = {b7eb2dc8-03b4-3acd-b08d-d68a60955634},
 created = {2024-09-16T07:04:07.908Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:04:14.349Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Grapevine (Vitis spp.) viral diseases cause substantial productivity and economic losses to the viticulture industry. Existing disease detection methods are both costly and labour-intensive, prompting a need within the industry for rapid and cost-effective detection methods. The present study evaluates the feasibility of unmanned aerial vehicle (UAV)-based hyperspectral sensing in the visible and near-infrared (VNIR) spectral bands to detect two economically significant viral diseases – Grapevine Leafroll Disease (GLD) and Shiraz Disease (SD) – in four popular winegrape cultivars in Australian vineyards. The partial least squares discriminant analysis (PLS-DA) and Receiver Operating Characteristics Curve (ROC) were used to discriminate diseased and healthy pixels and predict the disease for individual vines. The model predictions for red- and white-berried grapevine cultivars achieved an accuracy of 98% and 75%, respectively. For each viral disease, unique spectral regions and optimal detection times during the growing season were identified. Our work demonstrates the value of high-resolution hyperspectral remote sensing for the detection of viral disease symptoms in vineyards.},
 bibtype = {article},
 author = {Mickey Wang, Yeniu and Ostendorf, Bertram and Pagay, Vinay},
 doi = {10.1016/J.JAG.2024.103876},
 journal = {International Journal of Applied Earth Observation and Geoinformation}
}

Grapevine (Vitis spp.) viral diseases cause substantial productivity and economic losses to the viticulture industry. Existing disease detection methods are both costly and labour-intensive, prompting a need within the industry for rapid and cost-effective detection methods. The present study evaluates the feasibility of unmanned aerial vehicle (UAV)-based hyperspectral sensing in the visible and near-infrared (VNIR) spectral bands to detect two economically significant viral diseases – Grapevine Leafroll Disease (GLD) and Shiraz Disease (SD) – in four popular winegrape cultivars in Australian vineyards. The partial least squares discriminant analysis (PLS-DA) and Receiver Operating Characteristics Curve (ROC) were used to discriminate diseased and healthy pixels and predict the disease for individual vines. The model predictions for red- and white-berried grapevine cultivars achieved an accuracy of 98% and 75%, respectively. For each viral disease, unique spectral regions and optimal detection times during the growing season were identified. Our work demonstrates the value of high-resolution hyperspectral remote sensing for the detection of viral disease symptoms in vineyards.

@article{
 title = {AI-powered Solution for Plant Disease Detection in Viticulture},
 type = {article},
 year = {2024},
 pages = {468-475},
 volume = {238},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {27edbe9a-3a93-3afe-aaa2-95ba8deb9595},
 created = {2024-09-16T07:05:26.873Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:05:29.026Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 bibtype = {article},
 author = {Madeira, Miguel and Porfírio, Rui Pedro and Santos, Pedro Albuquerque and Madeira, Rui Neves},
 doi = {10.1016/J.PROCS.2024.06.049},
 journal = {Procedia Computer Science}
}

@article{
 title = {A novel framework for semi-automated system for grape leaf disease detection},
 type = {article},
 year = {2024},
 keywords = {Ensemble classification,Grape leaf disease detection,Law’s texture features},
 pages = {50733-50755},
 volume = {83},
 websites = {https://link.springer.com/article/10.1007/s11042-023-17629-3},
 month = {5},
 publisher = {Springer},
 day = {1},
 id = {d033ac4e-2b7f-3e25-afdc-a87d0797a38c},
 created = {2024-09-16T07:05:53.002Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:05:55.952Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Plants play a significant role in our lives. They are the primary source of food. But when these are diagnosed with diseases, productivity reduces. Grape is an important fruit, which is rich in potassium and hence is advantageous in balancing fluids in our body. Manually recognizing disease symptoms in plants is difficult due to several factors, involving time consumption and cost. A semi-automated system is proposed for grape lead disease detection. In existing systems, less accuracy has been observed and optimization of the segmentation process, hybridized feature extraction and ensemble classification has been ignored for detecting the leaf diseases. So, we have stressed on hybridization of different feature extraction algorithms and machine learning classifiers for the accurate prediction of diseases. Hence, a very novel approach is proposed that focuses on all the important stages in computerized leaf disease detection - segmentation, feature extraction and classification. To segment the diseased regions in mages, we have utilized K means clustering which is optimized using Grey Wolf Optimization (GWO). For the extraction of features, a hybrid of different algorithms has been utilized. Here, we have utilized Law’s mask, which involves a convolution process for extracting features and is an acknowledged feature extraction algorithm in machine learning. Grey Level Co-occurrence Matrix, Local Binary Pattern and Gabor features have been conjunct to create a hybrid approach for robust results. Later, a newly introduced powerful ensemble classifier is applied to classify the correct diseases. A total of four categories of grape diseases have been used for the research from the “PlantVillage” dataset – Leaf Blight, Black measles, Black rot and healthy. We have compared our approach to existing approaches for detecting diseases in grape leaves. Our approach has proved to be robust as compared to existing approaches and has revealed an accuracy of 95.69%.},
 bibtype = {article},
 author = {Kaur, Navneet and Devendran, V.},
 doi = {10.1007/S11042-023-17629-3/FIGURES/7},
 journal = {Multimedia Tools and Applications},
 number = {17}
}

Plants play a significant role in our lives. They are the primary source of food. But when these are diagnosed with diseases, productivity reduces. Grape is an important fruit, which is rich in potassium and hence is advantageous in balancing fluids in our body. Manually recognizing disease symptoms in plants is difficult due to several factors, involving time consumption and cost. A semi-automated system is proposed for grape lead disease detection. In existing systems, less accuracy has been observed and optimization of the segmentation process, hybridized feature extraction and ensemble classification has been ignored for detecting the leaf diseases. So, we have stressed on hybridization of different feature extraction algorithms and machine learning classifiers for the accurate prediction of diseases. Hence, a very novel approach is proposed that focuses on all the important stages in computerized leaf disease detection - segmentation, feature extraction and classification. To segment the diseased regions in mages, we have utilized K means clustering which is optimized using Grey Wolf Optimization (GWO). For the extraction of features, a hybrid of different algorithms has been utilized. Here, we have utilized Law’s mask, which involves a convolution process for extracting features and is an acknowledged feature extraction algorithm in machine learning. Grey Level Co-occurrence Matrix, Local Binary Pattern and Gabor features have been conjunct to create a hybrid approach for robust results. Later, a newly introduced powerful ensemble classifier is applied to classify the correct diseases. A total of four categories of grape diseases have been used for the research from the “PlantVillage” dataset – Leaf Blight, Black measles, Black rot and healthy. We have compared our approach to existing approaches for detecting diseases in grape leaves. Our approach has proved to be robust as compared to existing approaches and has revealed an accuracy of 95.69%.

@article{
 title = {GLDCNet: A novel convolutional neural network for grapevine leafroll disease recognition using UAV-based imagery},
 type = {article},
 year = {2024},
 keywords = {Deep learning,Grapevine leafroll disease,Multispectral,Spatial resolution,UAV,Vegetation index},
 pages = {108668},
 volume = {218},
 month = {3},
 publisher = {Elsevier},
 day = {1},
 id = {9e131e8a-39cf-3631-bbbf-bc63d1cbc53c},
 created = {2024-09-16T07:07:27.796Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:07:31.977Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {High-throughput phenotyping of grapevine leafroll disease (GLD) at the canopy scale helps develop fast and effective management in viticulture. However, detecting GLD efficiently in a vineyard is challenging owing to the limited adaptation of prior art. Therefore, we propose a novel convolutional neural network called GLDCNet to improve GLD recognition using unmanned aerial vehicle–based imagery. The effectiveness of the GLDCNet is attributed to the four new network designs used and is validated through ablation experiments. The GLDCNet achieves a classification accuracy of 99.57% using the RGB dataset and obtains more efficient and accurate results than nine other state-of-the-art methods. Furthermore, we systematically evaluated the impacts of image spatial resolution and vegetation indexes on the classification performance of the model. Experimental results suggest that improving image spatial resolution is more cost-effective than enhancing multispectral information for improving GLD recognition. Our proposed method offers a rapid, scalable, and accurate diagnostic protocol for detecting GLD in vineyards.},
 bibtype = {article},
 author = {Liu, Yixue and Su, Jinya and Zheng, Zhouzhou and Liu, Dizhu and Song, Yuyang and Fang, Yulin and Yang, Peng and Su, Baofeng},
 doi = {10.1016/J.COMPAG.2024.108668},
 journal = {Computers and Electronics in Agriculture}
}

High-throughput phenotyping of grapevine leafroll disease (GLD) at the canopy scale helps develop fast and effective management in viticulture. However, detecting GLD efficiently in a vineyard is challenging owing to the limited adaptation of prior art. Therefore, we propose a novel convolutional neural network called GLDCNet to improve GLD recognition using unmanned aerial vehicle–based imagery. The effectiveness of the GLDCNet is attributed to the four new network designs used and is validated through ablation experiments. The GLDCNet achieves a classification accuracy of 99.57% using the RGB dataset and obtains more efficient and accurate results than nine other state-of-the-art methods. Furthermore, we systematically evaluated the impacts of image spatial resolution and vegetation indexes on the classification performance of the model. Experimental results suggest that improving image spatial resolution is more cost-effective than enhancing multispectral information for improving GLD recognition. Our proposed method offers a rapid, scalable, and accurate diagnostic protocol for detecting GLD in vineyards.

@article{
 title = {Vineyard Zoning and Vine Detection Using Machine Learning in Unmanned Aerial Vehicle Imagery},
 type = {article},
 year = {2024},
 keywords = {K means,UAV,YOLO,multispectral images,neural networks,precision viticulture,remote sensing},
 pages = {584},
 volume = {16},
 websites = {https://www.mdpi.com/2072-4292/16/3/584/htm,https://www.mdpi.com/2072-4292/16/3/584},
 month = {2},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {3},
 id = {738bfeca-2609-3b6d-8182-247545290e28},
 created = {2024-09-16T07:08:50.062Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:14:47.744Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Precision viticulture systems are essential for enhancing traditional intensive viticulture, achieving high-quality results, and minimizing costs. This study explores the integration of Unmanned Aerial Vehicles (UAVs) and artificial intelligence in precision viticulture, focusing on vine detection and vineyard zoning. Vine detection employs the YOLO (You Only Look Once) deep learning algorithm, achieving a remarkable 90% accuracy by analysing UAV imagery with various spectral ranges from various phenological stages. Vineyard zoning, achieved through the application of the K-means algorithm, incorporates geospatial data such as the Normalized Difference Vegetation Index (NDVI) and the assessment of nitrogen, phosphorus, and potassium content in leaf blades and petioles. This approach enables efficient resource management tailored to each zone’s specific needs. The research aims to develop a decision-support model for precision viticulture. The proposed model demonstrates a high vine detection accuracy and defines management zones with variable weighting factors assigned to each variable while preserving location information, revealing significant differences in variables. The model’s advantages lie in its rapid results and minimal data requirements, offering profound insights into the benefits of UAV application for precise vineyard management. This approach has the potential to expedite decision making, allowing for adaptive strategies based on the unique conditions of each zone.},
 bibtype = {article},
 author = {Gavrilović, Milan and Jovanović, Dušan and Božović, Predrag and Benka, Pavel and Govedarica, Miro},
 doi = {10.3390/RS16030584},
 journal = {Remote Sensing 2024, Vol. 16, Page 584},
 number = {3}
}

Precision viticulture systems are essential for enhancing traditional intensive viticulture, achieving high-quality results, and minimizing costs. This study explores the integration of Unmanned Aerial Vehicles (UAVs) and artificial intelligence in precision viticulture, focusing on vine detection and vineyard zoning. Vine detection employs the YOLO (You Only Look Once) deep learning algorithm, achieving a remarkable 90% accuracy by analysing UAV imagery with various spectral ranges from various phenological stages. Vineyard zoning, achieved through the application of the K-means algorithm, incorporates geospatial data such as the Normalized Difference Vegetation Index (NDVI) and the assessment of nitrogen, phosphorus, and potassium content in leaf blades and petioles. This approach enables efficient resource management tailored to each zone’s specific needs. The research aims to develop a decision-support model for precision viticulture. The proposed model demonstrates a high vine detection accuracy and defines management zones with variable weighting factors assigned to each variable while preserving location information, revealing significant differences in variables. The model’s advantages lie in its rapid results and minimal data requirements, offering profound insights into the benefits of UAV application for precise vineyard management. This approach has the potential to expedite decision making, allowing for adaptive strategies based on the unique conditions of each zone.

@article{
 title = {Rapid Detection of Grapevine Viral Disease with High-Resolution Hyperspectral Remote Sensing Technology},
 type = {article},
 year = {2024},
 pages = {4303-4306},
 websites = {https://ieeexplore.ieee.org/document/10642180/},
 month = {7},
 publisher = {IEEE},
 day = {7},
 id = {12d3665a-b167-34ab-8274-015a33013c0e},
 created = {2024-09-16T07:15:48.068Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:15:53.089Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Pagay, Vinay},
 doi = {10.1109/IGARSS53475.2024.10642180},
 journal = {IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium}
}

@article{
 title = {Disease Detection in Grape Cultivation Using Strategically Placed Cameras and Machine Learning Algorithms with a Focus on Powdery Mildew and Blotches},
 type = {article},
 year = {2024},
 keywords = {Accuracy,Classification algorithms,Convolutional neural networks,Deep learning,Disease detection,Feature extraction,Grape cultivation,Image processing,Machine learning,Pipelines,Plant diseases},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {9624bf6b-36a8-308a-8ffb-edf76e788de9},
 created = {2024-09-16T07:16:13.077Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:16:46.254Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Grape cultivation faces various challenges, such as pests, management, fertilizer quality, and diseases caused by bacteria, fungi, and viruses. Notably, powdery mildew and blotches are significant diseases with different features, necessitating an accurate detection system to minimize crop losses. While traditional methods involve capturing images of diseased leaves, this research proposes a smart approach using deep learning and machine learning algorithms to analyze images taken by strategically placed cameras on farms. The research aims to design a system capable of detecting diseases that can provide information relevant to decisions, alert farmers, and allow authorized actions. Employing artificial intelligence algorithms such as support vector machines (SVM), convolutional neural networks (CNN), decision trees (DT), Naive Bayes (NB), and random forest (RF), the proposed model classifies datasets of powdery mildew, blotches, and healthy leaves when using augmented and histogram-oriented gradient (HOG) preprocessing. Following the classification of affected and healthy leaves, a stacking algorithm is used to select the optimal algorithm that provides the highest level of accuracy. The experimental results and analysis reveal that the CNN classifier outperforms others, achieving an accuracy of 96.1%. When transfer learning and fine tuning are applied to the CNN-based model, the accuracy of the model increases by 1.4% and 3.1%, respectively. SVM classification also provides a suitable level of accuracy of 96.0% for HOG augmented data.},
 bibtype = {article},
 author = {Khan, Kashif Hesham and Aljaedi, Amer and Ishtiaq, Muhammad Shakeel and Imam, Hassan and Bassfar, Zaid and Jamal, Sajjad Shaukat},
 doi = {10.1109/ACCESS.2024.3430190},
 journal = {IEEE Access}
}

Grape cultivation faces various challenges, such as pests, management, fertilizer quality, and diseases caused by bacteria, fungi, and viruses. Notably, powdery mildew and blotches are significant diseases with different features, necessitating an accurate detection system to minimize crop losses. While traditional methods involve capturing images of diseased leaves, this research proposes a smart approach using deep learning and machine learning algorithms to analyze images taken by strategically placed cameras on farms. The research aims to design a system capable of detecting diseases that can provide information relevant to decisions, alert farmers, and allow authorized actions. Employing artificial intelligence algorithms such as support vector machines (SVM), convolutional neural networks (CNN), decision trees (DT), Naive Bayes (NB), and random forest (RF), the proposed model classifies datasets of powdery mildew, blotches, and healthy leaves when using augmented and histogram-oriented gradient (HOG) preprocessing. Following the classification of affected and healthy leaves, a stacking algorithm is used to select the optimal algorithm that provides the highest level of accuracy. The experimental results and analysis reveal that the CNN classifier outperforms others, achieving an accuracy of 96.1%. When transfer learning and fine tuning are applied to the CNN-based model, the accuracy of the model increases by 1.4% and 3.1%, respectively. SVM classification also provides a suitable level of accuracy of 96.0% for HOG augmented data.

@article{
 title = {Vine Disease Detection UAV Multi Spectral Image using Segnet and Mobilenet Method},
 type = {article},
 year = {2024},
 keywords = {deep convolutional neural network,deep learning,mobilenet v2,vine disease detection},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {b2397b21-b39a-3877-8700-5c50a1842d39},
 created = {2024-09-16T07:17:26.795Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:17:29.795Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Deep Learning-based vine disease detection has garnered significant attention from the community, particularly with the utilization of UAV multispectral images for grapevine disease detection. However, identifying vine diseases in various crop and horticultural conditions remains a complex challenge, especially under mobile and edge computing conditions. The vine disease detection process makes use of the PlantVillage dataset, which includes unlabelled data. Data normalization is performed and UAVs are involved for data capture, while the SegNet architecture is utilized for segmentation. This enables the separation of healthy and unhealthy vines for detection. Subsequently, classification is performed using MobileNetV2, with layers split to detect diseases all combined with UAV spectral images and larger image sizes, greater than 32 × 32, resulting in better performance. The proposed method achieves high performance, with an accuracy achieve 99.50%, precision at 99.42%, recall at 99.39%, and mean average precision (MAP) at 99.20%. These metrics are compared to existing methods such as Deep Convolutional Neural Network (DCNN) and Inception V2.},
 bibtype = {article},
 author = {Aruna, M. G. and Silvia, Ensteih and Al-Fatlawy, Ramy Riad and Rao, Hanumanthakari Kalyan and Sowmya, M.},
 doi = {10.1109/ICDCOT61034.2024.10515972},
 journal = {International Conference on Distributed Computing and Optimization Techniques, ICDCOT 2024}
}

Deep Learning-based vine disease detection has garnered significant attention from the community, particularly with the utilization of UAV multispectral images for grapevine disease detection. However, identifying vine diseases in various crop and horticultural conditions remains a complex challenge, especially under mobile and edge computing conditions. The vine disease detection process makes use of the PlantVillage dataset, which includes unlabelled data. Data normalization is performed and UAVs are involved for data capture, while the SegNet architecture is utilized for segmentation. This enables the separation of healthy and unhealthy vines for detection. Subsequently, classification is performed using MobileNetV2, with layers split to detect diseases all combined with UAV spectral images and larger image sizes, greater than 32 × 32, resulting in better performance. The proposed method achieves high performance, with an accuracy achieve 99.50%, precision at 99.42%, recall at 99.39%, and mean average precision (MAP) at 99.20%. These metrics are compared to existing methods such as Deep Convolutional Neural Network (DCNN) and Inception V2.

@article{
 title = {GLAD: Advanced Attention Mechanism-Based Model for Grape Leaf Disease Detection},
 type = {article},
 year = {2024},
 keywords = {Yolov5-You only look once version5,agricultural disease detection,convolutional block attention module,convolutional neural network,efficient channel attention,multi-channel attention,shuffle attention},
 websites = {http://creativecommons.org/licenses/by/4.0/},
 id = {d4882bd1-6401-3314-abf3-132113a186e5},
 created = {2024-09-16T07:18:42.636Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:20:48.158Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Diseases affecting grape leaves can have a wide variety of symptoms and a complicated history in vineyards, making detection and diagnosis a formidable issue. The complexity of these problems is frequently too much for existing detection algorithms to handle. Hence, a new method called GLAD (Grape Leaf Disease Detection) was developed. GLAD makes use of the PLANT-VILLAGE dataset, which has been hand-picked to detect grape diseases. We added the self-attention mechanism to make it more effective, and it now can collect data on grape leaf illnesses all over the world better. Adaptively spatial feature fusion (ASFF) technology and BiFPN feature fusion network provide more robust models and improve grape leaf disease fusion by reducing complex background interference. The Shuffle Attention approach is also used to make identifying diseases in grape leaves easier. The dataset is enriched using data augmentation methodologies and transfer learning to identify diseases affecting grape leaves. As part of this process, the model's parameters are adjusted using data from other plant disease datasets. Despite several obstacles, the experimental findings show that the suggested model is intelligent enough to identify grape leaf disorders. Its real-time target detection capabilities are on full display when it outperforms state-of-the-art methods. A powerful and effective tool for the agricultural sector, GLAD is a major step forward in solving the problems associated with grape leaf disease identification.},
 bibtype = {article},
 author = {Thatha, Venkata Nagaraju and Mary, Polukonda and Kumari, Kamala and Sirisha, Uddagiri and Venkata, Valisetty and Manoj, Ram and Praveen, Surapaneni Phani},
 doi = {10.18280/isi.290230}
}

Diseases affecting grape leaves can have a wide variety of symptoms and a complicated history in vineyards, making detection and diagnosis a formidable issue. The complexity of these problems is frequently too much for existing detection algorithms to handle. Hence, a new method called GLAD (Grape Leaf Disease Detection) was developed. GLAD makes use of the PLANT-VILLAGE dataset, which has been hand-picked to detect grape diseases. We added the self-attention mechanism to make it more effective, and it now can collect data on grape leaf illnesses all over the world better. Adaptively spatial feature fusion (ASFF) technology and BiFPN feature fusion network provide more robust models and improve grape leaf disease fusion by reducing complex background interference. The Shuffle Attention approach is also used to make identifying diseases in grape leaves easier. The dataset is enriched using data augmentation methodologies and transfer learning to identify diseases affecting grape leaves. As part of this process, the model's parameters are adjusted using data from other plant disease datasets. Despite several obstacles, the experimental findings show that the suggested model is intelligent enough to identify grape leaf disorders. Its real-time target detection capabilities are on full display when it outperforms state-of-the-art methods. A powerful and effective tool for the agricultural sector, GLAD is a major step forward in solving the problems associated with grape leaf disease identification.

@article{
 title = {Comparison of different computer vision methods for vineyard canopy detection using UAV multispectral images},
 type = {article},
 year = {2024},
 keywords = {Deep learning,NDVI,OBIA,Unmanned aerial vehicles,Vitis vinifera L.},
 volume = {225},
 month = {10},
 publisher = {Elsevier B.V.},
 day = {1},
 id = {6c8a2906-835e-37e2-949a-75a491a4154b},
 created = {2024-09-16T07:21:58.705Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:21:59.344Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {In viticulture, the rapid and accurate acquisition of canopy spectral information through ultra-high spatial resolution imagery is increasingly demanded for decision support. The prevalent practice involves creating vigor maps using spectral data obtained from pure vine canopy pixels. Object-Based Image Analysis (OBIA) among conventional methods exhibits a reasonable efficiency in canopy classification due to its feature extraction capabilities. In recent years, deep learning (DL) techniques have demonstrated significant potential in orchard monitoring, leveraging their ability to automatically learn image features. This study assessed the performance of different methodologies, including Mask R-CNN, U-Net, OBIA and unsupervised methods, in identifying pure canopy pixels. The effectiveness of shadow and background detection methods and the impact of misclassified pixels on NDVI were compared. Results were compared with agronomic surveys conducted during the 2021 and 2022 growing seasons, focusing on two distinct phenological stages (BBCH65-BBCH85). Mask R-CNN and U-Net exhibited superior performance in terms of Overall Accuracy (OA), F1-score, and Intersection Over Union (IoU). Among OBIA methods, the Gaussian Mixture Model (GMM) proved to be the most effective classifier for canopy segmentation, and Support Vector Machine (SVM) also demonstrated reasonable stability. Conversely, Random Forest (RF) and K-Means yielded lower accuracy and higher error rates. As a result of the limited accuracy, it is noted for vineyard rows with low vigor canopies that NDVI was overestimated, while for high vigor canopies NDVI was underestimated. Significantly improved determination coefficients were observed for the comparison between Total Leaf Area (TLA) and NDVI data derived from Mask R-CNN and U-Net. Positive correlations were also found with NDVI data from GMM and SVM algorithms. Regarding leaf chlorophyll (Chl) and NDVI correlations, Mask R-CNN and U-Net methods showed superior performance. Additionally, the relationship between TLA and projected canopy area (PCA) was significantly better represented by U-Net and Mask R-CNN, while PCA was not recommended for estimating chlorophyll content. This investigation establishes that improved vine canopy delimitation contributes to improved vineyard vigour monitoring, providing winegrowers with more accurate and reliable agronomic information for management decisions.},
 bibtype = {article},
 author = {Ferro, Massimo Vincenzo and Sørensen, Claus Grøn and Catania, Pietro},
 doi = {10.1016/j.compag.2024.109277},
 journal = {Computers and Electronics in Agriculture}
}

In viticulture, the rapid and accurate acquisition of canopy spectral information through ultra-high spatial resolution imagery is increasingly demanded for decision support. The prevalent practice involves creating vigor maps using spectral data obtained from pure vine canopy pixels. Object-Based Image Analysis (OBIA) among conventional methods exhibits a reasonable efficiency in canopy classification due to its feature extraction capabilities. In recent years, deep learning (DL) techniques have demonstrated significant potential in orchard monitoring, leveraging their ability to automatically learn image features. This study assessed the performance of different methodologies, including Mask R-CNN, U-Net, OBIA and unsupervised methods, in identifying pure canopy pixels. The effectiveness of shadow and background detection methods and the impact of misclassified pixels on NDVI were compared. Results were compared with agronomic surveys conducted during the 2021 and 2022 growing seasons, focusing on two distinct phenological stages (BBCH65-BBCH85). Mask R-CNN and U-Net exhibited superior performance in terms of Overall Accuracy (OA), F1-score, and Intersection Over Union (IoU). Among OBIA methods, the Gaussian Mixture Model (GMM) proved to be the most effective classifier for canopy segmentation, and Support Vector Machine (SVM) also demonstrated reasonable stability. Conversely, Random Forest (RF) and K-Means yielded lower accuracy and higher error rates. As a result of the limited accuracy, it is noted for vineyard rows with low vigor canopies that NDVI was overestimated, while for high vigor canopies NDVI was underestimated. Significantly improved determination coefficients were observed for the comparison between Total Leaf Area (TLA) and NDVI data derived from Mask R-CNN and U-Net. Positive correlations were also found with NDVI data from GMM and SVM algorithms. Regarding leaf chlorophyll (Chl) and NDVI correlations, Mask R-CNN and U-Net methods showed superior performance. Additionally, the relationship between TLA and projected canopy area (PCA) was significantly better represented by U-Net and Mask R-CNN, while PCA was not recommended for estimating chlorophyll content. This investigation establishes that improved vine canopy delimitation contributes to improved vineyard vigour monitoring, providing winegrowers with more accurate and reliable agronomic information for management decisions.

@article{
 title = {Intelligent vineyard monitoring using YOLOv7},
 type = {article},
 year = {2024},
 keywords = {conference,earth sciences,energy,environment,proceedings},
 pages = {02002},
 volume = {548},
 websites = {https://www.e3s-conferences.org/articles/e3sconf/abs/2024/78/e3sconf_agritech-x_02002/e3sconf_agritech-x_02002.html},
 publisher = {EDP Sciences},
 id = {b26a1575-384f-37aa-b92a-79c3ff1cb885},
 created = {2024-09-16T07:22:46.875Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:22:50.812Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {The article discusses the technology for automated neural network monitoring of the vineyard’s physiological condition. The proposed solution is based on the integrated use of convolutional neural network method and machine vision technologies. The training of the YOLOv7 neural network was implemented in the Python environment using the PyTorch framework and the OpenCV computer vision library. The dataset consisting of 6320 images of grape leaves (including healthy and diseased ones) has been used for neural network training. The obtained results showed that the detection accuracy is at least 91%. Visualization of monitoring results has been carried out using heatmap, allowing to obtain information about vineyard physiological condition in dynamics. The proposed mathematical model allows to calculate the monitored vineyard’s area made by one complex per day.},
 bibtype = {article},
 author = {Kuznetsov, Pavel and Voronin, Dmitry and Kotelnikov, Dmitriy},
 doi = {10.1051/E3SCONF/202454802002},
 journal = {E3S Web of Conferences}
}

The article discusses the technology for automated neural network monitoring of the vineyard’s physiological condition. The proposed solution is based on the integrated use of convolutional neural network method and machine vision technologies. The training of the YOLOv7 neural network was implemented in the Python environment using the PyTorch framework and the OpenCV computer vision library. The dataset consisting of 6320 images of grape leaves (including healthy and diseased ones) has been used for neural network training. The obtained results showed that the detection accuracy is at least 91%. Visualization of monitoring results has been carried out using heatmap, allowing to obtain information about vineyard physiological condition in dynamics. The proposed mathematical model allows to calculate the monitored vineyard’s area made by one complex per day.

@article{
 title = {Deep Learning-based VGG16, VGG19, and ResNet Models for Grapevine Disease Classification},
 type = {article},
 year = {2024},
 keywords = {VGG16,VGG19,disease detection,grapevine,image classification,residual network},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {2ec91189-a635-3bb0-9455-09463ac0597d},
 created = {2024-09-16T07:23:02.375Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:23:09.552Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Identifying grape plant disease is very important to keep the vineyard healthy and make sure we can grow good grapes. This study looks into using deep learning models like VGG16, VGG19 and ResNet to automatically find out problems in grapevine health. The models were trained on a wide set of 9027 images from Kaggle. These pictures showed different types of disease, including powdery mildew and downy mildew among others. We used a learning rate of 0.01 to help the model come together during training faster. The results show that all three models work very well in finding grapevine diseases. VGG19 was the best one, reaching a correct rate of 98.39% at stage number 24. This better performance comes from its more complex design, allowing it to catch complicated disease parts. VGG16 also did very well, reaching 93.46% correctness, while ResNet had a strong accuracy of 89.33%. The results show that using machine learning deep models can help find grapevine diseases easily. This gives farmers a good tool to manage their vineyards well in advance. The side-by-side look shows the special good points of each design. This study helps to mix artificial intelligence with grape farming. It gives tips that could be helpful in choosing how to handle diseases better. Later research may look at easier ways to understand models, how well they work in different areas where vines grow, and if they can be used for more things. This is important for the grape and wine industry.},
 bibtype = {article},
 author = {Vats, Satvik and Anand, John and Kukreja, Vinay and Sharma, Rishabh},
 doi = {10.1109/I2CT61223.2024.10544168},
 journal = {2024 IEEE 9th International Conference for Convergence in Technology, I2CT 2024}
}

Identifying grape plant disease is very important to keep the vineyard healthy and make sure we can grow good grapes. This study looks into using deep learning models like VGG16, VGG19 and ResNet to automatically find out problems in grapevine health. The models were trained on a wide set of 9027 images from Kaggle. These pictures showed different types of disease, including powdery mildew and downy mildew among others. We used a learning rate of 0.01 to help the model come together during training faster. The results show that all three models work very well in finding grapevine diseases. VGG19 was the best one, reaching a correct rate of 98.39% at stage number 24. This better performance comes from its more complex design, allowing it to catch complicated disease parts. VGG16 also did very well, reaching 93.46% correctness, while ResNet had a strong accuracy of 89.33%. The results show that using machine learning deep models can help find grapevine diseases easily. This gives farmers a good tool to manage their vineyards well in advance. The side-by-side look shows the special good points of each design. This study helps to mix artificial intelligence with grape farming. It gives tips that could be helpful in choosing how to handle diseases better. Later research may look at easier ways to understand models, how well they work in different areas where vines grow, and if they can be used for more things. This is important for the grape and wine industry.

@article{
 title = {Proposed Fuzzy-Stranded-Neural Network Model That Utilizes IoT Plant-Level Sensory Monitoring and Distributed Services for the Early Detection of Downy Mildew in Viticulture},
 type = {article},
 year = {2024},
 keywords = {agriculture 4.0,decision support systems for the agriculture industry,deep learning models,distributed IoT systems and services,precision agriculture,viticulture decision making},
 pages = {63},
 volume = {13},
 websites = {https://www.mdpi.com/2073-431X/13/3/63/htm,https://www.mdpi.com/2073-431X/13/3/63},
 month = {2},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {28},
 id = {8547a55c-0ade-3ec3-a6d1-153632ac774b},
 created = {2024-09-16T07:23:49.551Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:24:18.949Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Novel monitoring architecture approaches are required to detect viticulture diseases early. Existing micro-climate decision support systems can only cope with late detection from empirical and semi-empirical models that provide less accurate results. Such models cannot alleviate precision viticulture planning and pesticide control actions, providing early reconnaissances that may trigger interventions. This paper presents a new plant-level monitoring architecture called thingsAI. The proposed system utilizes low-cost, autonomous, easy-to-install IoT sensors for vine-level monitoring, utilizing the low-power LoRaWAN protocol for sensory measurement acquisition. Facilitated by a distributed cloud architecture and open-source user interfaces, it provides state-of-the-art deep learning inference services and decision support interfaces. This paper also presents a new deep learning detection algorithm based on supervised fuzzy annotation processes, targeting downy mildew disease detection and, therefore, planning early interventions. The authors tested their proposed system and deep learning model on the grape variety of protected designation of origin called debina, cultivated in Zitsa, Greece. From their experimental results, the authors show that their proposed model can detect vine locations and timely breakpoints of mildew occurrences, which farmers can use as input for targeted intervention efforts.},
 bibtype = {article},
 author = {Kontogiannis, Sotirios and Koundouras, Stefanos and Pikridas, Christos},
 doi = {10.3390/COMPUTERS13030063},
 journal = {Computers 2024, Vol. 13, Page 63},
 number = {3}
}

Novel monitoring architecture approaches are required to detect viticulture diseases early. Existing micro-climate decision support systems can only cope with late detection from empirical and semi-empirical models that provide less accurate results. Such models cannot alleviate precision viticulture planning and pesticide control actions, providing early reconnaissances that may trigger interventions. This paper presents a new plant-level monitoring architecture called thingsAI. The proposed system utilizes low-cost, autonomous, easy-to-install IoT sensors for vine-level monitoring, utilizing the low-power LoRaWAN protocol for sensory measurement acquisition. Facilitated by a distributed cloud architecture and open-source user interfaces, it provides state-of-the-art deep learning inference services and decision support interfaces. This paper also presents a new deep learning detection algorithm based on supervised fuzzy annotation processes, targeting downy mildew disease detection and, therefore, planning early interventions. The authors tested their proposed system and deep learning model on the grape variety of protected designation of origin called debina, cultivated in Zitsa, Greece. From their experimental results, the authors show that their proposed model can detect vine locations and timely breakpoints of mildew occurrences, which farmers can use as input for targeted intervention efforts.

@article{
 title = {Grapevine Disease Identification Using Resnet−50},
 type = {article},
 year = {2024},
 keywords = {biology,conference,health,life,proceedings},
 pages = {01046},
 volume = {117},
 websites = {https://www.bio-conferences.org/articles/bioconf/abs/2024/36/bioconf_icolist2023_01046/bioconf_icolist2023_01046.html},
 month = {7},
 publisher = {EDP Sciences},
 day = {5},
 id = {dc584767-af3f-33d1-8707-b59cb1dc6971},
 created = {2024-09-16T07:24:18.193Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:24:22.721Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Visual identification of diseases in grapevines can be a difficult task for growers. The importance of farmers in the identification of grape diseases due to control the spread of disease and lower agricultural yield losses. In this study developed a disease identification system in plants using image processing. Images of leaves on grapevines infected with the disease were taken, extracted features from the images and applied the ResNet-50 algorithm. The dataset of grape leaf images taken was 200 images for four classes, including 3 classes of leaves identified as diseased and 1 class of healthy leaves. The experimental results show that the image processing system for identifying diseases in grapes identifies the types of disease in grapevines. This research has the potential to be implemented in a farm automation system to detect early diseases in grapevines and take appropriate preventive measures to increase productivity and crop quality.},
 bibtype = {article},
 author = {Badriyah, Asfiyatul and Sarosa, Moechammad and Asmara, Rosa Andrie and Wardani, Mila Kusuma and Al Riza, Dimas Firmanda},
 doi = {10.1051/BIOCONF/202411701046},
 journal = {BIO Web of Conferences}
}

Visual identification of diseases in grapevines can be a difficult task for growers. The importance of farmers in the identification of grape diseases due to control the spread of disease and lower agricultural yield losses. In this study developed a disease identification system in plants using image processing. Images of leaves on grapevines infected with the disease were taken, extracted features from the images and applied the ResNet-50 algorithm. The dataset of grape leaf images taken was 200 images for four classes, including 3 classes of leaves identified as diseased and 1 class of healthy leaves. The experimental results show that the image processing system for identifying diseases in grapes identifies the types of disease in grapevines. This research has the potential to be implemented in a farm automation system to detect early diseases in grapevines and take appropriate preventive measures to increase productivity and crop quality.

@article{
 title = {Image analysis with deep learning for early detection of downy mildew in grapevine},
 type = {article},
 year = {2024},
 keywords = {deep learning,early detection,explainable artificial intelligence,viticulture},
 pages = {113155},
 volume = {331},
 month = {5},
 publisher = {Elsevier},
 day = {1},
 id = {ccecb560-09d2-3c95-9ce7-d7bc970f0d0f},
 created = {2024-09-16T07:24:44.021Z},
 accessed = {2024-09-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-16T07:24:47.966Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3},
 private_publication = {false},
 abstract = {Downy mildew is a major disease of the grapevine that can severely reduce crop quality and yield. Its assessment in the laboratory is time-consuming, usually carried out by experts, and can require expensive and complex tools. For this reason, there is an opportunity to apply sensor technologies and artificial intelligence to plant disease detection. In this study, deep learning applied to RGB images was investigated to early detect downy mildew and the infection stage in grapevine leaf discs under laboratory conditions. Leaf discs of Tempranillo grapevine variety from 3 to 9 days post-inoculation located in Petri dishes were imaged using controlled conditions. Leaf disc images were extracted using computer vision techniques. Convolutional Neural Networks were used to classify the infected and healthy discs and to identify the disease infection. 10-fold cross-validation was used to evaluate the network results and Grad-CAM was used to interpret model prediction. An accuracy around 99% and a f1-score of 0.99 was achieved in downy mildew detection after DPI 3. An accuracy of 81% and a f1-score of 0.77 was obtained in infection stage identification. The developed method offered objective, rapid and accurate results, giving the possibility of early detecting downy mildew in grapevine leaf discs using low-cost techniques.},
 bibtype = {article},
 author = {Hernández, Inés and Gutiérrez, Salvador and Tardaguila, Javier},
 doi = {10.1016/J.SCIENTA.2024.113155},
 journal = {Scientia Horticulturae}
}

Downy mildew is a major disease of the grapevine that can severely reduce crop quality and yield. Its assessment in the laboratory is time-consuming, usually carried out by experts, and can require expensive and complex tools. For this reason, there is an opportunity to apply sensor technologies and artificial intelligence to plant disease detection. In this study, deep learning applied to RGB images was investigated to early detect downy mildew and the infection stage in grapevine leaf discs under laboratory conditions. Leaf discs of Tempranillo grapevine variety from 3 to 9 days post-inoculation located in Petri dishes were imaged using controlled conditions. Leaf disc images were extracted using computer vision techniques. Convolutional Neural Networks were used to classify the infected and healthy discs and to identify the disease infection. 10-fold cross-validation was used to evaluate the network results and Grad-CAM was used to interpret model prediction. An accuracy around 99% and a f1-score of 0.99 was achieved in downy mildew detection after DPI 3. An accuracy of 81% and a f1-score of 0.77 was obtained in infection stage identification. The developed method offered objective, rapid and accurate results, giving the possibility of early detecting downy mildew in grapevine leaf discs using low-cost techniques.

@article{
 title = {Dynamic Slicing and Reconstruction Algorithm for Precise Canopy Volume Estimation in 3D Citrus Tree Point Clouds},
 type = {article},
 year = {2024},
 keywords = {LiDAR,canopy volume,citrus tree,dynamic slicing,point cloud reconstruction},
 pages = {2142},
 volume = {16},
 websites = {https://www.mdpi.com/2072-4292/16/12/2142/htm,https://www.mdpi.com/2072-4292/16/12/2142},
 month = {6},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {13},
 id = {7d754336-6442-398e-b9d3-91e1143180ff},
 created = {2024-09-17T07:10:00.221Z},
 accessed = {2024-09-17},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-17T07:10:17.157Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Crop phenotyping data collection is the basis for precision agriculture and smart decision-making applications. Accurately obtaining the canopy volume of citrus trees is crucial for yield prediction, precise fertilization and cultivation management. To this end, we developed a dynamic slicing and reconstruction (DR) algorithm based on 3D point clouds. The algorithm dynamically slices nearby slices based on their proportional area change and density difference; for each slice point cloud, the average distance of each point from others is taken as the initial α value for the AS algorithm. This value is iteratively summed until it reconstructs the complete shape, allowing the volume of each slice shape to be determined. Compared with six point cloud-based reconstruction algorithms, the DR approach achieved the best results in removing perforations and lacunae (0.84) and exhibited volumetric consistency (1.53) that closely aligned with the growth pattern of citrus trees. The DR algorithm effectively addresses the challenges of adapting the thickness and number of canopy point cloud slices to the shape and size of the canopy in the ASBS and CHBS algorithms, as well as overcoming inaccuracies and incompleteness in reconstructed canopy models caused by limitations in capturing detailed features using the PCH algorithm. It offers improved adaptive ability, finer volume computations, better noise reduction, and anomaly removal.},
 bibtype = {article},
 author = {Li, Wenjie and Tang, Biyu and Hou, Zhen and Wang, Hongbo and Bing, Zongyu and Yang, Qiong and Zheng, Yongqiang},
 doi = {10.3390/RS16122142},
 journal = {Remote Sensing 2024, Vol. 16, Page 2142},
 number = {12}
}

Crop phenotyping data collection is the basis for precision agriculture and smart decision-making applications. Accurately obtaining the canopy volume of citrus trees is crucial for yield prediction, precise fertilization and cultivation management. To this end, we developed a dynamic slicing and reconstruction (DR) algorithm based on 3D point clouds. The algorithm dynamically slices nearby slices based on their proportional area change and density difference; for each slice point cloud, the average distance of each point from others is taken as the initial α value for the AS algorithm. This value is iteratively summed until it reconstructs the complete shape, allowing the volume of each slice shape to be determined. Compared with six point cloud-based reconstruction algorithms, the DR approach achieved the best results in removing perforations and lacunae (0.84) and exhibited volumetric consistency (1.53) that closely aligned with the growth pattern of citrus trees. The DR algorithm effectively addresses the challenges of adapting the thickness and number of canopy point cloud slices to the shape and size of the canopy in the ASBS and CHBS algorithms, as well as overcoming inaccuracies and incompleteness in reconstructed canopy models caused by limitations in capturing detailed features using the PCH algorithm. It offers improved adaptive ability, finer volume computations, better noise reduction, and anomaly removal.

@article{
 title = {The estimation of wheat yield combined with UAV canopy spectral and volumetric data},
 type = {article},
 year = {2024},
 keywords = {canopy volume,spectral,unmanned aerial vehicles (UAVs),wheat,yield estimation},
 pages = {e527},
 volume = {13},
 websites = {https://onlinelibrary.wiley.com/doi/full/10.1002/fes3.527,https://onlinelibrary.wiley.com/doi/abs/10.1002/fes3.527,https://onlinelibrary.wiley.com/doi/10.1002/fes3.527},
 month = {1},
 publisher = {John Wiley & Sons, Ltd},
 day = {1},
 id = {aaf9e70b-a8d9-3f01-9cf9-12485d562254},
 created = {2024-09-17T07:34:57.844Z},
 accessed = {2024-09-17},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-18T12:23:24.273Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {Estimating wheat yield accurately is crucial for efficient agricultural management. While canopy spectral information is widely used for this purpose, the incorporation of canopy volumetric features (CVFs) remains underexplored. This study bridges this gap by utilizing unmanned aerial vehicle (UAV) multispectral imaging to capture images and elevation data of wheat at key developmental stages—gestation and flowering stages. We innovatively leveraged the elevation differences between these stages to calculate canopy height, develop a novel CVF, and refine the wheat yield prediction model across various wheat varieties, nitrogen fertilizer levels, and planting densities. The integration of canopy volume information significantly enhanced the accuracy of our yield prediction model, as evidenced by an R2 of 0.8380, an RMSE of 313.3 kg/ha, and an nRMSE of 11.33%. This approach not only yielded more precise estimates than models relying solely on spectral data but also introduced a novel dimension to wheat yield estimation methodologies. Our findings suggest that incorporating canopy volume characteristics can substantially optimize wheat yield prediction models, presenting a groundbreaking perspective for agricultural yield estimation.},
 bibtype = {article},
 author = {Liu, Tao and Wu, Fei and Mou, Nana and Zhu, Shaolong and Yang, Tianle and Zhang, Weijun and Wang, Hui and Wu, Wei and Zhao, Yuanyuan and Sun, Chengming and Yao, Zhaosheng},
 doi = {10.1002/FES3.527},
 journal = {Food and Energy Security},
 number = {1}
}

Estimating wheat yield accurately is crucial for efficient agricultural management. While canopy spectral information is widely used for this purpose, the incorporation of canopy volumetric features (CVFs) remains underexplored. This study bridges this gap by utilizing unmanned aerial vehicle (UAV) multispectral imaging to capture images and elevation data of wheat at key developmental stages—gestation and flowering stages. We innovatively leveraged the elevation differences between these stages to calculate canopy height, develop a novel CVF, and refine the wheat yield prediction model across various wheat varieties, nitrogen fertilizer levels, and planting densities. The integration of canopy volume information significantly enhanced the accuracy of our yield prediction model, as evidenced by an R2 of 0.8380, an RMSE of 313.3 kg/ha, and an nRMSE of 11.33%. This approach not only yielded more precise estimates than models relying solely on spectral data but also introduced a novel dimension to wheat yield estimation methodologies. Our findings suggest that incorporating canopy volume characteristics can substantially optimize wheat yield prediction models, presenting a groundbreaking perspective for agricultural yield estimation.

@article{
 title = {A comparison and development of methods for estimating shrub volume using drone-imagery-derived point clouds},
 type = {article},
 year = {2024},
 keywords = {canopy volume,digital photogrammetry,from,fuel load,motion,rangeland,sagebrush,structure,unmanned aerial system (UAS),vegetation monitoring},
 pages = {e4877},
 volume = {15},
 websites = {https://onlinelibrary.wiley.com/doi/full/10.1002/ecs2.4877,https://onlinelibrary.wiley.com/doi/abs/10.1002/ecs2.4877,https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4877},
 month = {5},
 publisher = {John Wiley & Sons, Ltd},
 day = {1},
 id = {5b51708b-b9d7-34c1-b1c5-6dd1ab96a824},
 created = {2024-09-18T06:34:12.173Z},
 accessed = {2024-09-18},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-18T06:34:12.173Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Shrub volume is used to calculate numerous, essential ecological indicators in rangeland ecosystems such as biomass, fuel loading, wildlife habitat, site productivity, and ecosystem structure. Field techniques for biomass estimation, including destructive sampling, ocular estimates, and allometric techniques, use shrub height and canopy widths to estimate volume and translate it to biomass with species-specific allometric equations. These techniques are time-consuming and pose challenges, including removal of plant material and training of observers. We compared canopy volume estimates from field-based measurements with drone-collected canopy volume estimates for seven dominant shrub species within mountain big sagebrush (Artemisia tridentata subsp. vaseyana) plant communities in southern Idaho, USA. Canopy height and two perpendicular width measurements were taken from 103 shrubs of varying sizes, and volume was estimated using a traditional allometric equation. Overlapping aerial images captured with a DJI Mavic 2 Professional drone were used to create a 3D representation of the study area using structure-from-motion photogrammetry. Each shrub was extracted from the point cloud, and volume was estimated using allometric and volumetric methods. The volumetric method, which involved converting point clouds to raster canopy height models with 2.5- and 5-cm grid cells, outperformed the allometric method (R2 > 0.7) and was more reproducible and robust to user-related variability. Drone-estimated volume best-matched field-estimated volume (R2 > 0.9) for three larger species: A. tridentata subsp. tridentata, A. tridentata subsp. vaseyana, and Purshia tridentata. The volume of smaller shrubs (canopy widths <1 m) was slightly overestimated from drone-based models. We argue that drone-based models provide a suitable alternative to field methods, while having the added benefit of being less time-consuming, with fewer limitations, and more easily scaled to larger study areas than traditional field techniques. Finally, we demonstrate a proof of concept for automating canopy volume estimates using point-cloud-based automatic shrub detection algorithms. These findings demonstrate that drone-collected images can be used to assess shrub canopy volume for at least five upland sagebrush steppe shrub species and support the integration of drone data collection into rangeland vegetation monitoring.},
 bibtype = {article},
 author = {Harrison, Georgia R. and Shrestha, Abhinav and Strand, Eva K. and Karl, Jason W.},
 doi = {10.1002/ECS2.4877},
 journal = {Ecosphere},
 number = {5}
}

Shrub volume is used to calculate numerous, essential ecological indicators in rangeland ecosystems such as biomass, fuel loading, wildlife habitat, site productivity, and ecosystem structure. Field techniques for biomass estimation, including destructive sampling, ocular estimates, and allometric techniques, use shrub height and canopy widths to estimate volume and translate it to biomass with species-specific allometric equations. These techniques are time-consuming and pose challenges, including removal of plant material and training of observers. We compared canopy volume estimates from field-based measurements with drone-collected canopy volume estimates for seven dominant shrub species within mountain big sagebrush (Artemisia tridentata subsp. vaseyana) plant communities in southern Idaho, USA. Canopy height and two perpendicular width measurements were taken from 103 shrubs of varying sizes, and volume was estimated using a traditional allometric equation. Overlapping aerial images captured with a DJI Mavic 2 Professional drone were used to create a 3D representation of the study area using structure-from-motion photogrammetry. Each shrub was extracted from the point cloud, and volume was estimated using allometric and volumetric methods. The volumetric method, which involved converting point clouds to raster canopy height models with 2.5- and 5-cm grid cells, outperformed the allometric method (R2 > 0.7) and was more reproducible and robust to user-related variability. Drone-estimated volume best-matched field-estimated volume (R2 > 0.9) for three larger species: A. tridentata subsp. tridentata, A. tridentata subsp. vaseyana, and Purshia tridentata. The volume of smaller shrubs (canopy widths <1 m) was slightly overestimated from drone-based models. We argue that drone-based models provide a suitable alternative to field methods, while having the added benefit of being less time-consuming, with fewer limitations, and more easily scaled to larger study areas than traditional field techniques. Finally, we demonstrate a proof of concept for automating canopy volume estimates using point-cloud-based automatic shrub detection algorithms. These findings demonstrate that drone-collected images can be used to assess shrub canopy volume for at least five upland sagebrush steppe shrub species and support the integration of drone data collection into rangeland vegetation monitoring.

@article{
 title = {Deep learning method for leaf-density estimation based on wind-excited audio of fruit-tree canopies},
 type = {article},
 year = {2024},
 keywords = {Acoustic spectrogram,Attentional mechanism,Deep learning,Precision agriculture,Wind-excited audio},
 pages = {109062},
 volume = {222},
 month = {7},
 publisher = {Elsevier},
 day = {1},
 id = {e91e35ff-90d5-3307-bb7c-f4f2e4ae4324},
 created = {2024-09-18T06:36:19.368Z},
 accessed = {2024-09-18},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-18T12:23:46.911Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {It is essential for precision air-assisted ground sprayer to accurately detect the leaf density of fruit trees in agriculture. However, it is difficult to achieve real-time measurement of canopy leaf densities by using conventional sensors (such as LIDAR and camera). In fact, the interaction between wind and canopies of different leaf densities may generate different excited-audios during air-assisted spraying. Therefore, this paper proposes a method for leaf-density detection of fruit-tree canopies based on the variance of wind-excited audios. First, the fused spectrogram (FSP) was constructed by the short-time fast Fourier transform (STFT), consisting of spectral features and the spectrogram of audio signals. Then, the estimation model between the FSP and leaf densities was developed by using deep convolutional neural network (DCNN), which was enhanced by a spectral centroid attention (SCA) based on the distribution function of spectral centroids. Finally, the test results showed that: (1) the developed model achieved a 96.93% accuracy and a 97.96% precision in leaf-density recognition, and (2) compared with unimproved DCNN model, the accuracy and precision of the developed model were increased about 7.85% and 8.47%, respectively, which indicated that this method could achieve the prediction of leaf-density based on wind-excited audios. The study is expected to provide a reference for leaf-density detection of fruit tree canopies and real-time control of precision spray.},
 bibtype = {article},
 author = {Li, Wenwei and Yang, Shenghui and Zhao, Hangxing and Jiang, Shijie and Zheng, Yongjun and Liu, Xingxing and Tan, Yu},
 doi = {10.1016/J.COMPAG.2024.109062},
 journal = {Computers and Electronics in Agriculture}
}

It is essential for precision air-assisted ground sprayer to accurately detect the leaf density of fruit trees in agriculture. However, it is difficult to achieve real-time measurement of canopy leaf densities by using conventional sensors (such as LIDAR and camera). In fact, the interaction between wind and canopies of different leaf densities may generate different excited-audios during air-assisted spraying. Therefore, this paper proposes a method for leaf-density detection of fruit-tree canopies based on the variance of wind-excited audios. First, the fused spectrogram (FSP) was constructed by the short-time fast Fourier transform (STFT), consisting of spectral features and the spectrogram of audio signals. Then, the estimation model between the FSP and leaf densities was developed by using deep convolutional neural network (DCNN), which was enhanced by a spectral centroid attention (SCA) based on the distribution function of spectral centroids. Finally, the test results showed that: (1) the developed model achieved a 96.93% accuracy and a 97.96% precision in leaf-density recognition, and (2) compared with unimproved DCNN model, the accuracy and precision of the developed model were increased about 7.85% and 8.47%, respectively, which indicated that this method could achieve the prediction of leaf-density based on wind-excited audios. The study is expected to provide a reference for leaf-density detection of fruit tree canopies and real-time control of precision spray.

@article{
 title = {Measuring ornamental tree canopy attributes for precision spraying using drone technology and self-supervised segmentation},
 type = {article},
 year = {2024},
 keywords = {Canopy volume,Nursery crops,Precision spray applications,Segment anything model,Tree height measurement},
 websites = {https://doi.org/10.1016/j.compag.2024.109359},
 id = {ca5f1703-aa20-3eec-bd51-dbf3d1a2c120},
 created = {2024-09-18T06:38:08.270Z},
 accessed = {2024-09-18},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-18T12:24:14.165Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {Tree canopy attributes or characteristics, such as canopy volume and density, are important parameters for calculating the precise quantity of agrochemicals required in each tree. Assessing the agrochemical needs in a nursery can help growers effectively estimate expenses, mitigate the risks of overuse or underuse, enhance resource utilization, and promote sustainable agricultural practices. This study aimed to develop an unmanned aerial vehicle (UAV)-based system to measure tree canopy attributes like tree height, tree count, canopy area and canopy volume using an image processing and self-supervised zero-shot segmentation approach. A high-resolution red-green-blue (RGB) sensor mounted on a drone captured three aerial image datasets, D1 and D2 from the first experimental plot and D3 from the second experimental plot. The acquired aerial images were stitched together and processed to generate a digital surface model (DSM) and a digital terrain model (DTM). A self-supervised zero-shot segmentation model, Segment Anything Model (SAM), was applied for tree segmen-tation and counting. The tree canopy area was calculated by segmenting individual trees and applying a conversion factor to convert pixelwise areas into square meters. Height calculation was done using elevation data of each pixel from DSM-DTM imagery within SAM-derived bounding box coordinates, and the highest pixel was considered as the height of the tree. The drone-calculated heights of 24 randomly selected trees were compared with manually measured heights in all datasets. The results showed an average absolute error of 2.43% in D1, 7.09% in D2 and 12.58% in D3. The Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were noted to be 0.09 m and 0.23 m in D1, 0.13 m and 0.25 m in D2, and 0.16 m and 0.21 m in D3, respectively. In D3, manual area and volume calculations were performed for validation. The RMSE and MAE for area calculation were calculated as 0.18 m 2 and 0.15 m 2 , respectively, and the RMSE and MAE for volume measurements were calculated as 0.33 m 3 and 0.26 m 3 respectively, which indicated the level of agreement between manual and drone measurements. Using canopy volume information, calculated by summing pixel heights within each tree's bounding box and multiplying by the ground sample distance (GSD) of 0.025 m per pixel, the average quantity of agrochemicals needed for a Maple tree was estimated at 0.20 L. These results underscore the high potential of this method for accurate canopy characteristic calculations and precision spray applications in the ornamental industry.},
 bibtype = {article},
 author = {Rayamajhi, Aleena and Jahanifar, Hasan},
 doi = {10.1016/j.compag.2024.109359}
}

Tree canopy attributes or characteristics, such as canopy volume and density, are important parameters for calculating the precise quantity of agrochemicals required in each tree. Assessing the agrochemical needs in a nursery can help growers effectively estimate expenses, mitigate the risks of overuse or underuse, enhance resource utilization, and promote sustainable agricultural practices. This study aimed to develop an unmanned aerial vehicle (UAV)-based system to measure tree canopy attributes like tree height, tree count, canopy area and canopy volume using an image processing and self-supervised zero-shot segmentation approach. A high-resolution red-green-blue (RGB) sensor mounted on a drone captured three aerial image datasets, D1 and D2 from the first experimental plot and D3 from the second experimental plot. The acquired aerial images were stitched together and processed to generate a digital surface model (DSM) and a digital terrain model (DTM). A self-supervised zero-shot segmentation model, Segment Anything Model (SAM), was applied for tree segmen-tation and counting. The tree canopy area was calculated by segmenting individual trees and applying a conversion factor to convert pixelwise areas into square meters. Height calculation was done using elevation data of each pixel from DSM-DTM imagery within SAM-derived bounding box coordinates, and the highest pixel was considered as the height of the tree. The drone-calculated heights of 24 randomly selected trees were compared with manually measured heights in all datasets. The results showed an average absolute error of 2.43% in D1, 7.09% in D2 and 12.58% in D3. The Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were noted to be 0.09 m and 0.23 m in D1, 0.13 m and 0.25 m in D2, and 0.16 m and 0.21 m in D3, respectively. In D3, manual area and volume calculations were performed for validation. The RMSE and MAE for area calculation were calculated as 0.18 m 2 and 0.15 m 2 , respectively, and the RMSE and MAE for volume measurements were calculated as 0.33 m 3 and 0.26 m 3 respectively, which indicated the level of agreement between manual and drone measurements. Using canopy volume information, calculated by summing pixel heights within each tree's bounding box and multiplying by the ground sample distance (GSD) of 0.025 m per pixel, the average quantity of agrochemicals needed for a Maple tree was estimated at 0.20 L. These results underscore the high potential of this method for accurate canopy characteristic calculations and precision spray applications in the ornamental industry.

@article{
 title = {Maize height estimation using combined unmanned aerial vehicle oblique photography and LIDAR canopy dynamic characteristics},
 type = {article},
 year = {2024},
 keywords = {Canopy dynamic characteristics,LIDAR,Oblique photography,Plant height,UAV},
 pages = {108685},
 volume = {218},
 month = {3},
 publisher = {Elsevier},
 day = {1},
 id = {13acf507-eadb-3445-b1f8-c9a1675b217d},
 created = {2024-09-18T06:42:23.235Z},
 accessed = {2024-09-18},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-18T12:25:21.270Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {Plant height (PH) is a key indicator for assessing plant health and growth, as well as an important criterion for ideotype breeding, making efficient, accurate PH measurement essential. The traditional manual measurement method is small-scale, inefficient, and time-consuming, which has restricted crop breeding progress and agricultural production efficiency improvements. Therefore, in this study, we aimed to estimate the maize populations’ PH in the field using multisource unmanned aerial vehicle (UAV) images and to explore the factors that affect the accuracy of oblique photography (OP) and light detection and ranging (LIDAR) measurements of the populations’ PH. To this end, we acquired red–green–blue (RGB), LIDAR, and multispectral image data for a maize population canopy using a UAV in four growth stages in two maize production regions, and we obtained the maize PH via field measurement as validation data. First, we reconstructed the three-dimensional point cloud using OP and LIDAR, and we used the 2nd and 100th percentiles of the elevation information as the lower and upper boundaries, respectively, to estimate the maize height. In addition, we used multispectral-based vegetation indices (VIs) and RGB-based texture indices (TIs) to describe the population canopy's characteristics, and we compared and analyzed the accuracy of the PH estimation for four data types and ten data fusion methods. The results revealed that all four data types could accurately estimate the maize PH and that the R2 values were all greater than 0.75. The light detection and ranging elevation (LIDAR_elev) had the highest estimation accuracy, and the R2 values of the maximum, minimum, and average PH of the population were all greater than 0.90. We also found that the dynamic characteristics of the canopy growth were important factors affecting the estimation of the PH using the oblique photography elevation (OP_elev) and LIDAR_elev. Thus, after fusing the VIs and TIs, the R2 value reached a maximum of 0.98, and the root mean square error (RMSE) was 8 cm. The use of different regression models had little influence on the estimation of the PH. Our research provides an effective method for high-throughput evaluation of a maize population's PH growth and serves as a reference for studying other phenotype parameters.},
 bibtype = {article},
 author = {Liu, Tao and Zhu, Shaolong and Yang, Tianle and Zhang, Weijun and Xu, Yang and Zhou, Kai and Wu, Wei and Zhao, Yuanyuan and Yao, Zhaosheng and Yang, Guanshuo and Wang, Ying and Sun, Chengming and Sun, Jianjun},
 doi = {10.1016/J.COMPAG.2024.108685},
 journal = {Computers and Electronics in Agriculture}
}

Plant height (PH) is a key indicator for assessing plant health and growth, as well as an important criterion for ideotype breeding, making efficient, accurate PH measurement essential. The traditional manual measurement method is small-scale, inefficient, and time-consuming, which has restricted crop breeding progress and agricultural production efficiency improvements. Therefore, in this study, we aimed to estimate the maize populations’ PH in the field using multisource unmanned aerial vehicle (UAV) images and to explore the factors that affect the accuracy of oblique photography (OP) and light detection and ranging (LIDAR) measurements of the populations’ PH. To this end, we acquired red–green–blue (RGB), LIDAR, and multispectral image data for a maize population canopy using a UAV in four growth stages in two maize production regions, and we obtained the maize PH via field measurement as validation data. First, we reconstructed the three-dimensional point cloud using OP and LIDAR, and we used the 2nd and 100th percentiles of the elevation information as the lower and upper boundaries, respectively, to estimate the maize height. In addition, we used multispectral-based vegetation indices (VIs) and RGB-based texture indices (TIs) to describe the population canopy's characteristics, and we compared and analyzed the accuracy of the PH estimation for four data types and ten data fusion methods. The results revealed that all four data types could accurately estimate the maize PH and that the R2 values were all greater than 0.75. The light detection and ranging elevation (LIDAR_elev) had the highest estimation accuracy, and the R2 values of the maximum, minimum, and average PH of the population were all greater than 0.90. We also found that the dynamic characteristics of the canopy growth were important factors affecting the estimation of the PH using the oblique photography elevation (OP_elev) and LIDAR_elev. Thus, after fusing the VIs and TIs, the R2 value reached a maximum of 0.98, and the root mean square error (RMSE) was 8 cm. The use of different regression models had little influence on the estimation of the PH. Our research provides an effective method for high-throughput evaluation of a maize population's PH growth and serves as a reference for studying other phenotype parameters.

@article{
 title = {Improved voxel-based volume estimation and pruning severity mapping of apple trees during the pruning period},
 type = {article},
 year = {2024},
 keywords = {3D reconstruction,Multi-camera photogrammetry,Pruning severity,Tree volume,Voxelization},
 pages = {108834},
 volume = {219},
 month = {4},
 publisher = {Elsevier},
 day = {1},
 id = {257521bf-6b16-3690-98bc-58d062c7aa69},
 created = {2024-09-18T06:42:41.357Z},
 accessed = {2024-09-18},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-18T12:27:07.114Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {Estimating tree volume is crucial for managing apple orchards, as it reflects the nutritional status and vigor of the trees. However, accurate assessment of tree volume in apple orchards is challenging due to their complex structure. This study introduces a novel method for three-dimensional volume calculation of individual apple trees. Our approach facilitates the determination of pruning severity by analyzing the limb-to-trunk volume ratio and enables the creation of detailed pre- and post-pruning maps. We utilized a lightweight multi-camera system to reconstruct 3D point clouds of the trees and developed a voxel-based algorithm for tree volume calculation. This algorithm includes steps for interior filling, edge voxel thinning, and interior refilling. We validated our algorithm on seven apple trees by comparing the calculated volumes with the ground truth, determined using the water displacement method. The results showed that our voxel-based algorithm was highly effective in accurately calculating individual tree volumes from 3D point clouds. The algorithm also demonstrated a high coefficient of determination (0.994) and a mean absolute percentage error of 2.919% in a linear regression analysis against the ground truth. Furthermore, we produced detailed tree volume and pruning severity maps for individual trees, both before and after pruning. In conclusion, this study offers an effective solution combining 3D imaging and volume calculation techniques to accurately estimate individual apple tree volumes, providing a quantitative assessment of pruning severity.},
 bibtype = {article},
 author = {Dong, Xuhua and Kim, Woo Young and Yu, Zheng and Oh, Ju Youl and Ehsani, Reza and Lee, Kyeong Hwan},
 doi = {10.1016/J.COMPAG.2024.108834},
 journal = {Computers and Electronics in Agriculture}
}

Estimating tree volume is crucial for managing apple orchards, as it reflects the nutritional status and vigor of the trees. However, accurate assessment of tree volume in apple orchards is challenging due to their complex structure. This study introduces a novel method for three-dimensional volume calculation of individual apple trees. Our approach facilitates the determination of pruning severity by analyzing the limb-to-trunk volume ratio and enables the creation of detailed pre- and post-pruning maps. We utilized a lightweight multi-camera system to reconstruct 3D point clouds of the trees and developed a voxel-based algorithm for tree volume calculation. This algorithm includes steps for interior filling, edge voxel thinning, and interior refilling. We validated our algorithm on seven apple trees by comparing the calculated volumes with the ground truth, determined using the water displacement method. The results showed that our voxel-based algorithm was highly effective in accurately calculating individual tree volumes from 3D point clouds. The algorithm also demonstrated a high coefficient of determination (0.994) and a mean absolute percentage error of 2.919% in a linear regression analysis against the ground truth. Furthermore, we produced detailed tree volume and pruning severity maps for individual trees, both before and after pruning. In conclusion, this study offers an effective solution combining 3D imaging and volume calculation techniques to accurately estimate individual apple tree volumes, providing a quantitative assessment of pruning severity.

@article{
 title = {Limitations of estimating branch volume from terrestrial laser scanning},
 type = {article},
 year = {2024},
 keywords = {Above-ground volume,European beech,LiDAR,QSM,Remote sensing,TLS},
 pages = {687-702},
 volume = {143},
 websites = {https://link.springer.com/article/10.1007/s10342-023-01651-z},
 month = {4},
 publisher = {Springer Science and Business Media Deutschland GmbH},
 day = {1},
 id = {955dcab3-69c8-35cd-9e2d-73f3bab6674b},
 created = {2024-09-18T06:46:58.598Z},
 accessed = {2024-09-18},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-23T09:20:30.273Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {Quantitative structural models (QSMs) are frequently used to simplify single tree point clouds obtained by terrestrial laser scanning (TLS). QSMs use geometric primitives to derive topological and volumetric information about trees. Previous studies have shown a high agreement between TLS and QSM total volume estimates alongside field measured data for whole trees. Although already broadly applied, the uncertainties of the combination of TLS and QSM modelling are still largely unexplored. In our study, we investigated the effect of scanning distance on length and volume estimates of branches when deriving QSMs from TLS data. We scanned ten European beech (Fagus sylvatica L.) branches with an average length of 2.6 m. The branches were scanned from distances ranging from 5 to 45 m at step intervals of 5 m from three scan positions each. Twelve close-range scans were performed as a benchmark. For each distance and branch, QSMs were derived. We found that with increasing distance, the point cloud density and the cumulative length of the reconstructed branches decreased, whereas individual volumes increased. Dependent on the QSM hyperparameters, at a scanning distance of 45 m, cumulative branch length was on average underestimated by − 75%, while branch volume was overestimated by up to + 539%. We assume that the high deviations are related to point cloud quality. As the scanning distance increases, the size of the individual laser footprints and the distances between them increase, making it more difficult to fully capture small branches and to adjust suitable QSMs.},
 bibtype = {article},
 author = {Morhart, Christopher and Schindler, Zoe and Frey, Julian and Sheppard, Jonathan P. and Calders, Kim and Disney, Mathias and Morsdorf, Felix and Raumonen, Pasi and Seifert, Thomas},
 doi = {10.1007/S10342-023-01651-Z/TABLES/3},
 journal = {European Journal of Forest Research},
 number = {2}
}

Quantitative structural models (QSMs) are frequently used to simplify single tree point clouds obtained by terrestrial laser scanning (TLS). QSMs use geometric primitives to derive topological and volumetric information about trees. Previous studies have shown a high agreement between TLS and QSM total volume estimates alongside field measured data for whole trees. Although already broadly applied, the uncertainties of the combination of TLS and QSM modelling are still largely unexplored. In our study, we investigated the effect of scanning distance on length and volume estimates of branches when deriving QSMs from TLS data. We scanned ten European beech (Fagus sylvatica L.) branches with an average length of 2.6 m. The branches were scanned from distances ranging from 5 to 45 m at step intervals of 5 m from three scan positions each. Twelve close-range scans were performed as a benchmark. For each distance and branch, QSMs were derived. We found that with increasing distance, the point cloud density and the cumulative length of the reconstructed branches decreased, whereas individual volumes increased. Dependent on the QSM hyperparameters, at a scanning distance of 45 m, cumulative branch length was on average underestimated by − 75%, while branch volume was overestimated by up to + 539%. We assume that the high deviations are related to point cloud quality. As the scanning distance increases, the size of the individual laser footprints and the distances between them increase, making it more difficult to fully capture small branches and to adjust suitable QSMs.

@article{
 title = {ASIPNet: Orientation-Aware Learning Object Detection for Remote Sensing Images},
 type = {article},
 year = {2024},
 keywords = {aware learning,free detector,object detection,one,orientation,remote sensing,stage anchor},
 pages = {2992},
 volume = {16},
 websites = {https://www.mdpi.com/2072-4292/16/16/2992/htm,https://www.mdpi.com/2072-4292/16/16/2992},
 month = {8},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {15},
 id = {533065c9-0b56-3f77-9865-f615af0e81d2},
 created = {2024-09-26T07:46:24.806Z},
 accessed = {2024-09-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-26T07:46:46.408Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Remote sensing imagery poses significant challenges for object detection due to the presence of objects at multiple scales, dense target overlap, and the complexity of extracting features from small targets. This paper introduces an innovative Adaptive Spatial Information Perception Network (ASIPNet), designed to address the problem of detecting objects in complex remote sensing image scenes and significantly enhance detection accuracy. We first designed the core component of ASIPNet, an Adaptable Spatial Information Perception Module (ASIPM), which strengthens the feature extraction of multi-scale objects in remote sensing images by dynamically perceiving contextual background information. Secondly, To further refine the model’s accuracy in predicting oriented bounding boxes, we integrated the Skew Intersection over Union based on Kalman Filtering (KFIoU), which serves as an advanced loss function, surpassing the capabilities of the baseline model’s traditional loss function. Finally, we designed detailed experiments on the DOTAv1 and DIOR-R datasets, which are annotated with rotation, to comprehensively evaluate the performance of ASIPNet. The experimental results demonstrate that ASIPNet achieved mAP50 scores of 76.0% and 80.1%, respectively. These results not only validate the model’s effectiveness but also indicate that this method is significantly ahead of other most current state-of-the-art approaches.},
 bibtype = {article},
 author = {Dong, Ruchan and Yin, Shunyao and Jiao, Licheng and An, Jungang and Wu, Wenjing},
 doi = {10.3390/RS16162992},
 journal = {Remote Sensing 2024, Vol. 16, Page 2992},
 number = {16}
}

Remote sensing imagery poses significant challenges for object detection due to the presence of objects at multiple scales, dense target overlap, and the complexity of extracting features from small targets. This paper introduces an innovative Adaptive Spatial Information Perception Network (ASIPNet), designed to address the problem of detecting objects in complex remote sensing image scenes and significantly enhance detection accuracy. We first designed the core component of ASIPNet, an Adaptable Spatial Information Perception Module (ASIPM), which strengthens the feature extraction of multi-scale objects in remote sensing images by dynamically perceiving contextual background information. Secondly, To further refine the model’s accuracy in predicting oriented bounding boxes, we integrated the Skew Intersection over Union based on Kalman Filtering (KFIoU), which serves as an advanced loss function, surpassing the capabilities of the baseline model’s traditional loss function. Finally, we designed detailed experiments on the DOTAv1 and DIOR-R datasets, which are annotated with rotation, to comprehensively evaluate the performance of ASIPNet. The experimental results demonstrate that ASIPNet achieved mAP50 scores of 76.0% and 80.1%, respectively. These results not only validate the model’s effectiveness but also indicate that this method is significantly ahead of other most current state-of-the-art approaches.

@article{
 title = {A Comparative Study of Loss Functions for Arbitrary-Oriented Object Detection in Aerial Images},
 type = {article},
 year = {2024},
 keywords = {CRIoU,DOTA-PS,gaussian-based loss,oriented object detection,rotated loss},
 pages = {22-27},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {aba3a414-d4fe-34d0-9ef1-d4ce42f2b128},
 created = {2024-09-26T07:53:02.973Z},
 accessed = {2024-09-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-26T07:53:11.193Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Detecting objects with specific orientations in aerial imagery is a complex task due to the random rotations and high density of objects within the images. Additionally, boundary discontinuity from regression parametrization poses a major challenge in designing loss functions for rotated object detection. Several bounding box representations and their associated loss functions are being gradually developed to address this challenge. In this study, we compare loss functions for oriented bounding boxes including Probabilistic IoU (ProbIoU), KFIoU and others. The evaluation uses DOTA-PS, a subset of a large scale aerial benchmark dataset DOTAv1.0, which includes two categories, plane and ship. We also introduce Complete Rotated IoU (CRIoU) loss which is improved over Rotated IoU (RIoU) loss. ProbIoU and CRIoU loss achieved mAP score of 93.1% on the test set while RIoU and KFIoU losses got 92.8% and 90.6% respectively. CRIoU demonstrates marginal improvement in oriented bounding box localization, supported by the brief experimental and visual analysis.},
 bibtype = {article},
 author = {San, Kaung Htet and Kondo, Toshiaki and Marukatat, Sanparith and Hara-Azumi, Yuko},
 doi = {10.1109/JCSSE61278.2024.10613659},
 journal = {Proceedings - 21st International Joint Conference on Computer Science and Software Engineering, JCSSE 2024}
}

Detecting objects with specific orientations in aerial imagery is a complex task due to the random rotations and high density of objects within the images. Additionally, boundary discontinuity from regression parametrization poses a major challenge in designing loss functions for rotated object detection. Several bounding box representations and their associated loss functions are being gradually developed to address this challenge. In this study, we compare loss functions for oriented bounding boxes including Probabilistic IoU (ProbIoU), KFIoU and others. The evaluation uses DOTA-PS, a subset of a large scale aerial benchmark dataset DOTAv1.0, which includes two categories, plane and ship. We also introduce Complete Rotated IoU (CRIoU) loss which is improved over Rotated IoU (RIoU) loss. ProbIoU and CRIoU loss achieved mAP score of 93.1% on the test set while RIoU and KFIoU losses got 92.8% and 90.6% respectively. CRIoU demonstrates marginal improvement in oriented bounding box localization, supported by the brief experimental and visual analysis.

@inproceedings{
 title = {Attention-based mechanism for head-body fusion gaze estimation in dynamic scenes},
 type = {inproceedings},
 year = {2024},
 keywords = {Bi-LSTM,CBAM,feature fusion,gaze estimation},
 pages = {484-489},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {5ceb6585-5ca2-39ff-8649-c879a61d2e8d},
 created = {2024-11-06T13:03:59.513Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-07T07:33:10.444Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4cda9297-f98e-4246-88d6-ffeeade205c3},
 private_publication = {false},
 abstract = {The gaze of an individual can be understood as the information conveyed in the interaction between the individual and the surroundings. Computers face significant challenges in estimating the gaze direction of pedestrians in real-world scenarios. The presence of blurry images or occlusion of the tester's eyes makes the estimation particularly challenging. The majority of gaze estimation methodologies are contingent upon the availability of high-definition facial or ocular images. To address this issue, we propose a noval gaze estimation method that integrates head and torso features with the objective of achieving gaze estimation in authentic camera settings. In order to address the challenge of accurately extracting gaze features in the presence of background noise, we propose a CBAM-based feature extractor. Furthermore, we propose a linear regression method for multi-channel feature fusion using Bi-LSTM to address ambiguities due to changes in gaze time, enabling the estimation of the user's gaze direction. Our method accurately predicts the user's gaze direction, head deflection direction, body deflection direction, angular error and confidence by analysing seven consecutive frames of images.},
 bibtype = {inproceedings},
 author = {Zhang, Wulue and Xiong, Jianbin and Dong, Xiangjun and Wang, Qi and Quan, Guoyuan},
 doi = {10.1109/IoTAAI62601.2024.10692563},
 booktitle = {2024 6th International Conference on Internet of Things, Automation and Artificial Intelligence, IoTAAI 2024}
}

The gaze of an individual can be understood as the information conveyed in the interaction between the individual and the surroundings. Computers face significant challenges in estimating the gaze direction of pedestrians in real-world scenarios. The presence of blurry images or occlusion of the tester's eyes makes the estimation particularly challenging. The majority of gaze estimation methodologies are contingent upon the availability of high-definition facial or ocular images. To address this issue, we propose a noval gaze estimation method that integrates head and torso features with the objective of achieving gaze estimation in authentic camera settings. In order to address the challenge of accurately extracting gaze features in the presence of background noise, we propose a CBAM-based feature extractor. Furthermore, we propose a linear regression method for multi-channel feature fusion using Bi-LSTM to address ambiguities due to changes in gaze time, enabling the estimation of the user's gaze direction. Our method accurately predicts the user's gaze direction, head deflection direction, body deflection direction, angular error and confidence by analysing seven consecutive frames of images.

@article{
 title = {FRESH: Fusion-Based 3D Apple Recognition via Estimating Stem Direction Heading},
 type = {article},
 year = {2024},
 keywords = {3D detection,deep learning,smart agriculture,stem direction estimation},
 pages = {2161},
 volume = {14},
 websites = {https://www.mdpi.com/2077-0472/14/12/2161/htm,https://www.mdpi.com/2077-0472/14/12/2161},
 month = {11},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {27},
 id = {24edc8d4-59cf-39ce-8c74-196f396c2320},
 created = {2025-01-16T09:13:27.433Z},
 accessed = {2025-01-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2025-01-16T09:14:17.111Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {In 3D apple detection, the challenge of direction for apple stem harvesting for agricultural robotics has not yet been resolved. Addressing the issue of determining the stem direction of apples is essential for the harvesting processes employed by automated robots. This research proposes a 3D apple detection framework to identify stem direction. First, we constructed a dataset for 3D apple detection that considers the 3-axis rotation of apples based on stem direction. Secondly, we designed a 3D detection algorithm that not only recognizes the dimensions and location of apples, as existing methods do, but also predicts their 3-axis rotation. Furthermore, we effectively fused 3D point clouds with 2D images to leverage the geometric data from point clouds and the semantic information from images, enhancing the apple detection performance. Experimental results indicated that our method achieved AP@0.25 89.56% for 3D detection by considering apple rotation, surpassing the existing methods. Moreover, we experimentally validated that the proposed loss function most effectively estimated the rotation among the various approaches we explored. This study shows the effectiveness of 3D apple detection with consideration of rotation, emphasizing its potential for practical application in autonomous robotic systems.},
 bibtype = {article},
 author = {Son, Geonhwa and Lee, Seunghyeon and Choi, Yukyung},
 doi = {10.3390/AGRICULTURE14122161},
 journal = {Agriculture 2024, Vol. 14, Page 2161},
 number = {12}
}

In 3D apple detection, the challenge of direction for apple stem harvesting for agricultural robotics has not yet been resolved. Addressing the issue of determining the stem direction of apples is essential for the harvesting processes employed by automated robots. This research proposes a 3D apple detection framework to identify stem direction. First, we constructed a dataset for 3D apple detection that considers the 3-axis rotation of apples based on stem direction. Secondly, we designed a 3D detection algorithm that not only recognizes the dimensions and location of apples, as existing methods do, but also predicts their 3-axis rotation. Furthermore, we effectively fused 3D point clouds with 2D images to leverage the geometric data from point clouds and the semantic information from images, enhancing the apple detection performance. Experimental results indicated that our method achieved AP@0.25 89.56% for 3D detection by considering apple rotation, surpassing the existing methods. Moreover, we experimentally validated that the proposed loss function most effectively estimated the rotation among the various approaches we explored. This study shows the effectiveness of 3D apple detection with consideration of rotation, emphasizing its potential for practical application in autonomous robotic systems.

@article{
 title = {Mapping the spatial variability of Botrytis bunch rot risk in vineyards using UAV multispectral imagery},
 type = {article},
 year = {2023},
 pages = {126691},
 volume = {142},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {1e73afe1-1f7b-321d-88b5-ed8bec648865},
 created = {2022-12-12T06:49:27.644Z},
 accessed = {2022-12-12},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:53:04.946Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {article},
 author = {Vélez, Sergio and Ariza-Sentís, Mar and Valente, João},
 doi = {10.1016/J.EJA.2022.126691},
 journal = {European Journal of Agronomy},
 keywords = {velez2023mappingspatialvariability}
}

@article{
 title = {The Fast Detection of Crop Disease Leaves Based on Single-Channel Gravitational Kernel Density Clustering},
 type = {article},
 year = {2023},
 keywords = {clustering algorithm,color space,gravitational kernel density,image segmentation},
 pages = {1172},
 volume = {13},
 websites = {https://www.mdpi.com/2076-3417/13/2/1172/htm,https://www.mdpi.com/2076-3417/13/2/1172},
 month = {1},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {15},
 id = {47e6da0f-2ff4-3745-9e6d-cc2196fc1818},
 created = {2023-01-31T07:25:59.471Z},
 accessed = {2023-01-31},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-31T07:26:03.318Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Plant diseases and pests may seriously affect the yield of crops and even threaten the survival of human beings. The characteristics of plant diseases and insect pests are mainly reflected in the occurrence of lesions on crop leaves. Machine vision disease detection is of great significance for the early detection and prevention of plant diseases and insect pests. A fast detection method for lesions based on a single-channel gravitational kernel density clustering algorithm was designed to examine the complexity and ambiguity of diseased leaf images. Firstly, a polynomial was used to fit the R-channel feature histogram curve of a diseased leaf image in the RGB color space, and then the peak point and peak area of the fitted feature histogram curve were determined according to the derivative attribute. Secondly, the cluster numbers and the initial cluster center of the diseased leaf images were determined according to the peak area and peak point. Thirdly, according to the clustering center of the preliminarily determined diseased leaf images, the single-channel gravity kernel density clustering algorithm in this paper was used to achieve the rapid segmentation of the diseased leaf lesions. Finally, the experimental results showed that our method could segment the lesions quickly and accurately.},
 bibtype = {article},
 author = {Ren, Yifeng and Li, Qingyan and Liu, Zhe},
 doi = {10.3390/APP13021172},
 journal = {Applied Sciences 2023, Vol. 13, Page 1172},
 number = {2}
}

Plant diseases and pests may seriously affect the yield of crops and even threaten the survival of human beings. The characteristics of plant diseases and insect pests are mainly reflected in the occurrence of lesions on crop leaves. Machine vision disease detection is of great significance for the early detection and prevention of plant diseases and insect pests. A fast detection method for lesions based on a single-channel gravitational kernel density clustering algorithm was designed to examine the complexity and ambiguity of diseased leaf images. Firstly, a polynomial was used to fit the R-channel feature histogram curve of a diseased leaf image in the RGB color space, and then the peak point and peak area of the fitted feature histogram curve were determined according to the derivative attribute. Secondly, the cluster numbers and the initial cluster center of the diseased leaf images were determined according to the peak area and peak point. Thirdly, according to the clustering center of the preliminarily determined diseased leaf images, the single-channel gravity kernel density clustering algorithm in this paper was used to achieve the rapid segmentation of the diseased leaf lesions. Finally, the experimental results showed that our method could segment the lesions quickly and accurately.

@article{
 title = {Field-based robotic leaf angle detection and characterization of maize plants using stereo vision and deep convolutional neural networks},
 type = {article},
 year = {2023},
 keywords = {based plant phenotyping,convolutional neural network,field,keypoint detection,leaf angle,stereo vision},
 websites = {https://onlinelibrary.wiley.com/doi/full/10.1002/rob.22166,https://onlinelibrary.wiley.com/doi/abs/10.1002/rob.22166,https://onlinelibrary.wiley.com/doi/10.1002/rob.22166},
 month = {2},
 publisher = {John Wiley & Sons, Ltd},
 day = {27},
 id = {87c98ef2-9ed6-3a4f-be2d-e127d3f97548},
 created = {2023-03-09T08:51:03.766Z},
 accessed = {2023-03-09},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-03-09T08:52:59.386Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {238906f4-3e7c-4ebb-b571-5e94ea26a909,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Maize (Zea mays L.) is one of the three major cereal crops in the world. Leaf angle is an important architectural trait of crops due to its substantial role in light interception by the canopy and hence photosynthetic efficiency. Traditionally, leaf angle has been measured using a protractor, a process that is both slow and laborious. Efficiently measuring leaf angle under field conditions via imaging is challenging due to leaf density in the canopy and the resulting occlusions. However, advances in imaging technologies and machine learning have provided new tools for image acquisition and analysis that could be used to characterize leaf angle using three-dimensional (3D) models of field-grown plants. In this study, PhenoBot 3.0, a robotic vehicle designed to traverse between pairs of agronomically spaced rows of crops, was equipped with multiple tiers of PhenoStereo cameras to capture side-view images of maize plants in the field. PhenoStereo is a customized stereo camera module with integrated strobe lighting for high-speed stereoscopic image acquisition under variable outdoor lighting conditions. An automated image processing pipeline (AngleNet) was developed to measure leaf angles of nonoccluded leaves. In this pipeline, a novel representation form of leaf angle as a triplet of keypoints was proposed. The pipeline employs convolutional neural networks to detect each leaf angle in two-dimensional images and 3D modeling approaches to extract quantitative data from reconstructed models. Satisfactory accuracies in terms of correlation coefficient (r) and mean absolute error (MAE) were achieved for leaf angle (r MAE > 0.87, < 5°) and internode heights (r MAE > 0.99, < 3.5cm). Our study demonstrates the feasibility of using stereo vision to investigate the distribution of leaf angles in maize under field conditions. The proposed system is an efficient alternative to traditional leaf angle phenotyping and thus could accelerate breeding for improved plant architecture. K E Y W O R D S convolutional neural network, field-based plant phenotyping, keypoint detection, leaf angle, stereo vision J Field Robotics. 2023;1-20. wileyonlinelibrary.com/journal/rob | 1 This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.},
 bibtype = {article},
 author = {Xiang, Lirong and Gai, Jingyao and Bao, Yin and Yu, Jianming and Patrick, | and Schnable, S and Lie Tang, |},
 doi = {10.1002/ROB.22166},
 journal = {Journal of Field Robotics}
}

Maize (Zea mays L.) is one of the three major cereal crops in the world. Leaf angle is an important architectural trait of crops due to its substantial role in light interception by the canopy and hence photosynthetic efficiency. Traditionally, leaf angle has been measured using a protractor, a process that is both slow and laborious. Efficiently measuring leaf angle under field conditions via imaging is challenging due to leaf density in the canopy and the resulting occlusions. However, advances in imaging technologies and machine learning have provided new tools for image acquisition and analysis that could be used to characterize leaf angle using three-dimensional (3D) models of field-grown plants. In this study, PhenoBot 3.0, a robotic vehicle designed to traverse between pairs of agronomically spaced rows of crops, was equipped with multiple tiers of PhenoStereo cameras to capture side-view images of maize plants in the field. PhenoStereo is a customized stereo camera module with integrated strobe lighting for high-speed stereoscopic image acquisition under variable outdoor lighting conditions. An automated image processing pipeline (AngleNet) was developed to measure leaf angles of nonoccluded leaves. In this pipeline, a novel representation form of leaf angle as a triplet of keypoints was proposed. The pipeline employs convolutional neural networks to detect each leaf angle in two-dimensional images and 3D modeling approaches to extract quantitative data from reconstructed models. Satisfactory accuracies in terms of correlation coefficient (r) and mean absolute error (MAE) were achieved for leaf angle (r MAE > 0.87, < 5°) and internode heights (r MAE > 0.99, < 3.5cm). Our study demonstrates the feasibility of using stereo vision to investigate the distribution of leaf angles in maize under field conditions. The proposed system is an efficient alternative to traditional leaf angle phenotyping and thus could accelerate breeding for improved plant architecture. K E Y W O R D S convolutional neural network, field-based plant phenotyping, keypoint detection, leaf angle, stereo vision J Field Robotics. 2023;1-20. wileyonlinelibrary.com/journal/rob | 1 This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

@article{
 title = {Towards Computer-Vision Based Vineyard Navigation for Quadruped Robots},
 type = {article},
 year = {2023},
 keywords = {Computer-Vision,Index Terms-Agricultural Robotics,Quadruped Control,Vine-yard Navigation},
 websites = {https://arxiv.org/abs/2301.00887v1},
 month = {1},
 day = {2},
 id = {66219f83-86f9-3016-a39b-3c88a9521a8f},
 created = {2023-03-22T08:21:14.884Z},
 accessed = {2023-03-22},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-03-23T06:17:47.904Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4c7c81ce-f24b-44ae-bc2a-bf60600a3a24,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {There is a dramatic shortage of skilled labor for modern vineyards. The Vinum
project is developing a mobile robotic solution to autonomously navigate
through vineyards for winter grapevine pruning. This necessitates an autonomous
navigation stack for the robot pruning a vineyard. The Vinum project is using
the quadruped robot HyQReal. This paper introduces an architecture for a
quadruped robot to autonomously move through a vineyard by identifying and
approaching grapevines for pruning. The higher level control is a state machine
switching between searching for destination positions, autonomously navigating
towards those locations, and stopping for the robot to complete a task. The
destination points are determined by identifying grapevine trunks using
instance segmentation from a Mask Region-Based Convolutional Neural Network
(Mask-RCNN). These detections are sent through a filter to avoid redundancy and
remove noisy detections. The combination of these features is the basis for the
proposed architecture.},
 bibtype = {article},
 author = {Milburn, Lee and Gamba, Juan and Semini, Claudio},
 doi = {10.1016/j}
}

There is a dramatic shortage of skilled labor for modern vineyards. The Vinum project is developing a mobile robotic solution to autonomously navigate through vineyards for winter grapevine pruning. This necessitates an autonomous navigation stack for the robot pruning a vineyard. The Vinum project is using the quadruped robot HyQReal. This paper introduces an architecture for a quadruped robot to autonomously move through a vineyard by identifying and approaching grapevines for pruning. The higher level control is a state machine switching between searching for destination positions, autonomously navigating towards those locations, and stopping for the robot to complete a task. The destination points are determined by identifying grapevine trunks using instance segmentation from a Mask Region-Based Convolutional Neural Network (Mask-RCNN). These detections are sent through a filter to avoid redundancy and remove noisy detections. The combination of these features is the basis for the proposed architecture.

@article{
 title = {DualSeg: Fusing transformer and CNN structure for image segmentation in complex vineyard environment},
 type = {article},
 year = {2023},
 keywords = {CNN,DualSeg,FFM,Grape peduncle,Semantic segmentation,Transformer},
 pages = {107682},
 volume = {206},
 websites = {https://doi.org/10.1016/j.compag.2023.107682},
 publisher = {Elsevier B.V.},
 id = {abee0c52-1180-3bce-a363-ab99434df835},
 created = {2023-03-23T06:17:46.404Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-03-23T06:24:07.233Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {Semantic segmentation is a kind of classification at the pixel level, which has a remarkable effect in dealing with complex scenes. Most popular semantic segmentation models are based on Convolutional Neural Network (CNN) and Transformer structures. Convolution operations and attention modules are good at extracting local and global features. In this paper, we propose a parallel network structure, termed DualSeg, which leverages the advantages of CNN at local processing and Transformer at global interaction. And a feature fusion module (FFM) is designed to fuse parameters between modules in different branches to retain local and global features to the maximum extent. In the vineyards, the grape clusters and grape peduncles are often heavily obscured by leaves, branches and clusters, making it difficult to distinguish them accurately. Therefore, the scene is taken as an example to verify the segmentation effect of the model. In the experiment, the DualSeg model and mainstream segmentation models are used to test the segmentation effect of different models in this scene. And the experimental results showed that the DualSeg model had the best segmentation effect of all models. Specifically, the model had an IoU value of grape peduncle segmentation of 72.1%, which was more than 3.9% higher than that of other models in the case of 80 K iterations. In addition, the mIoU value of this model is 83.7%, which was the highest value compared with other models. The demonstrated performance of DualSeg model indicated that harvesting robots could be use it.},
 bibtype = {article},
 author = {Wang, Jinhai and Zhang, Zongyin and Luo, Lufeng and Wei, Huiling and Wang, Wei and Chen, Mingyou and Luo, Shaoming},
 doi = {10.1016/j.compag.2023.107682},
 journal = {Computers and Electronics in Agriculture},
 number = {January}
}

Semantic segmentation is a kind of classification at the pixel level, which has a remarkable effect in dealing with complex scenes. Most popular semantic segmentation models are based on Convolutional Neural Network (CNN) and Transformer structures. Convolution operations and attention modules are good at extracting local and global features. In this paper, we propose a parallel network structure, termed DualSeg, which leverages the advantages of CNN at local processing and Transformer at global interaction. And a feature fusion module (FFM) is designed to fuse parameters between modules in different branches to retain local and global features to the maximum extent. In the vineyards, the grape clusters and grape peduncles are often heavily obscured by leaves, branches and clusters, making it difficult to distinguish them accurately. Therefore, the scene is taken as an example to verify the segmentation effect of the model. In the experiment, the DualSeg model and mainstream segmentation models are used to test the segmentation effect of different models in this scene. And the experimental results showed that the DualSeg model had the best segmentation effect of all models. Specifically, the model had an IoU value of grape peduncle segmentation of 72.1%, which was more than 3.9% higher than that of other models in the case of 80 K iterations. In addition, the mIoU value of this model is 83.7%, which was the highest value compared with other models. The demonstrated performance of DualSeg model indicated that harvesting robots could be use it.

@article{
 title = {Designing a Proximal Sensing Camera Acquisition System for Vineyard Applications: Results and Feedback on 8 Years of Experiments},
 type = {article},
 year = {2023},
 keywords = {rancon2023designingproximalsensing},
 pages = {847},
 volume = {23},
 websites = {https://www.mdpi.com/1424-8220/23/2/847/htm,https://www.mdpi.com/1424-8220/23/2/847},
 month = {1},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {11},
 id = {eabbddde-c9a9-37da-ac49-4f47f2e73a56},
 created = {2023-03-23T06:21:22.030Z},
 accessed = {2023-03-23},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T11:30:22.566Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The potential of image proximal sensing for agricultural applications has been a prolific scientific subject in the recent literature. Its main appeal lies in the sensing of precise information about plant status, which is either harder or impossible to extract from lower-resolution downward-looking image sensors such as satellite or drone imagery. Yet, many theoretical and practical problems arise when dealing with proximal sensing, especially on perennial crops such as vineyards. Indeed, vineyards exhibit challenging physical obstacles and many degrees of variability in their layout. In this paper, we present the design of a mobile camera suited to vineyards and harsh experimental conditions, as well as the results and assessments of 8 years’ worth of studies using that camera. These projects ranged from in-field yield estimation (berry counting) to disease detection, providing new insights on typical viticulture problems that could also be generalized to orchard crops. Different recommendations are then provided using small case studies, such as the difficulties related to framing plots with different structures or the mounting of the sensor on a moving vehicle. While results stress the obvious importance and strong benefits of a thorough experimental design, they also indicate some inescapable pitfalls, illustrating the need for more robust image analysis algorithms and better databases. We believe sharing that experience with the scientific community can only benefit the future development of these innovative approaches.},
 bibtype = {article},
 author = {Rançon, Florian and Keresztes, Barna and Deshayes, Aymeric and Tardif, Malo and Abdelghafour, Florent and Fontaine, Gael and Da Costa, Jean Pierre and Germain, Christian},
 doi = {10.3390/S23020847},
 journal = {Sensors 2023, Vol. 23, Page 847},
 number = {2}
}

The potential of image proximal sensing for agricultural applications has been a prolific scientific subject in the recent literature. Its main appeal lies in the sensing of precise information about plant status, which is either harder or impossible to extract from lower-resolution downward-looking image sensors such as satellite or drone imagery. Yet, many theoretical and practical problems arise when dealing with proximal sensing, especially on perennial crops such as vineyards. Indeed, vineyards exhibit challenging physical obstacles and many degrees of variability in their layout. In this paper, we present the design of a mobile camera suited to vineyards and harsh experimental conditions, as well as the results and assessments of 8 years’ worth of studies using that camera. These projects ranged from in-field yield estimation (berry counting) to disease detection, providing new insights on typical viticulture problems that could also be generalized to orchard crops. Different recommendations are then provided using small case studies, such as the difficulties related to framing plots with different structures or the mounting of the sensor on a moving vehicle. While results stress the obvious importance and strong benefits of a thorough experimental design, they also indicate some inescapable pitfalls, illustrating the need for more robust image analysis algorithms and better databases. We believe sharing that experience with the scientific community can only benefit the future development of these innovative approaches.

@article{
 title = {DualSeg: Fusing transformer and CNN structure for image segmentation in complex vineyard environment},
 type = {article},
 year = {2023},
 keywords = {CNN,DualSeg,FFM,Grape peduncle,Semantic segmentation,Transformer},
 pages = {107682},
 volume = {206},
 websites = {https://doi.org/10.1016/j.compag.2023.107682},
 publisher = {Elsevier B.V.},
 id = {c4dd463c-82de-3c58-8e9c-74690bb11043},
 created = {2023-03-23T06:24:06.347Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-03-23T06:24:16.472Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Semantic segmentation is a kind of classification at the pixel level, which has a remarkable effect in dealing with complex scenes. Most popular semantic segmentation models are based on Convolutional Neural Network (CNN) and Transformer structures. Convolution operations and attention modules are good at extracting local and global features. In this paper, we propose a parallel network structure, termed DualSeg, which leverages the advantages of CNN at local processing and Transformer at global interaction. And a feature fusion module (FFM) is designed to fuse parameters between modules in different branches to retain local and global features to the maximum extent. In the vineyards, the grape clusters and grape peduncles are often heavily obscured by leaves, branches and clusters, making it difficult to distinguish them accurately. Therefore, the scene is taken as an example to verify the segmentation effect of the model. In the experiment, the DualSeg model and mainstream segmentation models are used to test the segmentation effect of different models in this scene. And the experimental results showed that the DualSeg model had the best segmentation effect of all models. Specifically, the model had an IoU value of grape peduncle segmentation of 72.1%, which was more than 3.9% higher than that of other models in the case of 80 K iterations. In addition, the mIoU value of this model is 83.7%, which was the highest value compared with other models. The demonstrated performance of DualSeg model indicated that harvesting robots could be use it.},
 bibtype = {article},
 author = {Wang, Jinhai and Zhang, Zongyin and Luo, Lufeng and Wei, Huiling and Wang, Wei and Chen, Mingyou and Luo, Shaoming},
 doi = {10.1016/j.compag.2023.107682},
 journal = {Computers and Electronics in Agriculture},
 number = {September 2022}
}

Semantic segmentation is a kind of classification at the pixel level, which has a remarkable effect in dealing with complex scenes. Most popular semantic segmentation models are based on Convolutional Neural Network (CNN) and Transformer structures. Convolution operations and attention modules are good at extracting local and global features. In this paper, we propose a parallel network structure, termed DualSeg, which leverages the advantages of CNN at local processing and Transformer at global interaction. And a feature fusion module (FFM) is designed to fuse parameters between modules in different branches to retain local and global features to the maximum extent. In the vineyards, the grape clusters and grape peduncles are often heavily obscured by leaves, branches and clusters, making it difficult to distinguish them accurately. Therefore, the scene is taken as an example to verify the segmentation effect of the model. In the experiment, the DualSeg model and mainstream segmentation models are used to test the segmentation effect of different models in this scene. And the experimental results showed that the DualSeg model had the best segmentation effect of all models. Specifically, the model had an IoU value of grape peduncle segmentation of 72.1%, which was more than 3.9% higher than that of other models in the case of 80 K iterations. In addition, the mIoU value of this model is 83.7%, which was the highest value compared with other models. The demonstrated performance of DualSeg model indicated that harvesting robots could be use it.

@article{
 title = {Machine-Learning Methods for the Identification of Key Predictors of Site-Specific Vineyard Yield and Vine Size},
 type = {article},
 year = {2023},
 pages = {ajev.2022.22050},
 websites = {https://www.ajevonline.org/content/early/2023/01/11/ajev.2022.22050,https://www.ajevonline.org/content/early/2023/01/11/ajev.2022.22050.abstract},
 month = {1},
 publisher = {American Journal of Enology and Viticulture},
 day = {17},
 id = {1f625568-08df-3a18-baa3-201381871406},
 created = {2023-03-23T07:14:15.118Z},
 accessed = {2023-03-23},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-03-23T09:28:40.666Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Background and goals Lake Erie Concord growers have access to high-resolution spatial soil and production data but lack protocols and information on the optimum time to collect these data. This study intends to provide clearer information regarding the type and timing of sensor information to support in-season management.

Methods and key findings A three-year study in a 2.6 ha vineyard collected yield, pruning mass, canopy vigor and soil data, including yield and pruning mass from the previous year, at 321 sites. Stepwise linear regression and random forest regression approaches were used to model site-specific yield and pruning mass using spatial historical production data, multi-temporal in-season canopy vigor and soil data. The more complex yield elaboration process was best modelled with non-linear random forest regression while the simpler development of pruning mass was best modelled by linear regression.

Conclusions and significance Canopy vigor in the weeks preceding bloom was the most important predictor of the current season’s yield and should be used to generate stratified sampling designs for crop estimation at 30 days after bloom. In contrast, pruning mass was not well predicted by canopy vigor, even late-season canopy vigor, which is widely advocated for pruning mass estimation in viticulture. The previous year’s pruning mass was the dominant predictor of pruning mass in the current season. To model pruning mass going forward, the best approach is to start measuring it. Further work is still needed to develop robust, local site-specific yield and pruning mass models for operational decision-making in concord vineyards.},
 bibtype = {article},
 author = {Taylor, James A. and Bates, Terence R. and Jakubowski, Rhiann and Jones, Hazaël},
 doi = {10.5344/AJEV.2022.22050},
 journal = {American Journal of Enology and Viticulture}
}

Background and goals Lake Erie Concord growers have access to high-resolution spatial soil and production data but lack protocols and information on the optimum time to collect these data. This study intends to provide clearer information regarding the type and timing of sensor information to support in-season management. Methods and key findings A three-year study in a 2.6 ha vineyard collected yield, pruning mass, canopy vigor and soil data, including yield and pruning mass from the previous year, at 321 sites. Stepwise linear regression and random forest regression approaches were used to model site-specific yield and pruning mass using spatial historical production data, multi-temporal in-season canopy vigor and soil data. The more complex yield elaboration process was best modelled with non-linear random forest regression while the simpler development of pruning mass was best modelled by linear regression. Conclusions and significance Canopy vigor in the weeks preceding bloom was the most important predictor of the current season’s yield and should be used to generate stratified sampling designs for crop estimation at 30 days after bloom. In contrast, pruning mass was not well predicted by canopy vigor, even late-season canopy vigor, which is widely advocated for pruning mass estimation in viticulture. The previous year’s pruning mass was the dominant predictor of pruning mass in the current season. To model pruning mass going forward, the best approach is to start measuring it. Further work is still needed to develop robust, local site-specific yield and pruning mass models for operational decision-making in concord vineyards.

@article{
 title = {Leaf area index estimation of pergola-trained vineyards in arid regions using classical and deep learning methods based on UAV-based RGB images},
 type = {article},
 year = {2023},
 keywords = {CNN,Data augmentation,Leaf area index,Machine learning,Spectral features,Textural features,UAV},
 pages = {107723},
 volume = {207},
 websites = {https://doi.org/10.1016/j.compag.2023.107723},
 publisher = {Elsevier B.V.},
 id = {273167f1-1ed5-37ca-b3c1-5273b3c459ff},
 created = {2023-03-23T09:28:39.759Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-03-23T09:28:49.839Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Timely and accurate mapping of leaf area index (LAI) in vineyards plays an important role for management choices in precision agricultural practices. However, only a little work has been done to extract the LAI of pergola-trained vineyards using higher spatial resolution remote sensing data. The main objective of this study was to evaluate the ability of unmanned aerial vehicle (UAV) imageries to estimate the LAI of pergola-trained vineyards using shallow and deep machine learning (ML) methods. Field trials were conducted in different growth seasons in 2021 by collecting 465 LAI samples. Firstly, this study trained five classical shallow ML models and an ensemble learning model by using different spectral and textural indices calculated from UAV imageries, and the most correlated or useful features for LAI estimations in different growth stages were differentiated. Then, due to the classical ML approaches need the arduous computation of multiple indices and feature selection procedures, another ResNet-based convolutional neural network (CNN) model was constructed which can be directly fed by cropped images. Furthermore, this study introduced a new image data augmentation method which is applicable to regression problems. Results indicated that the textural indices performed better than spectral indices, while the combination of them can improve estimation results, and the ensemble learning method showed the best among classical ML models. By choosing the optimal input image size, the CNN model we constructed estimated the LAI most effectively without extracting and selecting the features manually. The proposed image data augmentation method can generate new training images with new labels by mosaicking the original ones, and the CNN model showed improved performance after using this method compared to those using only the original images, or augmented by rotation and flipping methods. This data augmentation method can be applied to other regression models to extract growth parameters of crops using remote sensing data, and we conclude that the UAV imagery and deep learning methods are promising in LAI estimations of pergola-trained vineyards.},
 bibtype = {article},
 author = {Ilniyaz, Osman and Du, Qingyun and Shen, Huanfeng and He, Wenwen and Feng, Luwei and Azadi, Hossein and Kurban, Alishir and Chen, Xi},
 doi = {10.1016/j.compag.2023.107723},
 journal = {Computers and Electronics in Agriculture},
 number = {February}
}

Timely and accurate mapping of leaf area index (LAI) in vineyards plays an important role for management choices in precision agricultural practices. However, only a little work has been done to extract the LAI of pergola-trained vineyards using higher spatial resolution remote sensing data. The main objective of this study was to evaluate the ability of unmanned aerial vehicle (UAV) imageries to estimate the LAI of pergola-trained vineyards using shallow and deep machine learning (ML) methods. Field trials were conducted in different growth seasons in 2021 by collecting 465 LAI samples. Firstly, this study trained five classical shallow ML models and an ensemble learning model by using different spectral and textural indices calculated from UAV imageries, and the most correlated or useful features for LAI estimations in different growth stages were differentiated. Then, due to the classical ML approaches need the arduous computation of multiple indices and feature selection procedures, another ResNet-based convolutional neural network (CNN) model was constructed which can be directly fed by cropped images. Furthermore, this study introduced a new image data augmentation method which is applicable to regression problems. Results indicated that the textural indices performed better than spectral indices, while the combination of them can improve estimation results, and the ensemble learning method showed the best among classical ML models. By choosing the optimal input image size, the CNN model we constructed estimated the LAI most effectively without extracting and selecting the features manually. The proposed image data augmentation method can generate new training images with new labels by mosaicking the original ones, and the CNN model showed improved performance after using this method compared to those using only the original images, or augmented by rotation and flipping methods. This data augmentation method can be applied to other regression models to extract growth parameters of crops using remote sensing data, and we conclude that the UAV imagery and deep learning methods are promising in LAI estimations of pergola-trained vineyards.

@article{
 title = {Using deep learning for pruning region detection and plant organ segmentation in dormant spur-pruned grapevines},
 type = {article},
 year = {2023},
 keywords = {Agriculture,Atmospheric Sciences,Chemistry and Earth Sciences,Computer Science,Physics,Remote Sensing/Photogrammetry,Soil Science & Conservation,Statistics for Engineering},
 pages = {1-23},
 websites = {https://link.springer.com/article/10.1007/s11119-023-10006-y},
 month = {3},
 publisher = {Springer},
 day = {22},
 id = {09fb60d8-e775-3eb7-94eb-d244ea79398c},
 created = {2023-03-28T05:58:44.167Z},
 accessed = {2023-03-28},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-03-29T11:22:31.558Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Even though mechanization has dramatically decreased labor requirements, vineyard management costs are still affected by selective operations such as winter pruning. Robotic solutions are becoming more common in agriculture, however, few studies have focused on grapevines. This work aims at fine-tuning and testing two different deep neural networks for: (i) detecting pruning regions (PRs), and (ii) performing organ segmentation of spur-pruned dormant grapevines. The Faster R-CNN network was fine-tuned using 1215 RGB images collected in different vineyards and annotated through bounding boxes. The network was tested on 232 RGB images, PRs were categorized by wood type (W), orientation (Or) and visibility (V), and performance metrics were calculated. PR detection was dramatically affected by visibility. Highest detection was associated with visible intermediate complex spurs in Merlot (0.97), while most represented coplanar simple spurs allowed a 74% detection rate. The Mask R-CNN network was trained for grapevine organs (GOs) segmentation by using 119 RGB images annotated by distinguishing 5 classes (cordon, arm, spur, cane and node). The network was tested on 60 RGB images of light pruned (LP), shoot-thinned (ST) and unthinned control (C) grapevines. Nodes were the best segmented GOs (0.88) and general recall was higher for ST (0.85) compared to C (0.80) confirming the role of canopy management in improving performances of hi-tech solutions based on artificial intelligence. The two fine-tuned and tested networks are part of a larger control framework that is under development for autonomous winter pruning of grapevines.},
 bibtype = {article},
 author = {Guadagna, P and Fernandes, · M and Chen, · F and Santamaria, · A and Teng, · T and Frioni, · T and Caldwell, · D G and Poni, · S and Semini, · C and Gatti, · M and Gatti, M},
 doi = {10.1007/S11119-023-10006-Y},
 journal = {Precision Agriculture 2023}
}

Even though mechanization has dramatically decreased labor requirements, vineyard management costs are still affected by selective operations such as winter pruning. Robotic solutions are becoming more common in agriculture, however, few studies have focused on grapevines. This work aims at fine-tuning and testing two different deep neural networks for: (i) detecting pruning regions (PRs), and (ii) performing organ segmentation of spur-pruned dormant grapevines. The Faster R-CNN network was fine-tuned using 1215 RGB images collected in different vineyards and annotated through bounding boxes. The network was tested on 232 RGB images, PRs were categorized by wood type (W), orientation (Or) and visibility (V), and performance metrics were calculated. PR detection was dramatically affected by visibility. Highest detection was associated with visible intermediate complex spurs in Merlot (0.97), while most represented coplanar simple spurs allowed a 74% detection rate. The Mask R-CNN network was trained for grapevine organs (GOs) segmentation by using 119 RGB images annotated by distinguishing 5 classes (cordon, arm, spur, cane and node). The network was tested on 60 RGB images of light pruned (LP), shoot-thinned (ST) and unthinned control (C) grapevines. Nodes were the best segmented GOs (0.88) and general recall was higher for ST (0.85) compared to C (0.80) confirming the role of canopy management in improving performances of hi-tech solutions based on artificial intelligence. The two fine-tuned and tested networks are part of a larger control framework that is under development for autonomous winter pruning of grapevines.

@article{
 title = {Early yield prediction in different grapevine varieties using computer vision and machine learning},
 type = {article},
 year = {2023},
 keywords = {Digital viticulture,Non-invasive sensing technologies,SegNet architecture,Yield estimation},
 pages = {407-435},
 volume = {24},
 websites = {https://link.springer.com/article/10.1007/s11119-022-09950-y},
 month = {4},
 publisher = {Springer},
 day = {1},
 id = {52f774ad-e2fc-3b85-ae57-bdbce8ce989e},
 created = {2023-04-25T05:48:43.854Z},
 accessed = {2023-04-25},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-27T05:58:46.405Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Yield assessment is a highly relevant task for the wine industry. The goal of this work was to develop a new algorithm for early yield prediction in different grapevine varieties using computer vision and machine learning. Vines from six grapevine (Vitis vinifera L.) varieties were photographed using a mobile platform in a commercial vineyard at pea-size berry stage. A SegNet architecture was employed to detect the visible berries and canopy features. All features were used to train support vector regression (SVR) models for predicting number of actual berries and yield. Regarding the berries’ detection step, a F1-score average of 0.72 and coefficients of determination (R2) above 0.92 were achieved for all varieties between the number of estimated and the number of actual visible berries. The method yielded average values for root mean squared error (RMSE) of 195 berries, normalized RMSE (NRMSE) of 23.83% and R2 of 0.79 between the number of estimated and the number of actual berries per vine using the leave-one-out cross validation method. In terms of yield forecast, the correlation between the actual yield and its estimated value yielded R2 between 0.54 and 0.87 among different varieties and NRMSE between 16.47% and 39.17% while the global model (including all varieties) had a R2 equal to 0.83 and NRMSE of 29.77%. The number of actual berries and yield per vine can be predicted up to 60 days prior to harvest in several grapevine varieties using the new algorithm.},
 bibtype = {article},
 author = {Palacios, Fernando and Diago, Maria P. and Melo-Pinto, Pedro and Tardaguila, Javier},
 doi = {10.1007/S11119-022-09950-Y/FIGURES/7},
 journal = {Precision Agriculture},
 number = {2}
}

Yield assessment is a highly relevant task for the wine industry. The goal of this work was to develop a new algorithm for early yield prediction in different grapevine varieties using computer vision and machine learning. Vines from six grapevine (Vitis vinifera L.) varieties were photographed using a mobile platform in a commercial vineyard at pea-size berry stage. A SegNet architecture was employed to detect the visible berries and canopy features. All features were used to train support vector regression (SVR) models for predicting number of actual berries and yield. Regarding the berries’ detection step, a F1-score average of 0.72 and coefficients of determination (R2) above 0.92 were achieved for all varieties between the number of estimated and the number of actual visible berries. The method yielded average values for root mean squared error (RMSE) of 195 berries, normalized RMSE (NRMSE) of 23.83% and R2 of 0.79 between the number of estimated and the number of actual berries per vine using the leave-one-out cross validation method. In terms of yield forecast, the correlation between the actual yield and its estimated value yielded R2 between 0.54 and 0.87 among different varieties and NRMSE between 16.47% and 39.17% while the global model (including all varieties) had a R2 equal to 0.83 and NRMSE of 29.77%. The number of actual berries and yield per vine can be predicted up to 60 days prior to harvest in several grapevine varieties using the new algorithm.

@article{
 title = {Grape Cs-Ml Database-Informed Methods for Contemporary Vineyard Management},
 type = {article},
 year = {2023},
 keywords = {computer vision,image processing,machine,machine vision,visual computing,viticulture},
 pages = {121-126},
 id = {48f58971-3d7c-371e-8a09-6376f28c0b90},
 created = {2023-04-25T05:51:54.254Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-27T05:58:46.404Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {article},
 author = {Yakubreddy, Kancharakunt and Vellela, Sai Srinivas and Sk, Khader Basha and B, Venkateswara Reddy and Roja, D},
 doi = {10.56726/irjmets34069},
 journal = {International Research Journal of Modernization in Engineering Technology and Science},
 number = {03}
}

@article{
 title = {GrapesNet: Indian RGB & RGB-D vineyard image datasets for deep learning applications},
 type = {article},
 year = {2023},
 keywords = {Artificial intelligence,Deep learning,Grape bunch detection etc,Grape bunch segmentation,Vineyard dataset},
 pages = {109100},
 volume = {48},
 websites = {https://doi.org/10.1016/j.dib.2023.109100},
 publisher = {Elsevier Inc.},
 id = {976568e4-4810-31b2-ae89-98f389feb052},
 created = {2023-04-27T05:58:46.129Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-09-18T10:51:03.364Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {In most of the countries, grapes are considered as a cash crop. Currently huge research is going on in development of automated grape harvesting systems. Speedy and reliable grape bunch detection is prime need for various deep learning based automated systems which deals with object detection and object segmentation tasks. But currently very few datasets are available on grape bunches in vineyard, because of which there is restriction to the research in this area. In comparison to the vineyard in outside countries, Indian vineyard structure is more complex, so it becomes hard to work in real-time. To overcome these problems and to make vineyard dataset for suitable for Indian vineyard scenarios, this paper proposed four different datasets on grape bunches in vineyard. For creating all datasets in GrapesNet, natural environmental conditions have been considered. GrapesNet includes total 11000+ images of grape bunches. Necessary data for weight prediction of grape cluster is also provided with dataset like height, width and real weight of cluster present in image. Proposed datasets can be used for prime tasks like grape bunch detection, grape bunch segmentation, and grape bunch weight estimation etc. of future generation automated vineyard harvesting technologies.},
 bibtype = {article},
 author = {Barbole, Dhanashree K. and Jadhav, Parul M.},
 doi = {10.1016/j.dib.2023.109100},
 journal = {Data in Brief}
}

In most of the countries, grapes are considered as a cash crop. Currently huge research is going on in development of automated grape harvesting systems. Speedy and reliable grape bunch detection is prime need for various deep learning based automated systems which deals with object detection and object segmentation tasks. But currently very few datasets are available on grape bunches in vineyard, because of which there is restriction to the research in this area. In comparison to the vineyard in outside countries, Indian vineyard structure is more complex, so it becomes hard to work in real-time. To overcome these problems and to make vineyard dataset for suitable for Indian vineyard scenarios, this paper proposed four different datasets on grape bunches in vineyard. For creating all datasets in GrapesNet, natural environmental conditions have been considered. GrapesNet includes total 11000+ images of grape bunches. Necessary data for weight prediction of grape cluster is also provided with dataset like height, width and real weight of cluster present in image. Proposed datasets can be used for prime tasks like grape bunch detection, grape bunch segmentation, and grape bunch weight estimation etc. of future generation automated vineyard harvesting technologies.

@article{
 title = {Dataset on unmanned aerial vehicle multispectral images acquired over a vineyard affected by Botrytis cinerea in northern Spain},
 type = {article},
 year = {2023},
 keywords = {velez2023datasetuav},
 volume = {46},
 websites = {https://pubmed.ncbi.nlm.nih.gov/36660442/},
 month = {2},
 publisher = {Data Brief},
 day = {1},
 id = {745d3673-1aff-3790-a316-54c1d3ddb869},
 created = {2023-05-12T09:35:50.539Z},
 accessed = {2023-05-12},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:53:05.154Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,5a010301-acb6-4642-a6b2-8afaee1b741c,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Remote sensing makes it possible to gather data rapidly, precisely, accurately, and non-destructively, allowing it to assess grapevines accurately in near real-time. In addition, multispectral cameras capture information in different bands, which can be combined to generate vegetation indices useful in precision agriculture. This dataset contains 16,504 multispectral images from a 1.06 ha vineyard affected by Botrytis cinerea, in the north of Spain. The photos were taken throughout four UAV flights at 30 m height with varying camera angles on 16 September 2021, the same date as the grape harvest. The first flight took place with the camera tilted at 0° (nadir angle), the second flight at 30°, the third flight at 45°, and the fourth flight was also performed at 0° but was scheduled in the afternoon to capture the shadows of the plants projected on the ground. This dataset was created to support researchers interested in disease detection and, in general, UAV remote sensing in vineyards and other woody crops. Moreover, it allows digital photogrammetry and 3D reconstruction in the context of precision agriculture, enabling the study of the effect of different tilt angles on the 3D reconstruction of the vineyard and the generation of orthomosaics.},
 bibtype = {article},
 author = {Vélez, Sergio and Ariza-Sentís, Mar and Valente, João},
 doi = {10.1016/J.DIB.2022.108876},
 journal = {Data in brief}
}

Remote sensing makes it possible to gather data rapidly, precisely, accurately, and non-destructively, allowing it to assess grapevines accurately in near real-time. In addition, multispectral cameras capture information in different bands, which can be combined to generate vegetation indices useful in precision agriculture. This dataset contains 16,504 multispectral images from a 1.06 ha vineyard affected by Botrytis cinerea, in the north of Spain. The photos were taken throughout four UAV flights at 30 m height with varying camera angles on 16 September 2021, the same date as the grape harvest. The first flight took place with the camera tilted at 0° (nadir angle), the second flight at 30°, the third flight at 45°, and the fourth flight was also performed at 0° but was scheduled in the afternoon to capture the shadows of the plants projected on the ground. This dataset was created to support researchers interested in disease detection and, in general, UAV remote sensing in vineyards and other woody crops. Moreover, it allows digital photogrammetry and 3D reconstruction in the context of precision agriculture, enabling the study of the effect of different tilt angles on the 3D reconstruction of the vineyard and the generation of orthomosaics.

@article{
 title = {Fields2Cover: An Open-Source Coverage Path Planning Library for Unmanned Agricultural Vehicles},
 type = {article},
 year = {2023},
 keywords = {Agricultural automation,field robots,software architecture for robotic and automation},
 pages = {2166-2172},
 volume = {8},
 publisher = {IEEE},
 id = {d88d6fbd-a5ca-3b6b-84d8-933affd4d847},
 created = {2023-05-17T08:05:02.769Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-05-17T08:05:02.769Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {This letter describes Fields2Cover, a novel open source library for coverage path planning (CPP) for agricultural vehicles. While there are several CPP solutions nowadays, there have been limited efforts to unify them into an open source library and provide benchmarking tools to compare their performance. Fields2Cover provides a framework for planning coverage paths, developing novel techniques, and benchmarking state-of-the-art algorithms. The library features a modular and extensible architecture that supports various vehicles and can be used for a variety of applications, including farms. Its core modules are: a headland generator, a swath generator, a route planner and a path planner. An interface to the Robot Operating System (ROS) is also supplied as an add-on. In this letter, the functionalities of the library for planning a coverage path in agriculture are demonstrated using 8 state-of-the-art methods and 7 objective functions in simulation and field experiments.},
 bibtype = {article},
 author = {Mier, Gonzalo and Valente, Joao and De Bruin, Sytze},
 doi = {10.1109/LRA.2023.3248439},
 journal = {IEEE Robotics and Automation Letters},
 number = {4}
}

This letter describes Fields2Cover, a novel open source library for coverage path planning (CPP) for agricultural vehicles. While there are several CPP solutions nowadays, there have been limited efforts to unify them into an open source library and provide benchmarking tools to compare their performance. Fields2Cover provides a framework for planning coverage paths, developing novel techniques, and benchmarking state-of-the-art algorithms. The library features a modular and extensible architecture that supports various vehicles and can be used for a variety of applications, including farms. Its core modules are: a headland generator, a swath generator, a route planner and a path planner. An interface to the Robot Operating System (ROS) is also supplied as an add-on. In this letter, the functionalities of the library for planning a coverage path in agriculture are demonstrated using 8 state-of-the-art methods and 7 objective functions in simulation and field experiments.

@article{
 title = {An expertized grapevine disease image database including five grape varieties focused on Flavescence dorée and its confounding diseases, biotic and abiotic stresses},
 type = {article},
 year = {2023},
 keywords = {tardif2023expertizedgrapevinedisease},
 pages = {109230},
 volume = {48},
 month = {6},
 publisher = {Elsevier},
 day = {1},
 id = {a41d6550-5ce6-34ee-af7c-efdf2fc326d0},
 created = {2023-09-15T11:03:55.635Z},
 accessed = {2023-09-15},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:42:33.862Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0},
 private_publication = {false},
 abstract = {The grapevine is vulnerable to diseases, deficiencies, and pests, leading to significant yield losses. Current disease controls involve monitoring and spraying phytosanitary products at the vineyard block scale. However, automatic detection of disease symptoms could reduce the use of these products and treat diseases before they spread. Flavescence dorée (FD), a highly infectious disease that causes significant yield losses, is only diagnosed by identifying symptoms on three grapevine organs: leaf, shoot, and bunch. Its diagnosis is carried out by scouting experts, as many other diseases and stresses, either biotic or abiotic, imply similar symptoms (but not all at the same time). These experts need a decision support tool to improve their scouting efficiency. To address this, a dataset of 1483 RGB images of grapevines affected by various diseases and stresses, including FD, was acquired by proximal sensing. The images were taken in the field at a distance of 1-2 meters to capture entire grapevines and an industrial flash was ensuring a constant luminance on the images regardless of the environmental circumstances. Images of 5 grape varieties (Cabernet sauvignon, Cabernet franc, Merlot, Ugni blanc and Sauvignon blanc) were acquired during 2 years (2020 and 2021). Two types of annotations were made: expert diagnosis at the grapevine scale in the field and symptom annotations at the leaf, shoot, and bunch levels on computer. On 744 images, the leaves were annotated and divided into three classes: ‘FD symptomatic leaves’, ‘Esca symptomatic leaves’, and ‘Confounding leaves’. Symptomatic bunches and shoots were, in addition of leaves, annotated on 110 images using bounding boxes and broken lines, respectively. Additionally, 128 segmentation masks were created to allow the detection of the symptomatic shoots and bunches by segmentation algorithms and compare the results to those of the detection algorithms.},
 bibtype = {article},
 author = {Tardif, Malo and Amri, Ahmed and Deshayes, Aymeric and Greven, Marc and Keresztes, Barna and Fontaine, Gaël and Sicaud, Laetitia and Paulhac, Laetitia and Bentejac, Sophie and Da Costa, Jean Pierre},
 doi = {10.1016/J.DIB.2023.109230},
 journal = {Data in Brief}
}

The grapevine is vulnerable to diseases, deficiencies, and pests, leading to significant yield losses. Current disease controls involve monitoring and spraying phytosanitary products at the vineyard block scale. However, automatic detection of disease symptoms could reduce the use of these products and treat diseases before they spread. Flavescence dorée (FD), a highly infectious disease that causes significant yield losses, is only diagnosed by identifying symptoms on three grapevine organs: leaf, shoot, and bunch. Its diagnosis is carried out by scouting experts, as many other diseases and stresses, either biotic or abiotic, imply similar symptoms (but not all at the same time). These experts need a decision support tool to improve their scouting efficiency. To address this, a dataset of 1483 RGB images of grapevines affected by various diseases and stresses, including FD, was acquired by proximal sensing. The images were taken in the field at a distance of 1-2 meters to capture entire grapevines and an industrial flash was ensuring a constant luminance on the images regardless of the environmental circumstances. Images of 5 grape varieties (Cabernet sauvignon, Cabernet franc, Merlot, Ugni blanc and Sauvignon blanc) were acquired during 2 years (2020 and 2021). Two types of annotations were made: expert diagnosis at the grapevine scale in the field and symptom annotations at the leaf, shoot, and bunch levels on computer. On 744 images, the leaves were annotated and divided into three classes: ‘FD symptomatic leaves’, ‘Esca symptomatic leaves’, and ‘Confounding leaves’. Symptomatic bunches and shoots were, in addition of leaves, annotated on 110 images using bounding boxes and broken lines, respectively. Additionally, 128 segmentation masks were created to allow the detection of the symptomatic shoots and bunches by segmentation algorithms and compare the results to those of the detection algorithms.

@article{
 title = {AI-based Maize and Weeds Detection on the Edge with CornWeed Dataset},
 type = {article},
 year = {2023},
 keywords = {Index Terms-plant detection,agriculture,data acquisition,deep learning,maize data,vision transformer},
 id = {5e053b2f-e7dd-3afc-9d43-b08a200f3278},
 created = {2023-09-20T08:23:11.234Z},
 accessed = {2023-09-20},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-08T10:14:50.375Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Artificial intelligence (AI) is used more heavily in agricultural applications. Yet, the lack of wireless-fidelity (Wi-Fi) connections on agricultural fields makes AI cloud services unavailable. Consequently, AI models have to be processed directly on the edge. In this paper, we evaluate state-of-the-art detection algorithms for their use in agriculture, in particular plant detection. Thus, this paper presents the CornWeed data set, which has been recorded on farm machines, showing labelled maize crops and weeds for plant detection. The paper provides accuracies for the state-of-the-art detection algorithms on the CornWeed data set, as well as frames per second (FPS) metrics for the considered networks on multiple edge devices. Moreover, for the FPS analysis, the detection algorithms are converted to open neural network exchange (ONNX) and TensoRT engine files as they could be used as future standards for model exchange.},
 bibtype = {article},
 author = {Iqbal, Naeem and Manss, Christoph and Scholz, Christian and König, Daniel and Igelbrink, Matthias and Ruckelshausen, Arno},
 doi = {10.5281/zenodo.7961764}
}

Artificial intelligence (AI) is used more heavily in agricultural applications. Yet, the lack of wireless-fidelity (Wi-Fi) connections on agricultural fields makes AI cloud services unavailable. Consequently, AI models have to be processed directly on the edge. In this paper, we evaluate state-of-the-art detection algorithms for their use in agriculture, in particular plant detection. Thus, this paper presents the CornWeed data set, which has been recorded on farm machines, showing labelled maize crops and weeds for plant detection. The paper provides accuracies for the state-of-the-art detection algorithms on the CornWeed data set, as well as frames per second (FPS) metrics for the considered networks on multiple edge devices. Moreover, for the FPS analysis, the detection algorithms are converted to open neural network exchange (ONNX) and TensoRT engine files as they could be used as future standards for model exchange.

@article{
 title = {Automatic diagnosis of a multi-symptom grapevine disease by decision trees and Graph Neural Networks},
 type = {article},
 year = {2023},
 keywords = {tardif2023automaticdiagnosismultigraph},
 pages = {1011–1017},
 id = {4522f247-e38d-3c57-9146-584cf54f2aad},
 created = {2023-09-21T11:07:48.768Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T20:25:06.407Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The grapevine disease called "Flavescence dorée" (FD) is actively monitored in Europe as it decreases yield and kills grapevines while being highly contagious. In this study, three methods for its automatic diagnosis from images acquired by proximal sensing (RGB camera) are proposed, compared and discussed. Method A uses a Convolutional Neural Network (CNN) classifier applied on raw images. The two other methods both process in two steps: (i) individual symptom detection using a CNN-based box detector and a deep segmentation algorithm, (ii) symptom-based diagnosis based either on a Random Forest classifier (method B) or on a Graph Neural Network (method C). A 6- fold cross validation was performed on 787 images of vines suffering from FD or from other biotic or abiotic stress factors. Methods B and C reached almost equally good results and outperformed method A: they achieved respectively, in (precision, recall), (0.69, 0.81), (0.87, 0.88) and (0.88, 0.88).},
 bibtype = {article},
 author = {Tardif, Malo and Keresztes, Barna and Deshayes, Aymeric and Martin, D. and Greven, Marc and Da Costa, Jean-Pierre},
 journal = {Precision agriculture '23}
}

The grapevine disease called "Flavescence dorée" (FD) is actively monitored in Europe as it decreases yield and kills grapevines while being highly contagious. In this study, three methods for its automatic diagnosis from images acquired by proximal sensing (RGB camera) are proposed, compared and discussed. Method A uses a Convolutional Neural Network (CNN) classifier applied on raw images. The two other methods both process in two steps: (i) individual symptom detection using a CNN-based box detector and a deep segmentation algorithm, (ii) symptom-based diagnosis based either on a Random Forest classifier (method B) or on a Graph Neural Network (method C). A 6- fold cross validation was performed on 787 images of vines suffering from FD or from other biotic or abiotic stress factors. Methods B and C reached almost equally good results and outperformed method A: they achieved respectively, in (precision, recall), (0.69, 0.81), (0.87, 0.88) and (0.88, 0.88).

@article{
 title = {Automatic diagnosis of a multi-symptom grape vine disease using computer vision},
 type = {article},
 year = {2023},
 keywords = {tardif2023automaticdiagnosismulti},
 pages = {53-60},
 volume = {1360},
 id = {f3ba706c-2c98-3ba5-a4c7-8ffef672315d},
 created = {2023-09-21T11:07:48.771Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:42:35.262Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {“Flavescence dorée” (FD) is a highly transmissible disease very closely monitored in Europe as it reduces vine productivity and causes vine death. Currently, this disease is controlled by a two-pronged approach: spray insecticide on a regular basis to kill the vector and by experts surveying each row in a vineyard. Unfortunately, these experts are not able to carry out such a task every year on every vineyard and need an aid for planning their surveys. In this study, we propose and evaluate an original automatic method for the detection of FD, based on computer vision and artificial intelligence applied to images acquired by proximal sensing. A two-step approach is used, mimicking expert’s scouting in the vine rows: i) the three known isolated symptoms are detected, ii) isolated detections are combined to make a diagnosis at vine scale. To achieve this, a detection deep neural network is used to detect and classify non-healthy leaves into three classes – ‘FD symptomatic leaf', ‘Esca leaf’ and ‘Confounding leaf’ – while a segmentation network retrieves FD symptomatic shoots and bunches. Finally, the association of the detected symptoms is performed by a RandomForest classifier allowing a diagnosis at the image scale. The experimental evaluation is conducted on images collected on 14 blocks planted with 5 grape cultivars, allowing the study of the impact of acquisition conditions and variability of symptom expressions among grape cultivars.},
 bibtype = {article},
 author = {Tardif, Malo and Amri, Ahmed and Keresztes, Barna and Deshayes, Aymeric and Martin, Damian and Greven, Marc and Da Costa, Jean-Pierre},
 doi = {10.17660/ActaHortic.2023.1360.7},
 journal = {Acta Horticulturae}
}

“Flavescence dorée” (FD) is a highly transmissible disease very closely monitored in Europe as it reduces vine productivity and causes vine death. Currently, this disease is controlled by a two-pronged approach: spray insecticide on a regular basis to kill the vector and by experts surveying each row in a vineyard. Unfortunately, these experts are not able to carry out such a task every year on every vineyard and need an aid for planning their surveys. In this study, we propose and evaluate an original automatic method for the detection of FD, based on computer vision and artificial intelligence applied to images acquired by proximal sensing. A two-step approach is used, mimicking expert’s scouting in the vine rows: i) the three known isolated symptoms are detected, ii) isolated detections are combined to make a diagnosis at vine scale. To achieve this, a detection deep neural network is used to detect and classify non-healthy leaves into three classes – ‘FD symptomatic leaf', ‘Esca leaf’ and ‘Confounding leaf’ – while a segmentation network retrieves FD symptomatic shoots and bunches. Finally, the association of the detected symptoms is performed by a RandomForest classifier allowing a diagnosis at the image scale. The experimental evaluation is conducted on images collected on 14 blocks planted with 5 grape cultivars, allowing the study of the impact of acquisition conditions and variability of symptom expressions among grape cultivars.

@article{
 title = {The Use of Computer Vision to Improve the Affinity of Rootstock-Graft Combinations and Identify Diseases of Grape Seedlings},
 type = {article},
 year = {2023},
 keywords = {artificial intelligence,computer vision,environmental engineering,graft combinations’ affinity,grape diseases,grape seedlings,neural networks,rootstock,viticulture},
 pages = {92},
 volume = {8},
 websites = {https://www.mdpi.com/2411-5134/8/4/92/htm,https://www.mdpi.com/2411-5134/8/4/92},
 month = {7},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {19},
 id = {1235f0f2-9d34-3030-b753-a1416e1ce7ab},
 created = {2023-10-26T09:20:20.156Z},
 accessed = {2023-10-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:23.972Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This study explores the application of computer vision for enhancing the selection of rootstock-graft combinations and detecting diseases in grape seedlings. Computer vision has various applications in viticulture, but publications and research have not reported the use of computer vision in rootstock-graft selection, which defines the novelty of this research. This paper presents elements of the technology for applying computer vision to rootstock-graft combinations and includes an analysis of grape seedling cuttings. This analysis allows for a more accurate determination of the compatibility between rootstock and graft, as well as the detection of potential seedling diseases. The utilization of computer vision to automate the grafting process of grape cuttings offers significant benefits in terms of increased efficiency, improved quality, and reduced costs. This technology can replace manual labor and ensure economic efficiency and reliability, among other advantages. It also facilitates monitoring the development of seedlings to determine the appropriate planting time. Image processing algorithms play a vital role in automatically determining seedling characteristics such as trunk diameter and the presence of any damage. Furthermore, computer vision can aid in the identification of diseases and defects in seedlings, which is crucial for assessing their overall quality. The automation of these processes offers several advantages, including increased efficiency, improved quality, and reduced costs through the reduction of manual labor and waste. To fulfill these objectives, a unique robotic assembly line is planned for the grafting of grape cuttings. This line will be equipped with two conveyor belts, a delta robot, and a computer vision system. The use of computer vision in automating the grafting process for grape cuttings offers significant benefits in terms of efficiency, quality improvement, and cost reduction. By incorporating image processing algorithms and advanced robotics, this technology has the potential to revolutionize the viticulture industry. Thanks to training a computer vision system to analyze data on rootstock and graft grape varieties, it is possible to reduce the number of defects by half. The implementation of a semi-automated computer vision system can improve crossbreeding efficiency by 90%. Reducing the time spent on pairing selection is also a significant advantage. While manual selection takes between 1 and 2 min, reducing the time to 30 s using the semi-automated system, and the prospect of further automation reducing the time to 10–15 s, will significantly increase the productivity and efficiency of the process. In addition to the aforementioned benefits, the integration of computer vision technology in grape grafting processes brings several other advantages. One notable advantage is the increased accuracy and precision in pairing selection. Computer vision algorithms can analyze a wide range of factors, including size, shape, color, and structural characteristics, to make more informed decisions when matching rootstock and graft varieties. This can lead to better compatibility and improved overall grafting success rates.},
 bibtype = {article},
 author = {Rudenko, Marina and Plugatar, Yurij and Korzin, Vadim and Kazak, Anatoliy and Gallini, Nadezhda and Gorbunova, Natalia},
 doi = {10.3390/INVENTIONS8040092},
 journal = {Inventions 2023, Vol. 8, Page 92},
 number = {4}
}

This study explores the application of computer vision for enhancing the selection of rootstock-graft combinations and detecting diseases in grape seedlings. Computer vision has various applications in viticulture, but publications and research have not reported the use of computer vision in rootstock-graft selection, which defines the novelty of this research. This paper presents elements of the technology for applying computer vision to rootstock-graft combinations and includes an analysis of grape seedling cuttings. This analysis allows for a more accurate determination of the compatibility between rootstock and graft, as well as the detection of potential seedling diseases. The utilization of computer vision to automate the grafting process of grape cuttings offers significant benefits in terms of increased efficiency, improved quality, and reduced costs. This technology can replace manual labor and ensure economic efficiency and reliability, among other advantages. It also facilitates monitoring the development of seedlings to determine the appropriate planting time. Image processing algorithms play a vital role in automatically determining seedling characteristics such as trunk diameter and the presence of any damage. Furthermore, computer vision can aid in the identification of diseases and defects in seedlings, which is crucial for assessing their overall quality. The automation of these processes offers several advantages, including increased efficiency, improved quality, and reduced costs through the reduction of manual labor and waste. To fulfill these objectives, a unique robotic assembly line is planned for the grafting of grape cuttings. This line will be equipped with two conveyor belts, a delta robot, and a computer vision system. The use of computer vision in automating the grafting process for grape cuttings offers significant benefits in terms of efficiency, quality improvement, and cost reduction. By incorporating image processing algorithms and advanced robotics, this technology has the potential to revolutionize the viticulture industry. Thanks to training a computer vision system to analyze data on rootstock and graft grape varieties, it is possible to reduce the number of defects by half. The implementation of a semi-automated computer vision system can improve crossbreeding efficiency by 90%. Reducing the time spent on pairing selection is also a significant advantage. While manual selection takes between 1 and 2 min, reducing the time to 30 s using the semi-automated system, and the prospect of further automation reducing the time to 10–15 s, will significantly increase the productivity and efficiency of the process. In addition to the aforementioned benefits, the integration of computer vision technology in grape grafting processes brings several other advantages. One notable advantage is the increased accuracy and precision in pairing selection. Computer vision algorithms can analyze a wide range of factors, including size, shape, color, and structural characteristics, to make more informed decisions when matching rootstock and graft varieties. This can lead to better compatibility and improved overall grafting success rates.

@article{
 title = {Towards smart pruning: ViNet, a deep-learning approach for grapevine structure estimation},
 type = {article},
 year = {2023},
 keywords = {Convolutional network,Deep learning,Grapevine pruning,Plant skeleton,Precision viticulture,Vineyard},
 pages = {107736},
 volume = {207},
 month = {4},
 publisher = {Elsevier},
 day = {1},
 id = {cb56e55e-71e9-3e0b-b8ac-4284e305a41c},
 created = {2023-10-26T10:44:38.867Z},
 accessed = {2023-10-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:24.310Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Image and video tools for analyzing crop scenes and plants are essential for applying precision agriculture to crop maintenance, harvesting, or pruning. In this paper, we are interested in vine pruning, a task that requires a precise understanding of the vine structure with branch type identification, orientations, and node locations. However, estimating such a structure is highly challenging, given the large variety in grapevine appearances, lighting conditions, viewpoint, the interweaving of branches, occlusions, and the level of details needed. To address these challenges, we propose ViNet: a deep-learning approach for estimating the structure of grapevine, which comprises two main steps: The first one detects nodes and identifies the branch types of the plant, as well as the spatial relation between them, whilst the second one uses the extracted nodes and branches to build a graph, out of which the structure of the grapevine is inferred. In doing so, we make four main contributions: (i) we put forward for the first time a method for automatic segmentation and extraction of the grapevine structure from images; (ii) we propose a novel approach leveraging the powerful stacked hourglass network to infer node location, branch types and the spatial relationships between them; (iii) we propose a novel shortest path weighted graph optimization step to extract connections between nodes and infer the structure, allowing to address the problem of having an unknown number of branches in the tree; (iv) we publicly release a dataset of more than 1500 grapevine images fully annotated with the structure information. Extensive experiments on this dataset demonstrate the efficiency of our approach at predicting the structure of a grapevine, achieving a precision and recall for node prediction of 95% and 90%, respectively, as well as ablation studies validating our design choices.},
 bibtype = {article},
 author = {Gentilhomme, Theophile and Villamizar, Michael and Corre, Jerome and Odobez, Jean Marc},
 doi = {10.1016/J.COMPAG.2023.107736},
 journal = {Computers and Electronics in Agriculture}
}

Image and video tools for analyzing crop scenes and plants are essential for applying precision agriculture to crop maintenance, harvesting, or pruning. In this paper, we are interested in vine pruning, a task that requires a precise understanding of the vine structure with branch type identification, orientations, and node locations. However, estimating such a structure is highly challenging, given the large variety in grapevine appearances, lighting conditions, viewpoint, the interweaving of branches, occlusions, and the level of details needed. To address these challenges, we propose ViNet: a deep-learning approach for estimating the structure of grapevine, which comprises two main steps: The first one detects nodes and identifies the branch types of the plant, as well as the spatial relation between them, whilst the second one uses the extracted nodes and branches to build a graph, out of which the structure of the grapevine is inferred. In doing so, we make four main contributions: (i) we put forward for the first time a method for automatic segmentation and extraction of the grapevine structure from images; (ii) we propose a novel approach leveraging the powerful stacked hourglass network to infer node location, branch types and the spatial relationships between them; (iii) we propose a novel shortest path weighted graph optimization step to extract connections between nodes and infer the structure, allowing to address the problem of having an unknown number of branches in the tree; (iv) we publicly release a dataset of more than 1500 grapevine images fully annotated with the structure information. Extensive experiments on this dataset demonstrate the efficiency of our approach at predicting the structure of a grapevine, achieving a precision and recall for node prediction of 95% and 90%, respectively, as well as ablation studies validating our design choices.

@misc{
 title = {Vision-Based Monitoring of the Short-Term Dynamic Behaviour of Plants for Automated Phenotyping},
 type = {misc},
 year = {2023},
 pages = {624-633},
 id = {804abb0a-db90-31a1-9a72-f23b72a1314a},
 created = {2023-10-27T06:12:27.656Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:24.482Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Modern computer vision technology plays an increasingly important role in agriculture. Automated monitoring of plants for example is an essential task in several applications , such as high-throughput phenotyping or plant health monitoring. Under external influences like wind, plants typically exhibit dynamic behaviours which reveal important characteristics of their structure and condition. These behaviours , however, are typically not considered by state-of-the-art automated phenotyping methods which mostly observe static plant properties. In this paper, we propose an automated system for monitoring oscillatory plant movement from video sequences. We employ harmonic inversion for the purpose of efficiently and accurately estimating the eigenfrequency and damping parameters of individual plant parts. The achieved accuracy is compared against values obtained by performing the Discrete Fourier Transform (DFT), which we use as a baseline. We demonstrate the applicability of this approach on different plants and plant parts, like wheat ears, hanging vines, as well as stems and stalks, which exhibit a range of oscillatory motions. By utilising harmonic inversion, we are able to consistently obtain more accurate values for the eigenfrequencies compared to those obtained by DFT. We are furthermore able to directly estimate values for the damping coefficient, achieving a similar accuracy as via DFT-based methods, but without the additional computational effort required for the latter. With the approach presented in this paper, it is possible to obtain estimates of mechanical plant characteristics in an automated manner, enabling novel automated acquisition of novel traits for phenotyping.},
 bibtype = {misc},
 author = {Wagner, Nikolaus and Cielniak, Grzegorz}
}

Modern computer vision technology plays an increasingly important role in agriculture. Automated monitoring of plants for example is an essential task in several applications , such as high-throughput phenotyping or plant health monitoring. Under external influences like wind, plants typically exhibit dynamic behaviours which reveal important characteristics of their structure and condition. These behaviours , however, are typically not considered by state-of-the-art automated phenotyping methods which mostly observe static plant properties. In this paper, we propose an automated system for monitoring oscillatory plant movement from video sequences. We employ harmonic inversion for the purpose of efficiently and accurately estimating the eigenfrequency and damping parameters of individual plant parts. The achieved accuracy is compared against values obtained by performing the Discrete Fourier Transform (DFT), which we use as a baseline. We demonstrate the applicability of this approach on different plants and plant parts, like wheat ears, hanging vines, as well as stems and stalks, which exhibit a range of oscillatory motions. By utilising harmonic inversion, we are able to consistently obtain more accurate values for the eigenfrequencies compared to those obtained by DFT. We are furthermore able to directly estimate values for the damping coefficient, achieving a similar accuracy as via DFT-based methods, but without the additional computational effort required for the latter. With the approach presented in this paper, it is possible to obtain estimates of mechanical plant characteristics in an automated manner, enabling novel automated acquisition of novel traits for phenotyping.

@article{
 title = {Grape yield estimation with a smartphone’s colour and depth cameras using machine learning and computer vision techniques},
 type = {article},
 year = {2023},
 keywords = {Berry detection,Depth camera,Grapes,RGB-D,YOLO,Yield estimation},
 pages = {108174},
 volume = {213},
 month = {10},
 publisher = {Elsevier},
 day = {1},
 id = {0eb92be6-6b9a-3104-85d0-45a943b53730},
 created = {2023-10-27T06:13:29.083Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:24.634Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {A smartphone with both colour and time of flight depth cameras is used for automated grape yield estimation of Chardonnay grapes. A new technique is developed to automatically identify grape berries in the smartphone's depth maps. This utilises the distortion peaks in the depth map caused by diffused scattering of the light within each grape berry. This technique is then extended to allow unsupervised training of a YOLOv7 model for the detection of grape berries in the smartphone's colour images. A correlation coefficient (R2) of 0.946 was achieved when comparing the count of grape berries observed in RGB images to those accurately identified by YOLO. Additionally, an average precision score of 0.970 was attained. Two techniques are then presented to automatically estimate the size of the grape berries and generate 3D models of grape bunches using both colour and depth information.},
 bibtype = {article},
 author = {Parr, Baden and Legg, Mathew and Alam, Fakhrul},
 doi = {10.1016/J.COMPAG.2023.108174},
 journal = {Computers and Electronics in Agriculture}
}

A smartphone with both colour and time of flight depth cameras is used for automated grape yield estimation of Chardonnay grapes. A new technique is developed to automatically identify grape berries in the smartphone's depth maps. This utilises the distortion peaks in the depth map caused by diffused scattering of the light within each grape berry. This technique is then extended to allow unsupervised training of a YOLOv7 model for the detection of grape berries in the smartphone's colour images. A correlation coefficient (R2) of 0.946 was achieved when comparing the count of grape berries observed in RGB images to those accurately identified by YOLO. Additionally, an average precision score of 0.970 was attained. Two techniques are then presented to automatically estimate the size of the grape berries and generate 3D models of grape bunches using both colour and depth information.

@article{
 title = {Proximal sensing for geometric characterization of vines: A review of the latest advances},
 type = {article},
 year = {2023},
 pages = {107901},
 volume = {210},
 month = {7},
 publisher = {Elsevier},
 day = {1},
 id = {0e530cf4-c596-3685-a147-e26b52683013},
 created = {2023-10-27T06:36:38.933Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T14:59:44.683Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Several variables, including a rising human population, varying weather patterns in the context of ongoing climate change, and the rapid worldwide spread of epidemics, all contribute to boosting agricultural demand. To assure food availability, quality, and safety while increasing yields and profitability, precision agriculture must progress swiftly. Precision viticulture aims to optimize vineyard management in this setting by reducing resource consumption and environmental impact while simultaneously enhancing the yield, product quality, and oenological potential of vineyards. This comprehensive review article offers an overview of the real-world and laboratory applications of optical and non-optical sensors in precision viticulture for 3D modelling. Hence, there is a pressing need to track the development of crops at a wide range of spatial and temporal scales, in a wide variety of environments, and for a wide range of objectives in a non-destructive manner. Due to the intrinsic spatial heterogeneity of vineyards, the adoption of precision viticulture necessitates crop monitoring using contactless and non-invasive sensors such as ultrasonic, LiDAR (Light Detection and Ranging), depth, or RGB cameras to prevent low accuracy and sparse sampling. This study aims to assist researchers in gaining a broad understanding of the sensing technologies for precision viticulture, the present problems, and the advancement of the state of the art. The study focuses on sensors used for Proximal Sensing to geometrically characterize vines using statically or dynamically ground-based measurements through a wide range of mobile sensing platforms. The employed sensors, data extraction, and analysis procedures are described. Moreover, the present and future potential of Proximal Sensing and Remote Sensing in vineyards is discussed.},
 bibtype = {article},
 author = {Moreno, Hugo and Andújar, Dionisio},
 doi = {10.1016/J.COMPAG.2023.107901},
 journal = {Computers and Electronics in Agriculture},
 keywords = {moreno2023proximalsensinggeometric}
}

Several variables, including a rising human population, varying weather patterns in the context of ongoing climate change, and the rapid worldwide spread of epidemics, all contribute to boosting agricultural demand. To assure food availability, quality, and safety while increasing yields and profitability, precision agriculture must progress swiftly. Precision viticulture aims to optimize vineyard management in this setting by reducing resource consumption and environmental impact while simultaneously enhancing the yield, product quality, and oenological potential of vineyards. This comprehensive review article offers an overview of the real-world and laboratory applications of optical and non-optical sensors in precision viticulture for 3D modelling. Hence, there is a pressing need to track the development of crops at a wide range of spatial and temporal scales, in a wide variety of environments, and for a wide range of objectives in a non-destructive manner. Due to the intrinsic spatial heterogeneity of vineyards, the adoption of precision viticulture necessitates crop monitoring using contactless and non-invasive sensors such as ultrasonic, LiDAR (Light Detection and Ranging), depth, or RGB cameras to prevent low accuracy and sparse sampling. This study aims to assist researchers in gaining a broad understanding of the sensing technologies for precision viticulture, the present problems, and the advancement of the state of the art. The study focuses on sensors used for Proximal Sensing to geometrically characterize vines using statically or dynamically ground-based measurements through a wide range of mobile sensing platforms. The employed sensors, data extraction, and analysis procedures are described. Moreover, the present and future potential of Proximal Sensing and Remote Sensing in vineyards is discussed.

@article{
 title = {An improved lightweight network based on deep learning for grape recognition in unstructured environments},
 type = {article},
 year = {2023},
 keywords = {AlNet,Depthwise separable convolution,Grape recognition,ResBlock-M,YOLOX},
 month = {2},
 publisher = {Elsevier},
 day = {20},
 id = {2c84e3fc-fe5b-3a49-92cd-c31603b04189},
 created = {2023-10-27T06:43:20.203Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:25.239Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {In unstructured environments, dense grape fruit growth and the presence of occlusion cause difficult recognition problems, which will seriously affect the performance of grape picking robots. To address these problems, this study improves the YOLOX-Tiny model and proposes a new grape detection model, YOLOX-RA, which can quickly and accurately identify densely growing and occluded grape bunches. The proposed YOLOX-RA model uses a 3 × 3 convolutional layer with a step size of 2 to replace the focal layer to reduce the computational burden. The CBS layer in the ResBlock_Body module of the second, third, and fourth layers of the backbone layer is removed, and the CSPLayer module is replaced by the ResBlock-M module to speed up the detection. An auxiliary network (AlNet) with the remaining network blocks was added after the ResBlock-M module to improve the detection accuracy. Two depth-separable convolutions (DSC) are used in the neck module layer to replace the normal convolution to reduce the computational cost. We evaluated the detection performance of SSD, YOLOv4 SSD, YOLOv4-Tiny, YOLO-Grape, YOLOv5-X, YOLOX-Tiny, and YOLOX-RA on a grape test set. The results show that the YOLOX-RA model has the best detection performance, achieving 88.75 % mAP, a recognition speed of 84.88 FPS, and model size of 17.53 MB. It can accurately detect densely grown and shaded grape bunches, which can effectively improve the performance of the grape picking robot.},
 bibtype = {article},
 author = {Liu, Bingpiao and Zhang, Yunzhi and Wang, Jinhai and Luo, Lufeng and Lu, Qinghua and Wei, Huiling and Zhu, Wenbo},
 doi = {10.1016/J.INPA.2023.02.003},
 journal = {Information Processing in Agriculture}
}

In unstructured environments, dense grape fruit growth and the presence of occlusion cause difficult recognition problems, which will seriously affect the performance of grape picking robots. To address these problems, this study improves the YOLOX-Tiny model and proposes a new grape detection model, YOLOX-RA, which can quickly and accurately identify densely growing and occluded grape bunches. The proposed YOLOX-RA model uses a 3 × 3 convolutional layer with a step size of 2 to replace the focal layer to reduce the computational burden. The CBS layer in the ResBlock_Body module of the second, third, and fourth layers of the backbone layer is removed, and the CSPLayer module is replaced by the ResBlock-M module to speed up the detection. An auxiliary network (AlNet) with the remaining network blocks was added after the ResBlock-M module to improve the detection accuracy. Two depth-separable convolutions (DSC) are used in the neck module layer to replace the normal convolution to reduce the computational cost. We evaluated the detection performance of SSD, YOLOv4 SSD, YOLOv4-Tiny, YOLO-Grape, YOLOv5-X, YOLOX-Tiny, and YOLOX-RA on a grape test set. The results show that the YOLOX-RA model has the best detection performance, achieving 88.75 % mAP, a recognition speed of 84.88 FPS, and model size of 17.53 MB. It can accurately detect densely grown and shaded grape bunches, which can effectively improve the performance of the grape picking robot.

@article{
 title = {Weakly and semi-supervised detection, segmentation and tracking of table grapes with limited and noisy data},
 type = {article},
 year = {2023},
 keywords = {Computer vision,Deep learning,Fruit detection and segmentation,Self-supervised learning,Yield prediction},
 pages = {107624},
 volume = {205},
 month = {2},
 publisher = {Elsevier},
 day = {1},
 id = {06f46e4f-3763-3746-902c-5144a0b126b6},
 created = {2023-10-27T06:44:30.963Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:24.774Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Detection, segmentation and tracking of fruits and vegetables are three fundamental tasks for precision agriculture, enabling robotic harvesting and yield estimation applications. However, modern algorithms are data hungry and it is not always possible to gather enough data to apply the best performing supervised approaches. Since data collection is an expensive and cumbersome task, the enabling technologies for using computer vision in agriculture are often out of reach for small businesses. Following previous work in this context (Ciarfuglia et al., 2022), where we proposed an initial weakly supervised solution to reduce the data needed to get state-of-the-art detection and segmentation in precision agriculture applications, here we improve that system and explore the problem of tracking fruits in orchards. We present the case of vineyards of table grapes in southern Lazio (Italy) since grapes are a difficult fruit to segment due to occlusion, colour and general illumination conditions. We consider the case in which there is some initial labelled data that could work as source data (e.g. wine grape data), but it is considerably different from the target data (e.g. table grape data). To improve detection and segmentation on the target data, we propose to train the segmentation algorithm with a weak bounding box label, while for tracking we leverage 3D Structure from Motion algorithms to generate new labels from already labelled samples. Finally, the two systems are combined in a full semi-supervised approach. Comparisons with state-of-the-art supervised solutions show how our methods are able to train new models that achieve high performances with few labelled images and with very simple labelling.},
 bibtype = {article},
 author = {Ciarfuglia, Thomas A. and Motoi, Ionut M. and Saraceni, Leonardo and Fawakherji, Mulham and Sanfeliu, Alberto and Nardi, Daniele},
 doi = {10.1016/J.COMPAG.2023.107624},
 journal = {Computers and Electronics in Agriculture}
}

Detection, segmentation and tracking of fruits and vegetables are three fundamental tasks for precision agriculture, enabling robotic harvesting and yield estimation applications. However, modern algorithms are data hungry and it is not always possible to gather enough data to apply the best performing supervised approaches. Since data collection is an expensive and cumbersome task, the enabling technologies for using computer vision in agriculture are often out of reach for small businesses. Following previous work in this context (Ciarfuglia et al., 2022), where we proposed an initial weakly supervised solution to reduce the data needed to get state-of-the-art detection and segmentation in precision agriculture applications, here we improve that system and explore the problem of tracking fruits in orchards. We present the case of vineyards of table grapes in southern Lazio (Italy) since grapes are a difficult fruit to segment due to occlusion, colour and general illumination conditions. We consider the case in which there is some initial labelled data that could work as source data (e.g. wine grape data), but it is considerably different from the target data (e.g. table grape data). To improve detection and segmentation on the target data, we propose to train the segmentation algorithm with a weak bounding box label, while for tracking we leverage 3D Structure from Motion algorithms to generate new labels from already labelled samples. Finally, the two systems are combined in a full semi-supervised approach. Comparisons with state-of-the-art supervised solutions show how our methods are able to train new models that achieve high performances with few labelled images and with very simple labelling.

@article{
 title = {Real-time tracking and counting of grape clusters in the field based on channel pruning with YOLOv5s},
 type = {article},
 year = {2023},
 keywords = {Channel pruning,Grape,Multiple object tracking,Real-time counting,YOLOv5s},
 pages = {107662},
 volume = {206},
 month = {3},
 publisher = {Elsevier},
 day = {1},
 id = {28015a97-5be1-3629-8e36-31f19d0d5e27},
 created = {2023-10-27T06:45:46.225Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:24.922Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Accurate fruit counting helps grape wine industry make better logistics and decisions before harvest, and therefore produce higher quality wine. In view of poor real-time performance of the existing fruit tracking and counting methods, and a lack of effective counting methods for cluster-like fruits due to their huge shape variabilities. In this study, an end-to-end lightweight counting pipeline is developed to automate the processing of video data for real-time tracking and counting of grape clusters in field conditions. First, based on channel pruning algorithm, a more lightweight YOLOv5s cluster detection model is obtained, where number of model parameters, model size and floating-point operations (FLOPs) are reduced by 79 %, 76 %, and 58 %, respectively, and the pruned model size is only 3.4 MB. Secondly, the soft non-maximum suppression is introduced in prediction stage to improve detection performance for clusters with overlapping grapes. Test results show that mAP reaches 82.3 % and average inference time is 6.1 ms per image, which effectively reduces model parameters and complexity while ensuring detection accuracy. Finally, online multiple object tracking of clusters is implemented by integrating the detection results and SORT algorithm, where two counting modes are set by introducing counting lines. Test results on 8 videos indicated that the average counting accuracy of the proposed method reached 84.9 %, correlation coefficient with manual counting reached 0.9905, and speed of video processing reached up to 50.4 frames per second (FPS), meeting field real-time requirements. This study provides a timely technical reference for the development of orchard robots to achieve real-time automated yield estimation and accurate crop management decisions.},
 bibtype = {article},
 author = {Shen, Lei and Su, Jinya and He, Runtian and Song, Lijie and Huang, Rong and Fang, Yulin and Song, Yuyang and Su, Baofeng},
 doi = {10.1016/J.COMPAG.2023.107662},
 journal = {Computers and Electronics in Agriculture}
}

Accurate fruit counting helps grape wine industry make better logistics and decisions before harvest, and therefore produce higher quality wine. In view of poor real-time performance of the existing fruit tracking and counting methods, and a lack of effective counting methods for cluster-like fruits due to their huge shape variabilities. In this study, an end-to-end lightweight counting pipeline is developed to automate the processing of video data for real-time tracking and counting of grape clusters in field conditions. First, based on channel pruning algorithm, a more lightweight YOLOv5s cluster detection model is obtained, where number of model parameters, model size and floating-point operations (FLOPs) are reduced by 79 %, 76 %, and 58 %, respectively, and the pruned model size is only 3.4 MB. Secondly, the soft non-maximum suppression is introduced in prediction stage to improve detection performance for clusters with overlapping grapes. Test results show that mAP reaches 82.3 % and average inference time is 6.1 ms per image, which effectively reduces model parameters and complexity while ensuring detection accuracy. Finally, online multiple object tracking of clusters is implemented by integrating the detection results and SORT algorithm, where two counting modes are set by introducing counting lines. Test results on 8 videos indicated that the average counting accuracy of the proposed method reached 84.9 %, correlation coefficient with manual counting reached 0.9905, and speed of video processing reached up to 50.4 frames per second (FPS), meeting field real-time requirements. This study provides a timely technical reference for the development of orchard robots to achieve real-time automated yield estimation and accurate crop management decisions.

@article{
 title = {Exploratory approach for automatic detection of vine rows in terrace vineyards},
 type = {article},
 year = {2023},
 keywords = {Artificial Intelligence,Precision agriculture,Remote sensing,Terrace vineyards},
 pages = {139-144},
 volume = {219},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {957ecde5-2285-3250-bb4f-b4e50a33e778},
 created = {2023-10-27T07:00:02.237Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-27T07:00:24.059Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The Alto Douro Demarcated Region in Portugal is the oldest and most regulated wine-growing region in the world, formed by an ecosystem of unique value allowing the cultivation of vines on its characteristics terraces vineyards. The detection of vine rows in terrace vineyards constitutes an essential task regarding the achievement of important goals such as multi-Temporal crop evaluation and production estimation. Despite the advances and research in this field, most studies are limited to flat vineyards with straight vine rows. In this study an exploratory approach in the precision agriculture for automatic detection of vine rows in terrace vineyards is presented with remote sensing techniques associated with artificial intelligence such as Machine Learning and Deep learning. At the current stage the preliminary results are encouraging for the detection of vine rows in straight and curved lines considering the complexity of the terrain.},
 bibtype = {article},
 author = {Figueiredo, Nuno and Padua, Luis and Cunha, Antonio and Sousa, Joaquim J. and Sousa, Antonio},
 doi = {10.1016/J.PROCS.2023.01.274},
 journal = {Procedia Computer Science}
}

The Alto Douro Demarcated Region in Portugal is the oldest and most regulated wine-growing region in the world, formed by an ecosystem of unique value allowing the cultivation of vines on its characteristics terraces vineyards. The detection of vine rows in terrace vineyards constitutes an essential task regarding the achievement of important goals such as multi-Temporal crop evaluation and production estimation. Despite the advances and research in this field, most studies are limited to flat vineyards with straight vine rows. In this study an exploratory approach in the precision agriculture for automatic detection of vine rows in terrace vineyards is presented with remote sensing techniques associated with artificial intelligence such as Machine Learning and Deep learning. At the current stage the preliminary results are encouraging for the detection of vine rows in straight and curved lines considering the complexity of the terrain.

@article{
 title = {Computer-Vision Based Real Time Waypoint Generation for Autonomous Vineyard Navigation with Quadruped Robots},
 type = {article},
 year = {2023},
 keywords = {Agricultural Robotics,Autonomous Vineyard Navigation,Computer-Vision,Quadruped Control},
 pages = {239-244},
 publisher = {IEEE},
 id = {50d3d502-c446-33df-8937-a08ca8fa1aaf},
 created = {2023-10-27T07:07:26.026Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-27T07:07:33.908Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The VINUM project seeks to address the shortage of skilled labor in modern vineyards by introducing a cutting-edge mobile robotic solution. Leveraging the capabilities of the quadruped robot, HyQReal, this system, equipped with arm and vision sensors, offers autonomous navigation and winter pruning of grapevines reducing the need for human intervention. At the heart of this approach lies an architecture that empowers the robot to easily navigate vineyards, identify grapevines with unparalleled accuracy, and approach them for pruning with precision. A state machine drives the process, deftly switching between various stages to ensure seamless and efficient task completion. The system's performance was assessed through experimentation, focusing on waypoint precision and optimizing the robot's workspace for single-plant operations. Results indicate that the architecture is highly reliable, with a mean error of 21.5cm and a standard deviation of 17.6cm for HyQReal. However, improvements in grapevine detection accuracy are necessary for optimal performance. This work is based on a computer-vision-based navigation method for quadruped robots in vineyards, opening up new possibilities for selective task automation. The system's architecture works well in ideal weather conditions, generating and arriving at precise waypoints that maximize the attached robotic arm's workspace. This work is an extension of our short paper presented at the Italian Conference on Robotics and Intelligent Machines (I-RIM), 2022 [1].},
 bibtype = {article},
 author = {Milburn, Lee and Gamba, Juan and Fernandes, Miguel and Semini, Claudio},
 doi = {10.1109/ICARSC58346.2023.10129563},
 journal = {2023 IEEE International Conference on Autonomous Robot Systems and Competitions, ICARSC 2023}
}

The VINUM project seeks to address the shortage of skilled labor in modern vineyards by introducing a cutting-edge mobile robotic solution. Leveraging the capabilities of the quadruped robot, HyQReal, this system, equipped with arm and vision sensors, offers autonomous navigation and winter pruning of grapevines reducing the need for human intervention. At the heart of this approach lies an architecture that empowers the robot to easily navigate vineyards, identify grapevines with unparalleled accuracy, and approach them for pruning with precision. A state machine drives the process, deftly switching between various stages to ensure seamless and efficient task completion. The system's performance was assessed through experimentation, focusing on waypoint precision and optimizing the robot's workspace for single-plant operations. Results indicate that the architecture is highly reliable, with a mean error of 21.5cm and a standard deviation of 17.6cm for HyQReal. However, improvements in grapevine detection accuracy are necessary for optimal performance. This work is based on a computer-vision-based navigation method for quadruped robots in vineyards, opening up new possibilities for selective task automation. The system's architecture works well in ideal weather conditions, generating and arriving at precise waypoints that maximize the attached robotic arm's workspace. This work is an extension of our short paper presented at the Italian Conference on Robotics and Intelligent Machines (I-RIM), 2022 [1].

@article{
 title = {Object detection and tracking on UAV RGB videos for early extraction of grape phenotypic traits},
 type = {article},
 year = {2023},
 keywords = {Instance segmentation,MOTS,PointTrack,Spatial Embeddings,UAV,Video,Viticulture,YOLACT},
 pages = {108051},
 volume = {211},
 month = {8},
 publisher = {Elsevier},
 day = {1},
 id = {4aea33b6-c42e-325f-ae80-42ef172375ec},
 created = {2023-10-27T07:09:48.980Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:25.743Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Grapevine phenotyping is the process of determining the physical properties (e.g., size, shape, and number) of grape bunches and berries. Grapevine phenotyping information provides valuable characteristics to monitor the sanitary status of the vine. Knowing the number and dimensions of bunches and berries at an early stage of development provides relevant information to the winegrowers about the yield to be harvested. However, the process of counting and measuring is usually done manually, which is laborious and time-consuming. Previous studies have attempted to implement bunch detection on red bunches in vineyards with leaf removal and surveys have been done using ground vehicles and handled cameras. However, Unmanned Aerial Vehicles (UAV) mounted with RGB cameras, along with computer vision techniques offer a cheap, robust, and timesaving alternative. Therefore, Multi-object tracking and segmentation (MOTS) is utilized in this study to determine the traits of individual white grape bunches and berries from RGB videos obtained from a UAV acquired over a commercial vineyard with a high density of leaves. To achieve this goal two datasets with labelled images and phenotyping measurements were created and made available in a public repository. PointTrack algorithm was used for detecting and tracking the grape bunches, and two instance segmentation algorithms - YOLACT and Spatial Embeddings - have been compared for finding the most suitable approach to detect berries. It was found that the detection performs adequately for cluster detection with a MODSA of 93.85. For tracking, the results were not sufficient when trained with 679 frames.This study provides an automated pipeline for the extraction of several grape phenotyping traits described by the International Organization of Vine and Wine (OIV) descriptors. The selected OIV descriptors are the bunch length, width, and shape (codes 202, 203, and 208, respectively) and the berry length, width, and shape (codes 220, 221, and 223, respectively). Lastly, the comparison regarding the number of detected berries per bunch indicated that Spatial Embeddings assessed berry counting more accurately (79.5%) than YOLACT (44.6%).},
 bibtype = {article},
 author = {Ariza-Sentís, Mar and Baja, Hilmy and Vélez, Sergio and Valente, João},
 doi = {10.1016/J.COMPAG.2023.108051},
 journal = {Computers and Electronics in Agriculture}
}

Grapevine phenotyping is the process of determining the physical properties (e.g., size, shape, and number) of grape bunches and berries. Grapevine phenotyping information provides valuable characteristics to monitor the sanitary status of the vine. Knowing the number and dimensions of bunches and berries at an early stage of development provides relevant information to the winegrowers about the yield to be harvested. However, the process of counting and measuring is usually done manually, which is laborious and time-consuming. Previous studies have attempted to implement bunch detection on red bunches in vineyards with leaf removal and surveys have been done using ground vehicles and handled cameras. However, Unmanned Aerial Vehicles (UAV) mounted with RGB cameras, along with computer vision techniques offer a cheap, robust, and timesaving alternative. Therefore, Multi-object tracking and segmentation (MOTS) is utilized in this study to determine the traits of individual white grape bunches and berries from RGB videos obtained from a UAV acquired over a commercial vineyard with a high density of leaves. To achieve this goal two datasets with labelled images and phenotyping measurements were created and made available in a public repository. PointTrack algorithm was used for detecting and tracking the grape bunches, and two instance segmentation algorithms - YOLACT and Spatial Embeddings - have been compared for finding the most suitable approach to detect berries. It was found that the detection performs adequately for cluster detection with a MODSA of 93.85. For tracking, the results were not sufficient when trained with 679 frames.This study provides an automated pipeline for the extraction of several grape phenotyping traits described by the International Organization of Vine and Wine (OIV) descriptors. The selected OIV descriptors are the bunch length, width, and shape (codes 202, 203, and 208, respectively) and the berry length, width, and shape (codes 220, 221, and 223, respectively). Lastly, the comparison regarding the number of detected berries per bunch indicated that Spatial Embeddings assessed berry counting more accurately (79.5%) than YOLACT (44.6%).

@article{
 title = {Applying Knowledge Distillation on Pre-Trained Model for Early Grapevine Detection},
 type = {article},
 year = {2023},
 keywords = {Deep Learning,Fine-tuning,Knowledge Distillation,Pseudo-labelling,Yield forecast},
 pages = {149-156},
 websites = {https://ebooks.iospress.nl/doi/10.3233/AISE230024},
 month = {6},
 publisher = {IOS Press},
 day = {23},
 id = {2d7b8c7b-5e5c-3b73-a0dc-43befad392bb},
 created = {2023-10-27T07:11:58.041Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:25.911Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The development of Artificial Intelligence has raised interesting opportunities for improved automation in smart agriculture. Smart viticulture is one of the domains that can benefit from Computer-vision tasks through field sustainability. Computer-vision solutions present additional constraints as the amount of data for good training convergence has to be complex enough to cover sufficient features from desired inputs. In this paper, we present a study to implement a grapevine detection improvement for early grapes detection and grape yield prediction whose interest in Champagne and wine companies is undeniable. Earlier yield predictions allow a better market assessment, the harvest work's organization and help decision-making about plant management. Our goal is to carry estimations 5 to 6 weeks before the harvest. Furthermore, the grapevines growing condition and the large amount of data to process for yield estimation require an embedded device to acquire and compute deep learning inference. Thus, the grapes detection model has to be lightweight enough to run on an embedded device. These models were subsequently pre-trained on two different types of datasets and several layer depth of deep learning models to propose a pseudo-labelling Teacher-Student related Knowledge Distillation. Overall solutions proposed an improvement of 7.56%, 6.98, 8.279%, 7.934% and 13.63% for f1 score, precision, recall, mean average precision at 50 and mean average precision 50-95 respectively on BBCH77 phenological stage.},
 bibtype = {article},
 author = {Hollard, Lilian and Mohimont, Lucas},
 doi = {10.3233/AISE230024}
}

The development of Artificial Intelligence has raised interesting opportunities for improved automation in smart agriculture. Smart viticulture is one of the domains that can benefit from Computer-vision tasks through field sustainability. Computer-vision solutions present additional constraints as the amount of data for good training convergence has to be complex enough to cover sufficient features from desired inputs. In this paper, we present a study to implement a grapevine detection improvement for early grapes detection and grape yield prediction whose interest in Champagne and wine companies is undeniable. Earlier yield predictions allow a better market assessment, the harvest work's organization and help decision-making about plant management. Our goal is to carry estimations 5 to 6 weeks before the harvest. Furthermore, the grapevines growing condition and the large amount of data to process for yield estimation require an embedded device to acquire and compute deep learning inference. Thus, the grapes detection model has to be lightweight enough to run on an embedded device. These models were subsequently pre-trained on two different types of datasets and several layer depth of deep learning models to propose a pseudo-labelling Teacher-Student related Knowledge Distillation. Overall solutions proposed an improvement of 7.56%, 6.98, 8.279%, 7.934% and 13.63% for f1 score, precision, recall, mean average precision at 50 and mean average precision 50-95 respectively on BBCH77 phenological stage.

@article{
 title = {Toward Grapevine Digital Ampelometry Through Vision Deep Learning Models},
 type = {article},
 year = {2023},
 keywords = {Artificial neural networks (ANN),computer vision,image processing,precision agriculture,vine species identification},
 pages = {10132-10139},
 volume = {23},
 month = {5},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 day = {1},
 id = {e7b7c3b3-ee18-3021-adad-3cfc3b22dcc4},
 created = {2023-10-27T07:20:42.268Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:26.513Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Several thousand grapevine varieties exist, with even more naming identifiers. Adequate specialized labor is not available for proper classification or identification of grapevines, making the value of commercial vines uncertain. Traditional methods, such as genetic analysis or ampelometry, are time-consuming, expensive, and often require expert skills that are even rarer. New vision-based systems benefit from advanced and innovative technology and can be used by nonexperts in ampelometry. To this end, ac dl and ac ml approaches have been successfully applied for classification purposes. This work extends the state of the art by applying digital ampelometry techniques to larger grapevine varieties. We benchmarked MobileNet v2, ResNet-34, and VGG-11-BN DL classifiers to assess their ability for digital ampelography. In our experiment, all the models could identify the vines' varieties through the leaf with a weighted $F1$ score higher than 92%.},
 bibtype = {article},
 author = {Magalhaes, Sandro Costa and Castro, Luis and Rodrigues, Leandro and Padilha, Tiago Cerveira and De Carvalho, Frederico and Neves Dos Santos, Filipe and Pinho, Tatiana and Moreira, Germano and Cunha, Jorge and Cunha, Mario and Silva, Paulo and Moreira, Antonio Paulo},
 doi = {10.1109/JSEN.2023.3261544},
 journal = {IEEE Sensors Journal},
 number = {9}
}

Several thousand grapevine varieties exist, with even more naming identifiers. Adequate specialized labor is not available for proper classification or identification of grapevines, making the value of commercial vines uncertain. Traditional methods, such as genetic analysis or ampelometry, are time-consuming, expensive, and often require expert skills that are even rarer. New vision-based systems benefit from advanced and innovative technology and can be used by nonexperts in ampelometry. To this end, ac dl and ac ml approaches have been successfully applied for classification purposes. This work extends the state of the art by applying digital ampelometry techniques to larger grapevine varieties. We benchmarked MobileNet v2, ResNet-34, and VGG-11-BN DL classifiers to assess their ability for digital ampelography. In our experiment, all the models could identify the vines' varieties through the leaf with a weighted $F1$ score higher than 92%.

@article{
 title = {Segmentation Methods Evaluation on Grapevine Leaf Diseases},
 type = {article},
 year = {2023},
 pages = {1081-1085},
 websites = {https://github.com/MrD1360/deep_segmentation_vineyards_navigation},
 id = {f7691e2d-7d12-3b00-90a4-cc908e722e03},
 created = {2023-10-27T07:22:31.109Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:05.960Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The problem of vine disease detection (VDD) was addressed in a number of research papers, however, a generic solution is not yet available for this task in the community. The region of interest segmentation and object detection tasks are often complementary. A similar situation is encountered in VDD applications as well, in which crop or leaf detection can be done via instance segmentation techniques as well. The focus of this work is to validate the most suitable methods from the main literature on vine leaf segmentation and disease detection on a custom dataset containing leaves both from the laboratory environment and cropped from images in the field. We tested five promising methods including the Otsu's thresholding, Mask R-CNN, MobileNet, SegNet, and Feature Pyramid Network variants. The results of the comparison are available in Table I summarizing the accuracy and runtime of different methods.},
 bibtype = {article},
 author = {Molnár, Szilárd and Tamás, Levente},
 doi = {10.15439/2023F7053},
 journal = {FedCSIS},
 keywords = {molnar2023segmentationmethodsevaluation}
}

The problem of vine disease detection (VDD) was addressed in a number of research papers, however, a generic solution is not yet available for this task in the community. The region of interest segmentation and object detection tasks are often complementary. A similar situation is encountered in VDD applications as well, in which crop or leaf detection can be done via instance segmentation techniques as well. The focus of this work is to validate the most suitable methods from the main literature on vine leaf segmentation and disease detection on a custom dataset containing leaves both from the laboratory environment and cropped from images in the field. We tested five promising methods including the Otsu's thresholding, Mask R-CNN, MobileNet, SegNet, and Feature Pyramid Network variants. The results of the comparison are available in Table I summarizing the accuracy and runtime of different methods.

@article{
 title = {Early Yield Estimation in Viticulture Based on Grapevine Inflorescence Detection and Counting in Videos},
 type = {article},
 year = {2023},
 keywords = {Grapevine,computer vision,deep learning,early yield estimation,inflorescence detection,inflorescence tracking,viticulture},
 pages = {37790-37808},
 volume = {11},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {86d161e9-cf7a-3466-b8a6-52c5496412ad},
 created = {2023-10-27T07:26:01.046Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:26.212Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {In viticulture, yield estimation is a key activity, which is important throughout the wine industry value chain. The earlier that an accurate yield estimation can be made the greater its value, increasing management options for grape growers and commercial options for winemakers. For the yield estimate based on in- field measurements at scale, the number of inflorescences emerging after bud-burst offers the earliest practical signal, allowing a yield potential to be determined months before harvest. This paper presents an approach to automatically count the inflorescence number at the phenological stage E-L 12 using RGB video data and demonstrates its use for estimating yield. A dataset consisting of RGB videos was collected shortly after bud-burst from multiple vineyards, in conjunction with hand counts to produce a manual ground-truth for the inflorescence counting task. The video frames were annotated using bounding-boxes around the inflorescences to produce a digital ground-truth. A deep learning architecture was developed to learn features from the video frames during training and detect the inflorescences at the later inference stage. The detection results were fed to a tracking pipeline built using computer vision and deep learning techniques to generate numbers of inflorescences present in test videos. The visual and quantitative results are presented and evaluated for the inflorescence detection and counting tasks. The developed inflorescence detector achieves an average precision of 80.00%, a recall of 83.92%, and an F1-score of 80.48%, through a five-fold cross-validation on the annotated dataset. For the test videos, the developed automatic inflorescence counting model reports an absolute error of 11.03 inflorescences per panel, a normalized mean absolute error of 10.80%, and an R2 of 0.86, when the predicted per-panel counts were compared to the corresponding manual ground-truth. Based on the counting results, we estimate an early yield that is within 4% to 11% error when compared to the actual yield after harvest. Based on these results and a separate analysis of the relationship between hand counts of inflorescences and harvest yields in three vineyards over three growing seasons, we conclude that computer vision and machine learning based methods have the potential to provide early yield estimation in viticulture with a commercially viable accuracy.},
 bibtype = {article},
 author = {Khokher, Muhammad Rizwan and Liao, Qiyu and Smith, Adam L. and Sun, Changming and MacKenzie, Donald and Thomas, Mark R. and Wang, Dadong and Edwards, Everard J.},
 doi = {10.1109/ACCESS.2023.3263238},
 journal = {IEEE Access}
}

In viticulture, yield estimation is a key activity, which is important throughout the wine industry value chain. The earlier that an accurate yield estimation can be made the greater its value, increasing management options for grape growers and commercial options for winemakers. For the yield estimate based on in- field measurements at scale, the number of inflorescences emerging after bud-burst offers the earliest practical signal, allowing a yield potential to be determined months before harvest. This paper presents an approach to automatically count the inflorescence number at the phenological stage E-L 12 using RGB video data and demonstrates its use for estimating yield. A dataset consisting of RGB videos was collected shortly after bud-burst from multiple vineyards, in conjunction with hand counts to produce a manual ground-truth for the inflorescence counting task. The video frames were annotated using bounding-boxes around the inflorescences to produce a digital ground-truth. A deep learning architecture was developed to learn features from the video frames during training and detect the inflorescences at the later inference stage. The detection results were fed to a tracking pipeline built using computer vision and deep learning techniques to generate numbers of inflorescences present in test videos. The visual and quantitative results are presented and evaluated for the inflorescence detection and counting tasks. The developed inflorescence detector achieves an average precision of 80.00%, a recall of 83.92%, and an F1-score of 80.48%, through a five-fold cross-validation on the annotated dataset. For the test videos, the developed automatic inflorescence counting model reports an absolute error of 11.03 inflorescences per panel, a normalized mean absolute error of 10.80%, and an R2 of 0.86, when the predicted per-panel counts were compared to the corresponding manual ground-truth. Based on the counting results, we estimate an early yield that is within 4% to 11% error when compared to the actual yield after harvest. Based on these results and a separate analysis of the relationship between hand counts of inflorescences and harvest yields in three vineyards over three growing seasons, we conclude that computer vision and machine learning based methods have the potential to provide early yield estimation in viticulture with a commercially viable accuracy.

@article{
 title = {Comparison of deep learning methods for grapevine growth stage recognition},
 type = {article},
 year = {2023},
 keywords = {Computer vision,Deep learning,Grapes growth stages,Viticulture},
 pages = {107944},
 volume = {211},
 month = {8},
 publisher = {Elsevier},
 day = {1},
 id = {a8b05a7d-8e3a-3d41-9f3a-3b5c15032533},
 created = {2023-10-27T07:27:26.204Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:26.358Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Monitoring the phenological development stages of grapes represents a challenge in viticulture. It includes the phenological distinction of the growth stages of grapevines and the continuous technological developments, especially in computer vision, enabling a detailed classification of economically relevant development stages of grapes. In the present work, we show that based on a cascading computer vision approach, the development stages of grapes can be classified and distinguished at the micro level. In a comparative experiment (ResNet, DenseNet, InceptionV3), it could be shown that a ResNet architecture provides the best classification results with an average accuracy of 88.1%.},
 bibtype = {article},
 author = {Schieck, Martin and Krajsic, Philippe and Loos, Felix and Hussein, Abdulbaree and Franczyk, Bogdan and Kozierkiewicz, Adrianna and Pietranik, Marcin},
 doi = {10.1016/J.COMPAG.2023.107944},
 journal = {Computers and Electronics in Agriculture}
}

Monitoring the phenological development stages of grapes represents a challenge in viticulture. It includes the phenological distinction of the growth stages of grapevines and the continuous technological developments, especially in computer vision, enabling a detailed classification of economically relevant development stages of grapes. In the present work, we show that based on a cascading computer vision approach, the development stages of grapes can be classified and distinguished at the micro level. In a comparative experiment (ResNet, DenseNet, InceptionV3), it could be shown that a ResNet architecture provides the best classification results with an average accuracy of 88.1%.

@article{
 title = {Machine-Learning Methods to Identify Key Predictors of Site-Specific Vineyard Yield and Vine Size},
 type = {article},
 year = {2023},
 keywords = {Concord,proximal canopy sensing,random forests},
 pages = {740013},
 volume = {74},
 websites = {https://www.ajevonline.org/content/74/1/0740013,https://www.ajevonline.org/content/74/1/0740013.abstract},
 month = {1},
 publisher = {American Journal of Enology and Viticulture},
 day = {1},
 id = {27268e6e-3611-3319-815d-6a42915c03c8},
 created = {2023-10-27T07:30:00.849Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-27T07:30:06.542Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Background and goals Lake Erie Concord growers have access to high-resolution spatial soil and production data, but lack protocols and information on the optimum time to collect these data. This study examines the type and timing of sensor information to support in-season management.

Methods and key findings A three-year study in a 2.6 ha vineyard collected yield, pruning mass, canopy vigor, and soil data, including yield and pruning mass from the previous year, at 321 sites. Stepwise linear regression and random forest regression approaches were used to model site-specific yield and pruning mass using historical spatial production data, multi-temporal in-season canopy vigor, and soil data. The more complex yield elaboration process was best modelled with non-linear random forest regression, while the simpler development of pruning mass was best modelled by linear regression.

Conclusions and significance Canopy vigor in the weeks preceding bloom was the most important predictor of the current season’s yield and should be used to generate stratified sampling designs for crop estimation at 30 days after bloom. In contrast, pruning mass was not well-predicted by canopy vigor; even late-season canopy vigor, which is widely advocated to estimate pruning mass in viticulture. The previous year’s pruning mass was the dominant predictor of pruning mass in the current season. To model pruning mass going forward, the best approach is to start measuring it. Further work is still needed to develop robust, local site-specific yield and pruning mass models for operational decision-making in Concord vineyards.},
 bibtype = {article},
 author = {Taylor, James A. and Bates, Terence R. and Jakubowski, Rhiann and Jones, Hazaël},
 doi = {10.5344/AJEV.2022.22050},
 journal = {American Journal of Enology and Viticulture},
 number = {1}
}

Background and goals Lake Erie Concord growers have access to high-resolution spatial soil and production data, but lack protocols and information on the optimum time to collect these data. This study examines the type and timing of sensor information to support in-season management. Methods and key findings A three-year study in a 2.6 ha vineyard collected yield, pruning mass, canopy vigor, and soil data, including yield and pruning mass from the previous year, at 321 sites. Stepwise linear regression and random forest regression approaches were used to model site-specific yield and pruning mass using historical spatial production data, multi-temporal in-season canopy vigor, and soil data. The more complex yield elaboration process was best modelled with non-linear random forest regression, while the simpler development of pruning mass was best modelled by linear regression. Conclusions and significance Canopy vigor in the weeks preceding bloom was the most important predictor of the current season’s yield and should be used to generate stratified sampling designs for crop estimation at 30 days after bloom. In contrast, pruning mass was not well-predicted by canopy vigor; even late-season canopy vigor, which is widely advocated to estimate pruning mass in viticulture. The previous year’s pruning mass was the dominant predictor of pruning mass in the current season. To model pruning mass going forward, the best approach is to start measuring it. Further work is still needed to develop robust, local site-specific yield and pruning mass models for operational decision-making in Concord vineyards.

@article{
 title = {Machine-Learning Methods to Identify Key Predictors of Site-Specific Vineyard Yield and Vine Size},
 type = {article},
 year = {2023},
 keywords = {Concord,proximal canopy sensing,random forests},
 pages = {740013},
 volume = {74},
 websites = {https://www.ajevonline.org/content/74/1/0740013,https://www.ajevonline.org/content/74/1/0740013.abstract},
 month = {1},
 publisher = {American Journal of Enology and Viticulture},
 day = {1},
 id = {fb7ccd1e-a381-3cdc-969c-5b989e234133},
 created = {2023-10-27T07:32:15.680Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:26.667Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Background and goals Lake Erie Concord growers have access to high-resolution spatial soil and production data, but lack protocols and information on the optimum time to collect these data. This study examines the type and timing of sensor information to support in-season management.

Methods and key findings A three-year study in a 2.6 ha vineyard collected yield, pruning mass, canopy vigor, and soil data, including yield and pruning mass from the previous year, at 321 sites. Stepwise linear regression and random forest regression approaches were used to model site-specific yield and pruning mass using historical spatial production data, multi-temporal in-season canopy vigor, and soil data. The more complex yield elaboration process was best modelled with non-linear random forest regression, while the simpler development of pruning mass was best modelled by linear regression.

Conclusions and significance Canopy vigor in the weeks preceding bloom was the most important predictor of the current season’s yield and should be used to generate stratified sampling designs for crop estimation at 30 days after bloom. In contrast, pruning mass was not well-predicted by canopy vigor; even late-season canopy vigor, which is widely advocated to estimate pruning mass in viticulture. The previous year’s pruning mass was the dominant predictor of pruning mass in the current season. To model pruning mass going forward, the best approach is to start measuring it. Further work is still needed to develop robust, local site-specific yield and pruning mass models for operational decision-making in Concord vineyards.},
 bibtype = {article},
 author = {Taylor, James A. and Bates, Terence R. and Jakubowski, Rhiann and Jones, Hazaël},
 doi = {10.5344/AJEV.2022.22050},
 journal = {American Journal of Enology and Viticulture},
 number = {1}
}

Background and goals Lake Erie Concord growers have access to high-resolution spatial soil and production data, but lack protocols and information on the optimum time to collect these data. This study examines the type and timing of sensor information to support in-season management. Methods and key findings A three-year study in a 2.6 ha vineyard collected yield, pruning mass, canopy vigor, and soil data, including yield and pruning mass from the previous year, at 321 sites. Stepwise linear regression and random forest regression approaches were used to model site-specific yield and pruning mass using historical spatial production data, multi-temporal in-season canopy vigor, and soil data. The more complex yield elaboration process was best modelled with non-linear random forest regression, while the simpler development of pruning mass was best modelled by linear regression. Conclusions and significance Canopy vigor in the weeks preceding bloom was the most important predictor of the current season’s yield and should be used to generate stratified sampling designs for crop estimation at 30 days after bloom. In contrast, pruning mass was not well-predicted by canopy vigor; even late-season canopy vigor, which is widely advocated to estimate pruning mass in viticulture. The previous year’s pruning mass was the dominant predictor of pruning mass in the current season. To model pruning mass going forward, the best approach is to start measuring it. Further work is still needed to develop robust, local site-specific yield and pruning mass models for operational decision-making in Concord vineyards.

@article{
 title = {Assessment of vineyard vigour and yield spatio-temporal variability based on UAV high resolution multispectral images},
 type = {article},
 year = {2023},
 keywords = {Precision viticulture,Shoot pruning weight,Vegetation index},
 pages = {36-56},
 volume = {231},
 month = {7},
 publisher = {Academic Press},
 day = {1},
 id = {963e61b2-4599-34fa-a294-15fd9d6eae60},
 created = {2023-10-27T07:33:21.907Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:26.806Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Accurate, timely assessment of the vineyard on a field scale is essential for successful grape yield and quality. Remote sensing can be an effective and useful monitoring tool, as data from sensors on board Unmanned Aerial Vehicles (UAV) can measure vegetative and reproductive growth and thus directly or indirectly detect variability. Through the images obtained from UAV, the Vegetation Indices (VIs) can be calculated and compared with various agronomic characteristics of the vineyard. The objective of this study was to evaluate the multispectral response of the vineyard in three specific phenological phases and to analyse the spatial distribution of vegetative vigour. A multirotor UAV equipped with a camera featuring multispectral sensors was used. Four VIs namely Normalised Difference Vegetation Index (NDVI), Normalised Difference Red Edge (NDRE), Green Normalised Difference Vegetation Index (GNDVI), Modified Soil Adjusted Vegetation Index (MSAVI), were calculated using the georeferenced orthomosaic UAV images. Computer vision techniques were used to segment these orthoimages to extract only the vegetation canopy pixels. High level of agronomic variability within the vineyard was identified. Pearson's coefficient showed a significant correlation between NDVI and NDRE indices and yield since early phenological stages (r = 0.80 and 0.72 respectively), GNDVI at grape ripening (r = 0.83). Shoot pruning weight (SPW) shows the highest values of correlation (r = 0.84) with NDVI during the phenological stage of berries pea size. Simple linear regression techniques were evaluated using VIs as predictors of the SPW, and accurate predictive results were obtained for NDVI and NDRE with RMSE values of 0.18 and 0.24, respectively. Geostatistical analysis was applied to model the spatial variability of SPW, and thus vineyard vigour. Assessing spatial variability and appreciating the level of vigour enables improved vineyard management by increasing sustainability and production efficiency.},
 bibtype = {article},
 author = {Ferro, Massimo V. and Catania, Pietro and Miccichè, Daniele and Pisciotta, Antonino and Vallone, Mariangela and Orlando, Santo},
 doi = {10.1016/J.BIOSYSTEMSENG.2023.06.001},
 journal = {Biosystems Engineering}
}

Accurate, timely assessment of the vineyard on a field scale is essential for successful grape yield and quality. Remote sensing can be an effective and useful monitoring tool, as data from sensors on board Unmanned Aerial Vehicles (UAV) can measure vegetative and reproductive growth and thus directly or indirectly detect variability. Through the images obtained from UAV, the Vegetation Indices (VIs) can be calculated and compared with various agronomic characteristics of the vineyard. The objective of this study was to evaluate the multispectral response of the vineyard in three specific phenological phases and to analyse the spatial distribution of vegetative vigour. A multirotor UAV equipped with a camera featuring multispectral sensors was used. Four VIs namely Normalised Difference Vegetation Index (NDVI), Normalised Difference Red Edge (NDRE), Green Normalised Difference Vegetation Index (GNDVI), Modified Soil Adjusted Vegetation Index (MSAVI), were calculated using the georeferenced orthomosaic UAV images. Computer vision techniques were used to segment these orthoimages to extract only the vegetation canopy pixels. High level of agronomic variability within the vineyard was identified. Pearson's coefficient showed a significant correlation between NDVI and NDRE indices and yield since early phenological stages (r = 0.80 and 0.72 respectively), GNDVI at grape ripening (r = 0.83). Shoot pruning weight (SPW) shows the highest values of correlation (r = 0.84) with NDVI during the phenological stage of berries pea size. Simple linear regression techniques were evaluated using VIs as predictors of the SPW, and accurate predictive results were obtained for NDVI and NDRE with RMSE values of 0.18 and 0.24, respectively. Geostatistical analysis was applied to model the spatial variability of SPW, and thus vineyard vigour. Assessing spatial variability and appreciating the level of vigour enables improved vineyard management by increasing sustainability and production efficiency.

@article{
 title = {Deep Learning for Post-Harvest Grape Diseases Detection},
 type = {article},
 year = {2023},
 keywords = {classification,deep learning,fruit grading,grape disease,segmentation},
 volume = {0},
 id = {5c818afd-6b85-37db-85a3-09b23a3d471c},
 created = {2023-10-27T07:38:16.458Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-22T17:08:42.431Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Post-harvest fruit grading is a necessary step to avoid disease related loss in quality. This is relevant in the context of the Champagne industry where grapes can not be manipulated by machines to avoid crushing. Our team have been developing a computer vision based solution to automate this process. In this paper, our main contribution is the usage of a PSPnet segmentation model for real-time visible symptoms detection with a IoU score of 58%. The associated classification score reach 95%, which improved our previous work. We also study a MobileNet-V2 model’s ability to discriminate between different grape diseases in ideal condition.},
 bibtype = {article},
 author = {Mohimont, Lucas},
 doi = {10.3233/aise230025}
}

Post-harvest fruit grading is a necessary step to avoid disease related loss in quality. This is relevant in the context of the Champagne industry where grapes can not be manipulated by machines to avoid crushing. Our team have been developing a computer vision based solution to automate this process. In this paper, our main contribution is the usage of a PSPnet segmentation model for real-time visible symptoms detection with a IoU score of 58%. The associated classification score reach 95%, which improved our previous work. We also study a MobileNet-V2 model’s ability to discriminate between different grape diseases in ideal condition.

@article{
 title = {International Journal of INTELLIGENT SYSTEMS AND APPLICATIONS IN ENGINEERING Grape Vision: A CNN-Based System for Yield Component Analysis of Grape Clusters},
 type = {article},
 year = {2023},
 keywords = {CNN,Grape prediction,image processing,machine learning},
 pages = {239-244},
 volume = {2023},
 websites = {https://orcid.org/0000-0003-1814-591,},
 id = {245f990a-28b0-35d3-911c-59ec6aeb0006},
 created = {2023-10-27T07:43:19.741Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:27.138Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {The agricultural industry is adopting advanced technologies and applications like yield prediction, precision agriculture, and automated harvesting to enhance production and quality. Machine learning (ML) and computer vision are increasingly used for fruit detection, segmentation, and counting. Specifically, the use of Convolutional Neural Networks (CNN) in grape yield prediction and quality assessment is gaining popularity due to its high accuracy and cost efficiency. Additionally, a new methodology ba sed on image analysis has been developed for fast and inexpensive cluster yield component determination in the wine and table grape industry.},
 bibtype = {article},
 author = {Dange, B J and Kumar Mishra, Punit and Metre, Kalpana V and Gore, Santosh and Laxnamrao Kurkute, Sanjay and Khodke, H E and Gore, Sujata},
 journal = {Original Research Paper International Journal of Intelligent Systems and Applications in Engineering IJISAE},
 number = {9s}
}

The agricultural industry is adopting advanced technologies and applications like yield prediction, precision agriculture, and automated harvesting to enhance production and quality. Machine learning (ML) and computer vision are increasingly used for fruit detection, segmentation, and counting. Specifically, the use of Convolutional Neural Networks (CNN) in grape yield prediction and quality assessment is gaining popularity due to its high accuracy and cost efficiency. Additionally, a new methodology ba sed on image analysis has been developed for fast and inexpensive cluster yield component determination in the wine and table grape industry.

@article{
 title = {A Preliminary Method for Tracking In-Season Grapevine Cluster Closure Using Image Segmentation and Image Thresholding},
 type = {article},
 year = {2023},
 websites = {https://doi.org/10.1155/2023/3923839},
 id = {04103f3c-63de-3134-8a21-dda81d680a7e},
 created = {2023-10-27T07:46:25.965Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:27.794Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Mapping and monitoring cluster morphology provides insights for disease risk assessment, quality control in wine production, and understanding environmental innuences on cluster shape. During the progression of grapevine morphology, cluster closure (CC) (also called bunch closure) is the stage when berries touch one another. .is study used mobile phone images to develop a direct quantiication method for tracking CC in three grapevine cultivars (Riesling, Pinot gris, and Cabernet Franc). A total of 809 cluster images from fruit set to veraison were analyzed using two image segmentation methods: (i) a Pyramid Scene Parsing Network (PSPNet) to extract cluster boundaries and (ii) Otsu's image thresholding method to calculate % CC based on gaps between the berries. PSPNet produced high accuracy (mean accuracy = 0.98, mean intersection over union (mIoU) = 0.95) with mIoU > 0.90 for both cluster and noncluster classes. Otsu's thresholding method resulted in <2% falsely classiied gap and berry pixels aaecting quantiied % CC. .e progression of CC was described using basic statistics (mean and standard deviation) and using a curve et. .e CC curve showed an asymptotic trend, with a higher rate of progression observed in the erst three weeks, followed by a gradual approach towards an asymptote. We propose that the X value (in this example, number of weeks past berry set) at which the CC progression curve reaches the asymptote be considered as the oocial phenological stage of CC. .e developed method provides a continuous scale of CC throughout the season, potentially serving as a valuable open-source research tool for studying grapevine cluster phenology and factors aaecting CC.},
 bibtype = {article},
 author = {Trivedi, Manushi and Zhou, Yuwei and Moon, Jonathan Hyun and Meyers, James and Jiang, Yu and Lu, Guoyu and Heuvel, Justine Vanden},
 doi = {10.1155/2023/3923839}
}

Mapping and monitoring cluster morphology provides insights for disease risk assessment, quality control in wine production, and understanding environmental innuences on cluster shape. During the progression of grapevine morphology, cluster closure (CC) (also called bunch closure) is the stage when berries touch one another. .is study used mobile phone images to develop a direct quantiication method for tracking CC in three grapevine cultivars (Riesling, Pinot gris, and Cabernet Franc). A total of 809 cluster images from fruit set to veraison were analyzed using two image segmentation methods: (i) a Pyramid Scene Parsing Network (PSPNet) to extract cluster boundaries and (ii) Otsu's image thresholding method to calculate % CC based on gaps between the berries. PSPNet produced high accuracy (mean accuracy = 0.98, mean intersection over union (mIoU) = 0.95) with mIoU > 0.90 for both cluster and noncluster classes. Otsu's thresholding method resulted in <2% falsely classiied gap and berry pixels aaecting quantiied % CC. .e progression of CC was described using basic statistics (mean and standard deviation) and using a curve et. .e CC curve showed an asymptotic trend, with a higher rate of progression observed in the erst three weeks, followed by a gradual approach towards an asymptote. We propose that the X value (in this example, number of weeks past berry set) at which the CC progression curve reaches the asymptote be considered as the oocial phenological stage of CC. .e developed method provides a continuous scale of CC throughout the season, potentially serving as a valuable open-source research tool for studying grapevine cluster phenology and factors aaecting CC.

@article{
 title = {Evaluating Critical Disease Occurrence in Grapevine Leaves using CNN: Use-Case in Eastern Europe},
 type = {article},
 year = {2023},
 keywords = {oprea2023evaluatingcriticaldisease},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {382c45fa-abf6-3100-9f8d-38a85ca33118},
 created = {2023-10-27T07:47:55.667Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:06.161Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Convolutional Neural Networks are Deep Learning algorithms for image classification tasks in the Computer Vision area. Their efficiency was previously evaluated in medical areas, engineering fields and construction applications. Under this category, VGG16 and Avert-CNN (a modified VGG16 version) algorithms can perform real-time identification for diseases occurrence in agricultural plants with high accuracy and fast estimation time. Thus, this research addresses the identification of health status of grapevine leaves using specialized classification algorithms on images taken from PlantVillage dataset and images acquired in a vineyard in the South-East of Romania. The outcome of these classifications consists in predictions based on one of the 5 classes for which the convolutional network was trained, along with a prediction accuracy metric. The classes considered in this process correspond to the state of a healthy plant and a set of 4 distinct diseases that can affect the vine: Black rot, Esca, Leaf blight and Powdery mildew. Based on the achieved results, a novel convolutional neural network architecture is proposed to ensure reliable estimates on the disease's probability of occurrence. Its efficiency in reaching over 94 % prediction accuracy is demonstrated compared to the classic VGG16 which leads to 90.21 % data accuracy and surpasses Random Forest and Support Vector Machine algorithms that achieve 72.87% and 85.82% accuracy, respectively.},
 bibtype = {article},
 author = {Oprea, Cristina Claudia and Dragulinescu, Ana Maria Claudia and Marcu, Ioana Manuela and Pirnog, Ionut},
 doi = {10.1109/EMES58375.2023.10171678},
 journal = {2023 17th International Conference on Engineering of Modern Electric Systems, EMES 2023}
}

Convolutional Neural Networks are Deep Learning algorithms for image classification tasks in the Computer Vision area. Their efficiency was previously evaluated in medical areas, engineering fields and construction applications. Under this category, VGG16 and Avert-CNN (a modified VGG16 version) algorithms can perform real-time identification for diseases occurrence in agricultural plants with high accuracy and fast estimation time. Thus, this research addresses the identification of health status of grapevine leaves using specialized classification algorithms on images taken from PlantVillage dataset and images acquired in a vineyard in the South-East of Romania. The outcome of these classifications consists in predictions based on one of the 5 classes for which the convolutional network was trained, along with a prediction accuracy metric. The classes considered in this process correspond to the state of a healthy plant and a set of 4 distinct diseases that can affect the vine: Black rot, Esca, Leaf blight and Powdery mildew. Based on the achieved results, a novel convolutional neural network architecture is proposed to ensure reliable estimates on the disease's probability of occurrence. Its efficiency in reaching over 94 % prediction accuracy is demonstrated compared to the classic VGG16 which leads to 90.21 % data accuracy and surpasses Random Forest and Support Vector Machine algorithms that achieve 72.87% and 85.82% accuracy, respectively.

@article{
 title = {Benchmarking edge computing devices for grape bunches and trunks detection using accelerated object detection single shot multibox deep learning models},
 type = {article},
 year = {2023},
 keywords = {Embedded systems,Heterogeneous platforms,Object detection,RetinaNet resNet,SSD resNet},
 pages = {105604},
 volume = {117},
 month = {1},
 publisher = {Pergamon},
 day = {1},
 id = {7439a1c4-a827-3a2c-b671-870ad812b6b1},
 created = {2023-10-27T07:49:54.635Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:28.084Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Purpose: Visual perception enables robots to perceive the environment. Visual data is processed using computer vision algorithms that are usually time-expensive and require powerful devices to process the visual data in real-time, which is unfeasible for open-field robots with limited energy. This work benchmarks the performance of different heterogeneous platforms for object detection in real-time. This research benchmarks three architectures: embedded GPU—Graphical Processing Units (such as NVIDIA Jetson Nano 2GB and 4GB, and NVIDIA Jetson TX2), TPU—Tensor Processing Unit (such as Coral Dev Board TPU), and DPU—Deep Learning Processor Unit (such as in AMD/Xilinx ZCU104 Development Board, and AMD/Xilinx Kria KV260 Starter Kit). Methods: The authors used the RetinaNet ResNet-50 fine-tuned using the natural VineSet dataset. After the trained model was converted and compiled for target-specific hardware formats to improve the execution efficiency. Conclusions and Results: The platforms were assessed in terms of performance of the evaluation metrics and efficiency (time of inference). Graphical Processing Units (GPUs) were the slowest devices, running at 3FPS to 5FPS, and Field Programmable Gate Arrays (FPGAs) were the fastest devices, running at 14FPS to 25FPS. The efficiency of the Tensor Processing Unit (TPU) is irrelevant and similar to NVIDIA Jetson TX2. TPU and GPU are the most power-efficient, consuming about 5W. The performance differences, in the evaluation metrics, across devices are irrelevant and have an F1 of about 70% and mean Average Precision (mAP) of about 60%.},
 bibtype = {article},
 author = {Magalhães, Sandro Costa and dos Santos, Filipe Neves and Machado, Pedro and Moreira, António Paulo and Dias, Jorge},
 doi = {10.1016/J.ENGAPPAI.2022.105604},
 journal = {Engineering Applications of Artificial Intelligence}
}

Purpose: Visual perception enables robots to perceive the environment. Visual data is processed using computer vision algorithms that are usually time-expensive and require powerful devices to process the visual data in real-time, which is unfeasible for open-field robots with limited energy. This work benchmarks the performance of different heterogeneous platforms for object detection in real-time. This research benchmarks three architectures: embedded GPU—Graphical Processing Units (such as NVIDIA Jetson Nano 2GB and 4GB, and NVIDIA Jetson TX2), TPU—Tensor Processing Unit (such as Coral Dev Board TPU), and DPU—Deep Learning Processor Unit (such as in AMD/Xilinx ZCU104 Development Board, and AMD/Xilinx Kria KV260 Starter Kit). Methods: The authors used the RetinaNet ResNet-50 fine-tuned using the natural VineSet dataset. After the trained model was converted and compiled for target-specific hardware formats to improve the execution efficiency. Conclusions and Results: The platforms were assessed in terms of performance of the evaluation metrics and efficiency (time of inference). Graphical Processing Units (GPUs) were the slowest devices, running at 3FPS to 5FPS, and Field Programmable Gate Arrays (FPGAs) were the fastest devices, running at 14FPS to 25FPS. The efficiency of the Tensor Processing Unit (TPU) is irrelevant and similar to NVIDIA Jetson TX2. TPU and GPU are the most power-efficient, consuming about 5W. The performance differences, in the evaluation metrics, across devices are irrelevant and have an F1 of about 70% and mean Average Precision (mAP) of about 60%.

@article{
 title = {Using deep learning for pruning region detection and plant organ segmentation in dormant spur-pruned grapevines},
 type = {article},
 year = {2023},
 keywords = {Computer vision,Object detection,Robotics,Viticulture,Winter pruning},
 pages = {1547-1569},
 volume = {24},
 websites = {https://link.springer.com/article/10.1007/s11119-023-10006-y},
 month = {8},
 publisher = {Springer},
 day = {1},
 id = {2f305d7b-5281-3ddd-902e-68d989073a65},
 created = {2023-10-27T07:58:34.517Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:27.451Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Even though mechanization has dramatically decreased labor requirements, vineyard management costs are still affected by selective operations such as winter pruning. Robotic solutions are becoming more common in agriculture, however, few studies have focused on grapevines. This work aims at fine-tuning and testing two different deep neural networks for: (i) detecting pruning regions (PRs), and (ii) performing organ segmentation of spur-pruned dormant grapevines. The Faster R-CNN network was fine-tuned using 1215 RGB images collected in different vineyards and annotated through bounding boxes. The network was tested on 232 RGB images, PRs were categorized by wood type (W), orientation (Or) and visibility (V), and performance metrics were calculated. PR detection was dramatically affected by visibility. Highest detection was associated with visible intermediate complex spurs in Merlot (0.97), while most represented coplanar simple spurs allowed a 74% detection rate. The Mask R-CNN network was trained for grapevine organs (GOs) segmentation by using 119 RGB images annotated by distinguishing 5 classes (cordon, arm, spur, cane and node). The network was tested on 60 RGB images of light pruned (LP), shoot-thinned (ST) and unthinned control (C) grapevines. Nodes were the best segmented GOs (0.88) and general recall was higher for ST (0.85) compared to C (0.80) confirming the role of canopy management in improving performances of hi-tech solutions based on artificial intelligence. The two fine-tuned and tested networks are part of a larger control framework that is under development for autonomous winter pruning of grapevines.},
 bibtype = {article},
 author = {Guadagna, P. and Fernandes, M. and Chen, F. and Santamaria, A. and Teng, T. and Frioni, T. and Caldwell, D. G. and Poni, S. and Semini, C. and Gatti, M.},
 doi = {10.1007/S11119-023-10006-Y/TABLES/9},
 journal = {Precision Agriculture},
 number = {4}
}

Even though mechanization has dramatically decreased labor requirements, vineyard management costs are still affected by selective operations such as winter pruning. Robotic solutions are becoming more common in agriculture, however, few studies have focused on grapevines. This work aims at fine-tuning and testing two different deep neural networks for: (i) detecting pruning regions (PRs), and (ii) performing organ segmentation of spur-pruned dormant grapevines. The Faster R-CNN network was fine-tuned using 1215 RGB images collected in different vineyards and annotated through bounding boxes. The network was tested on 232 RGB images, PRs were categorized by wood type (W), orientation (Or) and visibility (V), and performance metrics were calculated. PR detection was dramatically affected by visibility. Highest detection was associated with visible intermediate complex spurs in Merlot (0.97), while most represented coplanar simple spurs allowed a 74% detection rate. The Mask R-CNN network was trained for grapevine organs (GOs) segmentation by using 119 RGB images annotated by distinguishing 5 classes (cordon, arm, spur, cane and node). The network was tested on 60 RGB images of light pruned (LP), shoot-thinned (ST) and unthinned control (C) grapevines. Nodes were the best segmented GOs (0.88) and general recall was higher for ST (0.85) compared to C (0.80) confirming the role of canopy management in improving performances of hi-tech solutions based on artificial intelligence. The two fine-tuned and tested networks are part of a larger control framework that is under development for autonomous winter pruning of grapevines.

@article{
 title = {Automated Grapevine Inflorescence Counting in a Vineyard Using Deep Learning and Multi-object Tracking},
 type = {article},
 year = {2023},
 keywords = {deep learning,high-throughput phenotyping,instance segmentation,multi-object tracking,precision viticulture},
 pages = {276-280},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {fa897f4e-6961-34c8-a524-6cbfa9e1e48e},
 created = {2023-10-27T08:04:57.476Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:27.940Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {To adjust management practices and improve wine marketing strategies, accurate vineyard yield estimation early in the growing season is essential. Conventional methods for yield forecasting rely on phenotypic features' manual assessment, which is time- and labor-intensive and often destructive. We combined a deep object segmentation method, mask region-based convolutional neural network (Mask R-CNN), with two potential multi-object tracking algorithms, simple online and real-time tracking (SORT) and intersection-over-union (IOU) trackers to develop a complete visual system that can automatically detect and track individual inflorescences, enabling the assessment of the number of inflorescences per vineyard row from vineyard video footage. The performance of the two tracking algorithms was evaluated using our vineyard dataset, which is more challenging than conventional tracking benchmark datasets owing to environmental factors. Our evaluation dataset consists of videos of four vineyard rows, including 221 vines that were automatically acquired under unprepared field conditions. We tracked individual inflorescences across video image frames with a 92.1% multi-object tracking accuracy (MOTA) and an 89.6% identity F1 score (IDF1). This allowed us to estimate inflorescence count per vineyard row with a 0.91 coefficient of determination (R2) between the estimated count and manual-annotated ground truth count. The impact of leaf occlusions on inflorescence visibility was lessened by processing multiple successive image frames with minimal displacements to construct multiple camera views. This study demonstrates the use of deep learning and multi-object tracking in creating a low-cost (requiring only an RGB camera), high-throughput phenotyping system for precision viticulture.},
 bibtype = {article},
 author = {Rahim, Umme Fawzia and Utsumi, Tomoyoshi and Iwaki, Yohei and Mineno, Hiroshi},
 doi = {10.1109/ICCAE56788.2023.10111243},
 journal = {2023 15th International Conference on Computer and Automation Engineering, ICCAE 2023}
}

To adjust management practices and improve wine marketing strategies, accurate vineyard yield estimation early in the growing season is essential. Conventional methods for yield forecasting rely on phenotypic features' manual assessment, which is time- and labor-intensive and often destructive. We combined a deep object segmentation method, mask region-based convolutional neural network (Mask R-CNN), with two potential multi-object tracking algorithms, simple online and real-time tracking (SORT) and intersection-over-union (IOU) trackers to develop a complete visual system that can automatically detect and track individual inflorescences, enabling the assessment of the number of inflorescences per vineyard row from vineyard video footage. The performance of the two tracking algorithms was evaluated using our vineyard dataset, which is more challenging than conventional tracking benchmark datasets owing to environmental factors. Our evaluation dataset consists of videos of four vineyard rows, including 221 vines that were automatically acquired under unprepared field conditions. We tracked individual inflorescences across video image frames with a 92.1% multi-object tracking accuracy (MOTA) and an 89.6% identity F1 score (IDF1). This allowed us to estimate inflorescence count per vineyard row with a 0.91 coefficient of determination (R2) between the estimated count and manual-annotated ground truth count. The impact of leaf occlusions on inflorescence visibility was lessened by processing multiple successive image frames with minimal displacements to construct multiple camera views. This study demonstrates the use of deep learning and multi-object tracking in creating a low-cost (requiring only an RGB camera), high-throughput phenotyping system for precision viticulture.

@article{
 title = {YOLO-based Multi-Modal Analysis of Vineyards using RGB-D Detections},
 type = {article},
 year = {2023},
 keywords = {RGB-D camera,RGB-D fusion,multi-modal dataset,multi-spectral camera,object detection,vineyard analysis,viticultural robotics},
 pages = {7-9},
 websites = {https://hal.science/hal-04218442,https://hal.science/hal-04218442/document},
 month = {6},
 day = {7},
 id = {91725ef7-c1e0-3717-a2c8-2634de9c8d45},
 created = {2023-10-27T08:07:32.715Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:27.627Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Agricultural robotics is a rapidly growing research area due to the need for new practices that are more environmentally responsible. It involves a range of technologies including autonomous vehicles, drones and robotic arms. These systems can be equipped with sensors and cameras to gather data and perform tasks autonomously or with minimal human intervention. For robot navigation and manipulation, and plant monitoring and analysis, perception is of prime importance and is still a challenging task today. For instance, visual perception using color images only for disease detection in vineyards, such as Mildew in which the symptoms manifest as small spots on or beneath the leaves, is still a hard task that does not allow to achieve high detection accuracy. To extract more representative features to improve the detection accuracy, other modalities must be used in addition to the Red Green and Blue (RGB) information of color images. In this paper, we present first a multimodal acquisition system that we have developed. It is composed of a multi-spectral (MS) camera and an RGB-D camera that are mounted on a mobile robot for data acquisition in a vineyard. Next, we describe the multi-modal dataset that we have built based on the data acquired with our system in a commercial vineyard. Finally, we implemented an Early RGB and depth data fusion technique together with the YOLOv5m Deep Learning network to detect the main parts of the vine: leaves, branches, and grapes using our dataset. The results that we have obtained, compared to those obtained using RGB images only with the YOLOv5m architecture, demonstrate the benefits of adding multi data fusion techniques to the object detection pipeline. These results are encouraging and show that multi-sensor data fusion is a technique that is worth considering as it can be useful for improving grapevine disease recognition technologies.},
 bibtype = {article},
 author = {Clamens, T and Rodriguez, J and Delamare, M and Lew-Yan-Voon, L and Fauvet, E and Fofi, D}
}

Agricultural robotics is a rapidly growing research area due to the need for new practices that are more environmentally responsible. It involves a range of technologies including autonomous vehicles, drones and robotic arms. These systems can be equipped with sensors and cameras to gather data and perform tasks autonomously or with minimal human intervention. For robot navigation and manipulation, and plant monitoring and analysis, perception is of prime importance and is still a challenging task today. For instance, visual perception using color images only for disease detection in vineyards, such as Mildew in which the symptoms manifest as small spots on or beneath the leaves, is still a hard task that does not allow to achieve high detection accuracy. To extract more representative features to improve the detection accuracy, other modalities must be used in addition to the Red Green and Blue (RGB) information of color images. In this paper, we present first a multimodal acquisition system that we have developed. It is composed of a multi-spectral (MS) camera and an RGB-D camera that are mounted on a mobile robot for data acquisition in a vineyard. Next, we describe the multi-modal dataset that we have built based on the data acquired with our system in a commercial vineyard. Finally, we implemented an Early RGB and depth data fusion technique together with the YOLOv5m Deep Learning network to detect the main parts of the vine: leaves, branches, and grapes using our dataset. The results that we have obtained, compared to those obtained using RGB images only with the YOLOv5m architecture, demonstrate the benefits of adding multi data fusion techniques to the object detection pipeline. These results are encouraging and show that multi-sensor data fusion is a technique that is worth considering as it can be useful for improving grapevine disease recognition technologies.

@article{
 title = {Localization of Mobile Manipulator in Vineyards for Autonomous Task Execution},
 type = {article},
 year = {2023},
 keywords = {autonomous mobile manipulator,localization,protective spraying,suckering,surrounding awareness,vineyards,viticulture},
 pages = {414},
 volume = {11},
 websites = {https://www.mdpi.com/2075-1702/11/4/414/htm,https://www.mdpi.com/2075-1702/11/4/414},
 month = {4},
 publisher = {MDPI},
 day = {1},
 id = {1fd176f0-39db-3a32-945f-78a35fd7755f},
 created = {2023-10-27T08:09:47.555Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-16T09:07:15.237Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4c7c81ce-f24b-44ae-bc2a-bf60600a3a24,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Although robotic systems have found their place in agriculture, there are still many challenges, especially in the area of localization in semi-structured environments. A robotic system has been developed and tested to perform various tasks in the steep vineyards of the Mediterranean region. In this paper, we describe a method for vine trunk localization, based solely on the visual recognition of vine trunks by neural networks fed by an RGB camera. Assuming that the height of the first wire in the vineyard is known, the proposed method is used to determine the location of vines in the immediate vicinity of the all-terrain mobile manipulator—ATMM-VIV—needed for spraying and bud suckering. The experiment was conducted in a slightly inclined vineyard to evaluate the proposed localization method.},
 bibtype = {article},
 author = {Hrabar, Ivan and Kovačić, Zdenko},
 doi = {10.3390/MACHINES11040414/S1},
 journal = {Machines},
 number = {4}
}

Although robotic systems have found their place in agriculture, there are still many challenges, especially in the area of localization in semi-structured environments. A robotic system has been developed and tested to perform various tasks in the steep vineyards of the Mediterranean region. In this paper, we describe a method for vine trunk localization, based solely on the visual recognition of vine trunks by neural networks fed by an RGB camera. Assuming that the height of the first wire in the vineyard is known, the proposed method is used to determine the location of vines in the immediate vicinity of the all-terrain mobile manipulator—ATMM-VIV—needed for spraying and bud suckering. The experiment was conducted in a slightly inclined vineyard to evaluate the proposed localization method.

@article{
 title = {Missing Plant Detection in Vineyards Using UAV Angled RGB Imagery Acquired in Dormant Period},
 type = {article},
 year = {2023},
 keywords = {UAV,missing plant detection,photogrammetry,point cloud,precision agriculture,vineyard},
 pages = {349},
 volume = {7},
 websites = {https://www.mdpi.com/2504-446X/7/6/349/htm,https://www.mdpi.com/2504-446X/7/6/349},
 month = {5},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {26},
 id = {48edb51a-87ab-3834-8241-2e0e44c28f1b},
 created = {2023-10-27T08:14:28.558Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:27.291Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Since 2010, more and more farmers have been using remote sensing data from unmanned aerial vehicles, which have a high spatial–temporal resolution, to determine the status of their crops and how their fields change. Imaging sensors, such as multispectral and RGB cameras, are the most widely used tool in vineyards to characterize the vegetative development of the canopy and detect the presence of missing vines along the rows. In this study, the authors propose different approaches to identify and locate each vine within a commercial vineyard using angled RGB images acquired during winter in the dormant period (without canopy leaves), thus minimizing any disturbance to the agronomic practices commonly conducted in the vegetative period. Using a combination of photogrammetric techniques and spatial analysis tools, a workflow was developed to extract each post and vine trunk from a dense point cloud and then assess the number and position of missing vines with high precision. In order to correctly identify the vines and missing vines, the performance of four methods was evaluated, and the best performing one achieved 95.10% precision and 92.72% overall accuracy. The results confirm that the methodology developed represents an effective support in the decision-making processes for the correct management of missing vines, which is essential for preserving a vineyard’s productive capacity and, more importantly, to ensure the farmer’s economic return.},
 bibtype = {article},
 author = {Di Gennaro, Salvatore Filippo and Vannini, Gian Luca and Berton, Andrea and Dainelli, Riccardo and Toscano, Piero and Matese, Alessandro},
 doi = {10.3390/DRONES7060349},
 journal = {Drones 2023, Vol. 7, Page 349},
 number = {6}
}

Since 2010, more and more farmers have been using remote sensing data from unmanned aerial vehicles, which have a high spatial–temporal resolution, to determine the status of their crops and how their fields change. Imaging sensors, such as multispectral and RGB cameras, are the most widely used tool in vineyards to characterize the vegetative development of the canopy and detect the presence of missing vines along the rows. In this study, the authors propose different approaches to identify and locate each vine within a commercial vineyard using angled RGB images acquired during winter in the dormant period (without canopy leaves), thus minimizing any disturbance to the agronomic practices commonly conducted in the vegetative period. Using a combination of photogrammetric techniques and spatial analysis tools, a workflow was developed to extract each post and vine trunk from a dense point cloud and then assess the number and position of missing vines with high precision. In order to correctly identify the vines and missing vines, the performance of four methods was evaluated, and the best performing one achieved 95.10% precision and 92.72% overall accuracy. The results confirm that the methodology developed represents an effective support in the decision-making processes for the correct management of missing vines, which is essential for preserving a vineyard’s productive capacity and, more importantly, to ensure the farmer’s economic return.

@article{
 title = {Disease detection in Okra plant and Grape vein using image processing},
 type = {article},
 year = {2023},
 keywords = {Convolution Neural Network (CNN),Deep Learning,Grape vein disease,Internet of Things (IoT),Pesticide Prediction,Tensorflow,okra plant disease},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {ca67c5a8-3430-35f3-b0ee-fb9b21937f77},
 created = {2023-10-27T08:18:15.566Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-22T17:08:42.107Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The control of early-stage disease in plants is an essential factor in agriculture. Identification of disease in plants at an early stage helps farmers to reduce the usage of pesticides and avoid economic losses. This also in turn helps in promoting high quality yield production. Convolutional neural networks (CNNs) are deep learning algorithms that is applied for high resolution image recognition. This study uses a deep convolution neural network algorithm to detect and classify plant diseases. RESNET50 network has been applied to improve the effectiveness. Tensor Flow algorithm is utilized for coding the CNN algorithm and for accurate classification of the disease in grape and okra plant leaf. The dataset employed comprises of 6 classes and includes 2500 images. Simulation results for the developed model had achieved an accuracy of 95.1% in training and 91.2% in validation class tests.},
 bibtype = {article},
 author = {Kavitha, R. and Harini, S. Sree and Akshatha, K.},
 doi = {10.1109/ICAECA56562.2023.10200150},
 journal = {2nd International Conference on Advancements in Electrical, Electronics, Communication, Computing and Automation, ICAECA 2023}
}

The control of early-stage disease in plants is an essential factor in agriculture. Identification of disease in plants at an early stage helps farmers to reduce the usage of pesticides and avoid economic losses. This also in turn helps in promoting high quality yield production. Convolutional neural networks (CNNs) are deep learning algorithms that is applied for high resolution image recognition. This study uses a deep convolution neural network algorithm to detect and classify plant diseases. RESNET50 network has been applied to improve the effectiveness. Tensor Flow algorithm is utilized for coding the CNN algorithm and for accurate classification of the disease in grape and okra plant leaf. The dataset employed comprises of 6 classes and includes 2500 images. Simulation results for the developed model had achieved an accuracy of 95.1% in training and 91.2% in validation class tests.

@article{
 title = {Autonomous Navigation and Crop Row Detection in Vineyards Using Machine Vision with 2D Camera},
 type = {article},
 year = {2023},
 keywords = {autonomous navigation,machine vision,vineyard navigation},
 pages = {309-326},
 volume = {4},
 websites = {https://www.mdpi.com/2673-4052/4/4/18/htm,https://www.mdpi.com/2673-4052/4/4/18},
 month = {9},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {24},
 id = {035c8c97-3afb-3397-93cc-1f474aba475b},
 created = {2023-10-27T08:19:51.145Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-16T09:07:16.694Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4c7c81ce-f24b-44ae-bc2a-bf60600a3a24,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {In order to improve agriculture productivity, autonomous navigation algorithms are being developed so that robots can navigate along agricultural environments to automatize tasks that are currently performed by hand. This work uses machine vision techniques such as the Otsu’s method, blob detection, and pixel counting to detect the center of the row. Additionally, a commutable control is implemented to autonomously navigate a vineyard. Experimental trials were conducted in an actual vineyard to validate the algorithm. In these trials show that the algorithm can successfully guide the robot through the row without any collisions. This algorithm offers a computationally efficient solution for vineyard row navigation, employing a 2D camera and the Otsu’s thresholding technique to ensure collision-free operation.},
 bibtype = {article},
 author = {Mendez, E and Camacho, Piña and Escobedo Cabello, J ; and Gómez-Espinosa, J A ; and Autonomous, A and Mendez, Enrico and Camacho, Javier Piña and Arturo, Jesús and Cabello, Escobedo and Gómez-Espinosa, Alfonso},
 doi = {10.3390/AUTOMATION4040018},
 journal = {Automation 2023, Vol. 4, Pages 309-326},
 number = {4}
}

In order to improve agriculture productivity, autonomous navigation algorithms are being developed so that robots can navigate along agricultural environments to automatize tasks that are currently performed by hand. This work uses machine vision techniques such as the Otsu’s method, blob detection, and pixel counting to detect the center of the row. Additionally, a commutable control is implemented to autonomously navigate a vineyard. Experimental trials were conducted in an actual vineyard to validate the algorithm. In these trials show that the algorithm can successfully guide the robot through the row without any collisions. This algorithm offers a computationally efficient solution for vineyard row navigation, employing a 2D camera and the Otsu’s thresholding technique to ensure collision-free operation.

@article{
 title = {Intelligent Monitoring System to Assess Plant Development State Based on Computer Vision in Viticulture},
 type = {article},
 year = {2023},
 keywords = {rudenko2023intelligentmonitoringsystem},
 pages = {171},
 volume = {11},
 websites = {https://www.mdpi.com/2079-3197/11/9/171/htm,https://www.mdpi.com/2079-3197/11/9/171},
 month = {9},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {3},
 id = {98ce58d3-cdfa-3d5d-ab7e-826000c4ecba},
 created = {2023-10-27T08:30:16.541Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T20:25:07.284Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,6b565182-74c4-44fd-98cc-10618152e2ae,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Plant health plays an important role in influencing agricultural yields and poor plant health can lead to significant economic losses. Grapes are an important and widely cultivated plant, especially in the southern regions of Russia. Grapes are subject to a number of diseases that require timely diagnosis and treatment. Incorrect identification of diseases can lead to large crop losses. A neural network deep learning dataset of 4845 grape disease images was created. Eight categories of common grape diseases typical of the Black Sea region were studied: Mildew, Oidium, Anthracnose, Esca, Gray rot, Black rot, White rot, and bacterial cancer of grapes. In addition, a set of healthy plants was included. In this paper, a new selective search algorithm for monitoring the state of plant development based on computer vision in viticulture, based on YOLOv5, was considered. The most difficult part of object detection is object localization. As a result, the fast and accurate detection of grape health status was realized. The test results showed that the accuracy was 97.5%, with a model size of 14.85 MB. An analysis of existing publications and patents found using the search “Computer vision in viticulture” showed that this technology is original and promising. The developed software package implements the best approaches to the control system in viticulture using computer vision technologies. A mobile application was developed for practical use by the farmer. The developed software and hardware complex can be installed in any vehicle. Such a mobile system will allow for real-time monitoring of the state of the vineyards and will display it on a map. The novelty of this study lies in the integration of software and hardware. Decision support system software can be adapted to solve other similar problems. The software product commercialization plan is focused on the automation and robotization of agriculture, and will form the basis for adding the next set of similar software.},
 bibtype = {article},
 author = {Rudenko, Marina and Kazak, Anatoliy and Oleinikov, Nikolay and Mayorova, Angela and Dorofeeva, Anna and Nekhaychuk, Dmitry and Shutova, Olga},
 doi = {10.3390/COMPUTATION11090171},
 journal = {Computation 2023, Vol. 11, Page 171},
 number = {9}
}

Plant health plays an important role in influencing agricultural yields and poor plant health can lead to significant economic losses. Grapes are an important and widely cultivated plant, especially in the southern regions of Russia. Grapes are subject to a number of diseases that require timely diagnosis and treatment. Incorrect identification of diseases can lead to large crop losses. A neural network deep learning dataset of 4845 grape disease images was created. Eight categories of common grape diseases typical of the Black Sea region were studied: Mildew, Oidium, Anthracnose, Esca, Gray rot, Black rot, White rot, and bacterial cancer of grapes. In addition, a set of healthy plants was included. In this paper, a new selective search algorithm for monitoring the state of plant development based on computer vision in viticulture, based on YOLOv5, was considered. The most difficult part of object detection is object localization. As a result, the fast and accurate detection of grape health status was realized. The test results showed that the accuracy was 97.5%, with a model size of 14.85 MB. An analysis of existing publications and patents found using the search “Computer vision in viticulture” showed that this technology is original and promising. The developed software package implements the best approaches to the control system in viticulture using computer vision technologies. A mobile application was developed for practical use by the farmer. The developed software and hardware complex can be installed in any vehicle. Such a mobile system will allow for real-time monitoring of the state of the vineyards and will display it on a map. The novelty of this study lies in the integration of software and hardware. Decision support system software can be adapted to solve other similar problems. The software product commercialization plan is focused on the automation and robotization of agriculture, and will form the basis for adding the next set of similar software.

@article{
 title = {Using a Camera System for the In-Situ Assessment of Cordon Dieback due to Grapevine Trunk Diseases},
 type = {article},
 year = {2023},
 websites = {https://doi.org/10.1155/2023/8634742},
 id = {d1119c87-7b74-3c51-b65c-6c9c5ac89357},
 created = {2023-10-27T08:32:17.044Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:42:35.062Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Background and Aims. .e assessment of grapevine trunk disease symptoms is a labour-intensive process that requires experience and is prone to bias. Methods that support the easy and accurate monitoring of trunk diseases will aid management decisions. Methods and Results. An algorithm was developed for the assessment of dieback symptoms due to trunk disease which is applied on a smartphone mounted on a vehicle driven through the vineyard. Vine images and corresponding expert ground truth assessments (of over 13,000 vines) were collected and correlated over two seasons in Shiraz vineyards in the Clare Valley, Barossa, and McLaren Vale, South Australia. .is dataset was used to train and verify YOLOv5 models to estimate the percentage dieback of cordons due to trunk diseases. .e performance of the models was evaluated on the metrics of highest conndence, highest dieback score, and average dieback score across multiple detections. Eighty-four percent of vines in a test set derived from an unseen vineyard were assigned a score by the model within 10% of the score given by experts in the vineyard. Conclusions. .e computer vision algorithms were implemented within the phone, allowing real-time assessment and row-level mapping with nothing more than a high-end mobile phone. Signiicance of the Study. .e algorithms form the basis of a system that will allow growers to scan their vineyards easily and regularly to monitor dieback due to grapevine trunk disease and will facilitate corrective interventions.},
 bibtype = {article},
 author = {Tang, Julie and Yem, Olivia and Russell, Finn and Stewart, Cameron A and Lin, Kangying and Jayakody, Hiranya and Ayres, Matthew R and Sosnowski, Mark R and Whitty, Mark and Petrie, Paul R},
 doi = {10.1155/2023/8634742},
 keywords = {tang2023usingcamerasystem}
}

Background and Aims. .e assessment of grapevine trunk disease symptoms is a labour-intensive process that requires experience and is prone to bias. Methods that support the easy and accurate monitoring of trunk diseases will aid management decisions. Methods and Results. An algorithm was developed for the assessment of dieback symptoms due to trunk disease which is applied on a smartphone mounted on a vehicle driven through the vineyard. Vine images and corresponding expert ground truth assessments (of over 13,000 vines) were collected and correlated over two seasons in Shiraz vineyards in the Clare Valley, Barossa, and McLaren Vale, South Australia. .is dataset was used to train and verify YOLOv5 models to estimate the percentage dieback of cordons due to trunk diseases. .e performance of the models was evaluated on the metrics of highest conndence, highest dieback score, and average dieback score across multiple detections. Eighty-four percent of vines in a test set derived from an unseen vineyard were assigned a score by the model within 10% of the score given by experts in the vineyard. Conclusions. .e computer vision algorithms were implemented within the phone, allowing real-time assessment and row-level mapping with nothing more than a high-end mobile phone. Signiicance of the Study. .e algorithms form the basis of a system that will allow growers to scan their vineyards easily and regularly to monitor dieback due to grapevine trunk disease and will facilitate corrective interventions.

@article{
 title = {Phenotyping grapevine red blotch virus and grapevine leafroll-associated viruses before and after symptom expression through machine-learning analysis of hyperspectral images},
 type = {article},
 year = {2023},
 keywords = {Vitis viniferaL,convolutional neural network,deep-learning,disease detection,phenomics,random forest,spectroscopy},
 pages = {1-15},
 volume = {14},
 id = {1d39ab76-7caf-3e37-8a32-afd42d7697da},
 created = {2023-10-27T08:35:48.176Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T11:26:17.964Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {Introduction: Grapevine leafroll-associated viruses (GLRaVs) and grapevine red blotch virus (GRBV) cause substantial economic losses and concern to North America’s grape and wine industries. Fast and accurate identification of these two groups of viruses is key to informing disease management strategies and limiting their spread by insect vectors in the vineyard. Hyperspectral imaging offers new opportunities for virus disease scouting. Methods: Here we used two machine learning methods, i.e., Random Forest (RF) and 3D-Convolutional Neural Network (CNN), to identify and distinguish leaves from red blotch-infected vines, leafroll-infected vines, and vines co-infected with both viruses using spatiospectral information in the visible domain (510-710nm). We captured hyperspectral images of about 500 leaves from 250 vines at two sampling times during the growing season (a pre-symptomatic stage at veraison and a symptomatic stage at mid-ripening). Concurrently, viral infections were determined in leaf petioles by polymerase chain reaction (PCR) based assays using virus-specific primers and by visual assessment of disease symptoms. Results: When binarily classifying infected vs. non-infected leaves, the CNN model reaches an overall maximum accuracy of 87% versus 82.8% for the RF model. Using the symptomatic dataset lowers the rate of false negatives. Based on a multiclass categorization of leaves, the CNN and RF models had a maximum accuracy of 77.7% and 76.9% (averaged across both healthy and infected leaf categories). Both CNN and RF outperformed visual assessment of symptoms by experts when using RGB segmented images. Interpretation of the RF data showed that the most important wavelengths were in the green, orange, and red subregions. Discussion: While differentiation between plants co-infected with GLRaVs and GRBV proved to be relatively challenging, both models showed promising accuracies across infection categories.},
 bibtype = {article},
 author = {Sawyer, Erica and Laroche-Pinel, Eve and Flasco, Madison and Cooper, Monica L. and Corrales, Benjamin and Fuchs, Marc and Brillante, Luca},
 doi = {10.3389/fpls.2023.1117869},
 journal = {Frontiers in Plant Science},
 number = {March}
}

Introduction: Grapevine leafroll-associated viruses (GLRaVs) and grapevine red blotch virus (GRBV) cause substantial economic losses and concern to North America’s grape and wine industries. Fast and accurate identification of these two groups of viruses is key to informing disease management strategies and limiting their spread by insect vectors in the vineyard. Hyperspectral imaging offers new opportunities for virus disease scouting. Methods: Here we used two machine learning methods, i.e., Random Forest (RF) and 3D-Convolutional Neural Network (CNN), to identify and distinguish leaves from red blotch-infected vines, leafroll-infected vines, and vines co-infected with both viruses using spatiospectral information in the visible domain (510-710nm). We captured hyperspectral images of about 500 leaves from 250 vines at two sampling times during the growing season (a pre-symptomatic stage at veraison and a symptomatic stage at mid-ripening). Concurrently, viral infections were determined in leaf petioles by polymerase chain reaction (PCR) based assays using virus-specific primers and by visual assessment of disease symptoms. Results: When binarily classifying infected vs. non-infected leaves, the CNN model reaches an overall maximum accuracy of 87% versus 82.8% for the RF model. Using the symptomatic dataset lowers the rate of false negatives. Based on a multiclass categorization of leaves, the CNN and RF models had a maximum accuracy of 77.7% and 76.9% (averaged across both healthy and infected leaf categories). Both CNN and RF outperformed visual assessment of symptoms by experts when using RGB segmented images. Interpretation of the RF data showed that the most important wavelengths were in the green, orange, and red subregions. Discussion: While differentiation between plants co-infected with GLRaVs and GRBV proved to be relatively challenging, both models showed promising accuracies across infection categories.

@article{
 title = {A Deep Learning Approach for Precision Viticulture, Assessing Grape Maturity via YOLOv7},
 type = {article},
 year = {2023},
 keywords = {YOLO,grape maturity detection,maturity estimation,object detection},
 pages = {8126},
 volume = {23},
 websites = {https://www.mdpi.com/1424-8220/23/19/8126/htm,https://www.mdpi.com/1424-8220/23/19/8126},
 month = {9},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {27},
 id = {28bb8779-4f34-30ab-8d9e-bca67bcfe59e},
 created = {2023-10-27T08:37:26.018Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:28.425Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {In the viticulture sector, robots are being employed more frequently to increase productivity and accuracy in operations such as vineyard mapping, pruning, and harvesting, especially in locations where human labor is in short supply or expensive. This paper presents the development of an algorithm for grape maturity estimation in the framework of vineyard management. An object detection algorithm is proposed based on You Only Look Once (YOLO) v7 and its extensions in order to detect grape maturity in a white variety of grape (Assyrtiko grape variety). The proposed algorithm was trained using images received over a period of six weeks from grapevines in Drama, Greece. Tests on high-quality images have demonstrated that the detection of five grape maturity stages is possible. Furthermore, the proposed approach has been compared against alternative object detection algorithms. The results showed that YOLO v7 outperforms other architectures both in precision and accuracy. This work paves the way for the development of an autonomous robot for grapevine management.},
 bibtype = {article},
 author = {Badeka, Eftichia and Karapatzak, Eleftherios and Karampatea, Aikaterini and Bouloumpasi, Elisavet and Kalathas, Ioannis and Lytridis, Chris and Tziolas, Emmanouil and Tsakalidou, Viktoria Nikoleta and Kaburlasos, Vassilis G},
 doi = {10.3390/S23198126},
 journal = {Sensors 2023, Vol. 23, Page 8126},
 number = {19}
}

In the viticulture sector, robots are being employed more frequently to increase productivity and accuracy in operations such as vineyard mapping, pruning, and harvesting, especially in locations where human labor is in short supply or expensive. This paper presents the development of an algorithm for grape maturity estimation in the framework of vineyard management. An object detection algorithm is proposed based on You Only Look Once (YOLO) v7 and its extensions in order to detect grape maturity in a white variety of grape (Assyrtiko grape variety). The proposed algorithm was trained using images received over a period of six weeks from grapevines in Drama, Greece. Tests on high-quality images have demonstrated that the detection of five grape maturity stages is possible. Furthermore, the proposed approach has been compared against alternative object detection algorithms. The results showed that YOLO v7 outperforms other architectures both in precision and accuracy. This work paves the way for the development of an autonomous robot for grapevine management.

@article{
 title = {Deep Learning YOLO-Based Solution for Grape Bunch Detection and Assessment of Biophysical Lesions},
 type = {article},
 year = {2023},
 keywords = {computer vision,machine learning,object detection,precision agriculture,viticulture},
 pages = {1120},
 volume = {13},
 websites = {https://www.mdpi.com/2073-4395/13/4/1120/htm,https://www.mdpi.com/2073-4395/13/4/1120},
 month = {4},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {14},
 id = {6ec3008c-82eb-3a79-8111-de95ce928fbb},
 created = {2023-10-27T08:38:57.257Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:28.246Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The world wine sector is a multi-billion dollar industry with a wide range of economic activities. Therefore, it becomes crucial to monitor the grapevine because it allows a more accurate estimation of the yield and ensures a high-quality end product. The most common way of monitoring the grapevine is through the leaves (preventive way) since the leaves first manifest biophysical lesions. However, this does not exclude the possibility of biophysical lesions manifesting in the grape berries. Thus, this work presents three pre-trained YOLO models (YOLOv5x6, YOLOv7-E6E, and YOLOR-CSP-X) to detect and classify grape bunches as healthy or damaged by the number of berries with biophysical lesions. Two datasets were created and made publicly available with original images and manual annotations to identify the complexity between detection (bunches) and classification (healthy or damaged) tasks. The datasets use the same 10,010 images with different classes. The Grapevine Bunch Detection Dataset uses the Bunch class, and The Grapevine Bunch Condition Detection Dataset uses the OptimalBunch and DamagedBunch classes. Regarding the three models trained for grape bunches detection, they obtained promising results, highlighting YOLOv7 with 77% of mAP and 94% of the F1-score. In the case of the task of detection and identification of the state of grape bunches, the three models obtained similar results, with YOLOv5 achieving the best ones with an mAP of 72% and an F1-score of 92%.},
 bibtype = {article},
 author = {Pinheiro, Isabel and Moreira, Germano and Queirós da Silva, Daniel and Magalhães, Sandro and Valente, António and Moura Oliveira, Paulo and Cunha, Mário and Santos, Filipe},
 doi = {10.3390/AGRONOMY13041120},
 journal = {Agronomy 2023, Vol. 13, Page 1120},
 number = {4}
}

The world wine sector is a multi-billion dollar industry with a wide range of economic activities. Therefore, it becomes crucial to monitor the grapevine because it allows a more accurate estimation of the yield and ensures a high-quality end product. The most common way of monitoring the grapevine is through the leaves (preventive way) since the leaves first manifest biophysical lesions. However, this does not exclude the possibility of biophysical lesions manifesting in the grape berries. Thus, this work presents three pre-trained YOLO models (YOLOv5x6, YOLOv7-E6E, and YOLOR-CSP-X) to detect and classify grape bunches as healthy or damaged by the number of berries with biophysical lesions. Two datasets were created and made publicly available with original images and manual annotations to identify the complexity between detection (bunches) and classification (healthy or damaged) tasks. The datasets use the same 10,010 images with different classes. The Grapevine Bunch Detection Dataset uses the Bunch class, and The Grapevine Bunch Condition Detection Dataset uses the OptimalBunch and DamagedBunch classes. Regarding the three models trained for grape bunches detection, they obtained promising results, highlighting YOLOv7 with 77% of mAP and 94% of the F1-score. In the case of the task of detection and identification of the state of grape bunches, the three models obtained similar results, with YOLOv5 achieving the best ones with an mAP of 72% and an F1-score of 92%.

@article{
 title = {Evolutionary conditional GANs for supervised data augmentation: The case of assessing berry number per cluster in grapevine},
 type = {article},
 year = {2023},
 pages = {110805},
 volume = {147},
 month = {11},
 publisher = {Elsevier},
 day = {1},
 id = {750acab6-e2fa-3983-b12e-1406e4d3895d},
 created = {2023-10-27T08:48:09.217Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:28.834Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 bibtype = {article},
 author = {Gutiérrez, Salvador and Tardaguila, Javier},
 doi = {10.1016/J.ASOC.2023.110805},
 journal = {Applied Soft Computing}
}

@article{
 title = {Surgical Fine-Tuning for Grape Bunch Segmentation under Visual Domain Shifts},
 type = {article},
 year = {2023},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {eff2178b-fe2e-3774-8071-deeedd767f20},
 created = {2023-10-27T08:58:59.204Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:28.639Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Mobile robots will play a crucial role in the transition towards sustainable agriculture. To autonomously and effectively monitor the state of plants, robots ought to be equipped with visual perception capabilities that are robust to the rapid changes that characterise agricultural settings. In this paper, we focus on the challenging task of segmenting grape bunches from images collected by mobile robots in vineyards. In this context, we present the first study that applies surgical fine-tuning to instance segmentation tasks. We show how selectively tuning only specific model layers can support the adaptation of pre-trained Deep Learning models to newly-collected grape images that introduce visual domain shifts, while also substantially reducing the number of tuned parameters.},
 bibtype = {article},
 author = {Chiatti, Agnese and Bertoglio, Riccardo and Catalano, Nico and Gatti, Matteo and Matteucci, Matteo},
 doi = {10.1109/ECMR59166.2023.10256348},
 journal = {Proceedings of the 11th European Conference on Mobile Robots, ECMR 2023}
}

Mobile robots will play a crucial role in the transition towards sustainable agriculture. To autonomously and effectively monitor the state of plants, robots ought to be equipped with visual perception capabilities that are robust to the rapid changes that characterise agricultural settings. In this paper, we focus on the challenging task of segmenting grape bunches from images collected by mobile robots in vineyards. In this context, we present the first study that applies surgical fine-tuning to instance segmentation tasks. We show how selectively tuning only specific model layers can support the adaptation of pre-trained Deep Learning models to newly-collected grape images that introduce visual domain shifts, while also substantially reducing the number of tuned parameters.

@article{
 title = {Modelling wine grapevines for autonomous robotic cane pruning},
 type = {article},
 year = {2023},
 keywords = {Horticulture,Robotics},
 websites = {http://creativecommons.org/licenses/by/4.0/},
 id = {7480fd04-a5cc-3825-9221-872da12f3cd0},
 created = {2023-10-27T09:06:43.156Z},
 accessed = {2023-10-27},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-27T09:13:03.144Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Machine Vision Pruning Orchard Vineyard Aotearoa (New Zealand) has a strong and growing winegrape industry struggling to access workers to complete skilled, seasonal tasks such as pruning. Maintaining high-producing vines requires training agricultural workers that can make quality cane pruning decisions , which can be difficult when workers are not readily available. A novel vision system for an autonomous cane pruning robot is presented that can assess a vine to make quality pruning decisions like an expert. The vision system is designed to generate an accurate digital 3D model of a vine with skeletonised cane structures to estimate key pruning metrics for each cane. The presented approach has been extensively evaluated in a real-world vineyard as a commercial platform would be expected to operate. The system is demonstrated to perform consistently at extracting dimensionally accurate digital models of the vines. Detailed evaluation of the digital models shows that 51.45% of the canes were modelled entirely, with a further 35.51% only missing a single internode connection. The quantified results demonstrate that the robotic platform can generate dimensionally accurate metrics of the canes for future decision-making and automation of pruning. ScienceDirect jo urnal homepage: www .e lsev ie r.com/ locate/issn/153 75110 b i o s y s t e m s e n g i n e e r i n g 2 3 5 (2 0 2 3) 3 1 e4 9 https://doi.org/10.1016/j.biosystemseng.2023.09.006 1537-5110/},
 bibtype = {article},
 author = {Williams, Henry and Smith, David and Shahabi, Jalil and Gee, Trevor and Nejati, Mahla and Mcguinness, Ben and Black, Kale and Tobias, Jonathan and Jangali, Rahul and Lim, Hin and Mcculloch, Josh and Green, Richard and O'connor, Mira and Gounder, Sandhiya and Ndaka, Angella and Burch, Karly and Fourie, Jaco and Hsiao, Jeffrey and Werner, Armin and Agnew, Rob and Oliver, Richard and Macdonald, Bruce A},
 doi = {10.1016/j.biosystemseng.2023.09.006}
}

Machine Vision Pruning Orchard Vineyard Aotearoa (New Zealand) has a strong and growing winegrape industry struggling to access workers to complete skilled, seasonal tasks such as pruning. Maintaining high-producing vines requires training agricultural workers that can make quality cane pruning decisions , which can be difficult when workers are not readily available. A novel vision system for an autonomous cane pruning robot is presented that can assess a vine to make quality pruning decisions like an expert. The vision system is designed to generate an accurate digital 3D model of a vine with skeletonised cane structures to estimate key pruning metrics for each cane. The presented approach has been extensively evaluated in a real-world vineyard as a commercial platform would be expected to operate. The system is demonstrated to perform consistently at extracting dimensionally accurate digital models of the vines. Detailed evaluation of the digital models shows that 51.45% of the canes were modelled entirely, with a further 35.51% only missing a single internode connection. The quantified results demonstrate that the robotic platform can generate dimensionally accurate metrics of the canes for future decision-making and automation of pruning. ScienceDirect jo urnal homepage: www .e lsev ie r.com/ locate/issn/153 75110 b i o s y s t e m s e n g i n e e r i n g 2 3 5 (2 0 2 3) 3 1 e4 9 https://doi.org/10.1016/j.biosystemseng.2023.09.006 1537-5110/

@article{
 title = {Modelling wine grapevines for autonomous robotic cane pruning},
 type = {article},
 year = {2023},
 keywords = {Horticulture,Robotics},
 websites = {http://creativecommons.org/licenses/by/4.0/},
 id = {62e53d0d-de91-34ab-b4f5-370b288a1028},
 created = {2023-10-27T09:07:10.046Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:29.135Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Machine Vision Pruning Orchard Vineyard Aotearoa (New Zealand) has a strong and growing winegrape industry struggling to access workers to complete skilled, seasonal tasks such as pruning. Maintaining high-producing vines requires training agricultural workers that can make quality cane pruning decisions , which can be difficult when workers are not readily available. A novel vision system for an autonomous cane pruning robot is presented that can assess a vine to make quality pruning decisions like an expert. The vision system is designed to generate an accurate digital 3D model of a vine with skeletonised cane structures to estimate key pruning metrics for each cane. The presented approach has been extensively evaluated in a real-world vineyard as a commercial platform would be expected to operate. The system is demonstrated to perform consistently at extracting dimensionally accurate digital models of the vines. Detailed evaluation of the digital models shows that 51.45% of the canes were modelled entirely, with a further 35.51% only missing a single internode connection. The quantified results demonstrate that the robotic platform can generate dimensionally accurate metrics of the canes for future decision-making and automation of pruning. ScienceDirect jo urnal homepage: www .e lsev ie r.com/ locate/issn/153 75110 b i o s y s t e m s e n g i n e e r i n g 2 3 5 (2 0 2 3) 3 1 e4 9 https://doi.org/10.1016/j.biosystemseng.2023.09.006 1537-5110/},
 bibtype = {article},
 author = {Williams, Henry and Smith, David and Shahabi, Jalil and Gee, Trevor and Nejati, Mahla and Mcguinness, Ben and Black, Kale and Tobias, Jonathan and Jangali, Rahul and Lim, Hin and Mcculloch, Josh and Green, Richard and O'connor, Mira and Gounder, Sandhiya and Ndaka, Angella and Burch, Karly and Fourie, Jaco and Hsiao, Jeffrey and Werner, Armin and Agnew, Rob and Oliver, Richard and Macdonald, Bruce A},
 doi = {10.1016/j.biosystemseng.2023.09.006}
}

Machine Vision Pruning Orchard Vineyard Aotearoa (New Zealand) has a strong and growing winegrape industry struggling to access workers to complete skilled, seasonal tasks such as pruning. Maintaining high-producing vines requires training agricultural workers that can make quality cane pruning decisions , which can be difficult when workers are not readily available. A novel vision system for an autonomous cane pruning robot is presented that can assess a vine to make quality pruning decisions like an expert. The vision system is designed to generate an accurate digital 3D model of a vine with skeletonised cane structures to estimate key pruning metrics for each cane. The presented approach has been extensively evaluated in a real-world vineyard as a commercial platform would be expected to operate. The system is demonstrated to perform consistently at extracting dimensionally accurate digital models of the vines. Detailed evaluation of the digital models shows that 51.45% of the canes were modelled entirely, with a further 35.51% only missing a single internode connection. The quantified results demonstrate that the robotic platform can generate dimensionally accurate metrics of the canes for future decision-making and automation of pruning. ScienceDirect jo urnal homepage: www .e lsev ie r.com/ locate/issn/153 75110 b i o s y s t e m s e n g i n e e r i n g 2 3 5 (2 0 2 3) 3 1 e4 9 https://doi.org/10.1016/j.biosystemseng.2023.09.006 1537-5110/

@article{
 title = {Correlation of the Grapevine (Vitis vinifera L.) Leaf Chlorophyll Concentration with RGB Color Indices},
 type = {article},
 year = {2023},
 keywords = {RGB,chlorophyll,precision viticulture,vegetation index},
 pages = {899},
 volume = {9},
 websites = {https://www.mdpi.com/2311-7524/9/8/899/htm,https://www.mdpi.com/2311-7524/9/8/899},
 month = {8},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {7},
 id = {a040356e-a472-3020-8304-27ec7fb84f2b},
 created = {2023-10-27T09:25:16.721Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:29.844Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Spectral investigation of the canopy has an increasing importance in precision viticulture to monitor the effect of biotic and abiotic stress factors. In this study, RGB (color model, red, green, blue)-based vegetation indices were evaluated to find a correlation with grapevine leaf chlorophyll concentration. ‘Hárslevelű’ (Vitis vinifera L.) leaf samples were obtained from a commercial vineyard and digitalized. The chlorophyll concentration of the samples was determined with a portable chlorophyll meter. Image processing and color analyses were performed to determine the RGB average values of the digitized samples. According to the RGB values, 31 vegetation indices were calculated and evaluated with a correlation test and multivariate regression. The Pearson correlation between the chlorophyll concentration and most of the indices was significant (p < 0.01), with some exceptions. The same results were obtained with the Spearman correlation as the relationship had high significance (p < 0.01) for most of the indices. The highest Pearson correlation was obtained with the index PCA2 (Principal Component Analysis 2), while Spearman correlation was the highest for RMB (difference between red and blue) and GMB (difference between green and blue). The multivariate regression model also showed a high correlation with the pigmentation. We consider that our results would be applicable in the future to receive information about the canopy physiological status monitored with on-the-go sensors.},
 bibtype = {article},
 author = {Bodor-Pesti, Péter and Taranyi, Dóra and Nyitrainé Sárdy, Diána Ágnes and Le Phuong Nguyen, Lien and Baranyai, László},
 doi = {10.3390/HORTICULTURAE9080899},
 journal = {Horticulturae 2023, Vol. 9, Page 899},
 number = {8}
}

Spectral investigation of the canopy has an increasing importance in precision viticulture to monitor the effect of biotic and abiotic stress factors. In this study, RGB (color model, red, green, blue)-based vegetation indices were evaluated to find a correlation with grapevine leaf chlorophyll concentration. ‘Hárslevelű’ (Vitis vinifera L.) leaf samples were obtained from a commercial vineyard and digitalized. The chlorophyll concentration of the samples was determined with a portable chlorophyll meter. Image processing and color analyses were performed to determine the RGB average values of the digitized samples. According to the RGB values, 31 vegetation indices were calculated and evaluated with a correlation test and multivariate regression. The Pearson correlation between the chlorophyll concentration and most of the indices was significant (p < 0.01), with some exceptions. The same results were obtained with the Spearman correlation as the relationship had high significance (p < 0.01) for most of the indices. The highest Pearson correlation was obtained with the index PCA2 (Principal Component Analysis 2), while Spearman correlation was the highest for RMB (difference between red and blue) and GMB (difference between green and blue). The multivariate regression model also showed a high correlation with the pigmentation. We consider that our results would be applicable in the future to receive information about the canopy physiological status monitored with on-the-go sensors.

@article{
 title = {Estimating soil and grapevine water status using ground based hyperspectral imaging under diffused lighting conditions: Addressing the effect of lighting variability in vineyards},
 type = {article},
 year = {2023},
 keywords = {Deficit irrigation,Diffused lighting,Grapevine,Hyperspectral imaging,Partial least square,Vitis,Water stress},
 pages = {108175},
 volume = {212},
 month = {9},
 publisher = {Elsevier},
 day = {1},
 id = {22c8f836-8108-3b8b-88fc-b434f6bb8e49},
 created = {2023-10-27T09:27:57.641Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:29.971Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {A timely and appropriate level of water deficit is desirable in wine grape production to optimize fruit quality for winemaking. Thus, it is crucial to find a robust and rapid method to assess grapevine water stress in real time. Hyperspectral imaging (HSI) has the potential to detect changes in leaf water status, but the robustness and accuracy are limited in field applications. This study focused on developing ground based approaches for detecting soil and grapevine water status using HSI obtained in diffused lighting conditions. During the 2021 growing season, leaf water potential (ΨL), stomatal conductance (gs) on the selected leaves and volumetric soil moisture (θv) in the root zone were measured as water status indicators. Spectral data from diffused and direct sunlight conditions were obtained to construct models to estimate plant and soil water status indicators. Partial least squares (PLS) regression models were individually developed to estimate ΨL, gs, and θv using spectra obtained from direct/diffused lighting conditions, respectively. The results indicated that the ΨL estimation model using spectral data from diffused lighting performed better than that obtained using direct sunlight, indicated by a higher R2 (0.89 vs. 0.82), a lower RMSE (0.12 vs. 0.15 MPa) and a lower MAE (0.10 vs. 0.11 MPa). The model developed for estimating θv using spectral data under diffused lighting achieved superior performance to the one in direct sunlight in terms of R2, RMSE and MAE (0.90 vs. 0.89 and 1.56 vs. 1.59 %, 1.26 vs. 1.29 %). These results demonstrated that spectral data obtained under diffused lighting can improve model performance by providing a more uniform illumination. Ground based HSI was capable of high-resolution sensing of grapevine water status by estimating ΨL and gs and map variability within canopies.},
 bibtype = {article},
 author = {Kang, Chenchen and Diverres, Geraldine and Achyut, Paudel and Karkee, Manoj and Zhang, Qin and Keller, Markus},
 doi = {10.1016/J.COMPAG.2023.108175},
 journal = {Computers and Electronics in Agriculture}
}

A timely and appropriate level of water deficit is desirable in wine grape production to optimize fruit quality for winemaking. Thus, it is crucial to find a robust and rapid method to assess grapevine water stress in real time. Hyperspectral imaging (HSI) has the potential to detect changes in leaf water status, but the robustness and accuracy are limited in field applications. This study focused on developing ground based approaches for detecting soil and grapevine water status using HSI obtained in diffused lighting conditions. During the 2021 growing season, leaf water potential (ΨL), stomatal conductance (gs) on the selected leaves and volumetric soil moisture (θv) in the root zone were measured as water status indicators. Spectral data from diffused and direct sunlight conditions were obtained to construct models to estimate plant and soil water status indicators. Partial least squares (PLS) regression models were individually developed to estimate ΨL, gs, and θv using spectra obtained from direct/diffused lighting conditions, respectively. The results indicated that the ΨL estimation model using spectral data from diffused lighting performed better than that obtained using direct sunlight, indicated by a higher R2 (0.89 vs. 0.82), a lower RMSE (0.12 vs. 0.15 MPa) and a lower MAE (0.10 vs. 0.11 MPa). The model developed for estimating θv using spectral data under diffused lighting achieved superior performance to the one in direct sunlight in terms of R2, RMSE and MAE (0.90 vs. 0.89 and 1.56 vs. 1.59 %, 1.26 vs. 1.29 %). These results demonstrated that spectral data obtained under diffused lighting can improve model performance by providing a more uniform illumination. Ground based HSI was capable of high-resolution sensing of grapevine water status by estimating ΨL and gs and map variability within canopies.

@article{
 title = {Potential detection of Flavescence dorée in the vineyard using close-range hyperspectral imaging},
 type = {article},
 year = {2023},
 keywords = {barjaktarovic2023potentialdetectionflavescence},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {b7c5c04a-6674-3559-80ac-2bc6da112ce7},
 created = {2023-10-27T09:33:29.495Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:05.342Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Grapevine is one of the most important crops cultivated across Europe. Climate factors and diseases constantly threaten its production. Recently, Flavescence dorée (FD), an incurable grapevine disease with the obligation to uproot each infected plant, has been widely spread in Europe. The symptoms of FD are visually expressed in the late summer. Currently, the adopted procedure consists in scouting for infected plants by trained experts, which is time-consuming and not frequent enough. As stress development causes subtle spectral changes before any visible symptoms appear, during the summer of 2022, hyperspectral and multispectral images were acquired in the two vineyards near Riva del Garda, Trentino, Italy. A classification accuracy between 90.2 % and 96.9 % in distinguishing between infected and healthy plants was obtained from the hyperspectral data. These findings justify further efforts to use an in-house developed, affordable multispectral camera, significantly reducing equipment cost and procedure complexity while mapping the relevant spectral channels.},
 bibtype = {article},
 author = {Barjaktarovic, Marko and Santoni, Massimo and Faralli, Michele and Bertamini, Massimo and Bruzzone, Lorenzo},
 doi = {10.1109/ICECCME57830.2023.10252351},
 journal = {International Conference on Electrical, Computer, Communications and Mechatronics Engineering, ICECCME 2023}
}

Grapevine is one of the most important crops cultivated across Europe. Climate factors and diseases constantly threaten its production. Recently, Flavescence dorée (FD), an incurable grapevine disease with the obligation to uproot each infected plant, has been widely spread in Europe. The symptoms of FD are visually expressed in the late summer. Currently, the adopted procedure consists in scouting for infected plants by trained experts, which is time-consuming and not frequent enough. As stress development causes subtle spectral changes before any visible symptoms appear, during the summer of 2022, hyperspectral and multispectral images were acquired in the two vineyards near Riva del Garda, Trentino, Italy. A classification accuracy between 90.2 % and 96.9 % in distinguishing between infected and healthy plants was obtained from the hyperspectral data. These findings justify further efforts to use an in-house developed, affordable multispectral camera, significantly reducing equipment cost and procedure complexity while mapping the relevant spectral channels.

@article{
 title = {Grapevine water status in a variably irrigated vineyard with NIR hyperspectral imaging from a UAV},
 type = {article},
 year = {2023},
 pages = {345–350},
 id = {54567139-8968-32e9-a191-74a8cb3ab39a},
 created = {2023-10-27T09:37:09.872Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:29.517Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 bibtype = {article},
 author = {Vasquez, K. and Laroche-Pinel, E. and Partida, G. and Brillante, L.},
 journal = {Precision agriculture '23}
}

@article{
 title = {Instance Segmentation and Berry Counting of Table Grape before Thinning Based on AS-SwinT},
 type = {article},
 year = {2023},
 volume = {5},
 websites = {https://spj.science.org/doi/10.34133/plantphenomics.0085},
 month = {8},
 publisher = {AAAS},
 day = {29},
 id = {50bc8402-ec7f-3a68-9762-cc082c870c78},
 created = {2023-10-27T09:40:15.912Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:30.128Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Berry thinning is one of the most important tasks in the management of high-quality table grapes. Farmers often thin the berries per cluster to a standard number by counting. With an aging populati...},
 bibtype = {article},
 author = {Du, Wensheng and Liu, Ping},
 doi = {10.34133/PLANTPHENOMICS.0085},
 journal = {Plant Phenomics}
}

Berry thinning is one of the most important tasks in the management of high-quality table grapes. Farmers often thin the berries per cluster to a standard number by counting. With an aging populati...

@article{
 title = {Detecting Grapevine Virus Infections in Red and White Winegrape Canopies Using Proximal Hyperspectral Sensing},
 type = {article},
 year = {2023},
 keywords = {wang2023detectinggrapevinevirus},
 pages = {2851},
 volume = {23},
 websites = {https://www.mdpi.com/1424-8220/23/5/2851/htm,https://www.mdpi.com/1424-8220/23/5/2851},
 month = {3},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {6},
 id = {121653f9-08d3-350d-af1d-17314e8b4858},
 created = {2023-10-27T09:47:18.224Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:53:05.578Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Grapevine virus-associated disease such as grapevine leafroll disease (GLD) affects grapevine health worldwide. Current diagnostic methods are either highly costly (laboratory-based diagnostics) or can be unreliable (visual assessments). Hyperspectral sensing technology is capable of measuring leaf reflectance spectra that can be used for the non-destructive and rapid detection of plant diseases. The present study used proximal hyperspectral sensing to detect virus infection in Pinot Noir (red-berried winegrape cultivar) and Chardonnay (white-berried winegrape cultivar) grapevines. Spectral data were collected throughout the grape growing season at six timepoints per cultivar. Partial least squares-discriminant analysis (PLS-DA) was used to build a predictive model of the presence or absence of GLD. The temporal change of canopy spectral reflectance showed that the harvest timepoint had the best prediction result. Prediction accuracies of 96% and 76% were achieved for Pinot Noir and Chardonnay, respectively. Our results provide valuable information on the optimal time for GLD detection. This hyperspectral method can also be deployed on mobile platforms including ground-based vehicles and unmanned aerial vehicles (UAV) for large-scale disease surveillance in vineyards.},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Pagay, Vinay},
 doi = {10.3390/S23052851},
 journal = {Sensors 2023, Vol. 23, Page 2851},
 number = {5}
}

Grapevine virus-associated disease such as grapevine leafroll disease (GLD) affects grapevine health worldwide. Current diagnostic methods are either highly costly (laboratory-based diagnostics) or can be unreliable (visual assessments). Hyperspectral sensing technology is capable of measuring leaf reflectance spectra that can be used for the non-destructive and rapid detection of plant diseases. The present study used proximal hyperspectral sensing to detect virus infection in Pinot Noir (red-berried winegrape cultivar) and Chardonnay (white-berried winegrape cultivar) grapevines. Spectral data were collected throughout the grape growing season at six timepoints per cultivar. Partial least squares-discriminant analysis (PLS-DA) was used to build a predictive model of the presence or absence of GLD. The temporal change of canopy spectral reflectance showed that the harvest timepoint had the best prediction result. Prediction accuracies of 96% and 76% were achieved for Pinot Noir and Chardonnay, respectively. Our results provide valuable information on the optimal time for GLD detection. This hyperspectral method can also be deployed on mobile platforms including ground-based vehicles and unmanned aerial vehicles (UAV) for large-scale disease surveillance in vineyards.

@article{
 title = {Detecting vineyard plants stress in situ using deep learning},
 type = {article},
 year = {2023},
 keywords = {Convolutional neural network,Grapevine health,Plant stress detection,Transfer learning},
 pages = {107837},
 volume = {210},
 month = {7},
 publisher = {Elsevier},
 day = {1},
 id = {5c94d87a-3539-3523-a19e-0dd768592a16},
 created = {2023-10-27T09:49:01.966Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:30.315Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Diseases and nutritional deficiencies have the potential to seriously impact the production yield and proper development of perennial species such as grapevine. The distinction between changes resulting from normal growth stages and plant alterations caused by biotic and abiotic stress is often drawn through visual inspection, where the observer's subjectivity could introduce human errors, despite the presence of experience and technical knowledge. This document presents an assessment of CNNs for detecting plant stress in grapevine RGB images captured in situ, under conditions that could include variations in light, shadows, insects, or the presence of scrubs. We evaluated five architectures for their ability to discriminate plants with stress symptoms in images captured through the annual grapevine cycle in field conditions. The best model exhibited a 97.2% accuracy, 0.996 ROC AUC, and 0.958 AP using the EfficientNetB3 architecture. Our methodology aims to support winegrowers in their decision-making by enhancing the information they collect through traditional visual inspection methods.},
 bibtype = {article},
 author = {Cándido-Mireles, Mayra and Hernández-Gama, Regina and Salas, Joaquín},
 doi = {10.1016/J.COMPAG.2023.107837},
 journal = {Computers and Electronics in Agriculture}
}

Diseases and nutritional deficiencies have the potential to seriously impact the production yield and proper development of perennial species such as grapevine. The distinction between changes resulting from normal growth stages and plant alterations caused by biotic and abiotic stress is often drawn through visual inspection, where the observer's subjectivity could introduce human errors, despite the presence of experience and technical knowledge. This document presents an assessment of CNNs for detecting plant stress in grapevine RGB images captured in situ, under conditions that could include variations in light, shadows, insects, or the presence of scrubs. We evaluated five architectures for their ability to discriminate plants with stress symptoms in images captured through the annual grapevine cycle in field conditions. The best model exhibited a 97.2% accuracy, 0.996 ROC AUC, and 0.958 AP using the EfficientNetB3 architecture. Our methodology aims to support winegrowers in their decision-making by enhancing the information they collect through traditional visual inspection methods.

@article{
 title = {Scalable Early Detection of Grapevine Viral Infection with Airborne Imaging Spectroscopy},
 type = {article},
 year = {2023},
 keywords = {AVIRIS next generation,early detection,grapevine leafroll-associated virus 3,imaging spectroscopy,scalable},
 websites = {https://apsjournals.apsnet.org/doi/10.1094/PHYTO-01-23-0030-R},
 month = {9},
 publisher = { The American Phytopathological Society },
 day = {20},
 id = {01adfe0d-c7e5-3448-9fbc-b08e35e80d9f},
 created = {2023-10-27T09:56:53.924Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-14T12:33:10.212Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The U.S. wine and grape industry loses $3B annually due to viral diseases including grapevine leafroll-associated virus complex 3 (GLRaV-3). Current detection methods are labor-intensive and expens...},
 bibtype = {article},
 author = {Romero Galvan, Fernando and Pavlick, Ryan and Trolley, Graham Richard and Aggarwal, Somil and Sousa, Daniel and Starr, Charlie and Forrestel, Elisabeth Jane and Bolton, Stephanie and Alsina, Maria del Mar and Dokoozlian, Nicholaus and Gold, Kaitlin Morey},
 doi = {10.1094/PHYTO-01-23-0030-R},
 journal = {https://doi.org/10.1094/PHYTO-01-23-0030-R}
}

The U.S. wine and grape industry loses $3B annually due to viral diseases including grapevine leafroll-associated virus complex 3 (GLRaV-3). Current detection methods are labor-intensive and expens...

@article{
 title = {A Grape Dataset for Instance Segmentation and Maturity Estimation},
 type = {article},
 year = {2023},
 keywords = {blekos2023grapedatasetinstance},
 pages = {1995},
 volume = {13},
 websites = {https://www.mdpi.com/2073-4395/13/8/1995/htm,https://www.mdpi.com/2073-4395/13/8/1995},
 month = {7},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {27},
 id = {45b21996-8c40-3316-a650-6ddb2cde7e51},
 created = {2023-10-27T10:02:43.010Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T15:32:09.826Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Grape maturity estimation is vital in precise agriculture as it enables informed decision making for disease control, harvest timing, grape quality, and quantity assurance. Despite its importance, there are few large publicly available datasets that can be used to train accurate and robust grape segmentation and maturity estimation algorithms. To this end, this work proposes the CERTH grape dataset, a new sizeable dataset that is designed explicitly for evaluating deep learning algorithms in grape segmentation and maturity estimation. The proposed dataset is one of the largest currently available grape datasets in the literature, consisting of around 2500 images and almost 10 k grape bunches, annotated with masks and maturity levels. The images in the dataset were captured under various illumination conditions and viewing angles and with significant occlusions between grape bunches and leaves, making it a valuable resource for the research community. Thorough experiments were conducted using a plethora of general object detection methods to provide a baseline for the future development of accurate and robust grape segmentation and maturity estimation algorithms that can significantly advance research in the field of viticulture.},
 bibtype = {article},
 author = {Blekos, Achilleas and Chatzis, Konstantinos and Kotaidou, Martha and Chatzis, Theocharis and Solachidis, Vassilios and Konstantinidis, Dimitrios and Dimitropoulos, Kosmas},
 doi = {10.3390/AGRONOMY13081995},
 journal = {Agronomy 2023, Vol. 13, Page 1995},
 number = {8}
}

Grape maturity estimation is vital in precise agriculture as it enables informed decision making for disease control, harvest timing, grape quality, and quantity assurance. Despite its importance, there are few large publicly available datasets that can be used to train accurate and robust grape segmentation and maturity estimation algorithms. To this end, this work proposes the CERTH grape dataset, a new sizeable dataset that is designed explicitly for evaluating deep learning algorithms in grape segmentation and maturity estimation. The proposed dataset is one of the largest currently available grape datasets in the literature, consisting of around 2500 images and almost 10 k grape bunches, annotated with masks and maturity levels. The images in the dataset were captured under various illumination conditions and viewing angles and with significant occlusions between grape bunches and leaves, making it a valuable resource for the research community. Thorough experiments were conducted using a plethora of general object detection methods to provide a baseline for the future development of accurate and robust grape segmentation and maturity estimation algorithms that can significantly advance research in the field of viticulture.

@article{
 title = {Scalable Early Detection of Grapevine Viral Infection with Airborne Imaging Spectroscopy},
 type = {article},
 year = {2023},
 keywords = {AVIRIS next generation,early detection,grapevine leafroll-associated virus 3,imaging spectroscopy,scalable},
 websites = {https://apsjournals.apsnet.org/doi/10.1094/PHYTO-01-23-0030-R},
 month = {9},
 publisher = { The American Phytopathological Society },
 day = {20},
 id = {e70316ac-78fe-384f-9690-906108e8f910},
 created = {2023-10-27T10:03:57.212Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-27T10:05:00.083Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The U.S. wine and grape industry loses $3B annually due to viral diseases including grapevine leafroll-associated virus complex 3 (GLRaV-3). Current detection methods are labor-intensive and expens...},
 bibtype = {article},
 author = {Romero Galvan, Fernando and Pavlick, Ryan and Trolley, Graham Richard and Aggarwal, Somil and Sousa, Daniel and Starr, Charlie and Forrestel, Elisabeth Jane and Bolton, Stephanie and Alsina, Maria del Mar and Dokoozlian, Nicholaus and Gold, Kaitlin Morey},
 doi = {10.1094/PHYTO-01-23-0030-R},
 journal = {https://doi.org/10.1094/PHYTO-01-23-0030-R}
}

The U.S. wine and grape industry loses $3B annually due to viral diseases including grapevine leafroll-associated virus complex 3 (GLRaV-3). Current detection methods are labor-intensive and expens...

@article{
 title = {Data Acquisition for Testing Potential Detection of Flavescence Dorée with a Designed, Affordable Multispectral Camera},
 type = {article},
 year = {2023},
 keywords = {barjaktarovic2023dataacquisitiontesting},
 volume = {15},
 id = {c0e4572b-9da8-3bed-b3e3-c3a6e5e1e43f},
 created = {2023-10-27T10:12:24.014Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:05.541Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {There is a constant push on agriculture to produce more food and other inputs for different industries. Precision agriculture is essential to meet these demands. The intake of this modern technology is rapidly increasing among large and medium-sized farms. However, small farms still struggle with their adaptation due to the expensive initial costs. A contribution in handling this challenge, this paper presents data gathering for testing an in-house made, cost-effective, multispectral camera to detect Flavescence dorée (FD). FD is a grapevine disease that, in the last few years, has become a major concern for grapevine producers across Europe. As a quarantine disease, mandatory control procedures, such as uprooting infected plants and removing all vineyard if the infection is higher than 20%, lead to an immense economic loss. Therefore, it is critical to detect each diseased plant promptly, thus reducing the expansion of Flavescence dorée. Data from two vineyards near Riva del Garda, Trentino, Italy, was acquired in 2022 using multispectral and hyperspectral cameras. The initial finding showed that there is a possibility to detect Flavescence dorée using Linear discriminant analysis (LDA) with hyperspectral data, obtaining an accuracy of 96.6 %. This result justifies future investigation on the use of multispectral images for Flavescence dorée detection.},
 bibtype = {article},
 author = {Barjaktarović, Marko and Santoni, Massimo and Faralli, Michele and Bertamini, Massimo and Bruzzone, Lorenzo},
 journal = {Telfor Journal},
 number = {1}
}

There is a constant push on agriculture to produce more food and other inputs for different industries. Precision agriculture is essential to meet these demands. The intake of this modern technology is rapidly increasing among large and medium-sized farms. However, small farms still struggle with their adaptation due to the expensive initial costs. A contribution in handling this challenge, this paper presents data gathering for testing an in-house made, cost-effective, multispectral camera to detect Flavescence dorée (FD). FD is a grapevine disease that, in the last few years, has become a major concern for grapevine producers across Europe. As a quarantine disease, mandatory control procedures, such as uprooting infected plants and removing all vineyard if the infection is higher than 20%, lead to an immense economic loss. Therefore, it is critical to detect each diseased plant promptly, thus reducing the expansion of Flavescence dorée. Data from two vineyards near Riva del Garda, Trentino, Italy, was acquired in 2022 using multispectral and hyperspectral cameras. The initial finding showed that there is a possibility to detect Flavescence dorée using Linear discriminant analysis (LDA) with hyperspectral data, obtaining an accuracy of 96.6 %. This result justifies future investigation on the use of multispectral images for Flavescence dorée detection.

@article{
 title = {NIR attribute selection for the development of vineyard water status predictive models},
 type = {article},
 year = {2023},
 keywords = {Grapevine,Interval Partial Least Squares,Manual wavelength selection,Stem water potential,Variable Importance in Projection scores},
 pages = {167-178},
 volume = {229},
 month = {5},
 publisher = {Academic Press},
 day = {1},
 id = {8aa48acf-a33d-3a3e-9ffc-917abc258ec0},
 created = {2023-10-27T10:16:58.488Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:30.474Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Near-Infrared spectroscopy (NIR) returns full spectra in the region between 750 and 2500 nm. Although a full spectrum provides extremely informative data, sometimes this enormous amount of detail is redundant and does not bring any additional information. In this work, different attribute selection methods for the development of vineyard water status predictive models are presented. Spectra from grapevine leaves were collected on-the-go (from a moving vehicle) along nine dates during the 2015 season in a commercial vineyard using a NIR spectrometer (1200–2100 nm). Contemporarily, the stem water potential (Ψstem) was also measured in the monitored vines. A manual selection, based on Variable Importance in Projection scores (VIP scores) to choose the spectrum intervals including the most important wavelengths (interval selection), the locally most important wavelengths in the spectrum (peak selection), as well as the Interval Partial Least Squares (IPLS) were tested as attribute selection methods. The results obtained for the estimation of Ψstem using the whole spectrum (R2P = 0.84, RMSEP = 0.167 MPa) were comparable to those yielded by the three attribute selection methods: the interval selection method (R2P = 0.80, RMSEP = 0.186 MPa), the peak selection method (R2P = 0.77, RMSEP = 0.201 MPa) and the IPLS (R2P ∼ 0.62–0.79, RMSEP ∼ 0.186–0.252 MPa). The highest simplification was provided by two IPLS models with three wavelengths and bandwidths of 20 and 4 nm that yielded R2P∼0.78 and RMSEP∼ 0.190 MPa. These results corroborate the suitability of a highly reduced selection of NIR wavelengths for the prediction of grapevine water status, and its utility to develop simpler multispectral devices for vineyard water status estimation.},
 bibtype = {article},
 author = {Marañón, Miguel and Fernández-Novales, Juan and Tardaguila, Javier and Gutiérrez, Salvador and Diago, Maria P.},
 doi = {10.1016/J.BIOSYSTEMSENG.2023.04.001},
 journal = {Biosystems Engineering}
}

Near-Infrared spectroscopy (NIR) returns full spectra in the region between 750 and 2500 nm. Although a full spectrum provides extremely informative data, sometimes this enormous amount of detail is redundant and does not bring any additional information. In this work, different attribute selection methods for the development of vineyard water status predictive models are presented. Spectra from grapevine leaves were collected on-the-go (from a moving vehicle) along nine dates during the 2015 season in a commercial vineyard using a NIR spectrometer (1200–2100 nm). Contemporarily, the stem water potential (Ψstem) was also measured in the monitored vines. A manual selection, based on Variable Importance in Projection scores (VIP scores) to choose the spectrum intervals including the most important wavelengths (interval selection), the locally most important wavelengths in the spectrum (peak selection), as well as the Interval Partial Least Squares (IPLS) were tested as attribute selection methods. The results obtained for the estimation of Ψstem using the whole spectrum (R2P = 0.84, RMSEP = 0.167 MPa) were comparable to those yielded by the three attribute selection methods: the interval selection method (R2P = 0.80, RMSEP = 0.186 MPa), the peak selection method (R2P = 0.77, RMSEP = 0.201 MPa) and the IPLS (R2P ∼ 0.62–0.79, RMSEP ∼ 0.186–0.252 MPa). The highest simplification was provided by two IPLS models with three wavelengths and bandwidths of 20 and 4 nm that yielded R2P∼0.78 and RMSEP∼ 0.190 MPa. These results corroborate the suitability of a highly reduced selection of NIR wavelengths for the prediction of grapevine water status, and its utility to develop simpler multispectral devices for vineyard water status estimation.

@article{
 title = {Evaluating the Potential of High-Resolution Visible Remote Sensing to Detect Shiraz Disease in Grapevines},
 type = {article},
 year = {2023},
 pages = {1-9},
 volume = {2023},
 month = {5},
 publisher = {Hindawi Limited},
 day = {5},
 id = {07dbd7c8-6a3d-3fcd-b3a4-5d3a3d9089c3},
 created = {2023-10-27T10:20:13.820Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:53:05.776Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Background and Aims. Shiraz disease (SD) is a viral disease associated with Grapevine virus A that causes significant yield loss in economically important grape cultivars in Australia such as Shiraz and Merlot. Current diagnostic methods are time-consuming and costly. This study evaluates an alternative methodology using visible remote sensing imagery to detect SD in Shiraz grapevines. Methods and Results. High-resolution visible remote sensing images were captured of Shiraz grapevines in two South Australian viticultural regions over two seasons. The projected leaf area (PLA) of individual grapevines was estimated from the images. Virus-infected vines had significantly lower PLA than healthy vines in the early season but fewer difference after veraison. The lower PLA was only observed in grapevines coinfected with grapevine leafroll-associated viruses (GLRaVs) and Grapevine virus A (GVA). Shiraz vines infected with either GLRaVs or GVA had similar PLA to healthy vines. Conclusions. High-resolution RGB remote sensing technology has the potential to rapidly estimate SD infection in Shiraz grapevines. Our observations of shoot devigouration only in coinfected vines calls into question the etiology of SD. Further validation of the PLA technique incorporating different regions, seasons, cultivars, and combinations of viruses is needed for improving the robustness of the method. Significance of the Study. This preliminary study presents a new rapid and low-cost surveillance method to estimate SD infections in Shiraz vineyards, which could significantly lower the cost for growers who conduct on-ground SD visual assessments or lab-based tissue testing at the vineyard scale.},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Pagay, Vinay},
 doi = {10.1155/2023/7376153},
 journal = {Australian Journal of Grape and Wine Research},
 keywords = {wang2023evaluatingpotentialhighresolution}
}

Background and Aims. Shiraz disease (SD) is a viral disease associated with Grapevine virus A that causes significant yield loss in economically important grape cultivars in Australia such as Shiraz and Merlot. Current diagnostic methods are time-consuming and costly. This study evaluates an alternative methodology using visible remote sensing imagery to detect SD in Shiraz grapevines. Methods and Results. High-resolution visible remote sensing images were captured of Shiraz grapevines in two South Australian viticultural regions over two seasons. The projected leaf area (PLA) of individual grapevines was estimated from the images. Virus-infected vines had significantly lower PLA than healthy vines in the early season but fewer difference after veraison. The lower PLA was only observed in grapevines coinfected with grapevine leafroll-associated viruses (GLRaVs) and Grapevine virus A (GVA). Shiraz vines infected with either GLRaVs or GVA had similar PLA to healthy vines. Conclusions. High-resolution RGB remote sensing technology has the potential to rapidly estimate SD infection in Shiraz grapevines. Our observations of shoot devigouration only in coinfected vines calls into question the etiology of SD. Further validation of the PLA technique incorporating different regions, seasons, cultivars, and combinations of viruses is needed for improving the robustness of the method. Significance of the Study. This preliminary study presents a new rapid and low-cost surveillance method to estimate SD infections in Shiraz vineyards, which could significantly lower the cost for growers who conduct on-ground SD visual assessments or lab-based tissue testing at the vineyard scale.

@article{
 title = {Unstructured road extraction and roadside fruit recognition in grape orchards based on a synchronous detection algorithm},
 type = {article},
 year = {2023},
 keywords = {deep learning,fruit harvesting robot,machine vision,non-structural environment,roadside fruits detection},
 pages = {1-22},
 volume = {14},
 id = {24a7aecc-0904-39e3-99a2-a16ebb199121},
 created = {2023-10-27T10:20:18.937Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-27T10:20:25.317Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Accurate road extraction and recognition of roadside fruit in complex orchard environments are essential prerequisites for robotic fruit picking and walking behavioral decisions. In this study, a novel algorithm was proposed for unstructured road extraction and roadside fruit synchronous recognition, with wine grapes and nonstructural orchards as research objects. Initially, a preprocessing method tailored to field orchards was proposed to reduce the interference of adverse factors in the operating environment. The preprocessing method contained 4 parts: interception of regions of interest, bilateral filter, logarithmic space transformation and image enhancement based on the MSRCR algorithm. Subsequently, the analysis of the enhanced image enabled the optimization of the gray factor, and a road region extraction method based on dual-space fusion was proposed by color channel enhancement and gray factor optimization. Furthermore, the YOLO model suitable for grape cluster recognition in the wild environment was selected, and its parameters were optimized to enhance the recognition performance of the model for randomly distributed grapes. Finally, a fusion recognition framework was innovatively established, wherein the road extraction result was taken as input, and the optimized parameter YOLO model was utilized to identify roadside fruits, thus realizing synchronous road extraction and roadside fruit detection. Experimental results demonstrated that the proposed method based on the pretreatment could reduce the impact of interfering factors in complex orchard environments and enhance the quality of road extraction. Using the optimized YOLOv7 model, the precision, recall, mAP, and F1-score for roadside fruit cluster detection were 88.9%, 89.7%, 93.4%, and 89.3%, respectively, all of which were higher than those of the YOLOv5 model and were more suitable for roadside grape recognition. Compared to the identification results obtained by the grape detection algorithm alone, the proposed synchronous algorithm increased the number of fruit identifications by 23.84% and the detection speed by 14.33%. This research enhanced the perception ability of robots and provided a solid support for behavioral decision systems.},
 bibtype = {article},
 author = {Zhou, Xinzhao and Zou, Xiangjun and Tang, Wei and Yan, Zhiwei and Meng, Hewei and Luo, Xiwen},
 doi = {10.3389/fpls.2023.1103276},
 journal = {Frontiers in Plant Science},
 number = {June}
}

Accurate road extraction and recognition of roadside fruit in complex orchard environments are essential prerequisites for robotic fruit picking and walking behavioral decisions. In this study, a novel algorithm was proposed for unstructured road extraction and roadside fruit synchronous recognition, with wine grapes and nonstructural orchards as research objects. Initially, a preprocessing method tailored to field orchards was proposed to reduce the interference of adverse factors in the operating environment. The preprocessing method contained 4 parts: interception of regions of interest, bilateral filter, logarithmic space transformation and image enhancement based on the MSRCR algorithm. Subsequently, the analysis of the enhanced image enabled the optimization of the gray factor, and a road region extraction method based on dual-space fusion was proposed by color channel enhancement and gray factor optimization. Furthermore, the YOLO model suitable for grape cluster recognition in the wild environment was selected, and its parameters were optimized to enhance the recognition performance of the model for randomly distributed grapes. Finally, a fusion recognition framework was innovatively established, wherein the road extraction result was taken as input, and the optimized parameter YOLO model was utilized to identify roadside fruits, thus realizing synchronous road extraction and roadside fruit detection. Experimental results demonstrated that the proposed method based on the pretreatment could reduce the impact of interfering factors in complex orchard environments and enhance the quality of road extraction. Using the optimized YOLOv7 model, the precision, recall, mAP, and F1-score for roadside fruit cluster detection were 88.9%, 89.7%, 93.4%, and 89.3%, respectively, all of which were higher than those of the YOLOv5 model and were more suitable for roadside grape recognition. Compared to the identification results obtained by the grape detection algorithm alone, the proposed synchronous algorithm increased the number of fruit identifications by 23.84% and the detection speed by 14.33%. This research enhanced the perception ability of robots and provided a solid support for behavioral decision systems.

@article{
 title = {Using a Camera System for the In-Situ Assessment of Cordon Dieback due to Grapevine Trunk Diseases},
 type = {article},
 year = {2023},
 websites = {https://doi.org/10.1155/2023/8634742},
 id = {4a68233e-bb72-34e9-a4be-7cc50d04653e},
 created = {2023-11-06T10:55:17.408Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-16T09:07:18.879Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Background and Aims. .e assessment of grapevine trunk disease symptoms is a labour-intensive process that requires experience and is prone to bias. Methods that support the easy and accurate monitoring of trunk diseases will aid management decisions. Methods and Results. An algorithm was developed for the assessment of dieback symptoms due to trunk disease which is applied on a smartphone mounted on a vehicle driven through the vineyard. Vine images and corresponding expert ground truth assessments (of over 13,000 vines) were collected and correlated over two seasons in Shiraz vineyards in the Clare Valley, Barossa, and McLaren Vale, South Australia. .is dataset was used to train and verify YOLOv5 models to estimate the percentage dieback of cordons due to trunk diseases. .e performance of the models was evaluated on the metrics of highest conndence, highest dieback score, and average dieback score across multiple detections. Eighty-four percent of vines in a test set derived from an unseen vineyard were assigned a score by the model within 10% of the score given by experts in the vineyard. Conclusions. .e computer vision algorithms were implemented within the phone, allowing real-time assessment and row-level mapping with nothing more than a high-end mobile phone. Signiicance of the Study. .e algorithms form the basis of a system that will allow growers to scan their vineyards easily and regularly to monitor dieback due to grapevine trunk disease and will facilitate corrective interventions.},
 bibtype = {article},
 author = {Tang, Julie and Yem, Olivia and Russell, Finn and Stewart, Cameron A and Lin, Kangying and Jayakody, Hiranya and Ayres, Matthew R and Sosnowski, Mark R and Whitty, Mark and Petrie, Paul R},
 doi = {10.1155/2023/8634742}
}

Background and Aims. .e assessment of grapevine trunk disease symptoms is a labour-intensive process that requires experience and is prone to bias. Methods that support the easy and accurate monitoring of trunk diseases will aid management decisions. Methods and Results. An algorithm was developed for the assessment of dieback symptoms due to trunk disease which is applied on a smartphone mounted on a vehicle driven through the vineyard. Vine images and corresponding expert ground truth assessments (of over 13,000 vines) were collected and correlated over two seasons in Shiraz vineyards in the Clare Valley, Barossa, and McLaren Vale, South Australia. .is dataset was used to train and verify YOLOv5 models to estimate the percentage dieback of cordons due to trunk diseases. .e performance of the models was evaluated on the metrics of highest conndence, highest dieback score, and average dieback score across multiple detections. Eighty-four percent of vines in a test set derived from an unseen vineyard were assigned a score by the model within 10% of the score given by experts in the vineyard. Conclusions. .e computer vision algorithms were implemented within the phone, allowing real-time assessment and row-level mapping with nothing more than a high-end mobile phone. Signiicance of the Study. .e algorithms form the basis of a system that will allow growers to scan their vineyards easily and regularly to monitor dieback due to grapevine trunk disease and will facilitate corrective interventions.

@article{
 title = {Exploratory approach for automatic detection of vine rows in terrace vineyards},
 type = {article},
 year = {2023},
 keywords = {Artificial Intelligence,Precision agriculture,Remote sensing,Terrace vineyards},
 pages = {139-144},
 volume = {219},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {07f3c769-1b1c-3f33-bf9d-870c047cca01},
 created = {2023-11-06T11:17:05.714Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-16T09:07:13.959Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4c7c81ce-f24b-44ae-bc2a-bf60600a3a24,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The Alto Douro Demarcated Region in Portugal is the oldest and most regulated wine-growing region in the world, formed by an ecosystem of unique value allowing the cultivation of vines on its characteristics terraces vineyards. The detection of vine rows in terrace vineyards constitutes an essential task regarding the achievement of important goals such as multi-Temporal crop evaluation and production estimation. Despite the advances and research in this field, most studies are limited to flat vineyards with straight vine rows. In this study an exploratory approach in the precision agriculture for automatic detection of vine rows in terrace vineyards is presented with remote sensing techniques associated with artificial intelligence such as Machine Learning and Deep learning. At the current stage the preliminary results are encouraging for the detection of vine rows in straight and curved lines considering the complexity of the terrain.},
 bibtype = {article},
 author = {Figueiredo, Nuno and Padua, Luis and Cunha, Antonio and Sousa, Joaquim J. and Sousa, Antonio},
 doi = {10.1016/J.PROCS.2023.01.274},
 journal = {Procedia Computer Science}
}

The Alto Douro Demarcated Region in Portugal is the oldest and most regulated wine-growing region in the world, formed by an ecosystem of unique value allowing the cultivation of vines on its characteristics terraces vineyards. The detection of vine rows in terrace vineyards constitutes an essential task regarding the achievement of important goals such as multi-Temporal crop evaluation and production estimation. Despite the advances and research in this field, most studies are limited to flat vineyards with straight vine rows. In this study an exploratory approach in the precision agriculture for automatic detection of vine rows in terrace vineyards is presented with remote sensing techniques associated with artificial intelligence such as Machine Learning and Deep learning. At the current stage the preliminary results are encouraging for the detection of vine rows in straight and curved lines considering the complexity of the terrain.

@article{
 title = {MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields},
 type = {article},
 year = {2023},
 keywords = {agricultural fields,edge detection,multi,remote sensing images,semantic segmentation,task learning},
 pages = {3934},
 volume = {15},
 websites = {https://www.mdpi.com/2072-4292/15/16/3934/htm,https://www.mdpi.com/2072-4292/15/16/3934},
 month = {8},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {8},
 id = {55df9797-5493-316c-945c-7358f59e50e7},
 created = {2023-11-07T08:51:09.069Z},
 accessed = {2023-11-07},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-07T08:51:15.288Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The implementation of precise agricultural fields can drive the intelligent development of agricultural production, and high-resolution remote sensing images provide convenience for obtaining precise fields. With the advancement of spatial resolution, the complexity and heterogeneity of land features are accentuated, making it challenging for existing methods to obtain structurally complete fields, especially in regions with blurred edges. Therefore, a multi-task learning network with attention-guided mechanism is introduced for segmenting agricultural fields. To be more specific, the attention-guided fusion module is used to learn complementary information layer by layer, while the multi-task learning scheme considers both edge detection and semantic segmentation task. Based on this, we further segmented the merged fields using broken edges, following the theory of connectivity perception. Finally, we chose three cities in The Netherlands as study areas for experimentation, and evaluated the extracted field regions and edges separately, the results showed that (1) The proposed method achieved the highest accuracy in three cities, with IoU of 91.27%, 93.05% and 89.76%, respectively. (2) The Qua metrics of the processed edges demonstrated improvements of 6%, 6%, and 5%, respectively. This work successfully segmented potential fields with blurred edges, indicating its potential for precision agriculture development.},
 bibtype = {article},
 author = {Mondino, Borgogno and Sarvia, Filippo and De Petris, Samuele and Orusa, Tommaso and Luo, Weiran and Zhang, Chengcai and Li, Ying and Yan, Yaning},
 doi = {10.3390/RS15163934},
 journal = {Remote Sensing 2023, Vol. 15, Page 3934},
 number = {16}
}

The implementation of precise agricultural fields can drive the intelligent development of agricultural production, and high-resolution remote sensing images provide convenience for obtaining precise fields. With the advancement of spatial resolution, the complexity and heterogeneity of land features are accentuated, making it challenging for existing methods to obtain structurally complete fields, especially in regions with blurred edges. Therefore, a multi-task learning network with attention-guided mechanism is introduced for segmenting agricultural fields. To be more specific, the attention-guided fusion module is used to learn complementary information layer by layer, while the multi-task learning scheme considers both edge detection and semantic segmentation task. Based on this, we further segmented the merged fields using broken edges, following the theory of connectivity perception. Finally, we chose three cities in The Netherlands as study areas for experimentation, and evaluated the extracted field regions and edges separately, the results showed that (1) The proposed method achieved the highest accuracy in three cities, with IoU of 91.27%, 93.05% and 89.76%, respectively. (2) The Qua metrics of the processed edges demonstrated improvements of 6%, 6%, and 5%, respectively. This work successfully segmented potential fields with blurred edges, indicating its potential for precision agriculture development.

@article{
 title = {MTA-YOLACT: Multitask-aware network on fruit bunch identification for cherry tomato robotic harvesting},
 type = {article},
 year = {2023},
 keywords = {Decision tree,Multitask-aware network,Tomato harvesting robot,YOLACT},
 pages = {126812},
 volume = {146},
 month = {5},
 publisher = {Elsevier},
 day = {1},
 id = {47e9274a-975f-3d54-bb52-eb3366c38eae},
 created = {2023-11-07T08:59:11.819Z},
 accessed = {2023-11-07},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-08T10:14:50.066Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Accurate and rapid perception of fruit bunch posture is necessary for the cherry tomato harvesting robot to successfully achieve the bunch's holding and separating. According to the postural relationship of the fruit bunch, bunch pedicel, and plant’ main-stem, the robotic end-effector's holding region and approach path could be determined, which were important for successful picking operation. The main goal of this research was to propose a multitask-aware network (MTA-YOLACT), which simultaneously performed region detection on fruit bunch, and region segmentation on pedicel and main-stem. The MTA-YOLACT extended from the pre-trained YOLACT model, included two detector branch networks for detection and instance segment, which shared the same backbone network, and the loss function with weighting coefficients of the two branches was adopted to balance the multi-task learning, according to multi-task's homoscedastic uncertainty during the model training. Furthermore, in order to cluster the fruit bunch, pedicel and main-stem from the same bunch target, a classification and regression tree (CART) model was built, based on the region's positional relationship from the MTA-YOLACT output. An image dataset of cherry tomato plants in China greenhouse was built to training and test the model. The results indicated a promising performance of the proposed network, with an F1-score of 95.4% on detecting fruit bunches and the mean Average Precision of 38.7% and 51.9% on the instance segmentation of pedicel and main-stem, which was 1.1% and 3.5% more than original YOLACT. Beyond that, our approach performed a real-time detection and instance segmentation of 13.3 frames per second (FPS). The whole bunch could be identified by the CART model with an average accuracy of 99.83% and the time cost of 9.53 ms. These results demonstrated the research could be a viable support to the harvesting robot's vision unit development and the end-effector's motion planning in the future research.},
 bibtype = {article},
 author = {Li, Yajun and Feng, Qingchun and Liu, Cheng and Xiong, Zicong and Sun, Yuhuan and Xie, Feng and Li, Tao and Zhao, Chunjiang},
 doi = {10.1016/J.EJA.2023.126812},
 journal = {European Journal of Agronomy}
}

Accurate and rapid perception of fruit bunch posture is necessary for the cherry tomato harvesting robot to successfully achieve the bunch's holding and separating. According to the postural relationship of the fruit bunch, bunch pedicel, and plant’ main-stem, the robotic end-effector's holding region and approach path could be determined, which were important for successful picking operation. The main goal of this research was to propose a multitask-aware network (MTA-YOLACT), which simultaneously performed region detection on fruit bunch, and region segmentation on pedicel and main-stem. The MTA-YOLACT extended from the pre-trained YOLACT model, included two detector branch networks for detection and instance segment, which shared the same backbone network, and the loss function with weighting coefficients of the two branches was adopted to balance the multi-task learning, according to multi-task's homoscedastic uncertainty during the model training. Furthermore, in order to cluster the fruit bunch, pedicel and main-stem from the same bunch target, a classification and regression tree (CART) model was built, based on the region's positional relationship from the MTA-YOLACT output. An image dataset of cherry tomato plants in China greenhouse was built to training and test the model. The results indicated a promising performance of the proposed network, with an F1-score of 95.4% on detecting fruit bunches and the mean Average Precision of 38.7% and 51.9% on the instance segmentation of pedicel and main-stem, which was 1.1% and 3.5% more than original YOLACT. Beyond that, our approach performed a real-time detection and instance segmentation of 13.3 frames per second (FPS). The whole bunch could be identified by the CART model with an average accuracy of 99.83% and the time cost of 9.53 ms. These results demonstrated the research could be a viable support to the harvesting robot's vision unit development and the end-effector's motion planning in the future research.

@article{
 title = {An efficient multi-task convolutional neural network for dairy farm object detection and segmentation},
 type = {article},
 year = {2023},
 keywords = {Dairy farm,GCS-MUL,Multi-task learning,Target identification},
 pages = {108000},
 volume = {211},
 month = {8},
 publisher = {Elsevier},
 day = {1},
 id = {6409dd1d-f148-3b11-a669-e360e37d5a6f},
 created = {2023-11-07T08:59:58.167Z},
 accessed = {2023-11-07},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-07T09:00:08.895Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Real-time and accurate detection of multiple types of targets and obstacles in dairy barns is a necessary function for autonomous pushing robots. To improve the efficiency of target recognition and to reduce the path extraction error of the pushing robot, on the basis of the high accuracy perception of every pixel collected with an embedded AI computer, a multi-task learning based dairy barn multi-type target recognition model Ghost CBAM Segmentation-Multi-task (GCS-MUL) was proposed, which could recognize dairy cows, obstacles and road targets in real-time and efficiently. Firstly, in order to enhance the ability to extract key features from the targets, the proposed model intergrades the Convolutional Block Attention Module (CBAM), a self-designed light-weight target feature extraction network Ghost CBAM Network (GCNet) as the backbone of the whole model. Secondly, to improve the model multi-scale feature fusion, Path Aggregation Network (PAN) and Feature Pyramid Network (FPN) structures with the GhostConv module were used in neck net. Finally, for real-time semantic segmentation dairy farms multiple targets, a Segmentation Head (Seg Head), which is composed of the Receptive Field Block (RFB), Pyramid Pooling Module (PPM) and Feature Fusion Module (FFM), was introduced. Experimental results showed that the mAP@0.5 (mean average precision IoU = 0.5) of the dairy farm target reached 94.86%. Compared to the YOLOv5 model, the precision and recall was improved by 7.47% and 6.85%, respectively. In comparison to the YOLOv7 model, the precision was improved by 5.1%. Furthermore, when compared to the SSD model, the proposed model have reduced the number of model parameters by 92.43%, and its average detection time was reduced by 84.37 ms, which is ideal for meeting the real-time target recognition requirements. The average detection time of the model is 66.43 ms, making it more suitable for deployment in embedded devices. Compared with Ghost CBAM-Detection (GC-Detect) without the introduction of the Seg Head, the precision, recall and mAP@0.5 was improved by 4.49%, 4.92% and 6.58%, respectively. The research results can provide accurate algorithms for real-time and efficient identification of dairy farm targets for pushing robots, and provide more effective road and environmental scene segmentation methods for autonomous walking.},
 bibtype = {article},
 author = {Tian, Fuyang and Hu, Guozheng and Yu, Sufang and Wang, Ruixue and Song, Zhanhua and Yan, Yinfa and Huang, Hailing and Wang, Qing and Wang, Zhonghua and Yu, Zhenwei},
 doi = {10.1016/J.COMPAG.2023.108000},
 journal = {Computers and Electronics in Agriculture}
}

Real-time and accurate detection of multiple types of targets and obstacles in dairy barns is a necessary function for autonomous pushing robots. To improve the efficiency of target recognition and to reduce the path extraction error of the pushing robot, on the basis of the high accuracy perception of every pixel collected with an embedded AI computer, a multi-task learning based dairy barn multi-type target recognition model Ghost CBAM Segmentation-Multi-task (GCS-MUL) was proposed, which could recognize dairy cows, obstacles and road targets in real-time and efficiently. Firstly, in order to enhance the ability to extract key features from the targets, the proposed model intergrades the Convolutional Block Attention Module (CBAM), a self-designed light-weight target feature extraction network Ghost CBAM Network (GCNet) as the backbone of the whole model. Secondly, to improve the model multi-scale feature fusion, Path Aggregation Network (PAN) and Feature Pyramid Network (FPN) structures with the GhostConv module were used in neck net. Finally, for real-time semantic segmentation dairy farms multiple targets, a Segmentation Head (Seg Head), which is composed of the Receptive Field Block (RFB), Pyramid Pooling Module (PPM) and Feature Fusion Module (FFM), was introduced. Experimental results showed that the mAP@0.5 (mean average precision IoU = 0.5) of the dairy farm target reached 94.86%. Compared to the YOLOv5 model, the precision and recall was improved by 7.47% and 6.85%, respectively. In comparison to the YOLOv7 model, the precision was improved by 5.1%. Furthermore, when compared to the SSD model, the proposed model have reduced the number of model parameters by 92.43%, and its average detection time was reduced by 84.37 ms, which is ideal for meeting the real-time target recognition requirements. The average detection time of the model is 66.43 ms, making it more suitable for deployment in embedded devices. Compared with Ghost CBAM-Detection (GC-Detect) without the introduction of the Seg Head, the precision, recall and mAP@0.5 was improved by 4.49%, 4.92% and 6.58%, respectively. The research results can provide accurate algorithms for real-time and efficient identification of dairy farm targets for pushing robots, and provide more effective road and environmental scene segmentation methods for autonomous walking.

@article{
 title = {Grape Cold Hardiness Prediction via Multi-Task Learning},
 type = {article},
 year = {2023},
 keywords = {Agriculture,Grapevines,Recurrent Neural Networks,Time Series Modelling,Transfer Learning},
 pages = {15717-15723},
 volume = {37},
 websites = {https://ojs.aaai.org/index.php/AAAI/article/view/26865},
 month = {9},
 publisher = {AAAI Press},
 day = {6},
 id = {9cfaa664-03f2-37ff-83be-15ddd02b6942},
 created = {2023-11-07T11:13:55.515Z},
 accessed = {2023-11-07},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-07T11:13:59.645Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Cold temperatures during fall and spring have the potential to cause frost damage to grapevines and other fruit plants, which can significantly decrease harvest yields. To help prevent these losses, farmers deploy expensive frost mitigation measures, such as, sprinklers, heaters, and wind machines, when they judge that damage may occur. This judgment, however, is challenging because the cold hardiness of plants changes throughout the dormancy period and it is difficult to directly measure. This has led scientists to develop cold hardiness prediction models that can be tuned to different grape cultivars based on laborious field measurement data. In this paper, we study whether deep-learning models can improve cold hardiness prediction for grapes based on data that has been collected over a 30-year time period. A key challenge is that the amount of data per cultivar is highly variable, with some cultivars having only a small amount. For this purpose, we investigate the use of multi-task learning to leverage data across cultivars in order to improve prediction performance for individual cultivars. We evaluate a number of multi-task learning approaches and show that the highest performing approach is able to significantly improve over learning for single cultivars and outperforms the current state-of-the-art scientific model for most cultivars.},
 bibtype = {article},
 author = {Saxena, Aseem and Pesantez-Cabrera, Paola and Ballapragada, Rohan and Lam, Kin Ho and Keller, Markus and Fern, Alan},
 doi = {10.1609/AAAI.V37I13.26865},
 journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
 number = {13}
}

Cold temperatures during fall and spring have the potential to cause frost damage to grapevines and other fruit plants, which can significantly decrease harvest yields. To help prevent these losses, farmers deploy expensive frost mitigation measures, such as, sprinklers, heaters, and wind machines, when they judge that damage may occur. This judgment, however, is challenging because the cold hardiness of plants changes throughout the dormancy period and it is difficult to directly measure. This has led scientists to develop cold hardiness prediction models that can be tuned to different grape cultivars based on laborious field measurement data. In this paper, we study whether deep-learning models can improve cold hardiness prediction for grapes based on data that has been collected over a 30-year time period. A key challenge is that the amount of data per cultivar is highly variable, with some cultivars having only a small amount. For this purpose, we investigate the use of multi-task learning to leverage data across cultivars in order to improve prediction performance for individual cultivars. We evaluate a number of multi-task learning approaches and show that the highest performing approach is able to significantly improve over learning for single cultivars and outperforms the current state-of-the-art scientific model for most cultivars.

@article{
 title = {Multi-task Transfer Learning Facilitated by Segmentation and Denoising for Anomaly Detection of Rail Fasteners},
 type = {article},
 year = {2023},
 keywords = {Anomaly detection,Denoising,Multi-task transfer learning,Railway fastener,Segmentation},
 pages = {2383-2394},
 volume = {18},
 websites = {https://link.springer.com/article/10.1007/s42835-022-01347-1},
 month = {5},
 publisher = {Korean Institute of Electrical Engineers},
 day = {1},
 id = {87040c64-d3ce-3623-b338-3a646954e781},
 created = {2023-11-08T10:16:08.570Z},
 accessed = {2023-11-08},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-14T10:14:35.248Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The rail fastener is the main component of rail tracks that contributes to safe train travel by fixing the rail to railroad ties. Anomalies such as the absence or damage should be checked regularly as they can lead to large-scale accidents. Image-based maintenance systems have been proposed as efficient management, but it is still difficult to detect anomalies due to occlusion by obstacles or noise in the image. Therefore, we propose a deep anomaly detection system for rail fasteners using multi-task transfer learning. First, the U-Net model is trained for an auxiliary task consisting of segmentation and denoising. Second, in the transfer learning process, a machine learning or deep learning-based classifier detects anomalies using the feature map obtained from the trained U-Net encoder. The proposed model is rigorously evaluated with our collected data. The experimental results show that the deep learning-based classifier detected anomalies with an accuracy of about 97.57%, and multi-task transfer learning contributes to the model focusing on the fastener region in the image. It suggests the potential of image-based automatic defect detection systems for many industrial applications.},
 bibtype = {article},
 author = {Kim, Beomjun and Jeon, Younghoon and Kang, Jeong Won and Gwak, Jeonghwan},
 doi = {10.1007/S42835-022-01347-1/FIGURES/19},
 journal = {Journal of Electrical Engineering and Technology},
 number = {3}
}

The rail fastener is the main component of rail tracks that contributes to safe train travel by fixing the rail to railroad ties. Anomalies such as the absence or damage should be checked regularly as they can lead to large-scale accidents. Image-based maintenance systems have been proposed as efficient management, but it is still difficult to detect anomalies due to occlusion by obstacles or noise in the image. Therefore, we propose a deep anomaly detection system for rail fasteners using multi-task transfer learning. First, the U-Net model is trained for an auxiliary task consisting of segmentation and denoising. Second, in the transfer learning process, a machine learning or deep learning-based classifier detects anomalies using the feature map obtained from the trained U-Net encoder. The proposed model is rigorously evaluated with our collected data. The experimental results show that the deep learning-based classifier detected anomalies with an accuracy of about 97.57%, and multi-task transfer learning contributes to the model focusing on the fastener region in the image. It suggests the potential of image-based automatic defect detection systems for many industrial applications.

@article{
 title = {Plant-water relations},
 type = {article},
 year = {2023},
 pages = {516-526},
 publisher = {Elsevier},
 id = {3eeac158-12ce-3096-b064-ef8b70a5ad0b},
 created = {2023-11-13T12:58:15.151Z},
 accessed = {2023-11-13},
 file_attached = {false},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-13T12:58:15.470Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {2bfb8d91-9fac-46f0-bd0c-93235d01dbed},
 private_publication = {false},
 bibtype = {article},
 author = {Dodd, Ian C. and Hirons, Andrew D. and Puértolas, Jaime},
 doi = {10.1016/B978-0-12-822974-3.00253-6},
 journal = {Encyclopedia of Soils in the Environment}
}

@article{
 title = {Scalable early detection of grapevine virus infection with airborne imaging spectroscopy},
 type = {article},
 year = {2023},
 keywords = {galvan2023scalableearlydetection},
 websites = {https://apsjournals.apsnet.org/doi/10.1094/PHYTO-01-23-0030-R},
 month = {8},
 publisher = {Scientific Societies},
 day = {25},
 id = {6dd24140-fee6-30db-a64f-0c8d61a86efc},
 created = {2023-11-14T11:17:40.710Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:05.752Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The US wine and grape industry suffers $3B in damages and losses annually due to viral diseases such as Grapevine Leafroll-associated Virus Complex 3 (GLRaV-3). Current detection methods are labor intensive and expensive. GLRaV-3 undergoes a latent period in which the vines are infected but do not yet display visible symptoms, making it an ideal model to evaluate the scalability of imaging spectroscopy-based disease detection. We deployed the NASA Airborne Visible and Infrared Imaging Spectrometer Next Generation (AVIRIS-NG) to detect GLRaV-3 in Cabernet Sauvignon grapevines in Lodi, CA in September 2020. Foliage was removed from the vines as part of mechanical harvest soon after imagery acquisition. In both Sept. 2020 and 2021, industry collaborators scouted 317ac on a vine-by-vine basis for visible viral symptoms and collected a subset for molecular confirmation testing. Grapevines identified as visibly diseased in 2021, but not 2020, were assumed to have been latently infected at time of acquisition. We trained spectral models with random forest and synthetic minority oversampling technique to distinguish non-infected and GLRaV-3-infected grapevines. Non-infected and GLRaV-3 infected vines could be differentiated both pre- and post-symptomatically at 1m through 5m resolution. The best-performing models had 87% accuracy distinguishing between non-infected and asymptomatic vines, and 85% accuracy distinguishing between non-infected and asymptomatic + symptomatic vines. The importance of non-visible wavelengths suggests this capacity is driven by disease-induced changes to overall plant physiology. Our work sets a foundation for using the forthcoming hyperspectral satellite Surface Biology and Geology for regional disease monitoring.},
 bibtype = {article},
 author = {Romero Galvan, Fernando and Pavlick, Ryan and Trolley, Graham Richard and Aggarwal, Somil and Sousa, Daniel and Starr, Charlie and Forrestel, Elisabeth Jane and Bolton, Stephanie and Alsina, Maria del Mar and Dokoozlian, Nicholaus and Gold, Kaitlin Morey},
 doi = {10.1094/PHYTO-01-23-0030-R/ASSET/IMAGES/LARGE/PHYTO-01-23-0030-RF5.JPEG},
 journal = {Phytopathology®}
}

The US wine and grape industry suffers $3B in damages and losses annually due to viral diseases such as Grapevine Leafroll-associated Virus Complex 3 (GLRaV-3). Current detection methods are labor intensive and expensive. GLRaV-3 undergoes a latent period in which the vines are infected but do not yet display visible symptoms, making it an ideal model to evaluate the scalability of imaging spectroscopy-based disease detection. We deployed the NASA Airborne Visible and Infrared Imaging Spectrometer Next Generation (AVIRIS-NG) to detect GLRaV-3 in Cabernet Sauvignon grapevines in Lodi, CA in September 2020. Foliage was removed from the vines as part of mechanical harvest soon after imagery acquisition. In both Sept. 2020 and 2021, industry collaborators scouted 317ac on a vine-by-vine basis for visible viral symptoms and collected a subset for molecular confirmation testing. Grapevines identified as visibly diseased in 2021, but not 2020, were assumed to have been latently infected at time of acquisition. We trained spectral models with random forest and synthetic minority oversampling technique to distinguish non-infected and GLRaV-3-infected grapevines. Non-infected and GLRaV-3 infected vines could be differentiated both pre- and post-symptomatically at 1m through 5m resolution. The best-performing models had 87% accuracy distinguishing between non-infected and asymptomatic vines, and 85% accuracy distinguishing between non-infected and asymptomatic + symptomatic vines. The importance of non-visible wavelengths suggests this capacity is driven by disease-induced changes to overall plant physiology. Our work sets a foundation for using the forthcoming hyperspectral satellite Surface Biology and Geology for regional disease monitoring.

@article{
 title = {Evaluating the Potential of High-Resolution Visible Remote Sensing to Detect Shiraz Disease in Grapevines},
 type = {article},
 year = {2023},
 pages = {1-9},
 volume = {2023},
 month = {5},
 publisher = {Hindawi Limited},
 day = {5},
 id = {c49d2169-01e3-33af-909e-e854e8c5e758},
 created = {2023-11-14T11:37:37.632Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-14T11:37:43.351Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Background and Aims. Shiraz disease (SD) is a viral disease associated with Grapevine virus A that causes significant yield loss in economically important grape cultivars in Australia such as Shiraz and Merlot. Current diagnostic methods are time-consuming and costly. This study evaluates an alternative methodology using visible remote sensing imagery to detect SD in Shiraz grapevines. Methods and Results. High-resolution visible remote sensing images were captured of Shiraz grapevines in two South Australian viticultural regions over two seasons. The projected leaf area (PLA) of individual grapevines was estimated from the images. Virus-infected vines had significantly lower PLA than healthy vines in the early season but fewer difference after veraison. The lower PLA was only observed in grapevines coinfected with grapevine leafroll-associated viruses (GLRaVs) and Grapevine virus A (GVA). Shiraz vines infected with either GLRaVs or GVA had similar PLA to healthy vines. Conclusions. High-resolution RGB remote sensing technology has the potential to rapidly estimate SD infection in Shiraz grapevines. Our observations of shoot devigouration only in coinfected vines calls into question the etiology of SD. Further validation of the PLA technique incorporating different regions, seasons, cultivars, and combinations of viruses is needed for improving the robustness of the method. Significance of the Study. This preliminary study presents a new rapid and low-cost surveillance method to estimate SD infections in Shiraz vineyards, which could significantly lower the cost for growers who conduct on-ground SD visual assessments or lab-based tissue testing at the vineyard scale.},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Pagay, Vinay},
 doi = {10.1155/2023/7376153},
 journal = {Australian Journal of Grape and Wine Research}
}

Background and Aims. Shiraz disease (SD) is a viral disease associated with Grapevine virus A that causes significant yield loss in economically important grape cultivars in Australia such as Shiraz and Merlot. Current diagnostic methods are time-consuming and costly. This study evaluates an alternative methodology using visible remote sensing imagery to detect SD in Shiraz grapevines. Methods and Results. High-resolution visible remote sensing images were captured of Shiraz grapevines in two South Australian viticultural regions over two seasons. The projected leaf area (PLA) of individual grapevines was estimated from the images. Virus-infected vines had significantly lower PLA than healthy vines in the early season but fewer difference after veraison. The lower PLA was only observed in grapevines coinfected with grapevine leafroll-associated viruses (GLRaVs) and Grapevine virus A (GVA). Shiraz vines infected with either GLRaVs or GVA had similar PLA to healthy vines. Conclusions. High-resolution RGB remote sensing technology has the potential to rapidly estimate SD infection in Shiraz grapevines. Our observations of shoot devigouration only in coinfected vines calls into question the etiology of SD. Further validation of the PLA technique incorporating different regions, seasons, cultivars, and combinations of viruses is needed for improving the robustness of the method. Significance of the Study. This preliminary study presents a new rapid and low-cost surveillance method to estimate SD infections in Shiraz vineyards, which could significantly lower the cost for growers who conduct on-ground SD visual assessments or lab-based tissue testing at the vineyard scale.

@article{
 title = {Low-Cost Handheld Spectrometry for Detecting Flavescence Dorée in Vineyards},
 type = {article},
 year = {2023},
 keywords = {cost spectrometer,feature selection,genetic algorithms (GA),hyperspectral remote sensing,low,machine learning algorithms,precision farming,vineyard},
 pages = {2388},
 volume = {13},
 websites = {https://www.mdpi.com/2076-3417/13/4/2388/htm,https://www.mdpi.com/2076-3417/13/4/2388},
 month = {2},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {13},
 id = {cb5fbf36-efe0-3ca0-bea3-5f234031cc82},
 created = {2023-11-14T12:13:43.997Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-22T17:08:41.948Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {This study was conducted to evaluate the potential of low-cost hyperspectral sensors for the early detection of Flavescence dorée (FD) from asymptomatic samples prior to symptom development. In total, 180 leaf spectra from 60 randomly selected plants (three leaves per plant) were collected by using two portable mini-spectrometers (Hamamatsu: 340–850 nm and NIRScan: 900–1700 nm) at five vegetative growth stages in a vineyard with grape variety Garganega. High differences in the Hamamatsu spectra of the two groups were found in the VIS-NIR (visible–near infrared) spectral region while very small differences were observed in the NIRScan spectra. We analyzed the spectral data of two sensors by using all bands, features reduced by an ensemble method, and by genetic algorithms (GA) to discriminate the asymptomatic healthy (FD negative) and diseased (FD positive) leaves using five different classifiers. Overall, high classification accuracies were found in case of the Hamamatsu sensor compared to the NIRScan sensor. The feature selection techniques performed better compared to all bands, and the highest classification accuracy of 96% was achieved when GA features of the Hamamatsu sensor were used with the logistic regression (LR) classifier on test samples. A slightly low accuracy of 85% was achieved when the features (selected by the ensemble method) of the Hamamatsu sensor were used with the support vector machine (SVM) classifier by using leave-one-out (LOO) cross-validation on the whole dataset. Results demonstrated that employing a feature selection technique can provide a valid tool for determining the optimal bands that can be used to identify FD disease in the vineyard. However, further validation studies are required, as this study was conducted using a small dataset and from the single grapevine variety.},
 bibtype = {article},
 author = {Imran, Hafiz Ali and Zeggada, Abdallah and Ianniello, Ivan and Melgani, Farid and Polverari, Annalisa and Baroni, Alice and Danzi, Davide and Goller, Rino},
 doi = {10.3390/APP13042388},
 journal = {Applied Sciences 2023, Vol. 13, Page 2388},
 number = {4}
}

This study was conducted to evaluate the potential of low-cost hyperspectral sensors for the early detection of Flavescence dorée (FD) from asymptomatic samples prior to symptom development. In total, 180 leaf spectra from 60 randomly selected plants (three leaves per plant) were collected by using two portable mini-spectrometers (Hamamatsu: 340–850 nm and NIRScan: 900–1700 nm) at five vegetative growth stages in a vineyard with grape variety Garganega. High differences in the Hamamatsu spectra of the two groups were found in the VIS-NIR (visible–near infrared) spectral region while very small differences were observed in the NIRScan spectra. We analyzed the spectral data of two sensors by using all bands, features reduced by an ensemble method, and by genetic algorithms (GA) to discriminate the asymptomatic healthy (FD negative) and diseased (FD positive) leaves using five different classifiers. Overall, high classification accuracies were found in case of the Hamamatsu sensor compared to the NIRScan sensor. The feature selection techniques performed better compared to all bands, and the highest classification accuracy of 96% was achieved when GA features of the Hamamatsu sensor were used with the logistic regression (LR) classifier on test samples. A slightly low accuracy of 85% was achieved when the features (selected by the ensemble method) of the Hamamatsu sensor were used with the support vector machine (SVM) classifier by using leave-one-out (LOO) cross-validation on the whole dataset. Results demonstrated that employing a feature selection technique can provide a valid tool for determining the optimal bands that can be used to identify FD disease in the vineyard. However, further validation studies are required, as this study was conducted using a small dataset and from the single grapevine variety.

@article{
 title = {Detecting Grapevine Virus Infections in Red and White Winegrape Canopies Using Proximal Hyperspectral Sensing},
 type = {article},
 year = {2023},
 keywords = {1,DA,GLD,GLRaV,GVA,PLS,disease detection,proximal sensing,spectroradiometer},
 pages = {2851},
 volume = {23},
 websites = {https://www.mdpi.com/1424-8220/23/5/2851/htm,https://www.mdpi.com/1424-8220/23/5/2851},
 month = {3},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {6},
 id = {180d072b-732a-3909-96c7-60bd222d226f},
 created = {2023-11-14T12:16:26.785Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T14:06:51.070Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Grapevine virus-associated disease such as grapevine leafroll disease (GLD) affects grapevine health worldwide. Current diagnostic methods are either highly costly (laboratory-based diagnostics) or can be unreliable (visual assessments). Hyperspectral sensing technology is capable of measuring leaf reflectance spectra that can be used for the non-destructive and rapid detection of plant diseases. The present study used proximal hyperspectral sensing to detect virus infection in Pinot Noir (red-berried winegrape cultivar) and Chardonnay (white-berried winegrape cultivar) grapevines. Spectral data were collected throughout the grape growing season at six timepoints per cultivar. Partial least squares-discriminant analysis (PLS-DA) was used to build a predictive model of the presence or absence of GLD. The temporal change of canopy spectral reflectance showed that the harvest timepoint had the best prediction result. Prediction accuracies of 96% and 76% were achieved for Pinot Noir and Chardonnay, respectively. Our results provide valuable information on the optimal time for GLD detection. This hyperspectral method can also be deployed on mobile platforms including ground-based vehicles and unmanned aerial vehicles (UAV) for large-scale disease surveillance in vineyards.},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Pagay, Vinay},
 doi = {10.3390/S23052851},
 journal = {Sensors 2023, Vol. 23, Page 2851},
 number = {5}
}

Grapevine virus-associated disease such as grapevine leafroll disease (GLD) affects grapevine health worldwide. Current diagnostic methods are either highly costly (laboratory-based diagnostics) or can be unreliable (visual assessments). Hyperspectral sensing technology is capable of measuring leaf reflectance spectra that can be used for the non-destructive and rapid detection of plant diseases. The present study used proximal hyperspectral sensing to detect virus infection in Pinot Noir (red-berried winegrape cultivar) and Chardonnay (white-berried winegrape cultivar) grapevines. Spectral data were collected throughout the grape growing season at six timepoints per cultivar. Partial least squares-discriminant analysis (PLS-DA) was used to build a predictive model of the presence or absence of GLD. The temporal change of canopy spectral reflectance showed that the harvest timepoint had the best prediction result. Prediction accuracies of 96% and 76% were achieved for Pinot Noir and Chardonnay, respectively. Our results provide valuable information on the optimal time for GLD detection. This hyperspectral method can also be deployed on mobile platforms including ground-based vehicles and unmanned aerial vehicles (UAV) for large-scale disease surveillance in vineyards.

@article{
 title = {Drones in Plant Disease Assessment, Efficient Monitoring, and Detection: A Way Forward to Smart Agriculture},
 type = {article},
 year = {2023},
 keywords = {abbas2023dronesplantdisease},
 pages = {1524},
 volume = {13},
 websites = {https://www.mdpi.com/2073-4395/13/6/1524/htm,https://www.mdpi.com/2073-4395/13/6/1524},
 month = {5},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {31},
 id = {b8f73b64-870f-3191-baaa-e6336eafd99e},
 created = {2023-11-14T13:34:25.055Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T15:03:47.116Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Plant diseases are one of the major threats to global food production. Efficient monitoring and detection of plant pathogens are instrumental in restricting and effectively managing the spread of the disease and reducing the cost of pesticides. Traditional, molecular, and serological methods that are widely used for plant disease detection are often ineffective if not applied during the initial stages of pathogenesis, when no or very weak symptoms appear. Moreover, they are almost useless in acquiring spatialized diagnostic results on plant diseases. On the other hand, remote sensing (RS) techniques utilizing drones are very effective for the rapid identification of plant diseases in their early stages. Currently, drones, play a pivotal role in the monitoring of plant pathogen spread, detection, and diagnosis to ensure crops’ health status. The advantages of drone technology include high spatial resolution (as several sensors are carried aboard), high efficiency, usage flexibility, and more significantly, quick detection of plant diseases across a large area with low cost, reliability, and provision of high-resolution data. Drone technology employs an automated procedure that begins with gathering images of diseased plants using various sensors and cameras. After extracting features, image processing approaches use the appropriate traditional machine learning or deep learning algorithms. Features are extracted from images of leaves using edge detection and histogram equalization methods. Drones have many potential uses in agriculture, including reducing manual labor and increasing productivity. Drones may be able to provide early warning of plant diseases, allowing farmers to prevent costly crop failures.},
 bibtype = {article},
 author = {Abbas, Aqleem and Zhang, Zhenhao and Zheng, Hongxia and Alami, Mohammad Murtaza and Alrefaei, Abdulmajeed F. and Abbas, Qamar and Naqvi, Syed Atif Hasan and Rao, Muhammad Junaid and Mosa, Walid F.A. and Abbas, Qamar and Hussain, Azhar and Hassan, Muhammad Zeeshan and Zhou, Lei},
 doi = {10.3390/AGRONOMY13061524},
 journal = {Agronomy 2023, Vol. 13, Page 1524},
 number = {6}
}

Plant diseases are one of the major threats to global food production. Efficient monitoring and detection of plant pathogens are instrumental in restricting and effectively managing the spread of the disease and reducing the cost of pesticides. Traditional, molecular, and serological methods that are widely used for plant disease detection are often ineffective if not applied during the initial stages of pathogenesis, when no or very weak symptoms appear. Moreover, they are almost useless in acquiring spatialized diagnostic results on plant diseases. On the other hand, remote sensing (RS) techniques utilizing drones are very effective for the rapid identification of plant diseases in their early stages. Currently, drones, play a pivotal role in the monitoring of plant pathogen spread, detection, and diagnosis to ensure crops’ health status. The advantages of drone technology include high spatial resolution (as several sensors are carried aboard), high efficiency, usage flexibility, and more significantly, quick detection of plant diseases across a large area with low cost, reliability, and provision of high-resolution data. Drone technology employs an automated procedure that begins with gathering images of diseased plants using various sensors and cameras. After extracting features, image processing approaches use the appropriate traditional machine learning or deep learning algorithms. Features are extracted from images of leaves using edge detection and histogram equalization methods. Drones have many potential uses in agriculture, including reducing manual labor and increasing productivity. Drones may be able to provide early warning of plant diseases, allowing farmers to prevent costly crop failures.

@article{
 title = {GrapesNet: Indian RGB & RGB-D vineyard image datasets for deep learning applications},
 type = {article},
 year = {2023},
 keywords = {barbole2023grapesnetindianrgb},
 pages = {109100},
 volume = {48},
 month = {6},
 publisher = {Elsevier},
 day = {1},
 id = {480d539c-ec12-354e-b98f-8f46a5851286},
 created = {2023-11-21T15:22:56.915Z},
 accessed = {2023-11-21},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T15:32:10.183Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {In most of the countries, grapes are considered as a cash crop. Currently huge research is going on in development of automated grape harvesting systems. Speedy and reliable grape bunch detection is prime need for various deep learning based automated systems which deals with object detection and object segmentation tasks. But currently very few datasets are available on grape bunches in vineyard, because of which there is restriction to the research in this area. In comparison to the vineyard in outside countries, Indian vineyard structure is more complex, so it becomes hard to work in real-time. To overcome these problems and to make vineyard dataset for suitable for Indian vineyard scenarios, this paper proposed four different datasets on grape bunches in vineyard. For creating all datasets in GrapesNet, natural environmental conditions have been considered. GrapesNet includes total 11000+ images of grape bunches. Necessary data for weight prediction of grape cluster is also provided with dataset like height, width and real weight of cluster present in image. Proposed datasets can be used for prime tasks like grape bunch detection, grape bunch segmentation, and grape bunch weight estimation etc. of future generation automated vineyard harvesting technologies.},
 bibtype = {article},
 author = {Barbole, Dhanashree K. and Jadhav, Parul M.},
 doi = {10.1016/J.DIB.2023.109100},
 journal = {Data in Brief}
}

In most of the countries, grapes are considered as a cash crop. Currently huge research is going on in development of automated grape harvesting systems. Speedy and reliable grape bunch detection is prime need for various deep learning based automated systems which deals with object detection and object segmentation tasks. But currently very few datasets are available on grape bunches in vineyard, because of which there is restriction to the research in this area. In comparison to the vineyard in outside countries, Indian vineyard structure is more complex, so it becomes hard to work in real-time. To overcome these problems and to make vineyard dataset for suitable for Indian vineyard scenarios, this paper proposed four different datasets on grape bunches in vineyard. For creating all datasets in GrapesNet, natural environmental conditions have been considered. GrapesNet includes total 11000+ images of grape bunches. Necessary data for weight prediction of grape cluster is also provided with dataset like height, width and real weight of cluster present in image. Proposed datasets can be used for prime tasks like grape bunch detection, grape bunch segmentation, and grape bunch weight estimation etc. of future generation automated vineyard harvesting technologies.

@article{
 title = {Deep Learning YOLO-Based Solution for Grape Bunch Detection and Assessment of Biophysical Lesions},
 type = {article},
 year = {2023},
 keywords = {computer vision,machine learning,object detection,precision agriculture,viticulture},
 pages = {1120},
 volume = {13},
 websites = {https://www.mdpi.com/2073-4395/13/4/1120/htm,https://www.mdpi.com/2073-4395/13/4/1120},
 month = {4},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {14},
 id = {bd30f66a-4817-310e-b81d-a4d2fa76ae36},
 created = {2023-11-21T15:31:54.214Z},
 accessed = {2023-11-21},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-22T12:32:57.248Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The world wine sector is a multi-billion dollar industry with a wide range of economic activities. Therefore, it becomes crucial to monitor the grapevine because it allows a more accurate estimation of the yield and ensures a high-quality end product. The most common way of monitoring the grapevine is through the leaves (preventive way) since the leaves first manifest biophysical lesions. However, this does not exclude the possibility of biophysical lesions manifesting in the grape berries. Thus, this work presents three pre-trained YOLO models (YOLOv5x6, YOLOv7-E6E, and YOLOR-CSP-X) to detect and classify grape bunches as healthy or damaged by the number of berries with biophysical lesions. Two datasets were created and made publicly available with original images and manual annotations to identify the complexity between detection (bunches) and classification (healthy or damaged) tasks. The datasets use the same 10,010 images with different classes. The Grapevine Bunch Detection Dataset uses the Bunch class, and The Grapevine Bunch Condition Detection Dataset uses the OptimalBunch and DamagedBunch classes. Regarding the three models trained for grape bunches detection, they obtained promising results, highlighting YOLOv7 with 77% of mAP and 94% of the F1-score. In the case of the task of detection and identification of the state of grape bunches, the three models obtained similar results, with YOLOv5 achieving the best ones with an mAP of 72% and an F1-score of 92%.},
 bibtype = {article},
 author = {Pinheiro, Isabel and Moreira, Germano and Queirós da Silva, Daniel and Magalhães, Sandro and Valente, António and Moura Oliveira, Paulo and Cunha, Mário and Santos, Filipe},
 doi = {10.3390/AGRONOMY13041120},
 journal = {Agronomy 2023, Vol. 13, Page 1120},
 number = {4}
}

The world wine sector is a multi-billion dollar industry with a wide range of economic activities. Therefore, it becomes crucial to monitor the grapevine because it allows a more accurate estimation of the yield and ensures a high-quality end product. The most common way of monitoring the grapevine is through the leaves (preventive way) since the leaves first manifest biophysical lesions. However, this does not exclude the possibility of biophysical lesions manifesting in the grape berries. Thus, this work presents three pre-trained YOLO models (YOLOv5x6, YOLOv7-E6E, and YOLOR-CSP-X) to detect and classify grape bunches as healthy or damaged by the number of berries with biophysical lesions. Two datasets were created and made publicly available with original images and manual annotations to identify the complexity between detection (bunches) and classification (healthy or damaged) tasks. The datasets use the same 10,010 images with different classes. The Grapevine Bunch Detection Dataset uses the Bunch class, and The Grapevine Bunch Condition Detection Dataset uses the OptimalBunch and DamagedBunch classes. Regarding the three models trained for grape bunches detection, they obtained promising results, highlighting YOLOv7 with 77% of mAP and 94% of the F1-score. In the case of the task of detection and identification of the state of grape bunches, the three models obtained similar results, with YOLOv5 achieving the best ones with an mAP of 72% and an F1-score of 92%.

@article{
 title = {The Use of Computer Vision to Improve the Affinity of Rootstock-Graft Combinations and Identify Diseases of Grape Seedlings},
 type = {article},
 year = {2023},
 keywords = {artificial intelligence,computer vision,environmental engineering,graft combinations’ affinity,grape diseases,grape seedlings,neural networks,rootstock,viticulture},
 pages = {92},
 volume = {8},
 websites = {https://www.mdpi.com/2411-5134/8/4/92/htm,https://www.mdpi.com/2411-5134/8/4/92},
 month = {7},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {19},
 id = {bd913b78-356d-3573-b259-f6fb292a6743},
 created = {2023-11-23T12:08:56.647Z},
 accessed = {2023-11-23},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:03.078Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This study explores the application of computer vision for enhancing the selection of rootstock-graft combinations and detecting diseases in grape seedlings. Computer vision has various applications in viticulture, but publications and research have not reported the use of computer vision in rootstock-graft selection, which defines the novelty of this research. This paper presents elements of the technology for applying computer vision to rootstock-graft combinations and includes an analysis of grape seedling cuttings. This analysis allows for a more accurate determination of the compatibility between rootstock and graft, as well as the detection of potential seedling diseases. The utilization of computer vision to automate the grafting process of grape cuttings offers significant benefits in terms of increased efficiency, improved quality, and reduced costs. This technology can replace manual labor and ensure economic efficiency and reliability, among other advantages. It also facilitates monitoring the development of seedlings to determine the appropriate planting time. Image processing algorithms play a vital role in automatically determining seedling characteristics such as trunk diameter and the presence of any damage. Furthermore, computer vision can aid in the identification of diseases and defects in seedlings, which is crucial for assessing their overall quality. The automation of these processes offers several advantages, including increased efficiency, improved quality, and reduced costs through the reduction of manual labor and waste. To fulfill these objectives, a unique robotic assembly line is planned for the grafting of grape cuttings. This line will be equipped with two conveyor belts, a delta robot, and a computer vision system. The use of computer vision in automating the grafting process for grape cuttings offers significant benefits in terms of efficiency, quality improvement, and cost reduction. By incorporating image processing algorithms and advanced robotics, this technology has the potential to revolutionize the viticulture industry. Thanks to training a computer vision system to analyze data on rootstock and graft grape varieties, it is possible to reduce the number of defects by half. The implementation of a semi-automated computer vision system can improve crossbreeding efficiency by 90%. Reducing the time spent on pairing selection is also a significant advantage. While manual selection takes between 1 and 2 min, reducing the time to 30 s using the semi-automated system, and the prospect of further automation reducing the time to 10–15 s, will significantly increase the productivity and efficiency of the process. In addition to the aforementioned benefits, the integration of computer vision technology in grape grafting processes brings several other advantages. One notable advantage is the increased accuracy and precision in pairing selection. Computer vision algorithms can analyze a wide range of factors, including size, shape, color, and structural characteristics, to make more informed decisions when matching rootstock and graft varieties. This can lead to better compatibility and improved overall grafting success rates.},
 bibtype = {article},
 author = {Rudenko, Marina and Plugatar, Yurij and Korzin, Vadim and Kazak, Anatoliy and Gallini, Nadezhda and Gorbunova, Natalia},
 doi = {10.3390/INVENTIONS8040092},
 journal = {Inventions 2023, Vol. 8, Page 92},
 number = {4}
}

This study explores the application of computer vision for enhancing the selection of rootstock-graft combinations and detecting diseases in grape seedlings. Computer vision has various applications in viticulture, but publications and research have not reported the use of computer vision in rootstock-graft selection, which defines the novelty of this research. This paper presents elements of the technology for applying computer vision to rootstock-graft combinations and includes an analysis of grape seedling cuttings. This analysis allows for a more accurate determination of the compatibility between rootstock and graft, as well as the detection of potential seedling diseases. The utilization of computer vision to automate the grafting process of grape cuttings offers significant benefits in terms of increased efficiency, improved quality, and reduced costs. This technology can replace manual labor and ensure economic efficiency and reliability, among other advantages. It also facilitates monitoring the development of seedlings to determine the appropriate planting time. Image processing algorithms play a vital role in automatically determining seedling characteristics such as trunk diameter and the presence of any damage. Furthermore, computer vision can aid in the identification of diseases and defects in seedlings, which is crucial for assessing their overall quality. The automation of these processes offers several advantages, including increased efficiency, improved quality, and reduced costs through the reduction of manual labor and waste. To fulfill these objectives, a unique robotic assembly line is planned for the grafting of grape cuttings. This line will be equipped with two conveyor belts, a delta robot, and a computer vision system. The use of computer vision in automating the grafting process for grape cuttings offers significant benefits in terms of efficiency, quality improvement, and cost reduction. By incorporating image processing algorithms and advanced robotics, this technology has the potential to revolutionize the viticulture industry. Thanks to training a computer vision system to analyze data on rootstock and graft grape varieties, it is possible to reduce the number of defects by half. The implementation of a semi-automated computer vision system can improve crossbreeding efficiency by 90%. Reducing the time spent on pairing selection is also a significant advantage. While manual selection takes between 1 and 2 min, reducing the time to 30 s using the semi-automated system, and the prospect of further automation reducing the time to 10–15 s, will significantly increase the productivity and efficiency of the process. In addition to the aforementioned benefits, the integration of computer vision technology in grape grafting processes brings several other advantages. One notable advantage is the increased accuracy and precision in pairing selection. Computer vision algorithms can analyze a wide range of factors, including size, shape, color, and structural characteristics, to make more informed decisions when matching rootstock and graft varieties. This can lead to better compatibility and improved overall grafting success rates.

@article{
 title = {The Use of Computer Vision to Improve the Affinity of Rootstock-Graft Combinations and Identify Diseases of Grape Seedlings},
 type = {article},
 year = {2023},
 keywords = {artificial intelligence,computer vision,environmental engineering,graft combinations’ affinity,grape diseases,grape seedlings,neural networks,rootstock,viticulture},
 pages = {92},
 volume = {8},
 websites = {https://www.mdpi.com/2411-5134/8/4/92/htm,https://www.mdpi.com/2411-5134/8/4/92},
 month = {7},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {19},
 id = {06a7b616-1426-3b7d-a39c-b12f78c0d74f},
 created = {2023-11-23T18:10:02.287Z},
 accessed = {2023-11-23},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:10:11.751Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {This study explores the application of computer vision for enhancing the selection of rootstock-graft combinations and detecting diseases in grape seedlings. Computer vision has various applications in viticulture, but publications and research have not reported the use of computer vision in rootstock-graft selection, which defines the novelty of this research. This paper presents elements of the technology for applying computer vision to rootstock-graft combinations and includes an analysis of grape seedling cuttings. This analysis allows for a more accurate determination of the compatibility between rootstock and graft, as well as the detection of potential seedling diseases. The utilization of computer vision to automate the grafting process of grape cuttings offers significant benefits in terms of increased efficiency, improved quality, and reduced costs. This technology can replace manual labor and ensure economic efficiency and reliability, among other advantages. It also facilitates monitoring the development of seedlings to determine the appropriate planting time. Image processing algorithms play a vital role in automatically determining seedling characteristics such as trunk diameter and the presence of any damage. Furthermore, computer vision can aid in the identification of diseases and defects in seedlings, which is crucial for assessing their overall quality. The automation of these processes offers several advantages, including increased efficiency, improved quality, and reduced costs through the reduction of manual labor and waste. To fulfill these objectives, a unique robotic assembly line is planned for the grafting of grape cuttings. This line will be equipped with two conveyor belts, a delta robot, and a computer vision system. The use of computer vision in automating the grafting process for grape cuttings offers significant benefits in terms of efficiency, quality improvement, and cost reduction. By incorporating image processing algorithms and advanced robotics, this technology has the potential to revolutionize the viticulture industry. Thanks to training a computer vision system to analyze data on rootstock and graft grape varieties, it is possible to reduce the number of defects by half. The implementation of a semi-automated computer vision system can improve crossbreeding efficiency by 90%. Reducing the time spent on pairing selection is also a significant advantage. While manual selection takes between 1 and 2 min, reducing the time to 30 s using the semi-automated system, and the prospect of further automation reducing the time to 10–15 s, will significantly increase the productivity and efficiency of the process. In addition to the aforementioned benefits, the integration of computer vision technology in grape grafting processes brings several other advantages. One notable advantage is the increased accuracy and precision in pairing selection. Computer vision algorithms can analyze a wide range of factors, including size, shape, color, and structural characteristics, to make more informed decisions when matching rootstock and graft varieties. This can lead to better compatibility and improved overall grafting success rates.},
 bibtype = {article},
 author = {Rudenko, Marina and Plugatar, Yurij and Korzin, Vadim and Kazak, Anatoliy and Gallini, Nadezhda and Gorbunova, Natalia},
 doi = {10.3390/INVENTIONS8040092},
 journal = {Inventions 2023, Vol. 8, Page 92},
 number = {4}
}

This study explores the application of computer vision for enhancing the selection of rootstock-graft combinations and detecting diseases in grape seedlings. Computer vision has various applications in viticulture, but publications and research have not reported the use of computer vision in rootstock-graft selection, which defines the novelty of this research. This paper presents elements of the technology for applying computer vision to rootstock-graft combinations and includes an analysis of grape seedling cuttings. This analysis allows for a more accurate determination of the compatibility between rootstock and graft, as well as the detection of potential seedling diseases. The utilization of computer vision to automate the grafting process of grape cuttings offers significant benefits in terms of increased efficiency, improved quality, and reduced costs. This technology can replace manual labor and ensure economic efficiency and reliability, among other advantages. It also facilitates monitoring the development of seedlings to determine the appropriate planting time. Image processing algorithms play a vital role in automatically determining seedling characteristics such as trunk diameter and the presence of any damage. Furthermore, computer vision can aid in the identification of diseases and defects in seedlings, which is crucial for assessing their overall quality. The automation of these processes offers several advantages, including increased efficiency, improved quality, and reduced costs through the reduction of manual labor and waste. To fulfill these objectives, a unique robotic assembly line is planned for the grafting of grape cuttings. This line will be equipped with two conveyor belts, a delta robot, and a computer vision system. The use of computer vision in automating the grafting process for grape cuttings offers significant benefits in terms of efficiency, quality improvement, and cost reduction. By incorporating image processing algorithms and advanced robotics, this technology has the potential to revolutionize the viticulture industry. Thanks to training a computer vision system to analyze data on rootstock and graft grape varieties, it is possible to reduce the number of defects by half. The implementation of a semi-automated computer vision system can improve crossbreeding efficiency by 90%. Reducing the time spent on pairing selection is also a significant advantage. While manual selection takes between 1 and 2 min, reducing the time to 30 s using the semi-automated system, and the prospect of further automation reducing the time to 10–15 s, will significantly increase the productivity and efficiency of the process. In addition to the aforementioned benefits, the integration of computer vision technology in grape grafting processes brings several other advantages. One notable advantage is the increased accuracy and precision in pairing selection. Computer vision algorithms can analyze a wide range of factors, including size, shape, color, and structural characteristics, to make more informed decisions when matching rootstock and graft varieties. This can lead to better compatibility and improved overall grafting success rates.

@article{
 title = {Domain Generalization for Crop Segmentation with Knowledge Distillation},
 type = {article},
 year = {2023},
 keywords = {Domain Generalization,Knowledge Distillation,Semantic Segmentation},
 websites = {https://arxiv.org/abs/2304.01029v2},
 month = {4},
 day = {3},
 id = {c276e262-94b0-3e54-a265-45792cf6834d},
 created = {2023-12-13T07:01:13.813Z},
 accessed = {2023-12-13},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-12-13T07:45:24.430Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {d25a2be2-b54f-400b-918b-b254e8044e39},
 private_publication = {false},
 abstract = {In recent years, precision agriculture has gradually oriented farming closer to automation processes to support all the activities related to field management. Service robotics plays a predominant role in this evolution by deploying autonomous agents that can navigate fields while performing tasks without human intervention, such as monitoring, spraying, and harvesting. To execute these precise actions, mobile robots need a real-time perception system that understands their surroundings and identifies their targets in the wild. Generalizing to new crops and environmental conditions is critical for practical applications, as labeled samples are rarely available. In this paper, we investigate the problem of crop segmentation and propose a novel approach to enhance domain generalization using knowledge distillation. In the proposed framework, we transfer knowledge from an ensemble of models individually trained on source domains to a student model that can adapt to unseen target domains. To evaluate the proposed method, we present a synthetic multi-domain dataset for crop segmentation containing plants of variegate shapes and covering different terrain styles, weather conditions, and light scenarios for more than 50,000 samples. We demonstrate significant improvements in performance over state-of-the-art methods and superior sim-to-real generalization. Our approach provides a promising solution for domain generalization in crop segmentation and has the potential to enhance a wide variety of precision agriculture applications.},
 bibtype = {article},
 author = {Angarano, Simone and Martini, Mauro and Navone, Alessandro and Chiaberge, Marcello}
}

In recent years, precision agriculture has gradually oriented farming closer to automation processes to support all the activities related to field management. Service robotics plays a predominant role in this evolution by deploying autonomous agents that can navigate fields while performing tasks without human intervention, such as monitoring, spraying, and harvesting. To execute these precise actions, mobile robots need a real-time perception system that understands their surroundings and identifies their targets in the wild. Generalizing to new crops and environmental conditions is critical for practical applications, as labeled samples are rarely available. In this paper, we investigate the problem of crop segmentation and propose a novel approach to enhance domain generalization using knowledge distillation. In the proposed framework, we transfer knowledge from an ensemble of models individually trained on source domains to a student model that can adapt to unseen target domains. To evaluate the proposed method, we present a synthetic multi-domain dataset for crop segmentation containing plants of variegate shapes and covering different terrain styles, weather conditions, and light scenarios for more than 50,000 samples. We demonstrate significant improvements in performance over state-of-the-art methods and superior sim-to-real generalization. Our approach provides a promising solution for domain generalization in crop segmentation and has the potential to enhance a wide variety of precision agriculture applications.

@article{
 title = {Applying Knowledge Distillation on Pre-Trained Model for Early Grapevine Detection},
 type = {article},
 year = {2023},
 keywords = {Deep Learning,Fine-tuning,Knowledge Distillation,Pseudo-labelling,Yield forecast},
 pages = {149-156},
 websites = {https://ebooks.iospress.nl/doi/10.3233/AISE230024},
 month = {6},
 publisher = {IOS Press},
 day = {23},
 id = {5bc00688-4c07-3da7-9171-c8c832e4c790},
 created = {2023-12-13T07:02:00.181Z},
 accessed = {2023-12-13},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-12-13T07:45:24.412Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {d25a2be2-b54f-400b-918b-b254e8044e39},
 private_publication = {false},
 abstract = {The development of Artificial Intelligence has raised interesting opportunities for improved automation in smart agriculture. Smart viticulture is one of the domains that can benefit from Computer-vision tasks through field sustainability. Computer-vision solutions present additional constraints as the amount of data for good training convergence has to be complex enough to cover sufficient features from desired inputs. In this paper, we present a study to implement a grapevine detection improvement for early grapes detection and grape yield prediction whose interest in Champagne and wine companies is undeniable. Earlier yield predictions allow a better market assessment, the harvest work's organization and help decision-making about plant management. Our goal is to carry estimations 5 to 6 weeks before the harvest. Furthermore, the grapevines growing condition and the large amount of data to process for yield estimation require an embedded device to acquire and compute deep learning inference. Thus, the grapes detection model has to be lightweight enough to run on an embedded device. These models were subsequently pre-trained on two different types of datasets and several layer depth of deep learning models to propose a pseudo-labelling Teacher-Student related Knowledge Distillation. Overall solutions proposed an improvement of 7.56%, 6.98, 8.279%, 7.934% and 13.63% for f1 score, precision, recall, mean average precision at 50 and mean average precision 50-95 respectively on BBCH77 phenological stage.},
 bibtype = {article},
 author = {Hollard, Lilian and Mohimont, Lucas},
 doi = {10.3233/AISE230024}
}

The development of Artificial Intelligence has raised interesting opportunities for improved automation in smart agriculture. Smart viticulture is one of the domains that can benefit from Computer-vision tasks through field sustainability. Computer-vision solutions present additional constraints as the amount of data for good training convergence has to be complex enough to cover sufficient features from desired inputs. In this paper, we present a study to implement a grapevine detection improvement for early grapes detection and grape yield prediction whose interest in Champagne and wine companies is undeniable. Earlier yield predictions allow a better market assessment, the harvest work's organization and help decision-making about plant management. Our goal is to carry estimations 5 to 6 weeks before the harvest. Furthermore, the grapevines growing condition and the large amount of data to process for yield estimation require an embedded device to acquire and compute deep learning inference. Thus, the grapes detection model has to be lightweight enough to run on an embedded device. These models were subsequently pre-trained on two different types of datasets and several layer depth of deep learning models to propose a pseudo-labelling Teacher-Student related Knowledge Distillation. Overall solutions proposed an improvement of 7.56%, 6.98, 8.279%, 7.934% and 13.63% for f1 score, precision, recall, mean average precision at 50 and mean average precision 50-95 respectively on BBCH77 phenological stage.

@misc{
 title = {Online Knowledge Distillation for Multi-Task Learning},
 type = {misc},
 year = {2023},
 pages = {2359-2368},
 id = {0489c563-6bb0-3117-8da2-b7731e835a4e},
 created = {2023-12-13T07:39:22.747Z},
 accessed = {2023-12-13},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-12-13T07:39:25.295Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {d25a2be2-b54f-400b-918b-b254e8044e39},
 private_publication = {false},
 abstract = {Multi-task learning (MTL) has found wide application in computer vision tasks. We train a backbone network to learn a shared representation for different tasks such as semantic segmentation, depth-and normal estimation. In many cases negative transfer, i.e. impaired performance in the target domain, causes the MTL accuracy to be lower than training the corresponding single-task networks. To mitigate this issue, we propose an online knowledge distillation method, where single-task networks are trained simultaneously with the MTL network to guide the optimization process. We propose selectively training layers for each task using an adaptive feature distillation (AFD) loss with an online task weighting (OTW) scheme. This task-wise feature distillation enables the MTL network to be trained in a similar way to the single-task networks. On the NYUv2 and Cityscapes datasets we show improvements over a baseline MTL model by 6.22% and 9.19%, respectively, outperform-ing recent MTL methods. We validate the design choices in ablative experiments, including the use of online task weighting and the adaptive feature distillation loss.},
 bibtype = {misc},
 author = {Jacob, Geethu Miriam and Agarwal, Vishal and Stenger, Björn}
}

Multi-task learning (MTL) has found wide application in computer vision tasks. We train a backbone network to learn a shared representation for different tasks such as semantic segmentation, depth-and normal estimation. In many cases negative transfer, i.e. impaired performance in the target domain, causes the MTL accuracy to be lower than training the corresponding single-task networks. To mitigate this issue, we propose an online knowledge distillation method, where single-task networks are trained simultaneously with the MTL network to guide the optimization process. We propose selectively training layers for each task using an adaptive feature distillation (AFD) loss with an online task weighting (OTW) scheme. This task-wise feature distillation enables the MTL network to be trained in a similar way to the single-task networks. On the NYUv2 and Cityscapes datasets we show improvements over a baseline MTL model by 6.22% and 9.19%, respectively, outperform-ing recent MTL methods. We validate the design choices in ablative experiments, including the use of online task weighting and the adaptive feature distillation loss.

@misc{
 title = {Multi-Task Learning with Knowledge Distillation for Dense Prediction},
 type = {misc},
 year = {2023},
 pages = {21550-21559},
 id = {ae105343-eeb4-308c-999b-a8324a26099f},
 created = {2023-12-13T07:45:14.990Z},
 accessed = {2023-12-13},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-12-13T07:45:17.960Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {d25a2be2-b54f-400b-918b-b254e8044e39},
 private_publication = {false},
 abstract = {While multi-task learning (MTL) has become an attractive topic, its training usually poses more difficulties than the single-task case. How to successfully apply knowledge distillation into MTL to improve training efficiency and model performance is still a challenging problem. In this paper, we introduce a new knowledge distillation procedure with an alternative match for MTL of dense prediction based on two simple design principles. First, for memory and training efficiency, we use a single strong multi-task model as a teacher during training instead of multiple teachers, as widely adopted in existing studies. Second, we employ a less sensitive Cauchy-Schwarz (CS) divergence instead of the Kullback-Leibler (KL) divergence and propose a CS distillation loss accordingly. With the less sensitive divergence, our knowledge distillation with an alternative match is applied for capturing inter-task and intra-task information between the teacher model and the student model of each task, thereby learning more "dark knowl-edge" for effective distillation. We conducted extensive experiments on dense prediction datasets, including NYUD-v2 and PASCAL-Context, for multiple vision tasks, such as semantic segmentation, human parts segmentation, depth estimation , surface normal estimation, and boundary detection. The results show that our proposed method decidedly improves model performance and the practical inference efficiency .},
 bibtype = {misc},
 author = {Xu, Yangyang and Yang, Yibo and Zhang, Lefei}
}

While multi-task learning (MTL) has become an attractive topic, its training usually poses more difficulties than the single-task case. How to successfully apply knowledge distillation into MTL to improve training efficiency and model performance is still a challenging problem. In this paper, we introduce a new knowledge distillation procedure with an alternative match for MTL of dense prediction based on two simple design principles. First, for memory and training efficiency, we use a single strong multi-task model as a teacher during training instead of multiple teachers, as widely adopted in existing studies. Second, we employ a less sensitive Cauchy-Schwarz (CS) divergence instead of the Kullback-Leibler (KL) divergence and propose a CS distillation loss accordingly. With the less sensitive divergence, our knowledge distillation with an alternative match is applied for capturing inter-task and intra-task information between the teacher model and the student model of each task, thereby learning more "dark knowl-edge" for effective distillation. We conducted extensive experiments on dense prediction datasets, including NYUD-v2 and PASCAL-Context, for multiple vision tasks, such as semantic segmentation, human parts segmentation, depth estimation , surface normal estimation, and boundary detection. The results show that our proposed method decidedly improves model performance and the practical inference efficiency .

@misc{
 title = {Research progress of autonomous navigation technology for multi-agricultural scenes},
 type = {misc},
 year = {2023},
 source = {Computers and Electronics in Agriculture},
 keywords = {Agricultural autonomous navigation,Agriculture,Differentiated,Scenes},
 volume = {211},
 month = {8},
 publisher = {Elsevier B.V.},
 day = {1},
 id = {308ff033-5c45-3017-8758-14f0ae00cc0e},
 created = {2024-01-09T11:55:49.228Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-09T14:11:54.625Z},
 read = {false},
 starred = {true},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {2bfb8d91-9fac-46f0-bd0c-93235d01dbed},
 private_publication = {false},
 abstract = {Due to the increasing demand for food, the global shortage of agricultural labor and the reduction in the utilization of agricultural resources, the demand for agricultural autonomous navigation technology and equipment has become increasingly urgent. Compared to the closed express structured environment for self-driving cars, the agricultural scene is complex and diverse, and its navigation modes and core technologies are also significantly different. In view of this, it is of great value to summarize the development characteristics and trends of autonomous navigation technology in fields, orchards, greenhouses and other scenarios. In this paper, (1) firstly, the types and characteristics of multi-agricultural scenes are analyzed, the principle and mode of agricultural autonomous navigation are expounded, and the agricultural autonomous navigation system and architecture are introduced; (2) secondly, the research status of autonomous navigation technology in open field agriculture, orchard agriculture, facility agriculture, livestock and poultry agriculture and aquatic agriculture is investigated; (3) finally, it is predicted that the differentiated autonomous navigation technology scheme for multi-scene agriculture, multi-dimensional perception, self-understanding navigation technology in agricultural scenarios, selective autonomous navigation technology for task self-matching fusion, the cooperative autonomous navigation technology for multi machine ad hoc network clusters, and the fault self-checking remote diagnosis navigation system are the future research hotspots and development trends. At the same time, it is clear that the in-depth practical application of artificial intelligence, the in-depth research and development of high-precision and low-cost sensors, and the collaborative integration technology of agricultural machinery and agronomy are the critical path to promote the breakthrough of agricultural autonomous navigation technology. The summary and prospect of this paper have positive significance for promoting the overall development of agricultural autonomous navigation technology.},
 bibtype = {misc},
 author = {Xie, Binbin and Jin, Yucheng and Faheem, Muhammad and Gao, Wenjie and Liu, Jizhan and Jiang, Houkang and Cai, Lianjiang and Li, Yuanxiang},
 doi = {10.1016/j.compag.2023.107963}
}

Due to the increasing demand for food, the global shortage of agricultural labor and the reduction in the utilization of agricultural resources, the demand for agricultural autonomous navigation technology and equipment has become increasingly urgent. Compared to the closed express structured environment for self-driving cars, the agricultural scene is complex and diverse, and its navigation modes and core technologies are also significantly different. In view of this, it is of great value to summarize the development characteristics and trends of autonomous navigation technology in fields, orchards, greenhouses and other scenarios. In this paper, (1) firstly, the types and characteristics of multi-agricultural scenes are analyzed, the principle and mode of agricultural autonomous navigation are expounded, and the agricultural autonomous navigation system and architecture are introduced; (2) secondly, the research status of autonomous navigation technology in open field agriculture, orchard agriculture, facility agriculture, livestock and poultry agriculture and aquatic agriculture is investigated; (3) finally, it is predicted that the differentiated autonomous navigation technology scheme for multi-scene agriculture, multi-dimensional perception, self-understanding navigation technology in agricultural scenarios, selective autonomous navigation technology for task self-matching fusion, the cooperative autonomous navigation technology for multi machine ad hoc network clusters, and the fault self-checking remote diagnosis navigation system are the future research hotspots and development trends. At the same time, it is clear that the in-depth practical application of artificial intelligence, the in-depth research and development of high-precision and low-cost sensors, and the collaborative integration technology of agricultural machinery and agronomy are the critical path to promote the breakthrough of agricultural autonomous navigation technology. The summary and prospect of this paper have positive significance for promoting the overall development of agricultural autonomous navigation technology.

@article{
 title = {MOT-DETR: 3D Single Shot Detection and Tracking with Transformers to build 3D representations for Agro-Food Robots},
 type = {article},
 year = {2023},
 websites = {http://arxiv.org/abs/2311.15674},
 month = {11},
 day = {27},
 id = {f51b55fa-1ca2-35eb-8255-dee4f4094980},
 created = {2024-01-12T12:17:14.270Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-12T12:17:14.778Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {2bfb8d91-9fac-46f0-bd0c-93235d01dbed},
 private_publication = {false},
 abstract = {In the current demand for automation in the agro-food industry, accurately detecting and localizing relevant objects in 3D is essential for successful robotic operations. However, this is a challenge due the presence of occlusions. Multi-view perception approaches allow robots to overcome occlusions, but a tracking component is needed to associate the objects detected by the robot over multiple viewpoints. Most multi-object tracking (MOT) algorithms are designed for high frame rate sequences and struggle with the occlusions generated by robots' motions and 3D environments. In this paper, we introduce MOT-DETR, a novel approach to detect and track objects in 3D over time using a combination of convolutional networks and transformers. Our method processes 2D and 3D data, and employs a transformer architecture to perform data fusion. We show that MOT-DETR outperforms state-of-the-art multi-object tracking methods. Furthermore, we prove that MOT-DETR can leverage 3D data to deal with long-term occlusions and large frame-to-frame distances better than state-of-the-art methods. Finally, we show how our method is resilient to camera pose noise that can affect the accuracy of point clouds. The implementation of MOT-DETR can be found here: https://github.com/drapado/mot-detr},
 bibtype = {article},
 author = {Rapado-Rincon, David and Nap, Henk and Smolenova, Katarina and van Henten, Eldert J. and Kootstra, Gert}
}

In the current demand for automation in the agro-food industry, accurately detecting and localizing relevant objects in 3D is essential for successful robotic operations. However, this is a challenge due the presence of occlusions. Multi-view perception approaches allow robots to overcome occlusions, but a tracking component is needed to associate the objects detected by the robot over multiple viewpoints. Most multi-object tracking (MOT) algorithms are designed for high frame rate sequences and struggle with the occlusions generated by robots' motions and 3D environments. In this paper, we introduce MOT-DETR, a novel approach to detect and track objects in 3D over time using a combination of convolutional networks and transformers. Our method processes 2D and 3D data, and employs a transformer architecture to perform data fusion. We show that MOT-DETR outperforms state-of-the-art multi-object tracking methods. Furthermore, we prove that MOT-DETR can leverage 3D data to deal with long-term occlusions and large frame-to-frame distances better than state-of-the-art methods. Finally, we show how our method is resilient to camera pose noise that can affect the accuracy of point clouds. The implementation of MOT-DETR can be found here: https://github.com/drapado/mot-detr

@article{
 title = {Robots in the Garden: Artificial Intelligence and Adaptive Landscapes},
 type = {article},
 year = {2023},
 keywords = {Resilience,artificial intelligence,climate change,machine learning,robotics,technology},
 pages = {264-272},
 volume = {2023},
 publisher = {VDE VERLAG GMBH},
 id = {82bae6f1-ed8b-3504-b125-7583e135ac2e},
 created = {2024-01-29T10:34:18.068Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-29T10:36:29.987Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {9e28b3ab-dc65-4cb8-8373-439f2758a335},
 private_publication = {false},
 abstract = {This paper introduces ELUA, the Ecological Laboratory for Urban Agriculture, a collaboration among landscape architects, architects and computer scientists who specialize in artificial intelligence, robotics and computer vision. ELUA has two gantry robots, one indoors and the other outside on the rooftop of a 6-story campus building. Each robot can seed, water, weed, and prune in its garden. To support responsive landscape research, ELUA also includes sensor arrays, an AI-powered camera, and an extensive network infrastructure. This project demonstrates a way to integrate artificial intelligence into an evolving urban ecosystem, and encourages landscape architects to develop an adaptive design framework where design becomes a long-term engagement with the environment.},
 bibtype = {article},
 author = {Zhang, Zihao and Epstein, Susan L. and Breen, Casey and Xia, Sophia and Zhu, Zhigang and Volkmann, Christian},
 doi = {10.14627/537740028},
 journal = {Journal of Digital Landscape Architecture},
 number = {8}
}

This paper introduces ELUA, the Ecological Laboratory for Urban Agriculture, a collaboration among landscape architects, architects and computer scientists who specialize in artificial intelligence, robotics and computer vision. ELUA has two gantry robots, one indoors and the other outside on the rooftop of a 6-story campus building. Each robot can seed, water, weed, and prune in its garden. To support responsive landscape research, ELUA also includes sensor arrays, an AI-powered camera, and an extensive network infrastructure. This project demonstrates a way to integrate artificial intelligence into an evolving urban ecosystem, and encourages landscape architects to develop an adaptive design framework where design becomes a long-term engagement with the environment.

@inproceedings{
 title = {Efficient Grapevine Structure Estimation in Vineyards Conditions},
 type = {inproceedings},
 year = {2023},
 pages = {712-720},
 id = {5138ec35-be6d-3fbf-8abe-d94b096f35dc},
 created = {2024-02-14T13:08:56.410Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.410Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {gentilhomme2023efficientgrapevinestructure},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Gentilhomme, Théophile and Villamizar, Michael and Corre, Jerome and Odobez, Jean-Marc},
 booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops}
}

@article{
 title = {Using a Camera System for the In-Situ Assessment of Cordon Dieback due to Grapevine Trunk Diseases},
 type = {article},
 year = {2023},
 pages = {8634742},
 volume = {2023},
 id = {b8c4317e-4a26-3de6-8b04-07bc437f8f4e},
 created = {2024-02-14T13:08:56.618Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.618Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {tang2023usingcamerasystem},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Tang, Julie and Yem, Olivia and Russell, Finn and Stewart, Cameron A and Lin, Kangying and Jayakody, Hiranya and Ayres, Matthew R and Sosnowski, Mark R and Whitty, Mark and Petrie, Paul R and others, undefined},
 journal = {Australian Journal of Grape and Wine Research}
}

@article{
 title = {Mapping the spatial variability of Botrytis bunch rot risk in vineyards using UAV multispectral imagery},
 type = {article},
 year = {2023},
 pages = {126691},
 volume = {142},
 id = {137b00b5-f7c4-3f89-aa59-28d74000c717},
 created = {2024-02-14T13:08:56.793Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.793Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {velez2023mappingspatialvariability},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Vélez, Sergio and Ariza-Sentís, Mar and Valente, João},
 journal = {European Journal of Agronomy}
}

@article{
 title = {Dataset on unmanned aerial vehicle multispectral images acquired over a vineyard affected by Botrytis cinerea in northern Spain},
 type = {article},
 year = {2023},
 pages = {108876},
 volume = {46},
 id = {45d0cfb7-d133-3eb9-81a2-af2ec5cfbf45},
 created = {2024-02-14T13:08:56.965Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.965Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {velez2023datasetuav},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Vélez, Sergio and Ariza-Sentís, Mar and Valente, João},
 journal = {Data in Brief}
}

@article{
 title = {Data acquisition for testing potential detection of Flavescence dorée with a designed, affordable multispectral camera},
 type = {article},
 year = {2023},
 pages = {2-7},
 volume = {2023},
 id = {2666545a-fbbd-3ebc-baeb-f2b6781a3134},
 created = {2024-02-14T13:08:57.344Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:57.344Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {barjaktarovic2023dataacquisitiontesting},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Barjaktarović, Marko and Santoni, Massimo and Faralli, Michele and Bertamini, Massimo and Bruzzone, Lorenzo and others, undefined},
 journal = {Telfor Journal},
 number = {1}
}

@article{
 title = {Detecting Grapevine Virus Infections in Red and White Winegrape Canopies Using Proximal Hyperspectral Sensing},
 type = {article},
 year = {2023},
 pages = {2851},
 volume = {23},
 id = {22c3356e-f390-3c90-ae8f-35b13ed96315},
 created = {2024-02-14T13:08:58.486Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:58.486Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {wang2023detectinggrapevinevirus},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Pagay, Vinay},
 journal = {Sensors},
 number = {5}
}

@article{
 title = {Drones in Plant Disease Assessment, Efficient Monitoring, and Detection: A Way Forward to Smart Agriculture},
 type = {article},
 year = {2023},
 pages = {1524},
 volume = {13},
 id = {bcea88b2-faac-3f75-b922-028485bc6ea7},
 created = {2024-02-14T13:08:58.499Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:58.499Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {abbas2023dronesplantdisease},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Abbas, Aqleem and Zhang, Zhenhao and Zheng, Hongxia and Alami, Mohammad Murtaza and Alrefaei, Abdulmajeed F and Abbas, Qamar and Naqvi, Syed Atif Hasan and Rao, Muhammad Junaid and Mosa, Walid F A and Abbas, Qamar and Hussain, Azhar and Hassan, Muhammad Zeeshan and Zhou, Lei},
 journal = {Agronomy},
 number = {6}
}

@article{
 title = {GrapesNet: Indian RGB \& RGB-D vineyard image datasets for deep learning applications},
 type = {article},
 year = {2023},
 pages = {109100},
 volume = {48},
 id = {d227a7ca-2b79-3b29-86b7-680ad8f52eb3},
 created = {2024-02-14T13:08:58.932Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:58.932Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {barbole2023grapesnetindian},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Barbole, Dhanashree K and Jadhav, Parul M},
 journal = {Data in Brief}
}

@inproceedings{
 title = {Segmentation Methods Evaluation on Grapevine Leaf Diseases},
 type = {inproceedings},
 year = {2023},
 pages = {1081-1085},
 volume = {35},
 series = {Annals of Computer Science and Information Systems},
 id = {f7deb781-84fd-3c01-bb41-e9ddeeaf4247},
 created = {2024-02-14T13:08:59.004Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.004Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {molnar2023segmentationmethodsevaluation},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Molnár, Szilárd and Tamás, Levente},
 booktitle = {Proceedings of the 18th Conference on Computer Science and Intelligence Systems, FedCSIS 2023, Warsaw, Poland, September 17-20, 2023}
}

@article{
 title = {A survey on deep learning-based identification of plant and crop diseases from UAV-based aerial images},
 type = {article},
 year = {2023},
 pages = {1297-1317},
 volume = {26},
 id = {278e00dc-e2f0-30f5-972b-40134694cf86},
 created = {2024-02-14T13:08:59.020Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.020Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {bouguettaya2022surveydeeplearning},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Bouguettaya, Abdelmalek and Zarzour, Hafed and Kechida, Ahmed and Taberkit, Amine Mohammed},
 journal = {Cluster Computing},
 number = {2}
}

@article{
 title = {Evaluating the Potential of High-Resolution Visible Remote Sensing to Detect Shiraz Disease in Grapevines},
 type = {article},
 year = {2023},
 volume = {2023},
 id = {15b6c3b5-5078-3e0f-9d3f-c2409fce4635},
 created = {2024-02-14T13:08:59.468Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.468Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {wang2023evaluatingpotentialhighresolution},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Pagay, Vinay and others, undefined},
 journal = {Australian Journal of Grape and Wine Research}
}

@misc{
 title = {Amazon SageMaker},
 type = {misc},
 year = {2023},
 month = {11},
 id = {76440691-a4e0-31ce-97e8-0a05f8ac765d},
 created = {2024-02-14T13:08:59.627Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.627Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {amazonsagemaker2023},
 source_type = {misc},
 medium = {https://aws.amazon.com/sagemaker/},
 private_publication = {false},
 bibtype = {misc},
 author = {AWS, undefined}
}

@article{
 title = {An expertized grapevine disease image database including five grape varieties focused on Flavescence dorée and its confounding diseases, biotic and abiotic stresses},
 type = {article},
 year = {2023},
 pages = {109230},
 volume = {48},
 id = {28afb888-298f-35fb-96b8-81159c8ca094},
 created = {2024-02-14T13:08:59.772Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.772Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {tardif2023expertizedgrapevinedisease},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Tardif, Malo and Amri, Ahmed and Deshayes, Aymeric and Greven, Marc and Keresztes, Barna and Fontaine, Gaël and Sicaud, Laetitia and Paulhac, Laetitia and Bentejac, Sophie and da Costa, Jean-Pierre},
 journal = {Data in Brief}
}

@article{
 title = {Scalable Early Detection of Grapevine Viral Infection with Airborne Imaging Spectroscopy},
 type = {article},
 year = {2023},
 pages = {1439-1446},
 volume = {113},
 id = {1daec3c3-127f-39b7-8b9b-53eeb98b03cb},
 created = {2024-02-14T13:09:00.034Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:00.034Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {galvan2023scalableearlydetection},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Galvan, Fernando E Romero and Pavlick, Ryan and Trolley, Graham and Aggarwal, Somil and Sousa, Daniel and Starr, Charles and Forrestel, Elisabeth and Bolton, Stephanie and Alsina, Maria del Mar and Dokoozlian, Nick and Gold, Kaitlin M},
 journal = {Phytopathology},
 number = {8}
}

@article{
 title = {Intelligent Monitoring System to Assess Plant Development State Based on Computer Vision in Viticulture},
 type = {article},
 year = {2023},
 pages = {171},
 volume = {11},
 id = {09397c77-6d21-361d-bbdb-9f071699e695},
 created = {2024-02-14T13:09:00.957Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:00.957Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {rudenko2023intelligentmonitoringsystem},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Rudenko, Marina and Kazak, Anatoliy and Oleinikov, Nikolay and Mayorova, Angela and Dorofeeva, Anna and Nekhaychuk, Dmitry and Shutova, Olga},
 journal = {Computation},
 number = {9}
}

@inproceedings{
 title = {Potential detection of Flavescence dorée in the vineyard using close-range hyperspectral imaging},
 type = {inproceedings},
 year = {2023},
 pages = {1-6},
 id = {2b58ffec-47dd-3310-9672-9315fb9b3eb5},
 created = {2024-02-14T13:09:01.891Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:01.891Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {barjaktarovic2023potentialdetectionflavescence},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Barjaktarović, Marko and Santoni, Massimo and Faralli, Michele and Bertamini, Massimo and Bruzzone, Lorenzo},
 booktitle = {2023 3rd International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME)}
}

@inproceedings{
 title = {Automatic diagnosis of a multi-symptom grape vine disease using computer vision},
 type = {inproceedings},
 year = {2023},
 pages = {53-60},
 publisher = {International Society for Horticultural Science (ISHS), Leuven, Belgium},
 id = {031bb480-497f-3641-a7cf-28dd16143f8d},
 created = {2024-02-14T13:09:02.071Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:02.071Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {tardif2023automaticdiagnosismulti},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Tardif, Malo and Amri, Ahmed and Keresztes, Barna and Deshayes, Aymeric and Martin, Damian and Greven, Marc and da Costa, Jean-Pierre},
 booktitle = {Acta Horticulturae}
}

@article{
 title = {Deep Learning YOLO-Based Solution for Grape Bunch Detection and Assessment of Biophysical Lesions},
 type = {article},
 year = {2023},
 pages = {1120},
 volume = {13},
 id = {067e5554-f72b-36dc-ab02-319f6c8a20a4},
 created = {2024-02-14T13:09:02.588Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:02.588Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {pinheiro2023deeplearningyolobased},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Pinheiro, Isabel and Moreira, Germano and da Silva, Daniel and Magalhães, Sandro and Valente, António and Moura Oliveira, Paulo and Cunha, Mário and Santos, Filipe},
 journal = {Agronomy},
 number = {4}
}

@inbook{
 type = {inbook},
 year = {2023},
 pages = {1011-1017},
 publisher = {Wageningen Academic},
 city = {Leiden, The Netherlands},
 id = {d3c70f35-47c4-37b2-890a-a236101fc08d},
 created = {2024-02-14T13:09:03.001Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:03.001Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {tardif2023automaticdiagnosismultigraph},
 source_type = {inbook},
 private_publication = {false},
 bibtype = {inbook},
 author = {Tardif, Malo and Keresztes, Barna and Deshayes, Aymeric and Martin, Damian and Greven, Marc and da Costa, Jean-Pierre},
 chapter = {Automatic diagnosis of a multi-symptom grapevine disease by decision trees and Graph Neural Networks},
 title = {Precision agriculture '23}
}

@article{
 title = {Dataset on UAV RGB videos acquired over a vineyard including bunch labels for object detection and tracking},
 type = {article},
 year = {2023},
 pages = {108848},
 volume = {46},
 id = {5ddb0849-78f8-3211-8ecb-9bae6d55eae4},
 created = {2024-02-14T13:09:03.112Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:03.112Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {arizasentis2023datasetuavrgb},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Ariza-Sentís, Mar and Vélez, Sergio and Valente, João},
 journal = {Data in Brief}
}

@article{
 title = {Implementation of drone technology for farm monitoring \& pesticide spraying: A review},
 type = {article},
 year = {2023},
 pages = {192-203},
 volume = {10},
 id = {b8b193bb-7e05-3c66-aae4-f45f24f541ea},
 created = {2024-02-14T13:09:03.187Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:03.187Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {hafeez2023implementationdronetechnology},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Hafeez, Abdul and Husain, Mohammed Aslam and Singh, S P and Chauhan, Anurag and Khan, Mohd. Tauseef and Kumar, Navneet and Chauhan, Abhishek and Soni, S K},
 journal = {Information Processing in Agriculture},
 number = {2}
}

@inproceedings{
 title = {Evaluating Critical Disease Occurrence in Grapevine Leaves using CNN: Use-Case in Eastern Europe},
 type = {inproceedings},
 year = {2023},
 pages = {1-4},
 id = {b7a00f2b-81ad-3a99-b33e-3643f80746c7},
 created = {2024-02-14T13:09:03.679Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:03.679Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {oprea2023evaluatingcriticaldisease},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Oprea, Cristina-Claudia and Drăgulinescu, Ana-Maria Claudia and Marcu, Ioana-Manuela and Pirnog, Ionuţ},
 booktitle = {2023 17th International Conference on Engineering of Modern Electric Systems (EMES)}
}

@article{
 title = {Proximal sensing for geometric characterization of vines: A review of the latest advances},
 type = {article},
 year = {2023},
 pages = {107901},
 volume = {210},
 id = {779e0de3-6eb2-314f-b3b0-a4076257a924},
 created = {2024-02-14T13:09:03.851Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:03.851Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {moreno2023proximalsensinggeometric},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Moreno, Hugo and Andújar, Dionisio},
 journal = {Computers and Electronics in Agriculture}
}

@article{
 title = {AOGC: Anchor-Free Oriented Object Detection Based on Gaussian Centerness},
 type = {article},
 year = {2023},
 keywords = {Gaussian kernal,anchor,free,one,orientated object detection,remote sensing images,stage},
 pages = {4690},
 volume = {15},
 websites = {https://www.mdpi.com/2072-4292/15/19/4690/htm,https://www.mdpi.com/2072-4292/15/19/4690},
 month = {9},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {25},
 id = {5d858a6f-91f7-32ae-9b36-df3d9e9948eb},
 created = {2024-02-16T10:13:40.203Z},
 accessed = {2024-02-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-04T11:58:02.971Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0,5a010301-acb6-4642-a6b2-8afaee1b741c},
 private_publication = {false},
 abstract = {Oriented object detection is a challenging task in scene text detection and remote sensing image analysis, and it has attracted extensive attention due to the development of deep learning in recent years. Currently, mainstream oriented object detectors are anchor-based methods. These methods increase the computational load of the network and cause a large amount of anchor box redundancy. In order to address this issue, we proposed an anchor-free oriented object detection method based on Gaussian centerness (AOGC), which is a single-stage anchor-free detection method. Our method uses contextual attention FPN (CAFPN) to obtain the contextual information of the target. Then, we designed a label assignment method for the oriented objects, which can select positive samples with higher quality and is suitable for large aspect ratio targets. Finally, we developed a Gaussian kernel-based centerness branch that can effectively determine the significance of different anchors. AOGC achieved a mAP of 74.30% on the DOTA-1.0 datasets and 89.80% on the HRSC2016 datasets, respectively. Our experimental results show that AOGC exhibits superior performance to other methods in single-stage oriented object detection and achieves similar performance to the two-stage methods.},
 bibtype = {article},
 author = {Xia, Gui-Song and Cheng, Gong and Feng, Jie and Mou, Lichao and Wang, Zechen and Bao, Chun and Cao, Jie and Hao, Qun},
 doi = {10.3390/RS15194690},
 journal = {Remote Sensing 2023, Vol. 15, Page 4690},
 number = {19}
}

Oriented object detection is a challenging task in scene text detection and remote sensing image analysis, and it has attracted extensive attention due to the development of deep learning in recent years. Currently, mainstream oriented object detectors are anchor-based methods. These methods increase the computational load of the network and cause a large amount of anchor box redundancy. In order to address this issue, we proposed an anchor-free oriented object detection method based on Gaussian centerness (AOGC), which is a single-stage anchor-free detection method. Our method uses contextual attention FPN (CAFPN) to obtain the contextual information of the target. Then, we designed a label assignment method for the oriented objects, which can select positive samples with higher quality and is suitable for large aspect ratio targets. Finally, we developed a Gaussian kernel-based centerness branch that can effectively determine the significance of different anchors. AOGC achieved a mAP of 74.30% on the DOTA-1.0 datasets and 89.80% on the HRSC2016 datasets, respectively. Our experimental results show that AOGC exhibits superior performance to other methods in single-stage oriented object detection and achieves similar performance to the two-stage methods.

@article{
 title = {A comprehensive survey of oriented object detection in remote sensing images},
 type = {article},
 year = {2023},
 keywords = {Anchor-free,Oriented object detection,Rotation invariance},
 pages = {119960},
 volume = {224},
 month = {8},
 publisher = {Pergamon},
 day = {15},
 id = {ca798cae-1aa3-32ab-9a6f-fb8bfbd9e8a5},
 created = {2024-03-14T10:28:19.463Z},
 accessed = {2024-03-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-15T07:41:31.204Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {With the rapid development of object detection, it is widely used in many scenes and images. However, the dense arrangement of objects with different dimensions, orientations and aspect ratios in remote sensing and aerial images undoubtedly poses many problems for detection. Anchor-based oriented object detection to maintain rotational invariance has to solve the problem of object orientation dimension and also to consider the calculation of angular periodicity in the regression calculation. To achieve accurate detection of objects, it is necessary to obtain the precise frame surrounding the object and the precise features. Anchor-free methods do not require a predefined anchor, but only need to learn the object feature parameters to get an accurate frame for detection. In this paper we first introduce the technical approaches to object detection, both traditional and deep learning-based methods. Then we summarize the main problems and methods solved in oriented object detection in anchor-based and anchor-free based detection. We present some datasets using oriented bounding box (OBB) annotation that are suitable for oriented object detection, as well as introduce the accepted benchmarks and evaluation metrics for object detection. Finally, we discuss potential trends in oriented object detection for the benefit of researchers who are new to the field.},
 bibtype = {article},
 author = {Wen, Long and Cheng, Yu and Fang, Yi and Li, Xinyu},
 doi = {10.1016/J.ESWA.2023.119960},
 journal = {Expert Systems with Applications}
}

With the rapid development of object detection, it is widely used in many scenes and images. However, the dense arrangement of objects with different dimensions, orientations and aspect ratios in remote sensing and aerial images undoubtedly poses many problems for detection. Anchor-based oriented object detection to maintain rotational invariance has to solve the problem of object orientation dimension and also to consider the calculation of angular periodicity in the regression calculation. To achieve accurate detection of objects, it is necessary to obtain the precise frame surrounding the object and the precise features. Anchor-free methods do not require a predefined anchor, but only need to learn the object feature parameters to get an accurate frame for detection. In this paper we first introduce the technical approaches to object detection, both traditional and deep learning-based methods. Then we summarize the main problems and methods solved in oriented object detection in anchor-based and anchor-free based detection. We present some datasets using oriented bounding box (OBB) annotation that are suitable for oriented object detection, as well as introduce the accepted benchmarks and evaluation metrics for object detection. Finally, we discuss potential trends in oriented object detection for the benefit of researchers who are new to the field.

@article{
 title = {ARS-DETR: Aspect Ratio Sensitive Oriented Object Detection with Transformer},
 type = {article},
 year = {2023},
 websites = {http://arxiv.org/abs/2303.04989},
 id = {8f79e926-8275-3bc4-8b5f-7b183c48d4d3},
 created = {2024-03-15T07:41:30.931Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-15T07:42:03.631Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Existing oriented object detection methods commonly use metric AP$_50$ to measure the performance of the model. We argue that AP$_50$ is inherently unsuitable for oriented object detection due to its large tolerance in angle deviation. Therefore, we advocate using high-precision metric, e.g. AP$_75$, to measure the performance of models. In this paper, we propose an Aspect Ratio Sensitive Oriented Object Detector with Transformer, termed ARS-DETR, which exhibits a competitive performance in high-precision oriented object detection. Specifically, a new angle classification method, calling Aspect Ratio aware Circle Smooth Label (AR-CSL), is proposed to smooth the angle label in a more reasonable way and discard the hyperparameter that introduced by previous work (e.g. CSL). Then, a rotated deformable attention module is designed to rotate the sampling points with the corresponding angles and eliminate the misalignment between region features and sampling points. Moreover, a dynamic weight coefficient according to the aspect ratio is adopted to calculate the angle loss. Comprehensive experiments on several challenging datasets show that our method achieves competitive performance on the high-precision oriented object detection task.},
 bibtype = {article},
 author = {Zeng, Ying and Yang, Xue and Li, Qingyun and Chen, Yushi and Yan, Junchi},
 doi = {10.1109/TGRS.2024.3364713}
}

Existing oriented object detection methods commonly use metric AP$_50$ to measure the performance of the model. We argue that AP$_50$ is inherently unsuitable for oriented object detection due to its large tolerance in angle deviation. Therefore, we advocate using high-precision metric, e.g. AP$_75$, to measure the performance of models. In this paper, we propose an Aspect Ratio Sensitive Oriented Object Detector with Transformer, termed ARS-DETR, which exhibits a competitive performance in high-precision oriented object detection. Specifically, a new angle classification method, calling Aspect Ratio aware Circle Smooth Label (AR-CSL), is proposed to smooth the angle label in a more reasonable way and discard the hyperparameter that introduced by previous work (e.g. CSL). Then, a rotated deformable attention module is designed to rotate the sampling points with the corresponding angles and eliminate the misalignment between region features and sampling points. Moreover, a dynamic weight coefficient according to the aspect ratio is adopted to calculate the angle loss. Comprehensive experiments on several challenging datasets show that our method achieves competitive performance on the high-precision oriented object detection task.

@article{
 title = {Benchmarking Generations of You Only Look Once Architectures for Detection of Defective and Normal Long Rod Insulators},
 type = {article},
 year = {2023},
 keywords = {Insulator detection,Insulator fault detection,YOLOv3,YOLOv4,YOLOv5},
 pages = {1093-1107},
 volume = {34},
 websites = {https://link.springer.com/article/10.1007/s40313-023-01023-3},
 month = {10},
 publisher = {Springer},
 day = {1},
 id = {751462e4-662d-3915-89e3-c97cafb131c6},
 created = {2024-03-16T12:43:07.789Z},
 accessed = {2024-03-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-16T12:43:11.012Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Effective infrastructure monitoring is a priority in all technical fields in this century. In high-voltage transmission networks, line inspection is one such task. Fault detection of insulators is crucial, and object detection algorithms can handle this problem. This work presents a comparison of You Only Look Once architectures. The different subtypes of the last three generations (v3, v4, and v5) are compared in terms of losses, precision, recall, and mean average precision on an open-source, augmented dataset of normal and defective insulators from the State Grid Corporation of China. The primary focus of this work is a comprehensive subtype analysis, providing a useful resource for academics and industry professionals involved in insulator detection and surveillance projects. This study aims to enhance the monitoring of insulator health and maintenance for industries relying on power grid stability. YOLOv5 subtypes are found to be the most suitable for this computer vision task, considering their mean average precision, which ranges between 98.1 and 99.0%, and a frame per second rate between 27.1 and 212.8, depending on the architecture size. While their predecessors are faster, they are less accurate. It is also discovered that, for all generations, normal-sized and large architectures generally demonstrate better accuracy. However, small architectures are noted for their significantly faster processing speeds.},
 bibtype = {article},
 author = {Békési, Gergő Bendegúz},
 doi = {10.1007/S40313-023-01023-3/TABLES/2},
 journal = {Journal of Control, Automation and Electrical Systems},
 number = {5}
}

Effective infrastructure monitoring is a priority in all technical fields in this century. In high-voltage transmission networks, line inspection is one such task. Fault detection of insulators is crucial, and object detection algorithms can handle this problem. This work presents a comparison of You Only Look Once architectures. The different subtypes of the last three generations (v3, v4, and v5) are compared in terms of losses, precision, recall, and mean average precision on an open-source, augmented dataset of normal and defective insulators from the State Grid Corporation of China. The primary focus of this work is a comprehensive subtype analysis, providing a useful resource for academics and industry professionals involved in insulator detection and surveillance projects. This study aims to enhance the monitoring of insulator health and maintenance for industries relying on power grid stability. YOLOv5 subtypes are found to be the most suitable for this computer vision task, considering their mean average precision, which ranges between 98.1 and 99.0%, and a frame per second rate between 27.1 and 212.8, depending on the architecture size. While their predecessors are faster, they are less accurate. It is also discovered that, for all generations, normal-sized and large architectures generally demonstrate better accuracy. However, small architectures are noted for their significantly faster processing speeds.

@article{
 title = {Investigation of You Only Look Once Networks for Vision-based Small Object Detection},
 type = {article},
 year = {2023},
 keywords = {YOLOv5,YOLOv6,YOLOv7,computer vision,jewellery detection,real-time detection,small object detection},
 pages = {69-82},
 volume = {14},
 websites = {www.ijacsa.thesai.org},
 month = {38},
 publisher = {The Science and Information (SAI) Organization Limited},
 day = {29},
 id = {ec6503e1-425e-36f5-b756-78d4bf05d3f1},
 created = {2024-03-16T12:50:26.559Z},
 accessed = {2024-03-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-04-10T06:59:40.618Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Small object detection is a challenging issue in computer vision-based algorithms. Although various methods have been investigated for common objects including person, car and others, small object are not addressed in this issue. Therefore, it is necessary to conduct more researches on them. This paper is focused on small object detection especially jewellery as current object detection methods suffer from low accuracy in this domain. This paper introduces a new dataset whose images were taken by a web camera from a jewellery store and data augmentation procedure. It comprises three classes, namely, ring, earrings, and pendant. In view of the small target of jewellery and the real-time detection, this study adopted the You Only Look Once (Yolo) algorithms. Different Yolo based model including eight versions are implemented and train them using our dataset to address most effective one. Evaluation criteria, including accuracy, F1 score, recall, and mAP, are used to evaluate the performance of the various YOLOv5, YOLOv6, and YOLOv7 versions. According to the experimental findings, utilizing YOLOv6 is significantly superior to YOLOv7 and marginally superior to YOLOv5.},
 bibtype = {article},
 author = {Yang, Li},
 doi = {10.14569/IJACSA.2023.0140410},
 journal = {International Journal of Advanced Computer Science and Applications},
 number = {4}
}

Small object detection is a challenging issue in computer vision-based algorithms. Although various methods have been investigated for common objects including person, car and others, small object are not addressed in this issue. Therefore, it is necessary to conduct more researches on them. This paper is focused on small object detection especially jewellery as current object detection methods suffer from low accuracy in this domain. This paper introduces a new dataset whose images were taken by a web camera from a jewellery store and data augmentation procedure. It comprises three classes, namely, ring, earrings, and pendant. In view of the small target of jewellery and the real-time detection, this study adopted the You Only Look Once (Yolo) algorithms. Different Yolo based model including eight versions are implemented and train them using our dataset to address most effective one. Evaluation criteria, including accuracy, F1 score, recall, and mAP, are used to evaluate the performance of the various YOLOv5, YOLOv6, and YOLOv7 versions. According to the experimental findings, utilizing YOLOv6 is significantly superior to YOLOv7 and marginally superior to YOLOv5.

@article{
 title = {Lightweight You Only Look Once v8: An Upgraded You Only Look Once v8 Algorithm for Small Object Identification in Unmanned Aerial Vehicle Images},
 type = {article},
 year = {2023},
 keywords = {YOLO v8,attention mechanism,small targets,target detection,unmanned aerial vehicle},
 pages = {12369},
 volume = {13},
 websites = {https://www.mdpi.com/2076-3417/13/22/12369/htm,https://www.mdpi.com/2076-3417/13/22/12369},
 month = {11},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {15},
 id = {1b3f9f34-4e59-3367-bad9-6fb9653a91d4},
 created = {2024-03-16T12:51:54.758Z},
 accessed = {2024-03-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-16T12:57:43.065Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {In order to solve the problems of high leakage rate, high false detection rate, low detection success rate and large model volume of small targets in the traditional target detection algorithm for Unmanned Aerial Vehicle (UAV) aerial images, a lightweight You Only Look Once (YOLO) v8 algorithm model Lightweight (LW)-YOLO v8 is proposed. By increasing the channel attention mechanism Squeeze-and-Excitation (SE) module, this method can adaptively improves the model’s ability to extract features from small targets; at the same time, the lightweight convolution technology is introduced into the Conv module, where the ordinary convolution is replaced by the GSConv module, which can effectively reduce the model computational volume; on the basis of the GSConv module, a single aggregation module VoV-GSCSPC is designed to optimize the model structure in order to achieve a higher computational cost-effectiveness. The experimental results show that the LW-YOLO v8 model’s mAP@0.5 metrics on the VisDrone2019 dataset are more favorable than those on the YOLO v8n model, improving by 3.8 percentage points, and the computational amount is reduced to 7.2 GFLOPs. The LW-YOLO v8 model proposed in this work can effectively accomplish the task of detecting small targets in aerial images from UAV at a lower cost.},
 bibtype = {article},
 author = {Huangfu, Zhongmin and Li, Shuqing},
 doi = {10.3390/APP132212369},
 journal = {Applied Sciences 2023, Vol. 13, Page 12369},
 number = {22}
}

In order to solve the problems of high leakage rate, high false detection rate, low detection success rate and large model volume of small targets in the traditional target detection algorithm for Unmanned Aerial Vehicle (UAV) aerial images, a lightweight You Only Look Once (YOLO) v8 algorithm model Lightweight (LW)-YOLO v8 is proposed. By increasing the channel attention mechanism Squeeze-and-Excitation (SE) module, this method can adaptively improves the model’s ability to extract features from small targets; at the same time, the lightweight convolution technology is introduced into the Conv module, where the ordinary convolution is replaced by the GSConv module, which can effectively reduce the model computational volume; on the basis of the GSConv module, a single aggregation module VoV-GSCSPC is designed to optimize the model structure in order to achieve a higher computational cost-effectiveness. The experimental results show that the LW-YOLO v8 model’s mAP@0.5 metrics on the VisDrone2019 dataset are more favorable than those on the YOLO v8n model, improving by 3.8 percentage points, and the computational amount is reduced to 7.2 GFLOPs. The LW-YOLO v8 model proposed in this work can effectively accomplish the task of detecting small targets in aerial images from UAV at a lower cost.

@article{
 title = {YOLO-SE: Improved YOLOv8 for Remote Sensing Object Detection and Recognition},
 type = {article},
 year = {2023},
 keywords = {loss functions,multi,object detection,remote sensing images,scale},
 pages = {12977},
 volume = {13},
 websites = {https://www.mdpi.com/2076-3417/13/24/12977/htm,https://www.mdpi.com/2076-3417/13/24/12977},
 month = {12},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {5},
 id = {29fa1003-d5cc-3fc0-9fca-cb7120e46000},
 created = {2024-03-16T15:17:53.484Z},
 accessed = {2024-03-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-16T15:17:58.593Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Object detection remains a pivotal aspect of remote sensing image analysis, and recent strides in Earth observation technology coupled with convolutional neural networks (CNNs) have propelled the field forward. Despite advancements, challenges persist, especially in detecting objects across diverse scales and pinpointing small-sized targets. This paper introduces YOLO-SE, a novel YOLOv8-based network that innovatively addresses these challenges. First, the introduction of a lightweight convolution SEConv in lieu of standard convolutions reduces the network’s parameter count, thereby expediting the detection process. To tackle multi-scale object detection, the paper proposes the SEF module, an enhancement based on SEConv. Second, an ingenious Efficient Multi-Scale Attention (EMA) mechanism is integrated into the network, forming the SPPFE module. This addition augments the network’s feature extraction capabilities, adeptly handling challenges in multi-scale object detection. Furthermore, a dedicated prediction head for tiny object detection is incorporated, and the original detection head is replaced by a transformer prediction head. To address adverse gradients stemming from low-quality instances in the target detection training dataset, the paper introduces the Wise-IoU bounding box loss function. YOLO-SE showcases remarkable performance, achieving an average precision at IoU threshold 0.5 (AP50) of 86.5% on the optical remote sensing dataset SIMD. This represents a noteworthy 2.1% improvement over YOLOv8 and YOLO-SE outperforms the state-of-the-art model by 0.91%. In further validation, experiments on the NWPU VHR-10 dataset demonstrated YOLO-SE’s superiority with an accuracy of 94.9%, surpassing that of YOLOv8 by 2.6%. The proposed advancements position YOLO-SE as a compelling solution in the realm of deep learning-based remote sensing image object detection.},
 bibtype = {article},
 author = {Wu, Tianyong and Dong, Youkou},
 doi = {10.3390/APP132412977},
 journal = {Applied Sciences 2023, Vol. 13, Page 12977},
 number = {24}
}

Object detection remains a pivotal aspect of remote sensing image analysis, and recent strides in Earth observation technology coupled with convolutional neural networks (CNNs) have propelled the field forward. Despite advancements, challenges persist, especially in detecting objects across diverse scales and pinpointing small-sized targets. This paper introduces YOLO-SE, a novel YOLOv8-based network that innovatively addresses these challenges. First, the introduction of a lightweight convolution SEConv in lieu of standard convolutions reduces the network’s parameter count, thereby expediting the detection process. To tackle multi-scale object detection, the paper proposes the SEF module, an enhancement based on SEConv. Second, an ingenious Efficient Multi-Scale Attention (EMA) mechanism is integrated into the network, forming the SPPFE module. This addition augments the network’s feature extraction capabilities, adeptly handling challenges in multi-scale object detection. Furthermore, a dedicated prediction head for tiny object detection is incorporated, and the original detection head is replaced by a transformer prediction head. To address adverse gradients stemming from low-quality instances in the target detection training dataset, the paper introduces the Wise-IoU bounding box loss function. YOLO-SE showcases remarkable performance, achieving an average precision at IoU threshold 0.5 (AP50) of 86.5% on the optical remote sensing dataset SIMD. This represents a noteworthy 2.1% improvement over YOLOv8 and YOLO-SE outperforms the state-of-the-art model by 0.91%. In further validation, experiments on the NWPU VHR-10 dataset demonstrated YOLO-SE’s superiority with an accuracy of 94.9%, surpassing that of YOLOv8 by 2.6%. The proposed advancements position YOLO-SE as a compelling solution in the realm of deep learning-based remote sensing image object detection.

@article{
 title = {G-Rep: Gaussian Representation for Arbitrary-Oriented Object Detection},
 type = {article},
 year = {2023},
 keywords = {Gaussian metrics,Gaussian representation,arbitrary-oriented object detection,convolutional neural networks},
 pages = {1-21},
 volume = {15},
 id = {545989fb-8ae1-39c1-aadd-31d1562b6398},
 created = {2024-03-27T10:37:21.871Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-04-08T06:04:34.526Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Typical representations for arbitrary-oriented object detection tasks include the oriented bounding box (OBB), the quadrilateral bounding box (QBB), and the point set (PointSet). Each representation encounters problems that correspond to its characteristics, such as boundary discontinuity, square-like problems, representation ambiguity, and isolated points, which lead to inaccurate detection. Although many effective strategies have been proposed for various representations, there is still no unified solution. Current detection methods based on Gaussian modeling have demonstrated the possibility of resolving this dilemma; however, they remain limited to OBB. To go further, in this paper, we propose a unified Gaussian representation called G-Rep to construct Gaussian distributions for OBB, QBB, and PointSet, which achieves a unified solution to various representations and problems. Specifically, PointSet- or QBB-based object representations are converted into Gaussian distributions and their parameters are optimized using the maximum likelihood estimation algorithm. Then, three optional Gaussian metrics are explored to optimize the regression loss of the detector because of their excellent parameter optimization mechanisms. Furthermore, we also use Gaussian metrics for sampling to align label assignment and regression loss. Experimental results obtained on several publicly available datasets, such as DOTA, HRSC2016, UCAS-AOD, and ICDAR2015, show the excellent performance of the proposed method for arbitrary-oriented object detection.},
 bibtype = {article},
 author = {Hou, Liping and Lu, Ke and Yang, Xue and Li, Yuqiu and Xue, Jian},
 doi = {10.3390/rs15030757},
 journal = {Remote Sensing},
 number = {3}
}

Typical representations for arbitrary-oriented object detection tasks include the oriented bounding box (OBB), the quadrilateral bounding box (QBB), and the point set (PointSet). Each representation encounters problems that correspond to its characteristics, such as boundary discontinuity, square-like problems, representation ambiguity, and isolated points, which lead to inaccurate detection. Although many effective strategies have been proposed for various representations, there is still no unified solution. Current detection methods based on Gaussian modeling have demonstrated the possibility of resolving this dilemma; however, they remain limited to OBB. To go further, in this paper, we propose a unified Gaussian representation called G-Rep to construct Gaussian distributions for OBB, QBB, and PointSet, which achieves a unified solution to various representations and problems. Specifically, PointSet- or QBB-based object representations are converted into Gaussian distributions and their parameters are optimized using the maximum likelihood estimation algorithm. Then, three optional Gaussian metrics are explored to optimize the regression loss of the detector because of their excellent parameter optimization mechanisms. Furthermore, we also use Gaussian metrics for sampling to align label assignment and regression loss. Experimental results obtained on several publicly available datasets, such as DOTA, HRSC2016, UCAS-AOD, and ICDAR2015, show the excellent performance of the proposed method for arbitrary-oriented object detection.

@article{
 title = {H2RBox-v2: Incorporating Symmetry for Boosting Horizontal Box Supervised Oriented Object Detection},
 type = {article},
 year = {2023},
 pages = {59137-59150},
 volume = {36},
 month = {12},
 day = {15},
 id = {3615db2d-cfc8-352d-91c2-a34c9a555d70},
 created = {2024-06-26T06:38:18.354Z},
 accessed = {2024-06-26},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-02T07:05:11.216Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {With the rapidly increasing demand for oriented object detection, e.g. in autonomous driving and remote sensing, the recently proposed paradigm involving weakly-supervised detector H2RBox for learning rotated box (RBox) from the more readily-available horizontal box (HBox) has shown promise. This paper presents H2RBox-v2, to further bridge the gap between HBox-supervised and RBox-supervised oriented object detection. Specifically, we propose to leverage the reflection symmetry via flip and rotate consistencies, using a weakly-supervised network branch similar to H2RBox, together with a novel self-supervised branch that learns orientations from the symmetry inherent in visual objects. The detector is further stabilized and enhanced by practical techniques to cope with peripheral issues e.g. angular periodicity. To our best knowledge, H2RBox-v2 is the first symmetry-aware self-supervised paradigm for oriented object detection. In particular, our method shows less susceptibility to low-quality annotation and insufficient training data compared to H2RBox. Specifically, H2RBox-v2 achieves very close performance to a rotation annotation trained counterpart-Rotated FCOS: 1) DOTA-v1.0/1.5/2.0: 72.31%/64.76%/50.33% vs. 72.44%/64.53%/51.77%; 2) HRSC: 89.66% vs. 88.99%; 3) FAIR1M: 42.27% vs. 41.25%.},
 bibtype = {article},
 author = {Yu, Yi and Yang, Xue and Li, Qingyun and Zhou, Yue and Zhang, Gefan and Da, Feipeng and Yan, Junchi},
 journal = {Advances in Neural Information Processing Systems}
}

With the rapidly increasing demand for oriented object detection, e.g. in autonomous driving and remote sensing, the recently proposed paradigm involving weakly-supervised detector H2RBox for learning rotated box (RBox) from the more readily-available horizontal box (HBox) has shown promise. This paper presents H2RBox-v2, to further bridge the gap between HBox-supervised and RBox-supervised oriented object detection. Specifically, we propose to leverage the reflection symmetry via flip and rotate consistencies, using a weakly-supervised network branch similar to H2RBox, together with a novel self-supervised branch that learns orientations from the symmetry inherent in visual objects. The detector is further stabilized and enhanced by practical techniques to cope with peripheral issues e.g. angular periodicity. To our best knowledge, H2RBox-v2 is the first symmetry-aware self-supervised paradigm for oriented object detection. In particular, our method shows less susceptibility to low-quality annotation and insufficient training data compared to H2RBox. Specifically, H2RBox-v2 achieves very close performance to a rotation annotation trained counterpart-Rotated FCOS: 1) DOTA-v1.0/1.5/2.0: 72.31%/64.76%/50.33% vs. 72.44%/64.53%/51.77%; 2) HRSC: 89.66% vs. 88.99%; 3) FAIR1M: 42.27% vs. 41.25%.

@article{
 title = {AgriDet: Plant Leaf Disease severity classification using agriculture detection framework},
 type = {article},
 year = {2023},
 keywords = {Classification,Deep learning,Occlusion,Plant disease detection,Segmentation,Severity classes},
 pages = {105754},
 volume = {119},
 month = {3},
 publisher = {Pergamon},
 day = {1},
 id = {c881c0e7-96e2-38c6-80b5-4b0ac2fdf8ca},
 created = {2024-09-20T11:52:45.872Z},
 accessed = {2024-09-20},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-20T11:52:49.023Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {In the field of modern agriculture, plant disease detection plays a vital role in improving crop productivity. To increase the yield on a large scale, it is necessary to predict the onset of the disease and give advice to farmers. Previous methods for detecting plant diseases rely on manual feature extraction, which is more expensive. Therefore, image-based techniques are gaining interest in the research area of plant disease detection. However, existing methods have several problems due to the improper nature of the captured image, including improper background conditions that lead to occlusion, illumination, orientation, and size. Also, cost complexity, misclassifications, and overfitting problems occur in several real-time applications. To solve these issues, we proposed an Agriculture Detection (AgriDet) framework that incorporates conventional Inception-Visual Geometry Group Network (INC-VGGN) and Kohonen-based deep learning networks to detect plant diseases and classify the severity level of diseased plants. In this framework, image pre-processing is done to remove all the constraints in the captured image. Then, the occlusion problem is tackled by the proposed multi-variate grabcut algorithm for effective segmentation. Furthermore, the framework performs accurate disease detection and classification by utilizing an improved base network, namely a pre-trained conventionally based INC-VGGN model. Here, the pre-trained INC-VGGN model is a deep convolutional neural network for prediction of plant diseases that was previously trained for the distinctive dataset. The pre-trained weights and the features learned in this base network are transferred into the newly developed neural network to perform the specific task of plant disease detection for our dataset. In order to overcome the overfitting problem, a dropout layer is introduced, and the deep learning of features is performed using the Kohonen learning layer. After percentage computation, the improved base network classifies the severity classes in the training sets. Finally, the performance of the framework is computed for different performance metrics and achieves better accuracy than previous models. Also, the performance of the statistical analysis is validated to prove the results in terms of accuracy, specificity, and sensitivity.},
 bibtype = {article},
 author = {Pal, Arunangshu and Kumar, Vinay},
 doi = {10.1016/J.ENGAPPAI.2022.105754},
 journal = {Engineering Applications of Artificial Intelligence}
}

In the field of modern agriculture, plant disease detection plays a vital role in improving crop productivity. To increase the yield on a large scale, it is necessary to predict the onset of the disease and give advice to farmers. Previous methods for detecting plant diseases rely on manual feature extraction, which is more expensive. Therefore, image-based techniques are gaining interest in the research area of plant disease detection. However, existing methods have several problems due to the improper nature of the captured image, including improper background conditions that lead to occlusion, illumination, orientation, and size. Also, cost complexity, misclassifications, and overfitting problems occur in several real-time applications. To solve these issues, we proposed an Agriculture Detection (AgriDet) framework that incorporates conventional Inception-Visual Geometry Group Network (INC-VGGN) and Kohonen-based deep learning networks to detect plant diseases and classify the severity level of diseased plants. In this framework, image pre-processing is done to remove all the constraints in the captured image. Then, the occlusion problem is tackled by the proposed multi-variate grabcut algorithm for effective segmentation. Furthermore, the framework performs accurate disease detection and classification by utilizing an improved base network, namely a pre-trained conventionally based INC-VGGN model. Here, the pre-trained INC-VGGN model is a deep convolutional neural network for prediction of plant diseases that was previously trained for the distinctive dataset. The pre-trained weights and the features learned in this base network are transferred into the newly developed neural network to perform the specific task of plant disease detection for our dataset. In order to overcome the overfitting problem, a dropout layer is introduced, and the deep learning of features is performed using the Kohonen learning layer. After percentage computation, the improved base network classifies the severity classes in the training sets. Finally, the performance of the framework is computed for different performance metrics and achieves better accuracy than previous models. Also, the performance of the statistical analysis is validated to prove the results in terms of accuracy, specificity, and sensitivity.

@article{
 title = {Plant Disease Severity Detection and Fertilizer Recommendation using Deep Learning Techniques},
 type = {article},
 year = {2023},
 keywords = {ANN,CNN,Deep Learning,KNN,Machine Learning,Prediction,SVM,Severity},
 pages = {215-221},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {a8f33efb-bc2f-36f7-84e5-6cc3b8e67971},
 created = {2024-09-20T11:59:05.739Z},
 accessed = {2024-09-20},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-23T09:20:38.430Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {In India, the agriculture industry plays a significant role in the economy and employs a sizable section of the workforce. The demand for food is increasing and analysis of agriculture data can help improve practices and increase productivity by providing insights into crop diseases and weather conditions. Plant diseases can greatly impact agricultural productivity, and early detection is crucial to avoiding losses. The proposed project makes use of different ML techniques such as KNN, SVM, and DL techniques such as CNN and ANN to detect plant diseases in an efficient and effective manner. These techniques can be trained on large datasets to learn patterns and make predictions, making them well suited for this task. The Deep Learning system includes a system that automatically scans leaf images and detects disease based on visual symptoms. This system also calculates severity level of disease and suggests suitable amount of fertilizer for disease to soak in their crop according to severity level. A user interface was created to help farmers and agriculture workers for easy usage by simple capturing leaf image and get suggestions, this helps farmers to increase their crop production and to maintain quality of crop.},
 bibtype = {article},
 author = {Sudhir, Buddepu and Teja, Devalaraju Charan and Sai, Kurra and Sridhar, Peddinti and Daniya, T.},
 doi = {10.1109/ICAAIC56838.2023.10140467},
 journal = {Proceedings of the 2nd International Conference on Applied Artificial Intelligence and Computing, ICAAIC 2023}
}

In India, the agriculture industry plays a significant role in the economy and employs a sizable section of the workforce. The demand for food is increasing and analysis of agriculture data can help improve practices and increase productivity by providing insights into crop diseases and weather conditions. Plant diseases can greatly impact agricultural productivity, and early detection is crucial to avoiding losses. The proposed project makes use of different ML techniques such as KNN, SVM, and DL techniques such as CNN and ANN to detect plant diseases in an efficient and effective manner. These techniques can be trained on large datasets to learn patterns and make predictions, making them well suited for this task. The Deep Learning system includes a system that automatically scans leaf images and detects disease based on visual symptoms. This system also calculates severity level of disease and suggests suitable amount of fertilizer for disease to soak in their crop according to severity level. A user interface was created to help farmers and agriculture workers for easy usage by simple capturing leaf image and get suggestions, this helps farmers to increase their crop production and to maintain quality of crop.

@article{
 title = {Apparatus and method for image-guided agriculture},
 type = {article},
 year = {2023},
 month = {2},
 day = {13},
 id = {8925d083-898f-3990-adb8-9393dde541d4},
 created = {2024-09-24T11:53:56.226Z},
 accessed = {2024-09-24},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-10-06T11:39:58.820Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {d1530310-ca4f-487a-aa15-294fe1ffea82},
 private_publication = {false},
 abstract = {A method for image-guided agriculture includes receiving images; processing the images to generate reflectance maps respectively corresponding to spectral bands; synthesizing the reflectance maps to generate a multispectral image including vegetation index information of a target area; receiving crop information in regions of the target area; and assessing crop conditions for the regions based on the identified crop information and the vegetation index information.},
 bibtype = {article},
 author = {}
}

A method for image-guided agriculture includes receiving images; processing the images to generate reflectance maps respectively corresponding to spectral bands; synthesizing the reflectance maps to generate a multispectral image including vegetation index information of a target area; receiving crop information in regions of the target area; and assessing crop conditions for the regions based on the identified crop information and the vegetation index information.

@article{
 title = {Behavior-Aware Pedestrian Trajectory Prediction in Ego-Centric Camera Views with Spatio-Temporal Ego-Motion Estimation †},
 type = {article},
 year = {2023},
 keywords = {autonomous driving,behavioral features,ego-motion compensation,pedestrian trajectory prediction},
 pages = {957-978},
 volume = {5},
 month = {9},
 publisher = {Multidisciplinary Digital Publishing Institute (MDPI)},
 day = {1},
 id = {061e221e-ce16-3509-b461-490878c857c2},
 created = {2024-11-06T13:04:01.465Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-07T10:52:37.948Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4cda9297-f98e-4246-88d6-ffeeade205c3},
 private_publication = {false},
 abstract = {With the ongoing development of automated driving systems, the crucial task of predicting pedestrian behavior is attracting growing attention. The prediction of future pedestrian trajectories from the ego-vehicle camera perspective is particularly challenging due to the dynamically changing scene. Therefore, we present Behavior-Aware Pedestrian Trajectory Prediction (BA-PTP), a novel approach to pedestrian trajectory prediction for ego-centric camera views. It incorporates behavioral features extracted from real-world traffic scene observations such as the body and head orientation of pedestrians, as well as their pose, in addition to positional information from body and head bounding boxes. For each input modality, we employed independent encoding streams that are combined through a modality attention mechanism. To account for the ego-motion of the camera in an ego-centric view, we introduced Spatio-Temporal Ego-Motion Module (STEMM), a novel approach to ego-motion prediction. Compared to the related works, it utilizes spatial goal points of the ego-vehicle that are sampled from its intended route. We experimentally validated the effectiveness of our approach using two datasets for pedestrian behavior prediction in urban traffic scenes. Based on ablation studies, we show the advantages of incorporating different behavioral features for pedestrian trajectory prediction in the image plane. Moreover, we demonstrate the benefit of integrating STEMM into our pedestrian trajectory prediction method, BA-PTP. BA-PTP achieves state-of-the-art performance on the PIE dataset, outperforming prior work by 7% in MSE-1.5 s and CMSE as well as 9% in CFMSE.},
 bibtype = {article},
 author = {Czech, Phillip and Braun, Markus and Kreßel, Ulrich and Yang, Bin},
 doi = {10.3390/make5030050},
 journal = {Machine Learning and Knowledge Extraction},
 number = {3}
}

With the ongoing development of automated driving systems, the crucial task of predicting pedestrian behavior is attracting growing attention. The prediction of future pedestrian trajectories from the ego-vehicle camera perspective is particularly challenging due to the dynamically changing scene. Therefore, we present Behavior-Aware Pedestrian Trajectory Prediction (BA-PTP), a novel approach to pedestrian trajectory prediction for ego-centric camera views. It incorporates behavioral features extracted from real-world traffic scene observations such as the body and head orientation of pedestrians, as well as their pose, in addition to positional information from body and head bounding boxes. For each input modality, we employed independent encoding streams that are combined through a modality attention mechanism. To account for the ego-motion of the camera in an ego-centric view, we introduced Spatio-Temporal Ego-Motion Module (STEMM), a novel approach to ego-motion prediction. Compared to the related works, it utilizes spatial goal points of the ego-vehicle that are sampled from its intended route. We experimentally validated the effectiveness of our approach using two datasets for pedestrian behavior prediction in urban traffic scenes. Based on ablation studies, we show the advantages of incorporating different behavioral features for pedestrian trajectory prediction in the image plane. Moreover, we demonstrate the benefit of integrating STEMM into our pedestrian trajectory prediction method, BA-PTP. BA-PTP achieves state-of-the-art performance on the PIE dataset, outperforming prior work by 7% in MSE-1.5 s and CMSE as well as 9% in CFMSE.

@article{
 title = {Learning power Gaussian modeling loss for dense rotated object detection in remote sensing images},
 type = {article},
 year = {2023},
 keywords = {Convolutional neural networks,Distribution metric,Losses,Remote sensing,Rotated object detection},
 pages = {353-365},
 volume = {36},
 month = {10},
 publisher = {Elsevier},
 day = {1},
 id = {6eb5e6c0-09cd-3c47-98ae-f865e97f7f5d},
 created = {2024-11-27T13:16:40.172Z},
 accessed = {2024-11-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-13T08:38:16.502Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Object detection in Remote Sensing (RS) has achieved tremendous advances in recent years, but it remains challenging for rotated object detection due to cluttered backgrounds, dense object arrangements and the wide range of size variations among objects. To tackle this problem, Dense Context Feature Pyramid Network (DCFPN) and a power α-Gaussian loss are designed for rotated object detection in this paper. The proposed DCFPN can extract multi-scale information densely and accurately by leveraging a dense multi-path dilation layer to cover all sizes of objects in remote sensing scenarios. For more accurate detection while avoiding bottlenecks such as boundary discontinuity in rotated bounding box regression, α-Gaussian loss, a unified power generalization of existing Gaussian modeling losses is proposed. Furthermore, the properties of α-Gaussian loss are analyzed comprehensively for a wider range of applications. Experimental results on four datasets (UCAS-AOD, HRSC2016, DIOR-R, and DOTA) show the effectiveness of the proposed method using different detectors, and are superior to the existing methods in both feature extraction and bounding box regression.},
 bibtype = {article},
 author = {LI, Yang and WANG, Haining and FANG, Yuqiang and WANG, Shengjin and LI, Zhi and JIANG, Bitao},
 doi = {10.1016/J.CJA.2023.04.022},
 journal = {Chinese Journal of Aeronautics},
 number = {10}
}

Object detection in Remote Sensing (RS) has achieved tremendous advances in recent years, but it remains challenging for rotated object detection due to cluttered backgrounds, dense object arrangements and the wide range of size variations among objects. To tackle this problem, Dense Context Feature Pyramid Network (DCFPN) and a power α-Gaussian loss are designed for rotated object detection in this paper. The proposed DCFPN can extract multi-scale information densely and accurately by leveraging a dense multi-path dilation layer to cover all sizes of objects in remote sensing scenarios. For more accurate detection while avoiding bottlenecks such as boundary discontinuity in rotated bounding box regression, α-Gaussian loss, a unified power generalization of existing Gaussian modeling losses is proposed. Furthermore, the properties of α-Gaussian loss are analyzed comprehensively for a wider range of applications. Experimental results on four datasets (UCAS-AOD, HRSC2016, DIOR-R, and DOTA) show the effectiveness of the proposed method using different detectors, and are superior to the existing methods in both feature extraction and bounding box regression.

@article{
 title = {Evaluating the Economic and Sustainability Impacts of Drones in Viticulture using BPMN-based Simulation},
 type = {article},
 year = {2023},
 keywords = {Business process management,digitalisation,plant protection,simulation,viticulture},
 pages = {892-901},
 volume = {225},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {c23d0251-7433-3afa-9fcc-16269d4785c9},
 created = {2024-12-06T10:29:24.434Z},
 accessed = {2024-12-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-06T10:29:27.091Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e},
 private_publication = {false},
 abstract = {This paper presents an investigation into the economic impact of drones in viticulture, an area that has not been previously researched. The authors calculate the economic impacts of drones in viticulture and use this to measure the overall sustainability impact assessment for this technology. The study aims to explore a method for investigating the impact of technological changes on business processes in viticulture. This involves selecting viticultural business processes, representing them using Business Process Model and Notation (BPMN), and considering their simulation. Backpack sprayers and trailed sprayers were considered conventional application methods. The application of crop protection products by drone was considered a digitalized variant. Literature research and guideline-based expert interviews with vinegrowers provided the information basis. The study focuses on plant protection in viticulture, but the results can be applied to other agricultural processes. By integrating all three indicators for sustainability, the study provides an evidence-based method for evaluating the impact of drones on viticultural business processes.},
 bibtype = {article},
 author = {Schieck, Martin and Roemer, Ingolf and Oertel, Anika and Franczyk, Bogdan},
 doi = {10.1016/J.PROCS.2023.10.076},
 journal = {Procedia Computer Science}
}

This paper presents an investigation into the economic impact of drones in viticulture, an area that has not been previously researched. The authors calculate the economic impacts of drones in viticulture and use this to measure the overall sustainability impact assessment for this technology. The study aims to explore a method for investigating the impact of technological changes on business processes in viticulture. This involves selecting viticultural business processes, representing them using Business Process Model and Notation (BPMN), and considering their simulation. Backpack sprayers and trailed sprayers were considered conventional application methods. The application of crop protection products by drone was considered a digitalized variant. Literature research and guideline-based expert interviews with vinegrowers provided the information basis. The study focuses on plant protection in viticulture, but the results can be applied to other agricultural processes. By integrating all three indicators for sustainability, the study provides an evidence-based method for evaluating the impact of drones on viticultural business processes.

@article{
 title = {Cost Analysis of Using UAV Sprayers for Olive Fruit Fly Control},
 type = {article},
 year = {2023},
 keywords = {UAV sprayers,economic feasibility,olive fruit fly},
 pages = {1925-1942},
 volume = {5},
 month = {12},
 publisher = {Multidisciplinary Digital Publishing Institute (MDPI)},
 day = {1},
 id = {6d74b4eb-7992-3c9d-b7b5-be10f91b43a0},
 created = {2024-12-06T10:30:17.395Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-06T10:30:18.316Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e},
 private_publication = {false},
 abstract = {Unmanned Aerial Vehicles (UAVs) are a novel up-and-coming technology with wide applicability and great potential to be used in agricultural systems for spraying applications. However, the cost-effectiveness of this application is still rather uncertain. The present study utilized actual data from field applications to analyze the critical components and parameters in the potential case of using UAV sprayers for the control of olive fruit flies in order to assess the operational costs. The results are compared with the costs of two traditional spraying methods: manual spaying by workers using backpack sprayers and manual spraying assisted by a tractor. The case of the olive fruit fly was selected because it involves costly, time consuming, and laborious manual spraying. Furthermore, the bait character of spraying in these applications does not require full canopy coverage, making it ideal for UAV applications. A parameterized computational model was developed to assess the costs of labor, capital spending, repair and maintenance, energy, licensees, fees and taxes, and storage for each of the three methods. In addition, the cost for surveillance was also accounted for with the UAV method. Consequently, a sensitivity analysis was performed to examine the impact of the most crucial parameters. The results showed that the cost of spraying with a UAV was 1.45 to 2 times higher than the traditional methods, mainly due to the high capital spending resulting from a low economic life. There are opportunities, however, of improving the economic performance, making it compatible to the traditional methods, by using a smaller UAV with longer lasting batteries and by expanding its annual use beyond the needs of olive fruit fly control.},
 bibtype = {article},
 author = {Cavalaris, Chris and Tagarakis, Aristotelis C. and Kateris, Dimitrios and Bochtis, Dionysis},
 doi = {10.3390/agriengineering5040118},
 journal = {AgriEngineering},
 number = {4}
}

Unmanned Aerial Vehicles (UAVs) are a novel up-and-coming technology with wide applicability and great potential to be used in agricultural systems for spraying applications. However, the cost-effectiveness of this application is still rather uncertain. The present study utilized actual data from field applications to analyze the critical components and parameters in the potential case of using UAV sprayers for the control of olive fruit flies in order to assess the operational costs. The results are compared with the costs of two traditional spraying methods: manual spaying by workers using backpack sprayers and manual spraying assisted by a tractor. The case of the olive fruit fly was selected because it involves costly, time consuming, and laborious manual spraying. Furthermore, the bait character of spraying in these applications does not require full canopy coverage, making it ideal for UAV applications. A parameterized computational model was developed to assess the costs of labor, capital spending, repair and maintenance, energy, licensees, fees and taxes, and storage for each of the three methods. In addition, the cost for surveillance was also accounted for with the UAV method. Consequently, a sensitivity analysis was performed to examine the impact of the most crucial parameters. The results showed that the cost of spraying with a UAV was 1.45 to 2 times higher than the traditional methods, mainly due to the high capital spending resulting from a low economic life. There are opportunities, however, of improving the economic performance, making it compatible to the traditional methods, by using a smaller UAV with longer lasting batteries and by expanding its annual use beyond the needs of olive fruit fly control.

@article{
 title = {An annotated image dataset of vegetable crops at an early stage of growth for proximal sensing applications},
 type = {article},
 year = {2022},
 keywords = {Deep learning,Neural network,Object detection,Precision agriculture,Tracking algorithm},
 pages = {108035},
 volume = {42},
 month = {1},
 publisher = {Elsevier B.V.},
 day = {1},
 id = {d8dcbddd-1461-33fd-984c-127f3836acc2},
 created = {2022-07-15T12:46:39.056Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-11-21T11:38:10.013Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {This article introduces a dataset of 2 801 images of vegetable crops. Maize ( Zea mays ), bean ( Phaseolus vulgaris ) and leek ( Allium ampeloprasum ) crops at an early stage of develop- ment (between 2 and 5 weeks from seeding of transplanting) are supported. Two kinds of annotations are provided: (i) bounding boxes enclosing the crops of interest or their stems, weeds being left apart, and (ii) crop structures in the form of star graphs whose vertices are the plant organs (stems and leaves) and whose edges represent the connections be- tween them. The images have been captured in various pro- duction and experimentation plots in France using an acqui- sition module which controls light conditions. They present a wide variety of soil conditions, weed infestation and growth stages. This dataset can benefit precision hoeing and in-field crop monitoring applications that are based on proximal im- agery.},
 bibtype = {article},
 author = {Lac, Louis and Keresztes, Barna and Louargant, Marine and Donias, Marc and Da Costa, Jean-Pierre},
 doi = {10.1016/j.compag.2021.106606},
 journal = {Data in Brief},
 number = {2352-3409}
}

This article introduces a dataset of 2 801 images of vegetable crops. Maize ( Zea mays ), bean ( Phaseolus vulgaris ) and leek ( Allium ampeloprasum ) crops at an early stage of develop- ment (between 2 and 5 weeks from seeding of transplanting) are supported. Two kinds of annotations are provided: (i) bounding boxes enclosing the crops of interest or their stems, weeds being left apart, and (ii) crop structures in the form of star graphs whose vertices are the plant organs (stems and leaves) and whose edges represent the connections be- tween them. The images have been captured in various pro- duction and experimentation plots in France using an acqui- sition module which controls light conditions. They present a wide variety of soil conditions, weed infestation and growth stages. This dataset can benefit precision hoeing and in-field crop monitoring applications that are based on proximal im- agery.

@article{
 title = {Spectral Comparison of UAV-Based Hyper and Multispectral Cameras for Precision Viticulture},
 type = {article},
 year = {2022},
 keywords = {Imaging sensor,Precision agriculture,Radiometric calibration,Remote sensing,Spectral signature,Vegetation indices},
 volume = {14},
 id = {24f8d72b-8dd2-3fc5-a2bd-c72ac953621f},
 created = {2022-08-14T07:28:08.590Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-08-14T07:28:13.039Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {Analysis of the spectral response of vegetation using optical sensors for non-destructive remote monitoring represents a key element for crop monitoring. Considering the wide presence on the market of unmanned aerial vehicle (UAVs) based commercial solutions, the need emerges for clear information on the performance of these products to guide the end-user in their choice and utilization for precision agriculture applications. This work aims to compare two UAV based commercial products, represented by DJI P4M and SENOP HSC-2 for the acquisition of multispectral and hyperspectral images, respectively, in vineyards. The accuracy of both cameras was evaluated on 6 different targets commonly found in vineyards, represented by bare soil, bare-stony soil, stony soil, soil with dry grass, partially grass covered soil and canopy. Given the importance of the radiometric calibration, four methods for multispectral images correction were evaluated, taking in account the irradiance sensor equipped on the camera (M1–M2) and the use of an empirical line model (ELM) based on reference reflectance panels (M3–M4). In addition, different DJI P4M exposure setups were evaluated. The performance of the cameras was evaluated by means of the calculation of three widely used vegetation indices (VIs), as percentage error (PE) with respect to ground truth spectroradiometer measurements. The results highlighted the importance of reference panels for the radiometric calibration of multispectral images (M1–M2 average PE = 21.8–100.0%; M3–M4 average PE = 11.9–29.5%). Generally, the hyperspectral camera provided the best accuracy with a PE ranging between 1.0% and 13.6%. Both cameras showed higher performance on the pure canopy pixel target, compared to mixed targets. However, this issue can be easily solved by applying widespread segmentation techniques for the row extraction. This work provides insights to assist end-users in the UAV spectral monitoring to obtain reliable information for the analysis of spatio-temporal variability within vineyards.},
 bibtype = {article},
 author = {Di Gennaro, Salvatore Filippo and Toscano, Piero and Gatti, Matteo and Poni, Stefano and Berton, Andrea and Matese, Alessandro},
 doi = {10.3390/rs14030449},
 journal = {Remote Sensing},
 number = {3}
}

Analysis of the spectral response of vegetation using optical sensors for non-destructive remote monitoring represents a key element for crop monitoring. Considering the wide presence on the market of unmanned aerial vehicle (UAVs) based commercial solutions, the need emerges for clear information on the performance of these products to guide the end-user in their choice and utilization for precision agriculture applications. This work aims to compare two UAV based commercial products, represented by DJI P4M and SENOP HSC-2 for the acquisition of multispectral and hyperspectral images, respectively, in vineyards. The accuracy of both cameras was evaluated on 6 different targets commonly found in vineyards, represented by bare soil, bare-stony soil, stony soil, soil with dry grass, partially grass covered soil and canopy. Given the importance of the radiometric calibration, four methods for multispectral images correction were evaluated, taking in account the irradiance sensor equipped on the camera (M1–M2) and the use of an empirical line model (ELM) based on reference reflectance panels (M3–M4). In addition, different DJI P4M exposure setups were evaluated. The performance of the cameras was evaluated by means of the calculation of three widely used vegetation indices (VIs), as percentage error (PE) with respect to ground truth spectroradiometer measurements. The results highlighted the importance of reference panels for the radiometric calibration of multispectral images (M1–M2 average PE = 21.8–100.0%; M3–M4 average PE = 11.9–29.5%). Generally, the hyperspectral camera provided the best accuracy with a PE ranging between 1.0% and 13.6%. Both cameras showed higher performance on the pure canopy pixel target, compared to mixed targets. However, this issue can be easily solved by applying widespread segmentation techniques for the row extraction. This work provides insights to assist end-users in the UAV spectral monitoring to obtain reliable information for the analysis of spatio-temporal variability within vineyards.

@article{
 title = {A Review of Multi-Sensor Fusion SLAM Systems Based on 3D LIDAR},
 type = {article},
 year = {2022},
 keywords = {LIDAR,SLAM,coupling methods,multi,sensor fusion},
 pages = {2835},
 volume = {14},
 websites = {https://www.mdpi.com/2072-4292/14/12/2835/htm,https://www.mdpi.com/2072-4292/14/12/2835},
 month = {6},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {13},
 id = {3e229525-f74a-3926-ba26-bdd1c9612cd7},
 created = {2022-08-17T15:00:18.543Z},
 accessed = {2022-08-17},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-12T12:11:15.356Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {The ability of intelligent unmanned platforms to achieve autonomous navigation and positioning in a large-scale environment has become increasingly demanding, in which LIDAR-based Simultaneous Localization and Mapping (SLAM) is the mainstream of research schemes. However, the LIDAR-based SLAM system will degenerate and affect the localization and mapping effects in extreme environments with high dynamics or sparse features. In recent years, a large number of LIDAR-based multi-sensor fusion SLAM works have emerged in order to obtain a more stable and robust system. In this work, the development process of LIDAR-based multi-sensor fusion SLAM and the latest research work are highlighted. After summarizing the basic idea of SLAM and the necessity of multi-sensor fusion, this paper introduces the basic principles and recent work of multi-sensor fusion in detail from four aspects based on the types of fused sensors and data coupling methods. Meanwhile, we review some SLAM datasets and compare the performance of five open-source algorithms using the UrbanNav dataset. Finally, the development trend and popular research directions of SLAM based on 3D LIDAR multi-sensor fusion are discussed and summarized.},
 bibtype = {article},
 author = {Jiang, Changhui and Meng, Qian and Xu, Bing and Gao, Wang and Wu, Panlong and Guan, Lianwu and Li, Zeyu and Xu, Xiaobin and Zhang, Lei and Yang, Jian and Cao, Chenfei and Wang, Wen and Ran, Yingying and Tan, Zhiying and Luo, Minzhou},
 doi = {10.3390/RS14122835},
 journal = {Remote Sensing 2022, Vol. 14, Page 2835},
 number = {12}
}

The ability of intelligent unmanned platforms to achieve autonomous navigation and positioning in a large-scale environment has become increasingly demanding, in which LIDAR-based Simultaneous Localization and Mapping (SLAM) is the mainstream of research schemes. However, the LIDAR-based SLAM system will degenerate and affect the localization and mapping effects in extreme environments with high dynamics or sparse features. In recent years, a large number of LIDAR-based multi-sensor fusion SLAM works have emerged in order to obtain a more stable and robust system. In this work, the development process of LIDAR-based multi-sensor fusion SLAM and the latest research work are highlighted. After summarizing the basic idea of SLAM and the necessity of multi-sensor fusion, this paper introduces the basic principles and recent work of multi-sensor fusion in detail from four aspects based on the types of fused sensors and data coupling methods. Meanwhile, we review some SLAM datasets and compare the performance of five open-source algorithms using the UrbanNav dataset. Finally, the development trend and popular research directions of SLAM based on 3D LIDAR multi-sensor fusion are discussed and summarized.

@article{
 title = {A Comprehensive Survey of Visual SLAM Algorithms},
 type = {article},
 year = {2022},
 keywords = {3D reconstruction,D SLAM,RGB,SLAM,embedded SLAM,evaluation criteria,inertial SLAM,visual},
 pages = {24},
 volume = {11},
 websites = {https://www.mdpi.com/2218-6581/11/1/24/htm,https://www.mdpi.com/2218-6581/11/1/24},
 month = {2},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {10},
 id = {14c29e02-3d1d-30c8-adaf-cf185bd207b5},
 created = {2022-08-17T15:08:11.482Z},
 accessed = {2022-08-17},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-20T10:22:03.061Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {Simultaneous localization and mapping (SLAM) techniques are widely researched, since they allow the simultaneous creation of a map and the sensors’ pose estimation in an unknown environment. Visual-based SLAM techniques play a significant role in this field, as they are based on a low-cost and small sensor system, which guarantees those advantages compared to other sensor-based SLAM techniques. The literature presents different approaches and methods to implement visual-based SLAM systems. Among this variety of publications, a beginner in this domain may find problems with identifying and analyzing the main algorithms and selecting the most appropriate one according to his or her project constraints. Therefore, we present the three main visual-based SLAM approaches (visual-only, visual-inertial, and RGB-D SLAM), providing a review of the main algorithms of each approach through diagrams and flowcharts, and highlighting the main advantages and disadvantages of each technique. Furthermore, we propose six criteria that ease the SLAM algorithm’s analysis and consider both the software and hardware levels. In addition, we present some major issues and future directions on visual-SLAM field, and provide a general overview of some of the existing benchmark datasets. This work aims to be the first step for those initiating a SLAM project to have a good perspective of SLAM techniques’ main elements and characteristics.},
 bibtype = {article},
 author = {Macario Barros, Andréa and Michel, Maugan and Moline, Yoann and Corre, Gwenolé and Carrel, Frédérick},
 doi = {10.3390/ROBOTICS11010024},
 journal = {Robotics 2022, Vol. 11, Page 24},
 number = {1}
}

Simultaneous localization and mapping (SLAM) techniques are widely researched, since they allow the simultaneous creation of a map and the sensors’ pose estimation in an unknown environment. Visual-based SLAM techniques play a significant role in this field, as they are based on a low-cost and small sensor system, which guarantees those advantages compared to other sensor-based SLAM techniques. The literature presents different approaches and methods to implement visual-based SLAM systems. Among this variety of publications, a beginner in this domain may find problems with identifying and analyzing the main algorithms and selecting the most appropriate one according to his or her project constraints. Therefore, we present the three main visual-based SLAM approaches (visual-only, visual-inertial, and RGB-D SLAM), providing a review of the main algorithms of each approach through diagrams and flowcharts, and highlighting the main advantages and disadvantages of each technique. Furthermore, we propose six criteria that ease the SLAM algorithm’s analysis and consider both the software and hardware levels. In addition, we present some major issues and future directions on visual-SLAM field, and provide a general overview of some of the existing benchmark datasets. This work aims to be the first step for those initiating a SLAM project to have a good perspective of SLAM techniques’ main elements and characteristics.

@article{
 title = {Simultaneous Localization and Mapping (SLAM) and Data Fusion in Unmanned Aerial Vehicles: Recent Advances and Challenges},
 type = {article},
 year = {2022},
 keywords = {multimodal sensor fusion,sensor data fusion,simultaneous localization and mapping,unmanned aerial vehicles},
 pages = {85},
 volume = {6},
 websites = {https://www.mdpi.com/2504-446X/6/4/85/htm,https://www.mdpi.com/2504-446X/6/4/85},
 month = {3},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {28},
 id = {3554808f-b2b8-3686-899d-45afa8007075},
 created = {2022-08-17T15:08:47.730Z},
 accessed = {2022-08-17},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-07T08:30:20.371Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {This article presents a survey of simultaneous localization and mapping (SLAM) and data fusion techniques for object detection and environmental scene perception in unmanned aerial vehicles (UAVs). We critically evaluate some current SLAM implementations in robotics and autonomous vehicles and their applicability and scalability to UAVs. SLAM is envisioned as a potential technique for object detection and scene perception to enable UAV navigation through continuous state estimation. In this article, we bridge the gap between SLAM and data fusion in UAVs while also comprehensively surveying related object detection techniques such as visual odometry and aerial photogrammetry. We begin with an introduction to applications where UAV localization is necessary, followed by an analysis of multimodal sensor data fusion to fuse the information gathered from different sensors mounted on UAVs. We then discuss SLAM techniques such as Kalman filters and extended Kalman filters to address scene perception, mapping, and localization in UAVs. The findings are summarized to correlate prevalent and futuristic SLAM and data fusion for UAV navigation, and some avenues for further research are discussed.},
 bibtype = {article},
 author = {Gupta, Abhishek and Fernando, Xavier},
 doi = {10.3390/DRONES6040085},
 journal = {Drones 2022, Vol. 6, Page 85},
 number = {4}
}

This article presents a survey of simultaneous localization and mapping (SLAM) and data fusion techniques for object detection and environmental scene perception in unmanned aerial vehicles (UAVs). We critically evaluate some current SLAM implementations in robotics and autonomous vehicles and their applicability and scalability to UAVs. SLAM is envisioned as a potential technique for object detection and scene perception to enable UAV navigation through continuous state estimation. In this article, we bridge the gap between SLAM and data fusion in UAVs while also comprehensively surveying related object detection techniques such as visual odometry and aerial photogrammetry. We begin with an introduction to applications where UAV localization is necessary, followed by an analysis of multimodal sensor data fusion to fuse the information gathered from different sensors mounted on UAVs. We then discuss SLAM techniques such as Kalman filters and extended Kalman filters to address scene perception, mapping, and localization in UAVs. The findings are summarized to correlate prevalent and futuristic SLAM and data fusion for UAV navigation, and some avenues for further research are discussed.

@article{
 title = {Visual SLAM Algorithms and their Application for AR, Mapping, Localization and Wayfinding},
 type = {article},
 year = {2022},
 keywords = {AR,simultaneous localization and mapping (SLAM),visual SLAM},
 pages = {100222},
 volume = {15},
 id = {bb7176ae-dddf-3563-b054-6ea58ed8c59b},
 created = {2022-08-19T09:11:13.027Z},
 accessed = {2022-08-19},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-06T13:33:32.078Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {Visual simultaneous localization and mapping (vSLAM) algorithms use device camera to estimate agent's position and reconstruct structures in an unknown environment. As an essential part of augmented reality (AR) experience, vS-LAM enhances the real-world environment through the addition of virtual objects , based on localization (location) and environment structure (mapping). From both technical and historical perspectives, this paper categorizes and summarizes some of the most recent visual SLAM algorithms proposed in research communities, while also discussing their applications in augmented reality, mapping , navigation, and localization.},
 bibtype = {article},
 author = {Theodorou, Charalambos and Velisavljevic, Vladan and Dyo, Vladimir and Nonyelu, Fredi},
 journal = {Array}
}

Visual simultaneous localization and mapping (vSLAM) algorithms use device camera to estimate agent's position and reconstruct structures in an unknown environment. As an essential part of augmented reality (AR) experience, vS-LAM enhances the real-world environment through the addition of virtual objects , based on localization (location) and environment structure (mapping). From both technical and historical perspectives, this paper categorizes and summarizes some of the most recent visual SLAM algorithms proposed in research communities, while also discussing their applications in augmented reality, mapping , navigation, and localization.

@article{
 title = {LCDNet: Deep Loop Closure Detection and Point Cloud Registration for LiDAR SLAM},
 type = {article},
 year = {2022},
 keywords = {Deep learning,Feature extraction,Laser radar,Point cloud compression,Protocols,Simultaneous localization and mapping,Standards,Task analysis,loop closure detection,place recognition,point cloud registration,simultaneous localization and mapping (SLAM)},
 pages = {2074-2093},
 volume = {38},
 id = {369f8d75-a41e-30e0-8079-38331480af8e},
 created = {2022-09-08T05:59:12.519Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-09T05:37:39.851Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {Loop closure detection is an essential component of simultaneous localization and mapping (SLAM) systems, which reduces the drift accumulated over time. Over the years, several deep learning approaches have been proposed to address this task; however, their performance has been subpar compared to handcrafted techniques, especially while dealing with reverse loops. In this article, we introduce the novel loop closure detection network (LCDNet) that effectively detects loop closures in light detection and ranging (LiDAR) point clouds by simultaneously identifying previously visited places and estimating the six degrees of freedom relative transformation between the current scan and the map. LCDNet is composed of a shared encoder, a place recognition head that extracts global descriptors, and a relative pose head that estimates the transformation between two point clouds. We introduce a novel relative pose head based on the unbalanced optimal transport theory that we implement in a differentiable manner to allow for end-to-end training. Extensive evaluations of LCDNet on multiple real-world autonomous driving datasets show that our approach outperforms state-of-the-art loop closure detection and point cloud registration techniques by a large margin, especially while dealing with reverse loops. Moreover, we integrate our proposed loop closure detection approach into a LiDAR SLAM library to provide a complete mapping system and demonstrate the generalization ability using different sensor setup in an unseen city.},
 bibtype = {article},
 author = {Cattaneo, Daniele and Vaghi, Matteo and Valada, Abhinav},
 doi = {10.1109/TRO.2022.3150683},
 journal = {IEEE Transactions on Robotics},
 number = {4}
}

Loop closure detection is an essential component of simultaneous localization and mapping (SLAM) systems, which reduces the drift accumulated over time. Over the years, several deep learning approaches have been proposed to address this task; however, their performance has been subpar compared to handcrafted techniques, especially while dealing with reverse loops. In this article, we introduce the novel loop closure detection network (LCDNet) that effectively detects loop closures in light detection and ranging (LiDAR) point clouds by simultaneously identifying previously visited places and estimating the six degrees of freedom relative transformation between the current scan and the map. LCDNet is composed of a shared encoder, a place recognition head that extracts global descriptors, and a relative pose head that estimates the transformation between two point clouds. We introduce a novel relative pose head based on the unbalanced optimal transport theory that we implement in a differentiable manner to allow for end-to-end training. Extensive evaluations of LCDNet on multiple real-world autonomous driving datasets show that our approach outperforms state-of-the-art loop closure detection and point cloud registration techniques by a large margin, especially while dealing with reverse loops. Moreover, we integrate our proposed loop closure detection approach into a LiDAR SLAM library to provide a complete mapping system and demonstrate the generalization ability using different sensor setup in an unseen city.

@article{
 title = {Depth camera based row-end detection and headland manuvering in orchard navigation without GNSS},
 type = {article},
 year = {2022},
 pages = {538-544},
 publisher = {IEEE},
 id = {3bd4953f-65b0-3702-a708-a15c16751103},
 created = {2022-09-10T14:11:45.509Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:00.555Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4c7c81ce-f24b-44ae-bc2a-bf60600a3a24,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {A robust navigation system is a prerequisite for a mobile robotic platform to carry out precision agriculture tasks in a modern orchard. In contrast to open fields, navigation based solely on the Global Navigation Satellite System (GNSS) is not stable in many orchards, where tree canopies may block the GNSS signal or introduce multipath. Many works have been done to localize the robot while traversing the inside a row, but navigating the robot to the next row on headland still relies on a reference map or artificial landmarks. In this work, we developed a row end detection method by exploiting drastic changes in the statistical distribution of points sensed by a depth camera compared to the points inside the row. Also, a robust way of row entry method is implemented by building a local environment map and reactive path tracker. The whole navigation system is tested and evaluated on a mobile robot in a vineyard. The experiment results show the robot can detect the tree row-end accurately and maneuver a U-Turn to the next row safely.},
 bibtype = {article},
 author = {Peng, Chen and Fei, Zhenghao and Vougioukas, Stavros G.},
 doi = {10.1109/MED54222.2022.9837270},
 journal = {2022 30th Mediterranean Conference on Control and Automation, MED 2022},
 keywords = {peng2022depthcamerabased}
}

A robust navigation system is a prerequisite for a mobile robotic platform to carry out precision agriculture tasks in a modern orchard. In contrast to open fields, navigation based solely on the Global Navigation Satellite System (GNSS) is not stable in many orchards, where tree canopies may block the GNSS signal or introduce multipath. Many works have been done to localize the robot while traversing the inside a row, but navigating the robot to the next row on headland still relies on a reference map or artificial landmarks. In this work, we developed a row end detection method by exploiting drastic changes in the statistical distribution of points sensed by a depth camera compared to the points inside the row. Also, a robust way of row entry method is implemented by building a local environment map and reactive path tracker. The whole navigation system is tested and evaluated on a mobile robot in a vineyard. The experiment results show the robot can detect the tree row-end accurately and maneuver a U-Turn to the next row safely.

@article{
 title = {Crop stem detection and tracking for precision hoeing using deep learning},
 type = {article},
 year = {2022},
 keywords = {Deep learning,Neural network,Object detection,Precision agriculture,Tracking algorithm},
 pages = {106606},
 volume = {192},
 websites = {https://doi.org/10.1016/j.compag.2021.106606},
 publisher = {Elsevier B.V.},
 id = {e211e23b-438e-3a1b-bae2-081ce65ad7db},
 created = {2022-09-12T10:25:30.018Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-11-18T12:35:41.721Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Developing alternatives to the chemical weeding process usually carried out in vegetable crop farming is necessary in order to reach a more sustainable agriculture. However, a precise mechanical weeding requires specific sensors and advanced computer vision algorithms to process crop and weed discrimination in real-time. In this paper we propose an algorithm able to detect, locate, and track the stem position of crops in images which is suitable for precision actions in vegetable fields such as mechanical hoeing within crop rows. The algorithm is twofold: (i) a deep neural network for object detection is first used to detect crop stems in individual RGB images and then (ii) an aggregation algorithm further refines the detections taking advantage of the temporal redundancy in consecutive frames. We evaluated the pipeline on images of maize and bean crops at an early stage of development, acquired in field conditions with a camera embedded in an experimental mechanical weeding system. We reported F1-scores of respectively 94.74% and 93.82% with a location accuracy around 0.7 cm when compared with human annotation. Moreover, this pipeline can operate in real-time on an embedded computer consuming as little power as 30 W.},
 bibtype = {article},
 author = {Lac, Louis and Da Costa, Jean Pierre and Donias, Marc and Keresztes, Barna and Bardet, Alain},
 doi = {10.1016/j.compag.2021.106606},
 journal = {Computers and Electronics in Agriculture},
 number = {December 2021}
}

Developing alternatives to the chemical weeding process usually carried out in vegetable crop farming is necessary in order to reach a more sustainable agriculture. However, a precise mechanical weeding requires specific sensors and advanced computer vision algorithms to process crop and weed discrimination in real-time. In this paper we propose an algorithm able to detect, locate, and track the stem position of crops in images which is suitable for precision actions in vegetable fields such as mechanical hoeing within crop rows. The algorithm is twofold: (i) a deep neural network for object detection is first used to detect crop stems in individual RGB images and then (ii) an aggregation algorithm further refines the detections taking advantage of the temporal redundancy in consecutive frames. We evaluated the pipeline on images of maize and bean crops at an early stage of development, acquired in field conditions with a camera embedded in an experimental mechanical weeding system. We reported F1-scores of respectively 94.74% and 93.82% with a location accuracy around 0.7 cm when compared with human annotation. Moreover, this pipeline can operate in real-time on an embedded computer consuming as little power as 30 W.

@inproceedings{
 title = {Pseudo-label Generation for Agricultural Robotics Applications},
 type = {inproceedings},
 year = {2022},
 pages = {1685-1693},
 id = {ffae0aa0-9cd7-3097-9976-720bce52843c},
 created = {2022-09-13T10:27:30.657Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T11:05:42.262Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Ciarfuglia, Thomas A. and Marian Motoi, Ionut and Saraceni, Leonardo and Nardi, Daniele},
 doi = {10.1109/cvprw56347.2022.00175},
 booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPR Workshops 2022, New Orleans, LA, USA, June 19-20, 2022},
 keywords = {ciarfuglia2022pseudolabelgeneration}
}

@article{
 title = {Integrating UAVs and Canopy Height Models in Vineyard Management: A Time-Space Approach},
 type = {article},
 year = {2022},
 keywords = {sassu2022integratinguavscanopy},
 pages = {1-14},
 volume = {14},
 id = {84eb32d4-9a28-3a70-819c-492fccca52cb},
 created = {2022-09-14T06:35:13.081Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-01T07:12:57.782Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The present study illustrates an operational approach estimating individual and aggregate vineyards’ canopy volume estimation through three years Tree-Row-Volume (TRV) measurements and remotely sensed imagery acquired with unmanned aerial vehicle (UAV) Red-Green-Blue (RGB) digital camera, processed with MATLAB scripts, and validated through ArcGIS tools. The TRV methodology was applied by sampling a different number of rows and plants (per row) each year with the aim of evaluating reliability and accuracy of this technique compared with a remote approach. The empirical results indicate that the estimated tree-row-volumes derived from a UAV Canopy Height Model (CHM) are up to 50% different from those measured on the field using the routinary technique of TRV in 2019. The difference is even much higher in the two 2016 dates. These empirical findings outline the importance of data integration among techniques that mix proximal and remote sensing in routine vineyards’ agronomic practices, helping to reduce management costs and increase the environmental sustainability of traditional cultivation systems.},
 bibtype = {article},
 author = {Sassu, Alberto and Ghiani, Luca and Salvati, Luca and Mercenaro, Luca and Deidda, Alessandro and Gambella, Filippo},
 doi = {10.3390/rs14010130},
 journal = {Remote Sensing},
 number = {1}
}

The present study illustrates an operational approach estimating individual and aggregate vineyards’ canopy volume estimation through three years Tree-Row-Volume (TRV) measurements and remotely sensed imagery acquired with unmanned aerial vehicle (UAV) Red-Green-Blue (RGB) digital camera, processed with MATLAB scripts, and validated through ArcGIS tools. The TRV methodology was applied by sampling a different number of rows and plants (per row) each year with the aim of evaluating reliability and accuracy of this technique compared with a remote approach. The empirical results indicate that the estimated tree-row-volumes derived from a UAV Canopy Height Model (CHM) are up to 50% different from those measured on the field using the routinary technique of TRV in 2019. The difference is even much higher in the two 2016 dates. These empirical findings outline the importance of data integration among techniques that mix proximal and remote sensing in routine vineyards’ agronomic practices, helping to reduce management costs and increase the environmental sustainability of traditional cultivation systems.

@article{
 title = {Semi-supervised deep learning and low-cost cameras for the semantic segmentation of natural images in viticulture},
 type = {article},
 year = {2022},
 keywords = {casado2022semisuperviseddeep},
 websites = {https://doi.org/10.1007/s11119-022-09929-9},
 publisher = {Springer US},
 id = {a2c92f19-f0c6-3b18-87e5-4571c72fd3ab},
 created = {2022-09-20T06:15:53.751Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T09:21:05.966Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Automatic yield monitoring and in-field robotic harvesting by low-cost cameras require object detection and segmentation solutions to tackle the poor quality of natural images and the lack of exactly-labeled datasets of consistent sizes. This work proposed the application of deep learning for semantic segmentation of natural images acquired by a low-cost RGB-D camera in a commercial vineyard. Several deep architectures were trained and compared on 85 labeled images. Three semi-supervised learning methods (PseudoLabeling, Distillation and Model Distillation) were proposed to take advantage of 320 non-annotated images. In these experiments, the DeepLabV3+ architecture with a ResNext50 backbone, trained with the set of labeled images, achieved the best overall accuracy of 84.78%. In contrast, the Manet architecture combined with the EfficientnetB3 backbone reached the highest accuracy for the bunch class (85.69%). The application of semi-supervised learning methods boosted the segmentation accuracy between 5.62 and 6.01%, on average. Further discussions are presented to show the effects of a fine-grained manual image annotation on the accuracy of the proposed methods and to compare time requirements.},
 bibtype = {article},
 author = {Casado-García, A. and Heras, J. and Milella, A. and Marani, R.},
 doi = {10.1007/s11119-022-09929-9},
 journal = {Precision Agriculture},
 number = {0123456789}
}

Automatic yield monitoring and in-field robotic harvesting by low-cost cameras require object detection and segmentation solutions to tackle the poor quality of natural images and the lack of exactly-labeled datasets of consistent sizes. This work proposed the application of deep learning for semantic segmentation of natural images acquired by a low-cost RGB-D camera in a commercial vineyard. Several deep architectures were trained and compared on 85 labeled images. Three semi-supervised learning methods (PseudoLabeling, Distillation and Model Distillation) were proposed to take advantage of 320 non-annotated images. In these experiments, the DeepLabV3+ architecture with a ResNext50 backbone, trained with the set of labeled images, achieved the best overall accuracy of 84.78%. In contrast, the Manet architecture combined with the EfficientnetB3 backbone reached the highest accuracy for the bunch class (85.69%). The application of semi-supervised learning methods boosted the segmentation accuracy between 5.62 and 6.01%, on average. Further discussions are presented to show the effects of a fine-grained manual image annotation on the accuracy of the proposed methods and to compare time requirements.

@article{
 title = {Remote sensing-based vineyard image segmentation with deep computer vision for precision agriculture},
 type = {article},
 year = {2022},
 id = {fdf4f9e5-0dd6-3e57-8b87-eeb1f5ae0821},
 created = {2022-09-21T05:45:27.466Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-11-18T12:35:42.204Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {article},
 author = {Sierra PARDO, Cesar Andres and CHIABERGE, Marcello and SALVETTI, Francesco and ANGARANO, Simone},
 number = {July}
}

@article{
 title = {A survey on deep learning-based identification of plant and crop diseases from UAV-based aerial images},
 type = {article},
 year = {2022},
 keywords = {bouguettaya2022surveydeeplearning},
 volume = {0123456789},
 websites = {https://doi.org/10.1007/s10586-022-03627-x},
 publisher = {Springer US},
 id = {85e319de-5fa8-3798-bf4a-f249f3eca0f3},
 created = {2022-09-21T05:45:27.712Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T11:07:27.535Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The agricultural crop productivity can be affected and reduced due to many factors such as weeds, pests, and diseases. Traditional methods that are based on terrestrial engines, devices, and farmers’ naked eyes are facing many limitations in terms of accuracy and the required time to cover large fields. Currently, precision agriculture that is based on the use of deep learning algorithms and Unmanned Aerial Vehicles (UAVs) provides an effective solution to achieve agriculture applications, including plant disease identification and treatment. In the last few years, plant disease monitoring using UAV platforms is one of the most important agriculture applications that have gained increasing interest by researchers. Accurate detection and treatment of plant diseases at early stages is crucial to improving agricultural production. To this end, in this review, we analyze the recent advances in the use of computer vision techniques that are based on deep learning algorithms and UAV technologies to identify and treat crop diseases.},
 bibtype = {article},
 author = {Bouguettaya, Abdelmalek and Zarzour, Hafed and Kechida, Ahmed and Taberkit, Amine Mohammed},
 doi = {10.1007/s10586-022-03627-x},
 journal = {Cluster Computing}
}

The agricultural crop productivity can be affected and reduced due to many factors such as weeds, pests, and diseases. Traditional methods that are based on terrestrial engines, devices, and farmers’ naked eyes are facing many limitations in terms of accuracy and the required time to cover large fields. Currently, precision agriculture that is based on the use of deep learning algorithms and Unmanned Aerial Vehicles (UAVs) provides an effective solution to achieve agriculture applications, including plant disease identification and treatment. In the last few years, plant disease monitoring using UAV platforms is one of the most important agriculture applications that have gained increasing interest by researchers. Accurate detection and treatment of plant diseases at early stages is crucial to improving agricultural production. To this end, in this review, we analyze the recent advances in the use of computer vision techniques that are based on deep learning algorithms and UAV technologies to identify and treat crop diseases.

@article{
 title = {Multispectral vineyard segmentation: A deep learning comparison study},
 type = {article},
 year = {2022},
 keywords = {barros2022multispectralvineyardsegmentation},
 pages = {106782},
 volume = {195},
 websites = {https://doi.org/10.1016/j.compag.2022.106782},
 publisher = {Elsevier B.V.},
 id = {624e5c81-f78e-3f49-a3bc-3d672a6b2f9f},
 created = {2022-09-21T05:45:27.790Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T11:05:43.096Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Digital agriculture has evolved significantly over the last few years due to the technological developments in automation and computational intelligence applied to the agricultural sector, including vineyards which are a relevant crop in the Mediterranean region. In this work, a study is presented of semantic segmentation for vine detection in real-world vineyards by exploring state-of-the-art deep segmentation networks and conventional unsupervised methods. Camera data have been collected on vineyards using an Unmanned Aerial System (UAS) equipped with a dual imaging sensor payload, namely a high-definition RGB camera and a five-band multispectral and thermal camera. Extensive experiments using deep-segmentation networks and unsupervised methods have been performed on multimodal datasets representing four distinct vineyards located in the central region of Portugal. The reported results indicate that SegNet, U-Net, and ModSegNet have equivalent overall performance in vine segmentation. The results also show that multimodality slightly improves the performance of vine segmentation, but the NIR spectrum alone generally is sufficient on most of the datasets. Furthermore, results suggest that high-definition RGB images produce equivalent or higher performance than any lower resolution multispectral band combination. Lastly, Deep Learning (DL) networks have higher overall performance than classical methods. The code and dataset are publicly available on https://github.com/Cybonic/DL_vineyard_segmentation_study.git.},
 bibtype = {article},
 author = {Barros, T. and Conde, P. and Gonçalves, G. and Premebida, C. and Monteiro, M. and Ferreira, C. S.S. and Nunes, U. J.},
 doi = {10.1016/j.compag.2022.106782},
 journal = {Computers and Electronics in Agriculture},
 number = {August 2021}
}

Digital agriculture has evolved significantly over the last few years due to the technological developments in automation and computational intelligence applied to the agricultural sector, including vineyards which are a relevant crop in the Mediterranean region. In this work, a study is presented of semantic segmentation for vine detection in real-world vineyards by exploring state-of-the-art deep segmentation networks and conventional unsupervised methods. Camera data have been collected on vineyards using an Unmanned Aerial System (UAS) equipped with a dual imaging sensor payload, namely a high-definition RGB camera and a five-band multispectral and thermal camera. Extensive experiments using deep-segmentation networks and unsupervised methods have been performed on multimodal datasets representing four distinct vineyards located in the central region of Portugal. The reported results indicate that SegNet, U-Net, and ModSegNet have equivalent overall performance in vine segmentation. The results also show that multimodality slightly improves the performance of vine segmentation, but the NIR spectrum alone generally is sufficient on most of the datasets. Furthermore, results suggest that high-definition RGB images produce equivalent or higher performance than any lower resolution multispectral band combination. Lastly, Deep Learning (DL) networks have higher overall performance than classical methods. The code and dataset are publicly available on https://github.com/Cybonic/DL_vineyard_segmentation_study.git.

@book{
 title = {A Survey of Deep Learning Methods for Fruit and Vegetable Detection and Yield Estimation},
 type = {book},
 year = {2022},
 source = {Data Analytics for Cybersecurity},
 pages = {60-77},
 issue = {September},
 id = {079c2567-9ef3-3c67-aa37-ee4548e6bf83},
 created = {2022-09-21T05:45:28.015Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-11-18T12:35:48.951Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {book},
 author = {Aslam, Faiza and Khan, Zia and Perveen, Kiran},
 doi = {10.1017/9781108231954.005}
}

@article{
 title = {Deep learning and computer vision for assessing the number of actual berries in commercial vineyards},
 type = {article},
 year = {2022},
 keywords = {palacios2022deeplearningcomputer},
 pages = {175-188},
 volume = {218},
 websites = {https://doi.org/10.1016/j.biosystemseng.2022.04.015},
 publisher = {The Author(s)},
 id = {6d89fe53-538d-3100-b90a-2b1735b83d56},
 created = {2022-09-21T05:45:28.132Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T09:08:47.773Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The number of berries is one of the most relevant yield components that drives grape production in viticulture. The goal of this work was to estimate the number of actual berries per grapevine using computer vision and deep learning in commercial vineyards. Images from the visible range (RGB) were acquired from a set of 96 grapevines (Vitis vinifera L.) at pea-size berry stage using a red, green and blue camera (RGB). At harvest, the number of berries and per vine was manually assessed as the ground-truth values. The algorithm involved computer vision to detect berries in the images and to extract canopy features, in order to gain information about canopy occlusion. These were used by the machine learning regression models built to estimate the number of actual berries per vine. A SegNet architecture was used to segment individual berries and several canopy related features. Four datasets were created combining the number of estimated visible berries and different canopy features. Three different regression models were tested on the four datasets. The best results were achieved with support vector regression (SVR) on a dataset including six canopy features. This method yielded a root mean squared error (RMSE) of 205 berries, a normalised root mean squared error (NRMSE) of 24.99% and a coefficient of determination (R2) of 0.83 between the number of estimated and the number of actual berries per vine. The results show that the number of actual berries in grapevines can be assessed with high accuracy up to 60 days prior to grape harvest, using the developed algorithm based on computer vision and deep learning.},
 bibtype = {article},
 author = {Palacios, Fernando and Melo-Pinto, Pedro and Diago, Maria P. and Tardaguila, Javier},
 doi = {10.1016/j.biosystemseng.2022.04.015},
 journal = {Biosystems Engineering}
}

The number of berries is one of the most relevant yield components that drives grape production in viticulture. The goal of this work was to estimate the number of actual berries per grapevine using computer vision and deep learning in commercial vineyards. Images from the visible range (RGB) were acquired from a set of 96 grapevines (Vitis vinifera L.) at pea-size berry stage using a red, green and blue camera (RGB). At harvest, the number of berries and per vine was manually assessed as the ground-truth values. The algorithm involved computer vision to detect berries in the images and to extract canopy features, in order to gain information about canopy occlusion. These were used by the machine learning regression models built to estimate the number of actual berries per vine. A SegNet architecture was used to segment individual berries and several canopy related features. Four datasets were created combining the number of estimated visible berries and different canopy features. Three different regression models were tested on the four datasets. The best results were achieved with support vector regression (SVR) on a dataset including six canopy features. This method yielded a root mean squared error (RMSE) of 205 berries, a normalised root mean squared error (NRMSE) of 24.99% and a coefficient of determination (R2) of 0.83 between the number of estimated and the number of actual berries per vine. The results show that the number of actual berries in grapevines can be assessed with high accuracy up to 60 days prior to grape harvest, using the developed algorithm based on computer vision and deep learning.

@article{
 title = {Fruit yield prediction and estimation in orchards: A state-of-the-art comprehensive review for both direct and indirect methods},
 type = {article},
 year = {2022},
 keywords = {he2022fruityieldprediction},
 pages = {106812},
 volume = {195},
 websites = {https://doi.org/10.1016/j.compag.2022.106812},
 publisher = {Elsevier B.V.},
 id = {071fea94-f2e1-30bd-a42b-b9dbdccebed3},
 created = {2022-09-21T05:45:28.646Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T11:05:42.540Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Orchard pre-harvest yield data is important for fruit growers, which can be used for economic benefit evaluation, management mode adjustment and so on. However, traditional manual operation by sampling estimation is quite an onerous and time-consuming task. The main approach of automatic yield monitoring is by establishing multi-information comprehensive prediction systems or using intelligent equipment. A review is performed to investigate and analyze the past 12 years (from 2010 to 2021) of research work regarding orchard yield prediction and estimation. According to our investigation, the most widely used input features in yield prediction systems are various vegetation indices information of plants, while machine learning is the most modeling method applied. In addition, machine vision systems based on image processing and deep learning have been developed rapidly in the field of agriculture and is also widely applied in orchard yield prediction. Finally, major challenges and countermeasures for orchard yield prediction and estimation are discussed. Complicated natural environments with inconsistent horticultural management practices are still critical challenges for large scale commercialization of researched yield prediction methods. Therefore, combination of engineering technology optimization and standardization of agronomic management practices are necessary to realize automatic operation in complex agricultural fields. The review is intended to summarize the development of yield prediction and estimation technologies and provided suggestions for orchard intelligent management.},
 bibtype = {article},
 author = {He, Leilei and Fang, Wentai and Zhao, Guanao and Wu, Zhenchao and Fu, Longsheng and Li, Rui and Majeed, Yaqoob and Dhupia, Jaspreet},
 doi = {10.1016/j.compag.2022.106812},
 journal = {Computers and Electronics in Agriculture},
 number = {September 2021}
}

Orchard pre-harvest yield data is important for fruit growers, which can be used for economic benefit evaluation, management mode adjustment and so on. However, traditional manual operation by sampling estimation is quite an onerous and time-consuming task. The main approach of automatic yield monitoring is by establishing multi-information comprehensive prediction systems or using intelligent equipment. A review is performed to investigate and analyze the past 12 years (from 2010 to 2021) of research work regarding orchard yield prediction and estimation. According to our investigation, the most widely used input features in yield prediction systems are various vegetation indices information of plants, while machine learning is the most modeling method applied. In addition, machine vision systems based on image processing and deep learning have been developed rapidly in the field of agriculture and is also widely applied in orchard yield prediction. Finally, major challenges and countermeasures for orchard yield prediction and estimation are discussed. Complicated natural environments with inconsistent horticultural management practices are still critical challenges for large scale commercialization of researched yield prediction methods. Therefore, combination of engineering technology optimization and standardization of agronomic management practices are necessary to realize automatic operation in complex agricultural fields. The review is intended to summarize the development of yield prediction and estimation technologies and provided suggestions for orchard intelligent management.

@article{
 title = {A CNN-SVM study based on selected deep features for grapevine leaves classification},
 type = {article},
 year = {2022},
 keywords = {koklu2022cnnscmstudy},
 pages = {110425},
 volume = {188},
 websites = {https://doi.org/10.1016/j.measurement.2021.110425},
 publisher = {Elsevier Ltd},
 id = {655d3003-c85a-3725-944e-987d43839961},
 created = {2022-09-21T05:45:28.663Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-01T07:12:57.831Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The main product of grapevines is grapes that are consumed fresh or processed. In addition, grapevine leaves are harvested once a year as a by-product. The species of grapevine leaves are important in terms of price and taste. In this study, deep learning-based classification is conducted by using images of grapevine leaves. For this purpose, images of 500 vine leaves belonging to 5 species were taken with a special self-illuminating system. Later, this number was increased to 2500 with data augmentation methods. The classification was conducted with a state-of-art CNN model fine-tuned MobileNetv2. As the second approach, features were extracted from pre-trained MobileNetv2′s Logits layer and classification was made using various SVM kernels. As the third approach, 1000 features extracted from MobileNetv2′s Logits layer were selected by the Chi-Squares method and reduced to 250. Then, classification was made with various SVM kernels using the selected features. The most successful method was obtained by extracting features from the Logits layer and reducing the feature with the Chi-Squares method. The most successful SVM kernel was Cubic. The classification success of the system has been determined as 97.60%. It was observed that feature selection increased the classification success although the number of features used in classification decreased.},
 bibtype = {article},
 author = {Koklu, Murat and Unlersen, M. Fahri and Ozkan, Ilker Ali and Aslan, M. Fatih and Sabanci, Kadir},
 doi = {10.1016/j.measurement.2021.110425},
 journal = {Measurement: Journal of the International Measurement Confederation},
 number = {November 2021}
}

The main product of grapevines is grapes that are consumed fresh or processed. In addition, grapevine leaves are harvested once a year as a by-product. The species of grapevine leaves are important in terms of price and taste. In this study, deep learning-based classification is conducted by using images of grapevine leaves. For this purpose, images of 500 vine leaves belonging to 5 species were taken with a special self-illuminating system. Later, this number was increased to 2500 with data augmentation methods. The classification was conducted with a state-of-art CNN model fine-tuned MobileNetv2. As the second approach, features were extracted from pre-trained MobileNetv2′s Logits layer and classification was made using various SVM kernels. As the third approach, 1000 features extracted from MobileNetv2′s Logits layer were selected by the Chi-Squares method and reduced to 250. Then, classification was made with various SVM kernels using the selected features. The most successful method was obtained by extracting features from the Logits layer and reducing the feature with the Chi-Squares method. The most successful SVM kernel was Cubic. The classification success of the system has been determined as 97.60%. It was observed that feature selection increased the classification success although the number of features used in classification decreased.

@article{
 title = {Implementation of drone technology for farm monitoring & pesticide spraying: A review},
 type = {article},
 year = {2022},
 keywords = {Agriculture drone,Crop monitoring,Pesticide spraying,Precision agriculture,Unmanned Aerial Vehicle},
 websites = {https://doi.org/10.1016/j.inpa.2022.02.002},
 publisher = {China Agricultural University},
 id = {c1bae125-ed1a-3511-befc-4cee6ce25cc1},
 created = {2022-09-21T05:45:28.846Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-12-02T11:07:12.672Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The world receives more than 200 thousand people in a day and it is expected that the total world population will reach 9.6 billion by the year 2050. This will result in extra food demand, which can only be met from enhanced crop yield. Therefore, modernization of the agricultural sector becomes the need of the hour. There are many constraints that are responsible for the low production of crops, which can be overcome by using drone technology in the agriculture sector. This paper presents an analysis of drone technologies and their modifications with time in the agriculture sector in the last decade. The application of drones in the area of crop monitoring, and pesticide spraying for Precision Agriculture (PA) has been covered. The work done related to drone structure, multiple sensor development, innovation in spot area spraying has been presented. Moreover, the use of Artificial Intelligent (AI) and deep learning for the remote monitoring of crops has been discussed.},
 bibtype = {article},
 author = {Hafeez, Abdul and Husain, Mohammed Aslam and Singh, S. P. and Chauhan, Anurag and Khan, Mohd Tauseef and Kumar, Navneet and Chauhan, Abhishek and Soni, S. K.},
 doi = {10.1016/j.inpa.2022.02.002},
 journal = {Information Processing in Agriculture},
 number = {xxxx}
}

The world receives more than 200 thousand people in a day and it is expected that the total world population will reach 9.6 billion by the year 2050. This will result in extra food demand, which can only be met from enhanced crop yield. Therefore, modernization of the agricultural sector becomes the need of the hour. There are many constraints that are responsible for the low production of crops, which can be overcome by using drone technology in the agriculture sector. This paper presents an analysis of drone technologies and their modifications with time in the agriculture sector in the last decade. The application of drones in the area of crop monitoring, and pesticide spraying for Precision Agriculture (PA) has been covered. The work done related to drone structure, multiple sensor development, innovation in spot area spraying has been presented. Moreover, the use of Artificial Intelligent (AI) and deep learning for the remote monitoring of crops has been discussed.

@article{
 title = {Convolutional Neural Networks in Detection of Plant Leaf Diseases: A Review},
 type = {article},
 year = {2022},
 keywords = {deep learning,machine learning,plant leaf diseases},
 pages = {1192},
 volume = {12},
 id = {ba5ce40d-562d-33bc-a26a-21dd3518115f},
 created = {2022-09-21T06:54:51.083Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-12-07T11:01:15.998Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Rapid improvements in deep learning (DL) techniques have made it possible to detect and recognize objects from images. DL approaches have recently entered various agricultural and farming applications after being successfully employed in various fields. Automatic identification of plant diseases can help farmers manage their crops more effectively, resulting in higher yields. Detecting plant disease in crops using images is an intrinsically difficult task. In addition to their detection, individual species identification is necessary for applying tailored control methods. A survey of research initiatives that use convolutional neural networks (CNN), a type of DL, to address various plant disease detection concerns was undertaken in the current publication. In this work, we have reviewed 100 of the most relevant CNN articles on detecting various plant leaf diseases over the last five years. In addition, we identified and summarized several problems and solutions corresponding to the CNN used in plant leaf disease detection. Moreover, Deep convolutional neural networks (DCNN) trained on image data were the most effective method for detecting early disease detection. We expressed the benefits and drawbacks of utilizing CNN in agriculture, and we discussed the direction of future developments in plant disease detection.},
 bibtype = {article},
 author = {Tugrul, Bulent and Elfatimi, Elhoucine and Eryigit, Recep},
 doi = {10.3390/agriculture12081192},
 journal = {Agriculture},
 number = {8}
}

Rapid improvements in deep learning (DL) techniques have made it possible to detect and recognize objects from images. DL approaches have recently entered various agricultural and farming applications after being successfully employed in various fields. Automatic identification of plant diseases can help farmers manage their crops more effectively, resulting in higher yields. Detecting plant disease in crops using images is an intrinsically difficult task. In addition to their detection, individual species identification is necessary for applying tailored control methods. A survey of research initiatives that use convolutional neural networks (CNN), a type of DL, to address various plant disease detection concerns was undertaken in the current publication. In this work, we have reviewed 100 of the most relevant CNN articles on detecting various plant leaf diseases over the last five years. In addition, we identified and summarized several problems and solutions corresponding to the CNN used in plant leaf disease detection. Moreover, Deep convolutional neural networks (DCNN) trained on image data were the most effective method for detecting early disease detection. We expressed the benefits and drawbacks of utilizing CNN in agriculture, and we discussed the direction of future developments in plant disease detection.

@article{
 title = {A Concurrent Multiscale Detector for End-to-End Image Matching},
 type = {article},
 year = {2022},
 keywords = {Concurrent detector,Deep learning,Detectors,Feature extraction,Image matching,Learning systems,Lighting,Optimization,end-to-end learning,image matching,key-point detection,multilevel features,multiscale features,rank consistent},
 pages = {1-15},
 id = {f01f1fef-c051-32f5-8ef4-f16225cd24de},
 created = {2022-09-22T17:08:07.507Z},
 file_attached = {true},
 profile_id = {276016a7-2c9d-3507-8888-093db7c54774},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-23T08:43:34.483Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {02fb5526-03ff-44ad-8d5c-42bd496c3100},
 private_publication = {false},
 abstract = {This article focuses on end-to-end image matching through joint key-point detection and descriptor extraction. To find repeatable and high discrimination key points, we improve the deep matching network from the perspectives of network structure and network optimization. First, we propose a concurrent multiscale detector (CS-det) network, which consists of several parallel convolutional networks to extract multiscale features and multilevel discriminative information for key-point detection. Moreover, we introduce an attention module to fuse the response maps of various features adaptively. Importantly, we propose two novel rank consistent losses (RC-losses) for network optimization, significantly improving image matching performances. On the one hand, we propose a score rank consistent loss (RC-S-loss) to ensure that the key points have high repeatability. Different from the score difference loss merely focusing on the absolute score of an individual key point, our proposed RC-S-loss pays more attention to the relative score of key points in the image. On the other hand, we propose a score-discrimination RC-loss to ensure that the key point has high discrimination, which can reduce the confusion from other key points in subsequent matching and then further enhance the accuracy of image matching. Extensive experimental results demonstrate that the proposed CS-det improves the mean matching result of deep detector by 1.4%–2.1%, and the proposed RC-losses can boost the matching performances by 2.7%–3.4% than score difference loss. Our source codes are available at https://github.com/iquandou/CS-Net.},
 bibtype = {article},
 author = {Quan, Dou and Wang, Shuang and Huyan, Ning and Li, Yi and Lei, Ruiqi and Chanussot, Jocelyn and Hou, Biao and Jiao, Licheng},
 doi = {10.1109/TNNLS.2022.3194079},
 journal = {IEEE Transactions on Neural Networks and Learning Systems}
}

This article focuses on end-to-end image matching through joint key-point detection and descriptor extraction. To find repeatable and high discrimination key points, we improve the deep matching network from the perspectives of network structure and network optimization. First, we propose a concurrent multiscale detector (CS-det) network, which consists of several parallel convolutional networks to extract multiscale features and multilevel discriminative information for key-point detection. Moreover, we introduce an attention module to fuse the response maps of various features adaptively. Importantly, we propose two novel rank consistent losses (RC-losses) for network optimization, significantly improving image matching performances. On the one hand, we propose a score rank consistent loss (RC-S-loss) to ensure that the key points have high repeatability. Different from the score difference loss merely focusing on the absolute score of an individual key point, our proposed RC-S-loss pays more attention to the relative score of key points in the image. On the other hand, we propose a score-discrimination RC-loss to ensure that the key point has high discrimination, which can reduce the confusion from other key points in subsequent matching and then further enhance the accuracy of image matching. Extensive experimental results demonstrate that the proposed CS-det improves the mean matching result of deep detector by 1.4%–2.1%, and the proposed RC-losses can boost the matching performances by 2.7%–3.4% than score difference loss. Our source codes are available at https://github.com/iquandou/CS-Net.

@article{
 title = {Sim2Real Object-Centric Keypoint Detection and Description},
 type = {article},
 year = {2022},
 pages = {5440-5449},
 volume = {36},
 id = {fe2ad67f-fd07-33fc-bf3f-e01ad982ac51},
 created = {2022-09-22T17:08:27.737Z},
 file_attached = {true},
 profile_id = {276016a7-2c9d-3507-8888-093db7c54774},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-23T09:20:30.287Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {02fb5526-03ff-44ad-8d5c-42bd496c3100,1608176d-7332-416b-a03b-89c3f2ae3ab6},
 private_publication = {false},
 abstract = {Keypoint detection and description play a central role in computer vision. Most existing methods are in the form of scene-level prediction, without returning the object classes of different keypoints. In this paper, we propose the object-centric formulation, which, beyond the conventional setting, requires further identifying which object each interest point belongs to. With such fine-grained information, our framework enables more downstream potentials, such as object-level matching and pose estimation in a clustered environment. To get around the difficulty of label collection in the real world, we develop a sim2real contrastive learning mechanism that can generalize the model trained in simulation to real-world applications. The novelties of our training method are three-fold: (i) we integrate the uncertainty into the learning framework to improve feature description of hard cases, e.g., less-textured or symmetric patches; (ii) we decouple the object descriptor into two independent branches, intra-object salience and inter-object distinctness, resulting in a better pixel-wise description; (iii) we enforce cross-view semantic consistency for enhanced robustness in representation learning. Comprehensive experiments on image matching and 6D pose estimation verify the encouraging generalization ability of our method. Particularly for 6D pose estimation, our method significantly outperforms typical unsupervised/sim2real methods, achieving a closer gap with the fully supervised counterpart.},
 bibtype = {article},
 author = {Zhong, Chengliang and Yang, Chao and Sun, Fuchun and Qi, Jinshan and Mu, Xiaodong and Liu, Huaping and Huang, Wenbing},
 doi = {10.1609/aaai.v36i5.20482},
 journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
 number = {5}
}

Keypoint detection and description play a central role in computer vision. Most existing methods are in the form of scene-level prediction, without returning the object classes of different keypoints. In this paper, we propose the object-centric formulation, which, beyond the conventional setting, requires further identifying which object each interest point belongs to. With such fine-grained information, our framework enables more downstream potentials, such as object-level matching and pose estimation in a clustered environment. To get around the difficulty of label collection in the real world, we develop a sim2real contrastive learning mechanism that can generalize the model trained in simulation to real-world applications. The novelties of our training method are three-fold: (i) we integrate the uncertainty into the learning framework to improve feature description of hard cases, e.g., less-textured or symmetric patches; (ii) we decouple the object descriptor into two independent branches, intra-object salience and inter-object distinctness, resulting in a better pixel-wise description; (iii) we enforce cross-view semantic consistency for enhanced robustness in representation learning. Comprehensive experiments on image matching and 6D pose estimation verify the encouraging generalization ability of our method. Particularly for 6D pose estimation, our method significantly outperforms typical unsupervised/sim2real methods, achieving a closer gap with the fully supervised counterpart.

@inproceedings{
 title = {Edible and Poisonous Mushrooms Classification by Machine Learning Algorithms},
 type = {inproceedings},
 year = {2022},
 keywords = {UCI machine learning repository,edible mushrooms,machine learning,mushroom dataset,poisonous mushrooms},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {a598a5bf-ae4e-3eb9-8c45-6212c9fbff92},
 created = {2022-09-29T07:47:54.960Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-12T10:14:11.859Z},
 read = {false},
 starred = {true},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 citation_key = {Tutuncu2022},
 folder_uuids = {6aec054e-977e-4d9f-8f01-b49e735da52b,5b1307f8-2af3-4763-ba1d-94a2bebe2114},
 private_publication = {false},
 abstract = {Of the millions of mushroom species growing all around the world, one type is edible, while the other is poisonous. It is not easy to distinguish edible and poisonous mushrooms from each other and it is a condition that requires expertise. The classification of poisonous and edible mushrooms is therefore important. Machine learning algorithms are an alternative method for classifying poisonous and edible mushrooms using morphological or physical features of fungi. The dataset used in this study is the Mushroom dataset available in the UC Irvine Machine Learning Repository. Based on 22 features in the Mushroom dataset and four different machine learning algorithms, models have been created for the classification of edible and poisonous fungi. The classification success rates of these models were obtained from Naive Bayes, Decision Tree, Support Vector Machine and AdaBoost algorithms with 90.99%, 98.82%, 99.98% and 100%, respectively. When these results were examined, taking into account the physical appearance features of the mushrooms, it was determined whether the mushrooms were edible and poisonous by 100% with the AdaBoost model.},
 bibtype = {inproceedings},
 author = {Tutuncu, Kemal and Cinar, Ilkay and Kursun, Ramazan and Koklu, Murat},
 doi = {10.1109/MECO55406.2022.9797212},
 booktitle = {2022 11th Mediterranean Conference on Embedded Computing, MECO 2022}
}

Of the millions of mushroom species growing all around the world, one type is edible, while the other is poisonous. It is not easy to distinguish edible and poisonous mushrooms from each other and it is a condition that requires expertise. The classification of poisonous and edible mushrooms is therefore important. Machine learning algorithms are an alternative method for classifying poisonous and edible mushrooms using morphological or physical features of fungi. The dataset used in this study is the Mushroom dataset available in the UC Irvine Machine Learning Repository. Based on 22 features in the Mushroom dataset and four different machine learning algorithms, models have been created for the classification of edible and poisonous fungi. The classification success rates of these models were obtained from Naive Bayes, Decision Tree, Support Vector Machine and AdaBoost algorithms with 90.99%, 98.82%, 99.98% and 100%, respectively. When these results were examined, taking into account the physical appearance features of the mushrooms, it was determined whether the mushrooms were edible and poisonous by 100% with the AdaBoost model.

@article{
 title = {Efficient Large-scale Localization by Global Instance Recognition},
 type = {article},
 year = {2022},
 pages = {17348-17357},
 id = {743b55ce-71e4-3314-9472-63904fab24bd},
 created = {2022-10-04T09:36:42.751Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-10-05T10:42:42.967Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {02fb5526-03ff-44ad-8d5c-42bd496c3100},
 private_publication = {false},
 abstract = {Hierarchical frameworks consisting of both coarse and fine localization are often used as the standard pipeline for large-scale visual localization. Despite their promising performance in simple environments, they still suffer from low efficiency and accuracy in large-scale scenes, especially under challenging conditions. In this paper, we propose an efficient and accurate large-scale localization framework based on the recognition of buildings, which are not only discriminative for coarse localization but also robust for fine localization. Specifically, we assign each building instance a global ID and perform pixel-wise recognition of these global instances in the localization process. For coarse localization, we employ an efficient reference search strategy to find candidates progressively from the local map observing recognized instances instead of the whole database. For fine localization, predicted labels are further used for instance-wise feature detection and matching , allowing our model to focus on fewer but more robust keypoints for establishing correspondences. The experiments in long-term large-scale localization datasets including Aachen and RobotCar-Seasons demonstrate that our method outperforms previous approaches consistently in terms of both efficiency and accuracy.},
 bibtype = {article},
 author = {Xue, Fei and Budvytis, Ignas and Olmeda Reino, Daniel and Cipolla, Roberto},
 journal = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}
}

Hierarchical frameworks consisting of both coarse and fine localization are often used as the standard pipeline for large-scale visual localization. Despite their promising performance in simple environments, they still suffer from low efficiency and accuracy in large-scale scenes, especially under challenging conditions. In this paper, we propose an efficient and accurate large-scale localization framework based on the recognition of buildings, which are not only discriminative for coarse localization but also robust for fine localization. Specifically, we assign each building instance a global ID and perform pixel-wise recognition of these global instances in the localization process. For coarse localization, we employ an efficient reference search strategy to find candidates progressively from the local map observing recognized instances instead of the whole database. For fine localization, predicted labels are further used for instance-wise feature detection and matching , allowing our model to focus on fewer but more robust keypoints for establishing correspondences. The experiments in long-term large-scale localization datasets including Aachen and RobotCar-Seasons demonstrate that our method outperforms previous approaches consistently in terms of both efficiency and accuracy.

@article{
 title = {Grapevine Segmentation in RGB Images using Deep Learning},
 type = {article},
 year = {2022},
 keywords = {carneiro2021grapevinesegmentationrgb},
 pages = {101-106},
 volume = {196},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {2d745254-3e61-3077-a348-ef23337ef217},
 created = {2022-11-10T10:19:53.593Z},
 accessed = {2022-11-10},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:01.276Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Wine is the most important product from the Douro Region, in Portugal. Ampelographs are disappearing, and farmers need new solutions to identify grapevine varieties to ensure high-quality standards. The development of methodology capable of automatically identify grapevine are in need. In the scenario, deep learning based methods are emerging as the state-of-art in grapevines classification tasks. In previous work, we verify the deep learning models would benefit from focus classification patches in leaves images areas. Deep learning segmentation methods can be used to find grapevine leaves areas. This paper presents a methodology to segment grapevines images automatically based on the U-net model. A private dataset was used, composed of 733 grapevines images frames extracted from 236 videos collected in a natural environment. The trained model obtained a Dice of 95.6% and an Intersection over Union of 91.6%, results that fully satisfy the need of localise grapevine leaves.},
 bibtype = {article},
 author = {Carneiro, Gabriel A. and Magalhães, Rafaela and Neto, Alexandre and Sousa, Joaquim J. and Cunha, António},
 doi = {10.1016/J.PROCS.2021.11.078},
 journal = {Procedia Computer Science}
}

Wine is the most important product from the Douro Region, in Portugal. Ampelographs are disappearing, and farmers need new solutions to identify grapevine varieties to ensure high-quality standards. The development of methodology capable of automatically identify grapevine are in need. In the scenario, deep learning based methods are emerging as the state-of-art in grapevines classification tasks. In previous work, we verify the deep learning models would benefit from focus classification patches in leaves images areas. Deep learning segmentation methods can be used to find grapevine leaves areas. This paper presents a methodology to segment grapevines images automatically based on the U-net model. A private dataset was used, composed of 733 grapevines images frames extracted from 236 videos collected in a natural environment. The trained model obtained a Dice of 95.6% and an Intersection over Union of 91.6%, results that fully satisfy the need of localise grapevine leaves.

@article{
 title = {PyTorch-OOD: A Library for Out-of-Distribution Detection based on PyTorch},
 type = {article},
 year = {2022},
 pages = {4350-4359},
 volume = {2022-June},
 publisher = {IEEE Computer Society},
 id = {0ba55f41-685f-3f71-9d30-1dd32173a28a},
 created = {2022-11-18T10:39:42.207Z},
 accessed = {2022-11-18},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-03T10:17:19.003Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,4b11b59b-846d-45b5-82b7-81d8bc3108b2,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Machine Learning models based on Deep Neural Networks behave unpredictably when presented with inputs that do not stem from the training distribution and sometimes make egregiously wrong predictions with high confidence. This property undermines the trustworthiness of systems depending on such models and potentially threatens the safety of their users. Out-of-Distribution (OOD) detection mechanisms can be used to prevent errors by detecting inputs that are so dissimilar from the training set that the model can not be expected to make reliable predictions. In this paper, we present PyTorch-OOD, a Python library for OOD detection based on PyTorch. Its primary goals are to accelerate OOD detection research and improve the reproducibility and comparability of experiments. PyTorch-OOD provides well-tested and documented implementations of OOD detection methods with a unified interface, as well as training and benchmark datasets, architectures, pre-trained models, and utility functions. The library is available online 1 under the permissive Apache 2.0 license and can be installed via Python Package Index (PyPI).},
 bibtype = {article},
 author = {Kirchheim, Konstantin and Filax, Marco and Ortmeier, Frank},
 doi = {10.1109/CVPRW56347.2022.00481},
 journal = {IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops}
}

Machine Learning models based on Deep Neural Networks behave unpredictably when presented with inputs that do not stem from the training distribution and sometimes make egregiously wrong predictions with high confidence. This property undermines the trustworthiness of systems depending on such models and potentially threatens the safety of their users. Out-of-Distribution (OOD) detection mechanisms can be used to prevent errors by detecting inputs that are so dissimilar from the training set that the model can not be expected to make reliable predictions. In this paper, we present PyTorch-OOD, a Python library for OOD detection based on PyTorch. Its primary goals are to accelerate OOD detection research and improve the reproducibility and comparability of experiments. PyTorch-OOD provides well-tested and documented implementations of OOD detection methods with a unified interface, as well as training and benchmark datasets, architectures, pre-trained models, and utility functions. The library is available online 1 under the permissive Apache 2.0 license and can be installed via Python Package Index (PyPI).

@article{
 title = {Comparison of Deep Neural Networks in Detecting Field Grapevine Diseases Using Transfer Learning},
 type = {article},
 year = {2022},
 keywords = {morellos2022comparisondeepneural},
 pages = {4648},
 volume = {14},
 websites = {https://www.mdpi.com/2072-4292/14/18/4648/htm,https://www.mdpi.com/2072-4292/14/18/4648},
 month = {9},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {17},
 id = {4aa3ddff-aef9-3503-ad9a-7724f466f0bc},
 created = {2022-12-08T08:27:01.307Z},
 accessed = {2022-12-08},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:03.908Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Plants diseases constitute a substantial threat for farmers given the high economic and environmental impact of their treatment. Detecting possible pathogen threats in plants based on non-destructive remote sensing and computer vision methods offers an alternative to existing laboratory methods and leads to improved crop management. Vine is an important crop that is mainly affected by fungal diseases. In this study, photos from healthy leaves and leaves infected by a fungal disease of vine are used to create disease identification classifiers. The transfer learning technique was employed in this study and was used to train three different deep convolutional neural network (DCNN) approaches that were compared according to their classification accuracy, namely AlexNet, VGG-19, and Inception v3. The above-mentioned models were trained on the open-source PlantVillage dataset using two training approaches: feature extraction, where the weights of the base deep neural network model were frozen and only the ones on the newly added layers were updated, and fine tuning, where the weights of the base model were also updated during training. Then, the created models were validated on the PlantVillage dataset and retrained using a custom field-grown vine photo dataset. The results showed that the fine-tuning approach showed better validation and testing accuracy, for all DCNNs, compared to the feature extraction approach. As far as the results of DCNNs are concerned, the Inception v3 algorithm outperformed VGG-19 and AlexNet in almost all the cases, demonstrating a validation performance of 100% for the fine-tuned strategy on the PlantVillage dataset and an accuracy of 83.3% for the respective strategy on a custom vine disease use case dataset, while AlexNet achieved 87.5% validation and 66.7% accuracy for the respective scenarios. Regarding VGG-19, the validation performance reached 100%, with an accuracy of 76.7%.},
 bibtype = {article},
 author = {Morellos, Antonios and Pantazi, Xanthoula Eirini and Paraskevas, Charalampos and Moshou, Dimitrios},
 doi = {10.3390/RS14184648},
 journal = {Remote Sensing 2022, Vol. 14, Page 4648},
 number = {18}
}

Plants diseases constitute a substantial threat for farmers given the high economic and environmental impact of their treatment. Detecting possible pathogen threats in plants based on non-destructive remote sensing and computer vision methods offers an alternative to existing laboratory methods and leads to improved crop management. Vine is an important crop that is mainly affected by fungal diseases. In this study, photos from healthy leaves and leaves infected by a fungal disease of vine are used to create disease identification classifiers. The transfer learning technique was employed in this study and was used to train three different deep convolutional neural network (DCNN) approaches that were compared according to their classification accuracy, namely AlexNet, VGG-19, and Inception v3. The above-mentioned models were trained on the open-source PlantVillage dataset using two training approaches: feature extraction, where the weights of the base deep neural network model were frozen and only the ones on the newly added layers were updated, and fine tuning, where the weights of the base model were also updated during training. Then, the created models were validated on the PlantVillage dataset and retrained using a custom field-grown vine photo dataset. The results showed that the fine-tuning approach showed better validation and testing accuracy, for all DCNNs, compared to the feature extraction approach. As far as the results of DCNNs are concerned, the Inception v3 algorithm outperformed VGG-19 and AlexNet in almost all the cases, demonstrating a validation performance of 100% for the fine-tuned strategy on the PlantVillage dataset and an accuracy of 83.3% for the respective strategy on a custom vine disease use case dataset, while AlexNet achieved 87.5% validation and 66.7% accuracy for the respective scenarios. Regarding VGG-19, the validation performance reached 100%, with an accuracy of 76.7%.

@article{
 title = {A Novel Enhanced VGG16 Model to Tackle Grapevine Leaves Diseases With Automatic Method},
 type = {article},
 year = {2022},
 keywords = {mousavi2022novelenhancedvgg16},
 pages = {111564-111578},
 volume = {10},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {c57835b5-b492-3333-9dc3-78be1814caed},
 created = {2022-12-08T08:40:38.289Z},
 accessed = {2022-12-08},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:04.129Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Diseases of grape leaves are destructive because they spread very quickly. Therefore, they should be detected and tackled fast. This study proposes a novel Enhanced VGG16 model for detecting, identifying, and classifying three widespread diseases in grape leaves. These diseases include Downey Mildew, Anthracnose, and Powdery Mildew. The data are automatically gathered with the assistance of photos taken with a quadcopter from the grapevine garden. The main central system analyzes the images received from the quadcopter to identify diseases. If it finds any of the three diseases, it sends the command to the hexacopter to spray pesticide into the grapevine garden location where the disease is found. The proposed method using the Enhanced VGG16 model with Faster Region-based Convolutional Neural Networks (R-CNN) is compared with different networks, including VGG16, GoogLeNet, ResNet50, and AlexNet. The experimental results on the grape leaf diseases demonstrated that the proposed method achieves the mean Average Precision) mAP (criterion improvements of 0.53%, 0.912%, 2.759%, and 7.268% compared with the ResNet50, VGG16, GoogLeNet, and AlexNet networks, respectively. Also, the average accuracy of the proposed Enhanced VGG16 model is 99.6%, which is 0.437-1.91% higher than other models. The Enhanced VGG16 has the best precision, and the number of layers is acceptable, although it is not less.},
 bibtype = {article},
 author = {Mousavi, Seyedamirhossein and Farahani, Gholamreza},
 doi = {10.1109/ACCESS.2022.3215639},
 journal = {IEEE Access}
}

Diseases of grape leaves are destructive because they spread very quickly. Therefore, they should be detected and tackled fast. This study proposes a novel Enhanced VGG16 model for detecting, identifying, and classifying three widespread diseases in grape leaves. These diseases include Downey Mildew, Anthracnose, and Powdery Mildew. The data are automatically gathered with the assistance of photos taken with a quadcopter from the grapevine garden. The main central system analyzes the images received from the quadcopter to identify diseases. If it finds any of the three diseases, it sends the command to the hexacopter to spray pesticide into the grapevine garden location where the disease is found. The proposed method using the Enhanced VGG16 model with Faster Region-based Convolutional Neural Networks (R-CNN) is compared with different networks, including VGG16, GoogLeNet, ResNet50, and AlexNet. The experimental results on the grape leaf diseases demonstrated that the proposed method achieves the mean Average Precision) mAP (criterion improvements of 0.53%, 0.912%, 2.759%, and 7.268% compared with the ResNet50, VGG16, GoogLeNet, and AlexNet networks, respectively. Also, the average accuracy of the proposed Enhanced VGG16 model is 99.6%, which is 0.437-1.91% higher than other models. The Enhanced VGG16 has the best precision, and the number of layers is acceptable, although it is not less.

@article{
 title = {Fields2Cover: An open-source coverage path planning library for unmanned agricultural vehicles},
 type = {article},
 year = {2022},
 keywords = {Agricultural Automation,Field Robots,Software Architecture for Robotic and Automation},
 websites = {https://arxiv.org/abs/2210.07838v1},
 month = {10},
 day = {14},
 id = {f8603cf3-c625-39d4-83f0-3ef1af949010},
 created = {2022-12-12T06:50:10.295Z},
 accessed = {2022-12-12},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-12-12T12:51:04.141Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {This paper describes Fields2Cover, a novel open source library for coverage
path planning (CPP) for agricultural vehicles. While there are several CPP
solutions nowadays, there have been limited efforts to unify them into an open
source library and provide benchmarking tools to compare their performance.
Fields2Cover provides a framework for planning coverage paths, developing novel
techniques, and benchmarking state-of-the-art algorithms. The library features
a modular and extensible architecture that supports various vehicles and can be
used for a variety of applications, including farms. Its core modules are: a
headland generator, a swath generator, a route planner and a path planner. An
interface to the Robot Operating System (ROS) is also supplied as an add-on. In
this paper, the functionalities of the library for planning a coverage path in
agriculture are demonstrated using 8 state-of-the-art methods and 7 objective
functions in simulation and field experiments.},
 bibtype = {article},
 author = {Mier, Gonzalo and Valente, João and de Bruin, Sytze},
 doi = {10.48550/arxiv.2210.07838}
}

This paper describes Fields2Cover, a novel open source library for coverage path planning (CPP) for agricultural vehicles. While there are several CPP solutions nowadays, there have been limited efforts to unify them into an open source library and provide benchmarking tools to compare their performance. Fields2Cover provides a framework for planning coverage paths, developing novel techniques, and benchmarking state-of-the-art algorithms. The library features a modular and extensible architecture that supports various vehicles and can be used for a variety of applications, including farms. Its core modules are: a headland generator, a swath generator, a route planner and a path planner. An interface to the Robot Operating System (ROS) is also supplied as an add-on. In this paper, the functionalities of the library for planning a coverage path in agriculture are demonstrated using 8 state-of-the-art methods and 7 objective functions in simulation and field experiments.

@article{
 title = {Deep semantic segmentation for the quantification of grape foliar diseases in the vineyard},
 type = {article},
 year = {2022},
 keywords = {liu2022deepsemanticsegmentation},
 volume = {13},
 id = {157de504-46d7-3142-b8de-8bb0c390e702},
 created = {2022-12-13T09:33:41.076Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T14:59:49.901Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Plant disease evaluation is crucial to pathogen management and plant breeding. Human field scouting has been widely used to monitor disease progress and provide qualitative and quantitative evaluation, which is costly, laborious, subjective, and often imprecise. To improve disease evaluation accuracy, throughput, and objectiveness, an image-based approach with a deep learning-based analysis pipeline was developed to calculate infection severity of grape foliar diseases. The image-based approach used a ground imaging system for field data acquisition, consisting of a custom stereo camera with strobe light for consistent illumination and real time kinematic (RTK) GPS for accurate localization. The deep learning-based pipeline used the hierarchical multiscale attention semantic segmentation (HMASS) model for disease infection segmentation, color filtering for grapevine canopy segmentation, and depth and location information for effective region masking. The resultant infection, canopy, and effective region masks were used to calculate the severity rate of disease infections in an image sequence collected in a given unit (e.g., grapevine panel). Fungicide trials for grape downy mildew (DM) and powdery mildew (PM) were used as case studies to evaluate the developed approach and pipeline. Experimental results showed that the HMASS model achieved acceptable to good segmentation accuracy of DM (mIoU > 0.84) and PM (mIoU > 0.74) infections in testing images, demonstrating the model capability for symptomatic disease segmentation. With the consistent image quality and multimodal metadata provided by the imaging system, the color filter and overlapping region removal could accurately and reliably segment grapevine canopies and identify repeatedly imaged regions between consecutive image frames, leading to critical information for infection severity calculation. Image-derived severity rates were highly correlated (r > 0.95) with human-assessed values, and had comparable statistical power in differentiating fungicide treatment efficacy in both case studies. Therefore, the developed approach and pipeline can be used as an effective and efficient tool to quantify the severity of foliar disease infections, enabling objective, high-throughput disease evaluation for fungicide trial evaluation, genetic mapping, and breeding programs.},
 bibtype = {article},
 author = {Liu, Ertai and Gold, Kaitlin M. and Combs, David and Cadle-Davidson, Lance and Jiang, Yu},
 doi = {10.3389/fpls.2022.978761},
 journal = {Frontiers in Plant Science}
}

Plant disease evaluation is crucial to pathogen management and plant breeding. Human field scouting has been widely used to monitor disease progress and provide qualitative and quantitative evaluation, which is costly, laborious, subjective, and often imprecise. To improve disease evaluation accuracy, throughput, and objectiveness, an image-based approach with a deep learning-based analysis pipeline was developed to calculate infection severity of grape foliar diseases. The image-based approach used a ground imaging system for field data acquisition, consisting of a custom stereo camera with strobe light for consistent illumination and real time kinematic (RTK) GPS for accurate localization. The deep learning-based pipeline used the hierarchical multiscale attention semantic segmentation (HMASS) model for disease infection segmentation, color filtering for grapevine canopy segmentation, and depth and location information for effective region masking. The resultant infection, canopy, and effective region masks were used to calculate the severity rate of disease infections in an image sequence collected in a given unit (e.g., grapevine panel). Fungicide trials for grape downy mildew (DM) and powdery mildew (PM) were used as case studies to evaluate the developed approach and pipeline. Experimental results showed that the HMASS model achieved acceptable to good segmentation accuracy of DM (mIoU > 0.84) and PM (mIoU > 0.74) infections in testing images, demonstrating the model capability for symptomatic disease segmentation. With the consistent image quality and multimodal metadata provided by the imaging system, the color filter and overlapping region removal could accurately and reliably segment grapevine canopies and identify repeatedly imaged regions between consecutive image frames, leading to critical information for infection severity calculation. Image-derived severity rates were highly correlated (r > 0.95) with human-assessed values, and had comparable statistical power in differentiating fungicide treatment efficacy in both case studies. Therefore, the developed approach and pipeline can be used as an effective and efficient tool to quantify the severity of foliar disease infections, enabling objective, high-throughput disease evaluation for fungicide trial evaluation, genetic mapping, and breeding programs.

@article{
 title = {VineInspector: The Vineyard Assistant},
 type = {article},
 year = {2022},
 keywords = {mendes2022vineinspector},
 pages = {730},
 volume = {12},
 websites = {https://www.mdpi.com/2077-0472/12/5/730/htm,https://www.mdpi.com/2077-0472/12/5/730},
 month = {5},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {22},
 id = {4effeeca-910f-3a75-b672-01416d23460f},
 created = {2022-12-13T09:35:28.093Z},
 accessed = {2022-12-13},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:03.308Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Proximity sensing approaches with a wide array of sensors available for use in precision viticulture contexts can nowadays be considered both well-know and mature technologies. Still, several in-field practices performed throughout different crops rely on direct visual observation supported on gained experience to assess aspects of plants’ phenological development, as well as indicators relating to the onset of common plagues and diseases. Aiming to mimic in-field direct observation, this paper presents VineInspector: a low-cost, self-contained and easy-to-install system, which is able to measure microclimatic parameters, and also to acquire images using multiple cameras. It is built upon a stake structure, rendering it suitable for deployment across a vineyard. The approach through which distinguishable attributes are detected, classified and tallied in the periodically acquired images, makes use of artificial intelligence approaches. Furthermore, it is made available through an IoT cloud-based support system. VineInspector was field-tested under real operating conditions to assess not only the robustness and the operating functionality of the hardware solution, but also the AI approaches’ accuracy. Two applications were developed to evaluate VineInspector’s consistency while a viticulturist’ assistant in everyday practices. One was intended to determine the size of the very first grapevines’ shoots, one of the required parameters of the well known 3–10 rule to predict primary downy mildew infection. The other was developed to tally grapevine moth males captured in sex traps. Results show that VineInspector is a logical step in smart proximity monitoring by mimicking direct visual observation from experienced viticulturists. While the latter traditionally are responsible for a set of everyday practices in the field, these are time and resource consuming. VineInspector was proven to be effective in two of these practices, performing them automatically. Therefore, it enables both the continuous monitoring and assessment of a vineyard’s phenological development in a more efficient manner, making way to more assertive and timely practices against pests and diseases.},
 bibtype = {article},
 author = {Mendes, Jorge and Peres, Emanuel and Neves Dos Santos, Filipe and Silva, Nuno and Silva, Renato and Sousa, Joaquim João and Cortez, Isabel and Morais, Raul},
 doi = {10.3390/AGRICULTURE12050730},
 journal = {Agriculture 2022, Vol. 12, Page 730},
 number = {5}
}

Proximity sensing approaches with a wide array of sensors available for use in precision viticulture contexts can nowadays be considered both well-know and mature technologies. Still, several in-field practices performed throughout different crops rely on direct visual observation supported on gained experience to assess aspects of plants’ phenological development, as well as indicators relating to the onset of common plagues and diseases. Aiming to mimic in-field direct observation, this paper presents VineInspector: a low-cost, self-contained and easy-to-install system, which is able to measure microclimatic parameters, and also to acquire images using multiple cameras. It is built upon a stake structure, rendering it suitable for deployment across a vineyard. The approach through which distinguishable attributes are detected, classified and tallied in the periodically acquired images, makes use of artificial intelligence approaches. Furthermore, it is made available through an IoT cloud-based support system. VineInspector was field-tested under real operating conditions to assess not only the robustness and the operating functionality of the hardware solution, but also the AI approaches’ accuracy. Two applications were developed to evaluate VineInspector’s consistency while a viticulturist’ assistant in everyday practices. One was intended to determine the size of the very first grapevines’ shoots, one of the required parameters of the well known 3–10 rule to predict primary downy mildew infection. The other was developed to tally grapevine moth males captured in sex traps. Results show that VineInspector is a logical step in smart proximity monitoring by mimicking direct visual observation from experienced viticulturists. While the latter traditionally are responsible for a set of everyday practices in the field, these are time and resource consuming. VineInspector was proven to be effective in two of these practices, performing them automatically. Therefore, it enables both the continuous monitoring and assessment of a vineyard’s phenological development in a more efficient manner, making way to more assertive and timely practices against pests and diseases.

@article{
 title = {TOWARD A LOW-COST, MULTISPECTRAL, HIGH ACCURACY MAPPING SYSTEM FOR VINEYARD INSPECTION},
 type = {article},
 year = {2022},
 keywords = {beniaouf2022towardlowcostmultispectral},
 pages = {849--854},
 volume = {43},
 websites = {https://doi.org/10.5194/isprs-archives-XLIII-B3-2022-849-2022},
 id = {e8236093-ffca-3078-adb7-39d672e3430e},
 created = {2022-12-13T10:32:41.494Z},
 accessed = {2022-12-13},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:00.815Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Confronted with the climate change challenge and the territorial constraints, agriculture has to modernize itself. The use of georef-erenced data and remote-sensing imagery is a major step in this direction. This precision mapping of crops requires powerful and accurate acquisition systems, while remaining financially attractive. The development of multispectral sensors and low-cost GNSS makes it possible to consider systems that will be able to map at the plant scale. However, these positioning systems do not yet guarantee a precise overlap of data acquired at different times. Thus, we propose in this paper a method to register terrestrial image data, acquired on vineyard plots. Our method seeks to avoid image registration problems, such as illumination changes, by detecting the vine stocks, reconstructing them in 3D, and registering them individually. The 3D detection method is based on an image-based object detection method (Faster R-CNN) and a structure-from-motion reconstruction of object-masked images. The results that we obtained on a vineyard plot, allowed us to validate the method, with a precision of less than 10 cm, making it possible to map the vine by stock.},
 bibtype = {article},
 author = {Beniaouf, S and Mabillard, R and Gressin, A},
 doi = {10.5194/isprs-archives-XLIII-B3-2022-849-2022},
 journal = {The International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences}
}

Confronted with the climate change challenge and the territorial constraints, agriculture has to modernize itself. The use of georef-erenced data and remote-sensing imagery is a major step in this direction. This precision mapping of crops requires powerful and accurate acquisition systems, while remaining financially attractive. The development of multispectral sensors and low-cost GNSS makes it possible to consider systems that will be able to map at the plant scale. However, these positioning systems do not yet guarantee a precise overlap of data acquired at different times. Thus, we propose in this paper a method to register terrestrial image data, acquired on vineyard plots. Our method seeks to avoid image registration problems, such as illumination changes, by detecting the vine stocks, reconstructing them in 3D, and registering them individually. The 3D detection method is based on an image-based object detection method (Faster R-CNN) and a structure-from-motion reconstruction of object-masked images. The results that we obtained on a vineyard plot, allowed us to validate the method, with a precision of less than 10 cm, making it possible to map the vine by stock.

@article{
 title = {LDD: A Dataset for Grape Diseases Object Detection and Instance Segmentation},
 type = {article},
 year = {2022},
 keywords = {Grape diseases,Instance segmentation,Learn-ing ·,Object,Object detection,Segmentation ·,Supervised,Supervised learning},
 pages = {383-393},
 volume = {13232 LNCS},
 websites = {http://arxiv.org/abs/2206.10192,http://dx.doi.org/10.1007/978-3-031-06430-2_32},
 month = {6},
 publisher = {Springer Science and Business Media Deutschland GmbH},
 day = {21},
 id = {4920fca3-1de7-3b20-8267-8a382de5bea7},
 created = {2022-12-13T11:48:52.969Z},
 accessed = {2022-12-13},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-13T13:49:04.963Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The Instance Segmentation task, an extension of the well-known Object Detection task, is of great help in many areas, such as precision agriculture: being able to automatically identify plant organs and the possible diseases associated with them, allows to effectively scale and automate crop monitoring and its diseases control. To address the problem related to early disease detection and diagnosis on vines plants, a new dataset has been created with the goal of advancing the state-of-the-art of diseases recognition via instance segmentation approaches. This was achieved by gathering images of leaves and clusters of grapes affected by diseases in their natural context. The dataset contains photos of 10 object types which include leaves and grapes with and without symptoms of the eight more common grape diseases, with a total of 17,706 labeled instances in 1,092 images. Multiple statistical measures are proposed in order to offer a complete view on the characteristics of the dataset. Preliminary results for the object detection and instance segmentation tasks reached by the models Mask R-CNN and R^3-CNN are provided as baseline, demonstrating that the procedure is able to reach promising results about the objective of automatic diseases' symptoms recognition.},
 bibtype = {article},
 author = {Rossi, Leonardo and Valenti, Marco and Legler, Sara Elisabetta and Prati, Andrea},
 doi = {10.1007/978-3-031-06430-2_32},
 journal = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}
}

The Instance Segmentation task, an extension of the well-known Object Detection task, is of great help in many areas, such as precision agriculture: being able to automatically identify plant organs and the possible diseases associated with them, allows to effectively scale and automate crop monitoring and its diseases control. To address the problem related to early disease detection and diagnosis on vines plants, a new dataset has been created with the goal of advancing the state-of-the-art of diseases recognition via instance segmentation approaches. This was achieved by gathering images of leaves and clusters of grapes affected by diseases in their natural context. The dataset contains photos of 10 object types which include leaves and grapes with and without symptoms of the eight more common grape diseases, with a total of 17,706 labeled instances in 1,092 images. Multiple statistical measures are proposed in order to offer a complete view on the characteristics of the dataset. Preliminary results for the object detection and instance segmentation tasks reached by the models Mask R-CNN and R^3-CNN are provided as baseline, demonstrating that the procedure is able to reach promising results about the objective of automatic diseases' symptoms recognition.

@inproceedings{
 title = {Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation},
 type = {inproceedings},
 year = {2022},
 pages = {7181-7190},
 websites = {https://github.com/xiaoachen98/DALN},
 id = {4cac66c7-fa19-38b6-bae1-d3dcc2048340},
 created = {2023-01-06T14:07:13.088Z},
 accessed = {2023-01-06},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-13T07:12:08.126Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Adversarial learning has achieved remarkable performances for unsupervised domain adaptation (UDA). Existing adversarial UDA methods typically adopt an additional discriminator to play the min-max game with a feature ex-tractor. However, most of these methods failed to effectively leverage the predicted discriminative information, and thus cause mode collapse for generator. In this work, we address this problem from a different perspective and design a simple yet effective adversarial paradigm in the form of a discriminator-free adversarial learning network (DALN), wherein the category classifier is reused as a discrimina-tor, which achieves explicit domain alignment and category distinguishment through a unified objective, enabling the DALN to leverage the predicted discriminative information for sufficient feature alignment. Basically, we introduce a Nuclear-norm Wasserstein discrepancy (NWD) that has definite guidance meaning for performing discrimination. Such NWD can be coupled with the classifier to serve as a discrim-inator satisfying the K-Lipschitz constraint without the requirements of additional weight clipping or gradient penalty strategy. Without bells and whistles, DALN compares favorably against the existing state-of-the-art (SOTA) methods on a variety of public datasets. Moreover, as a plug-and-play technique, NWD can be directly used as a generic regular-izer to benefit existing UDA algorithms. Code is available at https://github.com/xiaoachen98/DALN .},
 bibtype = {inproceedings},
 author = {Chen, Lin and Chen, Huaian and Wei, Zhixiang and Jin, Xin and Tan, Xiao and Jin, Yi and Chen, Enhong},
 booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022}
}

Adversarial learning has achieved remarkable performances for unsupervised domain adaptation (UDA). Existing adversarial UDA methods typically adopt an additional discriminator to play the min-max game with a feature ex-tractor. However, most of these methods failed to effectively leverage the predicted discriminative information, and thus cause mode collapse for generator. In this work, we address this problem from a different perspective and design a simple yet effective adversarial paradigm in the form of a discriminator-free adversarial learning network (DALN), wherein the category classifier is reused as a discrimina-tor, which achieves explicit domain alignment and category distinguishment through a unified objective, enabling the DALN to leverage the predicted discriminative information for sufficient feature alignment. Basically, we introduce a Nuclear-norm Wasserstein discrepancy (NWD) that has definite guidance meaning for performing discrimination. Such NWD can be coupled with the classifier to serve as a discrim-inator satisfying the K-Lipschitz constraint without the requirements of additional weight clipping or gradient penalty strategy. Without bells and whistles, DALN compares favorably against the existing state-of-the-art (SOTA) methods on a variety of public datasets. Moreover, as a plug-and-play technique, NWD can be directly used as a generic regular-izer to benefit existing UDA algorithms. Code is available at https://github.com/xiaoachen98/DALN .

@article{
 title = {Deliberated Domain Bridging for Domain Adaptive Semantic Segmentation},
 type = {article},
 year = {2022},
 websites = {https://arxiv.org/abs/2209.07695v3},
 month = {9},
 day = {16},
 id = {6fe4cfdb-b254-372f-bb5c-add3edcf2f0f},
 created = {2023-01-06T14:08:32.653Z},
 accessed = {2023-01-06},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-10T06:32:32.549Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {In unsupervised domain adaptation (UDA), directly adapting from the source to
the target domain usually suffers significant discrepancies and leads to
insufficient alignment. Thus, many UDA works attempt to vanish the domain gap
gradually and softly via various intermediate spaces, dubbed domain bridging
(DB). However, for dense prediction tasks such as domain adaptive semantic
segmentation (DASS), existing solutions have mostly relied on rough style
transfer and how to elegantly bridge domains is still under-explored. In this
work, we resort to data mixing to establish a deliberated domain bridging (DDB)
for DASS, through which the joint distributions of source and target domains
are aligned and interacted with each in the intermediate space. At the heart of
DDB lies a dual-path domain bridging step for generating two intermediate
domains using the coarse-wise and the fine-wise data mixing techniques,
alongside a cross-path knowledge distillation step for taking two complementary
models trained on generated intermediate samples as 'teachers' to develop a
superior 'student' in a multi-teacher distillation manner. These two
optimization steps work in an alternating way and reinforce each other to give
rise to DDB with strong adaptation power. Extensive experiments on adaptive
segmentation tasks with different settings demonstrate that our DDB
significantly outperforms state-of-the-art methods. Code is available at
https://github.com/xiaoachen98/DDB.git.},
 bibtype = {article},
 author = {Chen, Lin and Wei, Zhixiang and Jin, Xin and Chen, Huaian and Miao Zheng, † and Chen, Kai and Jin, Yi},
 doi = {10.48550/arxiv.2209.07695}
}

In unsupervised domain adaptation (UDA), directly adapting from the source to the target domain usually suffers significant discrepancies and leads to insufficient alignment. Thus, many UDA works attempt to vanish the domain gap gradually and softly via various intermediate spaces, dubbed domain bridging (DB). However, for dense prediction tasks such as domain adaptive semantic segmentation (DASS), existing solutions have mostly relied on rough style transfer and how to elegantly bridge domains is still under-explored. In this work, we resort to data mixing to establish a deliberated domain bridging (DDB) for DASS, through which the joint distributions of source and target domains are aligned and interacted with each in the intermediate space. At the heart of DDB lies a dual-path domain bridging step for generating two intermediate domains using the coarse-wise and the fine-wise data mixing techniques, alongside a cross-path knowledge distillation step for taking two complementary models trained on generated intermediate samples as 'teachers' to develop a superior 'student' in a multi-teacher distillation manner. These two optimization steps work in an alternating way and reinforce each other to give rise to DDB with strong adaptation power. Extensive experiments on adaptive segmentation tasks with different settings demonstrate that our DDB significantly outperforms state-of-the-art methods. Code is available at https://github.com/xiaoachen98/DDB.git.

@article{
 title = {Two-stage automatic diagnosis of Flavescence Dorée based on proximal imaging and artificial intelligence: a multi-year and multi-variety experimental study},
 type = {article},
 year = {2022},
 keywords = {artificial intelligence,computer vision,deep learning,flavescence dorée,image processing,proximal sensing,vine disease},
 pages = {371-384},
 volume = {56},
 id = {b8d9e7d6-f439-3f54-8d94-04d75061f05d},
 created = {2023-03-23T06:17:46.519Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:05.451Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {“Flavescence dorée” (FD) is a grape vine disease caused by the bacterial agent “Candidatus Phytoplasma vitis” and spread by the leafhopper Scaphoideus titanus Ball (Hemiptera: Cicadellidae). The disease is very closely monitored in Europe, as it reduces vine productivity and causes vine death and is also highly transmissible. Currently, the control method used against this disease is a two-pronged approach: i) the spraying of insecticide on a regular basis to kill the vector, and ii) a survey of each row in a vineyard by experts in this disease. Unfortunately, these experts are not able to carry out such a task every year on every vineyard and need an aid for planning their survey. In this study, we propose and evaluate an original automatic method for the detection of FD based on computer vision and artificial intelligence algorithms applied to images acquired by proximal sensing. A two-step approach was used, mimicking an expert’s scouting in the vine rows: (i) the three known isolated symptoms (red or yellow leaves depending on variety, together with a lack of shoot lignification and the presence of desiccated bunches) were detected, (ii) isolated detections were combined to make a diagnosis at image scale; i.e., vine scale. A detection network was used to detect and classify non-healthy leaves into three classes: ‘FD symptomatic leaf’, ‘Esca leaf’ and ‘Confounding leaf’; while a segmentation network was used for the retrieval of FD symptomatic shoots and bunches. Finally, the association of detected symptoms was performed by a RandomForest classifier for diagnosis at the image scale. The experimental evaluation was conducted on more than 1000 images collected from 14 blocks planted with five different grape varieties. The detection of the isolated symptoms achieved a precision of between 0.67 and 0.82 and a recall of between 0.39 and 0.59. The classification at the image scale obtained very good results when applied to images acquired under the same conditions, with the same grape varieties as the training images (precision and recall of more than 0.89). The results of the tests on the other grape varieties show the importance of having some of them in the training base in these AI-based approaches.},
 bibtype = {article},
 author = {Tardif, Malo and Amri, Ahmed and Keresztes, Barna and Deshayes, Aymeric and Martin, Damian and Greven, Marc and Da Costa, Jean Pierre},
 doi = {10.20870/oeno-one.2022.56.3.5460},
 journal = {Oeno One},
 number = {3}
}

“Flavescence dorée” (FD) is a grape vine disease caused by the bacterial agent “Candidatus Phytoplasma vitis” and spread by the leafhopper Scaphoideus titanus Ball (Hemiptera: Cicadellidae). The disease is very closely monitored in Europe, as it reduces vine productivity and causes vine death and is also highly transmissible. Currently, the control method used against this disease is a two-pronged approach: i) the spraying of insecticide on a regular basis to kill the vector, and ii) a survey of each row in a vineyard by experts in this disease. Unfortunately, these experts are not able to carry out such a task every year on every vineyard and need an aid for planning their survey. In this study, we propose and evaluate an original automatic method for the detection of FD based on computer vision and artificial intelligence algorithms applied to images acquired by proximal sensing. A two-step approach was used, mimicking an expert’s scouting in the vine rows: (i) the three known isolated symptoms (red or yellow leaves depending on variety, together with a lack of shoot lignification and the presence of desiccated bunches) were detected, (ii) isolated detections were combined to make a diagnosis at image scale; i.e., vine scale. A detection network was used to detect and classify non-healthy leaves into three classes: ‘FD symptomatic leaf’, ‘Esca leaf’ and ‘Confounding leaf’; while a segmentation network was used for the retrieval of FD symptomatic shoots and bunches. Finally, the association of detected symptoms was performed by a RandomForest classifier for diagnosis at the image scale. The experimental evaluation was conducted on more than 1000 images collected from 14 blocks planted with five different grape varieties. The detection of the isolated symptoms achieved a precision of between 0.67 and 0.82 and a recall of between 0.39 and 0.59. The classification at the image scale obtained very good results when applied to images acquired under the same conditions, with the same grape varieties as the training images (precision and recall of more than 0.89). The results of the tests on the other grape varieties show the importance of having some of them in the training base in these AI-based approaches.

@article{
 title = {Two-stage automatic diagnosis of Flavescence Dorée based on proximal imaging and artificial intelligence: a multi-year and multi-variety experimental study},
 type = {article},
 year = {2022},
 keywords = {tardif2022twostageautomatic},
 pages = {371-384},
 volume = {56},
 websites = {https://oeno-one.eu/article/view/5460/24052,https://oeno-one.eu/article/view/5460},
 month = {9},
 publisher = {International Viticulture and Enology Society},
 day = {19},
 id = {315fbcb3-2aa8-3652-9f78-81c4dbfa874d},
 created = {2023-03-23T06:20:00.327Z},
 accessed = {2023-03-23},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:42:35.682Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {“Flavescence dorée” (FD) is a grape vine disease caused by the bacterial agent “Candidatus Phytoplasma vitis” and spread by the leafhopper Scaphoideus titanus Ball (Hemiptera: Cicadellidae). The disease is very closely monitored in Europe, as it reduces vine productivity and causes vine death and is also highly transmissible. Currently, the control method used against this disease is a two-pronged approach: i) the spraying of insecticide on a regular basis to kill the vector, and ii) a survey of each row in a vineyard by experts in this disease. Unfortunately, these experts are not able to carry out such a task every year on every vineyard and need an aid for planning their survey.In this study, we propose and evaluate an original automatic method for the detection of FD based on computer vision and artificial intelligence algorithms applied to images acquired by proximal sensing. A two-step approach was used, mimicking an expert’s scouting in the vine rows: (i) the three known isolated symptoms (red or yellow leaves depending on variety, together with a lack of shoot lignification and the presence of desiccated bunches) were detected, (ii) isolated detections were combined to make a diagnosis at image scale; i.e., vine scale. A detection network was used to detect and classify non-healthy leaves into three classes: ‘FD symptomatic leaf', 'Esca leaf' and 'Confounding leaf'; while a segmentation network was used for the retrieval of FD symptomatic shoots and bunches. Finally, the association of detected symptoms was performed by a RandomForest classifier for diagnosis at the image scale. The experimental evaluation was conducted on more than 1000 images collected from 14 blocks planted with five different grape varieties. The detection of the isolated symptoms achieved a precision of between 0.67 and 0.82 and a recall of between 0.39 and 0.59. The classification at the image scale obtained very good results when applied to images acquired under the same conditions, with the same grape varieties as the training images (precision and recall of more than 0.89). The results of the tests on the other grape varieties show the importance of having some of them in the training base in these AI-based approaches.},
 bibtype = {article},
 author = {Tardif, Malo and Amri, Ahmed and Keresztes, Barna and Deshayes, Aymeric and Martin, Damian and Greven, Marc and Da Costa, Jean Pierre},
 doi = {10.20870/OENO-ONE.2022.56.3.5460},
 journal = {OENO One},
 number = {3}
}

“Flavescence dorée” (FD) is a grape vine disease caused by the bacterial agent “Candidatus Phytoplasma vitis” and spread by the leafhopper Scaphoideus titanus Ball (Hemiptera: Cicadellidae). The disease is very closely monitored in Europe, as it reduces vine productivity and causes vine death and is also highly transmissible. Currently, the control method used against this disease is a two-pronged approach: i) the spraying of insecticide on a regular basis to kill the vector, and ii) a survey of each row in a vineyard by experts in this disease. Unfortunately, these experts are not able to carry out such a task every year on every vineyard and need an aid for planning their survey.In this study, we propose and evaluate an original automatic method for the detection of FD based on computer vision and artificial intelligence algorithms applied to images acquired by proximal sensing. A two-step approach was used, mimicking an expert’s scouting in the vine rows: (i) the three known isolated symptoms (red or yellow leaves depending on variety, together with a lack of shoot lignification and the presence of desiccated bunches) were detected, (ii) isolated detections were combined to make a diagnosis at image scale; i.e., vine scale. A detection network was used to detect and classify non-healthy leaves into three classes: ‘FD symptomatic leaf', 'Esca leaf' and 'Confounding leaf'; while a segmentation network was used for the retrieval of FD symptomatic shoots and bunches. Finally, the association of detected symptoms was performed by a RandomForest classifier for diagnosis at the image scale. The experimental evaluation was conducted on more than 1000 images collected from 14 blocks planted with five different grape varieties. The detection of the isolated symptoms achieved a precision of between 0.67 and 0.82 and a recall of between 0.39 and 0.59. The classification at the image scale obtained very good results when applied to images acquired under the same conditions, with the same grape varieties as the training images (precision and recall of more than 0.89). The results of the tests on the other grape varieties show the importance of having some of them in the training base in these AI-based approaches.

@article{
 title = {Computer Vision and Deep Learning for Precision Viticulture},
 type = {article},
 year = {2022},
 keywords = {mohimont2022computervisiondeep},
 pages = {1-31},
 volume = {12},
 id = {1a33973b-958e-33b9-86fa-07ff69e1db25},
 created = {2023-03-28T05:59:35.417Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:11.825Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {During the last decades, researchers have developed novel computing methods to help viticulturists solve their problems, primarily those linked to yield estimation of their crops. This article aims to summarize the existing research associated with computer vision and viticulture. It focuses on approaches that use RGB images directly obtained from parcels, ranging from classic image analysis methods to Machine Learning, including novel Deep Learning techniques. We intend to produce a complete analysis accessible to everyone, including non-specialized readers, to discuss the recent progress of artificial intelligence (AI) in viticulture. To this purpose, we present work focusing on detecting grapevine flowers, grapes, and berries in the first sections of this article. In the last sections, we present different methods for yield estimation and the problems that arise with this task.},
 bibtype = {article},
 author = {Mohimont, Lucas and Alin, François and Rondeau, Marine and Gaveau, Nathalie and Steffenel, Luiz Angelo},
 doi = {10.3390/agronomy12102463},
 journal = {Agronomy},
 number = {10}
}

During the last decades, researchers have developed novel computing methods to help viticulturists solve their problems, primarily those linked to yield estimation of their crops. This article aims to summarize the existing research associated with computer vision and viticulture. It focuses on approaches that use RGB images directly obtained from parcels, ranging from classic image analysis methods to Machine Learning, including novel Deep Learning techniques. We intend to produce a complete analysis accessible to everyone, including non-specialized readers, to discuss the recent progress of artificial intelligence (AI) in viticulture. To this purpose, we present work focusing on detecting grapevine flowers, grapes, and berries in the first sections of this article. In the last sections, we present different methods for yield estimation and the problems that arise with this task.

@article{
 title = {Is it all a cluster game? - Exploring Out-of-Distribution Detection based on Clustering in the Embedding Space},
 type = {article},
 year = {2022},
 volume = {3087},
 id = {375d6c80-02cf-34e6-910b-976d73a3dc3e},
 created = {2023-04-21T09:42:40.788Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-21T09:43:04.773Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4b11b59b-846d-45b5-82b7-81d8bc3108b2},
 private_publication = {false},
 abstract = {It is essential for safety-critical applications of deep neural networks to determine when new inputs are significantly different from the training distribution. In this paper, we explore this out-of-distribution (OOD) detection problem for image classification using clusters of semantically similar embeddings of the training data and exploit the differences in distance relationships to these clusters between in- and out-of-distribution data. We study the structure and separation of clusters in the embedding space and find that the supervised contrastive learning leads to well separated clusters while its self-supervised counterpart fails to do so. In our extensive analysis of different training methods, clustering strategies, distance metrics and thresholding approaches, we observe that there is no clear winner. The optimal approach depends on the model architecture and selected datasets for in- and out-of-distribution. While we could reproduce the outstanding results for contrastive training on CIFAR-10 as in-distribution data, we find standard cross-entropy paired with cosine similarity outperforms all contrastive training methods when training on CIFAR-100 instead. Cross-entropy provides competitive results as compared to expensive contrastive training methods.},
 bibtype = {article},
 author = {Sinhamahapatra, Poulami and Koner, Rajat and Roscher, Karsten and Günnemann, Stephan},
 journal = {CEUR Workshop Proceedings}
}

It is essential for safety-critical applications of deep neural networks to determine when new inputs are significantly different from the training distribution. In this paper, we explore this out-of-distribution (OOD) detection problem for image classification using clusters of semantically similar embeddings of the training data and exploit the differences in distance relationships to these clusters between in- and out-of-distribution data. We study the structure and separation of clusters in the embedding space and find that the supervised contrastive learning leads to well separated clusters while its self-supervised counterpart fails to do so. In our extensive analysis of different training methods, clustering strategies, distance metrics and thresholding approaches, we observe that there is no clear winner. The optimal approach depends on the model architecture and selected datasets for in- and out-of-distribution. While we could reproduce the outstanding results for contrastive training on CIFAR-10 as in-distribution data, we find standard cross-entropy paired with cosine similarity outperforms all contrastive training methods when training on CIFAR-100 instead. Cross-entropy provides competitive results as compared to expensive contrastive training methods.

@article{
 title = {Semantic Novelty Detection via Relational Reasoning},
 type = {article},
 year = {2022},
 keywords = {Domain generalization,Novelty detection,Open set learning,Relational reasoning,Representation learning},
 pages = {183-200},
 volume = {13685 LNCS},
 id = {b833397b-81a2-3494-9ed8-83832f4cc92c},
 created = {2023-04-21T09:42:40.792Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-21T09:43:07.755Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4b11b59b-846d-45b5-82b7-81d8bc3108b2},
 private_publication = {false},
 abstract = {Semantic novelty detection aims at discovering unknown categories in the test data. This task is particularly relevant in safety-critical applications, such as autonomous driving or healthcare, where it is crucial to recognize unknown objects at deployment time and issue a warning to the user accordingly. Despite the impressive advancements of deep learning research, existing models still need a finetuning stage on the known categories in order to recognize the unknown ones. This could be prohibitive when privacy rules limit data access, or in case of strict memory and computational constraints (e.g. edge computing). We claim that a tailored representation learning strategy may be the right solution for effective and efficient semantic novelty detection. Besides extensively testing state-of-the-art approaches for this task, we propose a novel representation learning paradigm based on relational reasoning. It focuses on learning how to measure semantic similarity rather than recognizing known categories. Our experiments show that this knowledge is directly transferable to a wide range of scenarios, and it can be exploited as a plug-and-play module to convert closed-set recognition models into reliable open-set ones.},
 bibtype = {article},
 author = {Borlino, Francesco Cappio and Bucci, Silvia and Tommasi, Tatiana},
 doi = {10.1007/978-3-031-19806-9_11},
 journal = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}
}

Semantic novelty detection aims at discovering unknown categories in the test data. This task is particularly relevant in safety-critical applications, such as autonomous driving or healthcare, where it is crucial to recognize unknown objects at deployment time and issue a warning to the user accordingly. Despite the impressive advancements of deep learning research, existing models still need a finetuning stage on the known categories in order to recognize the unknown ones. This could be prohibitive when privacy rules limit data access, or in case of strict memory and computational constraints (e.g. edge computing). We claim that a tailored representation learning strategy may be the right solution for effective and efficient semantic novelty detection. Besides extensively testing state-of-the-art approaches for this task, we propose a novel representation learning paradigm based on relational reasoning. It focuses on learning how to measure semantic similarity rather than recognizing known categories. Our experiments show that this knowledge is directly transferable to a wide range of scenarios, and it can be exploited as a plug-and-play module to convert closed-set recognition models into reliable open-set ones.

@article{
 title = {A System-Level View on Out-of-Distribution Data in Robotics},
 type = {article},
 year = {2022},
 websites = {http://arxiv.org/abs/2212.14020},
 id = {da78495d-92bd-3f08-8f5e-f88647dd0023},
 created = {2023-04-21T09:42:40.862Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-24T08:55:32.972Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4b11b59b-846d-45b5-82b7-81d8bc3108b2},
 private_publication = {false},
 abstract = {When testing conditions differ from those represented in training data, so-called out-of-distribution (OOD) inputs can mar the reliability of black-box learned components in the modern robot autonomy stack. Therefore, coping with OOD data is an important challenge on the path towards trustworthy learning-enabled open-world autonomy. In this paper, we aim to demystify the topic of OOD data and its associated challenges in the context of data-driven robotic systems, drawing connections to emerging paradigms in the ML community that study the effect of OOD data on learned models in isolation. We argue that as roboticists, we should reason about the overall system-level competence of a robot as it performs tasks in OOD conditions. We highlight key research questions around this system-level view of OOD problems to guide future research toward safe and reliable learning-enabled autonomy.},
 bibtype = {article},
 author = {Sinha, Rohan and Sharma, Apoorva and Banerjee, Somrita and Lew, Thomas and Luo, Rachel and Richards, Spencer M. and Sun, Yixiao and Schmerling, Edward and Pavone, Marco}
}

When testing conditions differ from those represented in training data, so-called out-of-distribution (OOD) inputs can mar the reliability of black-box learned components in the modern robot autonomy stack. Therefore, coping with OOD data is an important challenge on the path towards trustworthy learning-enabled open-world autonomy. In this paper, we aim to demystify the topic of OOD data and its associated challenges in the context of data-driven robotic systems, drawing connections to emerging paradigms in the ML community that study the effect of OOD data on learned models in isolation. We argue that as roboticists, we should reason about the overall system-level competence of a robot as it performs tasks in OOD conditions. We highlight key research questions around this system-level view of OOD problems to guide future research toward safe and reliable learning-enabled autonomy.

@article{
 title = {Data Invariants to Understand Unsupervised Out-of-Distribution Detection},
 type = {article},
 year = {2022},
 keywords = {Out-of-distribution detection,Unsupervised learning},
 pages = {133-150},
 volume = {13691 LNCS},
 id = {4761f0c5-8abe-33a0-920e-6867fc08f747},
 created = {2023-04-21T09:42:41.024Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-25T10:23:33.179Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4b11b59b-846d-45b5-82b7-81d8bc3108b2},
 private_publication = {false},
 abstract = {Unsupervised out-of-distribution (U-OOD) detection has recently attracted much attention due to its importance in mission-critical systems and broader applicability over its supervised counterpart. Despite this increased attention, U-OOD methods suffer from important shortcomings. By performing a large-scale evaluation on different benchmarks and image modalities, we show in this work that most popular state-of-the-art methods are unable to consistently outperform a simple anomaly detector based on pre-trained features and the Mahalanobis distance (MahaAD). A key reason for the inconsistencies of these methods is the lack of a formal description of U-OOD. Motivated by a simple thought experiment, we propose a characterization of U-OOD based on the invariants of the training dataset. We show how this characterization is unknowingly embodied in the top-scoring MahaAD method, thereby explaining its quality. Furthermore, our approach can be used to interpret predictions of U-OOD detectors and provides insights into good practices for evaluating future U-OOD methods.},
 bibtype = {article},
 author = {Doorenbos, Lars and Sznitman, Raphael and Márquez-Neila, Pablo},
 doi = {10.1007/978-3-031-19821-2_8},
 journal = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}
}

Unsupervised out-of-distribution (U-OOD) detection has recently attracted much attention due to its importance in mission-critical systems and broader applicability over its supervised counterpart. Despite this increased attention, U-OOD methods suffer from important shortcomings. By performing a large-scale evaluation on different benchmarks and image modalities, we show in this work that most popular state-of-the-art methods are unable to consistently outperform a simple anomaly detector based on pre-trained features and the Mahalanobis distance (MahaAD). A key reason for the inconsistencies of these methods is the lack of a formal description of U-OOD. Motivated by a simple thought experiment, we propose a characterization of U-OOD based on the invariants of the training dataset. We show how this characterization is unknowingly embodied in the top-scoring MahaAD method, thereby explaining its quality. Furthermore, our approach can be used to interpret predictions of U-OOD detectors and provides insights into good practices for evaluating future U-OOD methods.

@article{
 title = {Detection of Anomalous Grapevine Berries Using Variational Autoencoders},
 type = {article},
 year = {2022},
 keywords = {miranda2022detectionanomalousgrapevine},
 pages = {1483},
 volume = {13},
 month = {6},
 publisher = {Frontiers Media S.A.},
 day = {1},
 id = {83084a4b-f81b-3d2e-933d-0e8982bbc211},
 created = {2023-06-06T06:58:16.296Z},
 accessed = {2023-06-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:03.510Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,6b565182-74c4-44fd-98cc-10618152e2ae,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Grapevine is one of the economically most important quality crops. The monitoring of the plant performance during the growth period is, therefore, important to ensure a high quality end-product. This includes the observation, detection, and respective reduction of unhealthy berries (physically damaged, or diseased). At harvest, it is not necessary to know the exact cause of the damage, but rather if the damage is apparent or not. Since a manual screening and selection before harvest is time-consuming and expensive, we propose an automatic, image-based machine learning approach, which can lead observers directly to anomalous areas without the need to monitor every plant manually. Specifically, we train a fully convolutional variational autoencoder with a feature perceptual loss on images with healthy berries only and consider image areas with deviations from this model as damaged berries. We use heatmaps which visualize the results of the trained neural network and, therefore, support the decision making for farmers. We compare our method against a convolutional autoencoder that was successfully applied to a similar task and show that our approach outperforms it.},
 bibtype = {article},
 author = {Miranda, Miro and Zabawa, Laura and Kicherer, Anna and Strothmann, Laurenz and Rascher, Uwe and Roscher, Ribana},
 doi = {10.3389/FPLS.2022.729097/BIBTEX},
 journal = {Frontiers in Plant Science}
}

Grapevine is one of the economically most important quality crops. The monitoring of the plant performance during the growth period is, therefore, important to ensure a high quality end-product. This includes the observation, detection, and respective reduction of unhealthy berries (physically damaged, or diseased). At harvest, it is not necessary to know the exact cause of the damage, but rather if the damage is apparent or not. Since a manual screening and selection before harvest is time-consuming and expensive, we propose an automatic, image-based machine learning approach, which can lead observers directly to anomalous areas without the need to monitor every plant manually. Specifically, we train a fully convolutional variational autoencoder with a feature perceptual loss on images with healthy berries only and consider image areas with deviations from this model as damaged berries. We use heatmaps which visualize the results of the trained neural network and, therefore, support the decision making for farmers. We compare our method against a convolutional autoencoder that was successfully applied to a similar task and show that our approach outperforms it.

@article{
 title = {Anomalib: a Deep Learning Library for Anomaly Detection},
 type = {article},
 year = {2022},
 keywords = {Unsupervised Anomaly detection,localization},
 pages = {1706-1710},
 id = {46e22201-5e91-3c56-989c-0bbf7a35ec27},
 created = {2023-06-06T07:01:36.423Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-27T12:11:37.203Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,4b11b59b-846d-45b5-82b7-81d8bc3108b2},
 private_publication = {false},
 abstract = {This paper introduces anomalib, a novel library for unsupervised anomaly detection and localization. With reproducibility and modularity in mind, this open-source library provides algorithms from the literature and a set of tools to design custom anomaly detection algorithms via a plug-and-play approach. Anomalib comprises state-of-the-art anomaly detection algorithms that achieve top performance on the benchmarks and that can be used off-the-shelf. In addition, the library provides components to design custom algorithms that could be tailored towards specific needs. Additional tools, including experiment trackers, visualizers, and hyperparameter optimizers, make it simple to design and implement anomaly detection models. The library also supports OpenVINO model-optimization and quantization for real-time deployment. Overall, anomalib is an extensive library for the design, implementation, and deployment of unsupervised anomaly detection models from data to the edge.},
 bibtype = {article},
 author = {Akcay, Samet and Ameln, Dick and Vaidya, Ashwin and Lakshmanan, Barath and Ahuja, Nilesh and Genc, Utku},
 doi = {10.1109/ICIP46576.2022.9897283},
 journal = {Proceedings - International Conference on Image Processing, ICIP}
}

This paper introduces anomalib, a novel library for unsupervised anomaly detection and localization. With reproducibility and modularity in mind, this open-source library provides algorithms from the literature and a set of tools to design custom anomaly detection algorithms via a plug-and-play approach. Anomalib comprises state-of-the-art anomaly detection algorithms that achieve top performance on the benchmarks and that can be used off-the-shelf. In addition, the library provides components to design custom algorithms that could be tailored towards specific needs. Additional tools, including experiment trackers, visualizers, and hyperparameter optimizers, make it simple to design and implement anomaly detection models. The library also supports OpenVINO model-optimization and quantization for real-time deployment. Overall, anomalib is an extensive library for the design, implementation, and deployment of unsupervised anomaly detection models from data to the edge.

@article{
 title = {Anomaly Detection via Reverse Distillation from One-Class Embedding},
 type = {article},
 year = {2022},
 keywords = {Self- & semi- & meta- Representation learning,Transfer/low-shot/long-tail learning},
 pages = {9727-9736},
 volume = {2022-June},
 websites = {https://arxiv.org/abs/2201.10703v2},
 month = {1},
 publisher = {IEEE Computer Society},
 day = {26},
 id = {b511a525-0139-31cb-a9a2-8146a34ec8f7},
 created = {2023-06-12T12:12:58.942Z},
 accessed = {2023-06-12},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-26T09:18:51.773Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,4b11b59b-846d-45b5-82b7-81d8bc3108b2,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Knowledge distillation (KD) achieves promising results on the challenging
problem of unsupervised anomaly detection (AD).The representation discrepancy
of anomalies in the teacher-student (T-S) model provides essential evidence for
AD. However, using similar or identical architectures to build the teacher and
student models in previous studies hinders the diversity of anomalous
representations. To tackle this problem, we propose a novel T-S model
consisting of a teacher encoder and a student decoder and introduce a simple
yet effective "reverse distillation" paradigm accordingly. Instead of receiving
raw images directly, the student network takes teacher model's one-class
embedding as input and targets to restore the teacher's multiscale
representations. Inherently, knowledge distillation in this study starts from
abstract, high-level presentations to low-level features. In addition, we
introduce a trainable one-class bottleneck embedding (OCBE) module in our T-S
model. The obtained compact embedding effectively preserves essential
information on normal patterns, but abandons anomaly perturbations. Extensive
experimentation on AD and one-class novelty detection benchmarks shows that our
method surpasses SOTA performance, demonstrating our proposed approach's
effectiveness and generalizability.},
 bibtype = {article},
 author = {Deng, Hanqiu and Li, Xingyu},
 doi = {10.1109/CVPR52688.2022.00951},
 journal = {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition}
}

Knowledge distillation (KD) achieves promising results on the challenging problem of unsupervised anomaly detection (AD).The representation discrepancy of anomalies in the teacher-student (T-S) model provides essential evidence for AD. However, using similar or identical architectures to build the teacher and student models in previous studies hinders the diversity of anomalous representations. To tackle this problem, we propose a novel T-S model consisting of a teacher encoder and a student decoder and introduce a simple yet effective "reverse distillation" paradigm accordingly. Instead of receiving raw images directly, the student network takes teacher model's one-class embedding as input and targets to restore the teacher's multiscale representations. Inherently, knowledge distillation in this study starts from abstract, high-level presentations to low-level features. In addition, we introduce a trainable one-class bottleneck embedding (OCBE) module in our T-S model. The obtained compact embedding effectively preserves essential information on normal patterns, but abandons anomaly perturbations. Extensive experimentation on AD and one-class novelty detection benchmarks shows that our method surpasses SOTA performance, demonstrating our proposed approach's effectiveness and generalizability.

@article{
 title = {Real-time detection and identification of plant leaf diseases using convolutional neural networks on an embedded platform},
 type = {article},
 year = {2022},
 keywords = {Convolutional neural network,Crop diseases identification,Embedded Platform,Machine Learning,Precision Agriculture,Real-time detection},
 pages = {2923-2938},
 volume = {38},
 websites = {https://link.springer.com/article/10.1007/s00371-021-02164-9},
 month = {8},
 publisher = {Springer Science and Business Media Deutschland GmbH},
 day = {1},
 id = {d58cc36d-0f8f-39bd-b57b-4e052bf6cb66},
 created = {2023-08-07T08:03:32.301Z},
 accessed = {2023-08-07},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-08-07T08:03:35.726Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Early identification of crop disease can aid the farmers to take timely precautions and countermeasures for its removal. In this paper, a real-time system to identify the type of disease present in a crop based on leaf images using machine learning is proposed. A deep convolutional neural network architecture is proposed to classify the crop disease, and a single shot detector is used for identification and localization of the leaf. These models are deployed on an embedded hardware, Nvidia Jetson TX1, for real-time in-field plant disease detection and identification. The disease classification accuracy achieved is around 96.88%, and the classification results are compared with existing convolutional neural network architectures. Also, the high success rate of the proposed system in the actual field test makes the proposed system a completely deployable system.},
 bibtype = {article},
 author = {Gajjar, Ruchi and Gajjar, Nagendra and Thakor, Vaibhavkumar Jigneshkumar and Patel, Nikhilkumar Pareshbhai and Ruparelia, Stavan},
 doi = {10.1007/S00371-021-02164-9/FIGURES/12},
 journal = {Visual Computer},
 number = {8}
}

Early identification of crop disease can aid the farmers to take timely precautions and countermeasures for its removal. In this paper, a real-time system to identify the type of disease present in a crop based on leaf images using machine learning is proposed. A deep convolutional neural network architecture is proposed to classify the crop disease, and a single shot detector is used for identification and localization of the leaf. These models are deployed on an embedded hardware, Nvidia Jetson TX1, for real-time in-field plant disease detection and identification. The disease classification accuracy achieved is around 96.88%, and the classification results are compared with existing convolutional neural network architectures. Also, the high success rate of the proposed system in the actual field test makes the proposed system a completely deployable system.

@article{
 title = {ResGANet: Residual group attention network for medical image classification and segmentation},
 type = {article},
 year = {2022},
 keywords = {Deep learning,Image classification,Image segmentation,Medical image analysis,Residual group attention network},
 pages = {102313},
 volume = {76},
 websites = {https://doi.org/10.1016/j.media.2021.102313},
 publisher = {Elsevier B.V.},
 id = {76835e5f-acd2-3de0-bf20-e0389f4bd4b7},
 created = {2023-09-13T11:36:00.962Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-09T07:27:31.283Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126},
 private_publication = {false},
 abstract = {In recent years, deep learning technology has shown superior performance in different fields of medical image analysis. Some deep learning architectures have been proposed and used for computational pathology classification, segmentation, and detection tasks. Due to their simple, modular structure, most downstream applications still use ResNet and its variants as the backbone network. This paper proposes a modular group attention block that can capture feature dependencies in medical images in two independent dimensions: channel and space. By stacking these group attention blocks in ResNet-style, we obtain a new ResNet variant called ResGANet. The stacked ResGANet architecture has 1.51–3.47 times fewer parameters than the original ResNet and can be directly used for downstream medical image segmentation tasks. Many experiments show that the proposed ResGANet is superior to state-of-the-art backbone models in medical image classification tasks. Applying it to different segmentation networks can improve the baseline model in medical image segmentation tasks without changing the network architecture. We hope that this work provides a promising method for enhancing the feature representation of convolutional neural networks (CNNs) in the future.},
 bibtype = {article},
 author = {Cheng, Junlong and Tian, Shengwei and Yu, Long and Gao, Chengrui and Kang, Xiaojing and Ma, Xiang and Wu, Weidong and Liu, Shijia and Lu, Hongchun},
 doi = {10.1016/j.media.2021.102313},
 journal = {Medical Image Analysis}
}

In recent years, deep learning technology has shown superior performance in different fields of medical image analysis. Some deep learning architectures have been proposed and used for computational pathology classification, segmentation, and detection tasks. Due to their simple, modular structure, most downstream applications still use ResNet and its variants as the backbone network. This paper proposes a modular group attention block that can capture feature dependencies in medical images in two independent dimensions: channel and space. By stacking these group attention blocks in ResNet-style, we obtain a new ResNet variant called ResGANet. The stacked ResGANet architecture has 1.51–3.47 times fewer parameters than the original ResNet and can be directly used for downstream medical image segmentation tasks. Many experiments show that the proposed ResGANet is superior to state-of-the-art backbone models in medical image classification tasks. Applying it to different segmentation networks can improve the baseline model in medical image segmentation tasks without changing the network architecture. We hope that this work provides a promising method for enhancing the feature representation of convolutional neural networks (CNNs) in the future.

@article{
 title = {A 3D grape bunch reconstruction pipeline based on constraint-based optimisation and restricted reconstruction grammar},
 type = {article},
 year = {2022},
 keywords = {xin20223dgrapebunchreconstruction},
 pages = {106840},
 volume = {196},
 month = {5},
 publisher = {Elsevier},
 day = {1},
 id = {822faa78-2d90-37be-a667-353429b1c9aa},
 created = {2023-10-27T06:47:29.647Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:25.087Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Phenotypic traits of grapevines are known to be closely related to grapevine yield, wine flavour and sensitivity to disease. Traditional phenotyping methods based on manual measurements face the bottleneck of intensely repetitive and time consuming measurement. As viticulturists turning their focus to the 3D automatic phenotyping domain, existing work on grape bunch phenotyping indicates deficiencies in incomplete reconstruction information, poor element coincidence with the ground truth and poor performance under field conditions. To this end, the proposed work introduces a novel reconstruction pipeline by dividing it into sub-problems of visible-berry-related reconstruction and invisible element prediction. By taking a 2D image of the target bunch as the only sensor input, visible berries are detected using image processing algorithms. With the detected berry information, their morphological positions are predicted, from which internodes that are associated with the detected berries are derived. Parameters of derived internodes are then estimated employing constraint-based optimisation theory, from which the visible-berry-related reconstruction is able to be achieved. Invisible element prediction is then conducted by filling elements according to a Restricted Reconstruction Grammar (RRG). A fully reconstructed bunch model is finally presented. Compared with existing work, the proposed grape bunch reconstruction pipeline achieved an improvement in quantity estimation of rachis internodes, tertiary internodes and pedicels, whose percentage errors were indicated as 21.2, 42.2 and 31.2% respectively. A better performance was also revealed in length estimations of secondary internodes and pedicels with percentage errors of 3.5 and 0.5%. This may largely facilitate related studies on disease control of grape bunches since internode numbers and lengths are closely related to bunch compactness which is the indicator of disease sensitivity. Especially, the proposed reconstruction pipeline shows an promising improvement in element coincidence with F1 scores of 0.90, 0.77, 0.45, 0.43 for respective element types. Knowing that element coincidence may influence the inner space utilisation of a bunch, the proposed work provides a better option for 3D grape bunch phenotyping.},
 bibtype = {article},
 author = {Xin, Bolai and Whitty, Mark},
 doi = {10.1016/J.COMPAG.2022.106840},
 journal = {Computers and Electronics in Agriculture}
}

Phenotypic traits of grapevines are known to be closely related to grapevine yield, wine flavour and sensitivity to disease. Traditional phenotyping methods based on manual measurements face the bottleneck of intensely repetitive and time consuming measurement. As viticulturists turning their focus to the 3D automatic phenotyping domain, existing work on grape bunch phenotyping indicates deficiencies in incomplete reconstruction information, poor element coincidence with the ground truth and poor performance under field conditions. To this end, the proposed work introduces a novel reconstruction pipeline by dividing it into sub-problems of visible-berry-related reconstruction and invisible element prediction. By taking a 2D image of the target bunch as the only sensor input, visible berries are detected using image processing algorithms. With the detected berry information, their morphological positions are predicted, from which internodes that are associated with the detected berries are derived. Parameters of derived internodes are then estimated employing constraint-based optimisation theory, from which the visible-berry-related reconstruction is able to be achieved. Invisible element prediction is then conducted by filling elements according to a Restricted Reconstruction Grammar (RRG). A fully reconstructed bunch model is finally presented. Compared with existing work, the proposed grape bunch reconstruction pipeline achieved an improvement in quantity estimation of rachis internodes, tertiary internodes and pedicels, whose percentage errors were indicated as 21.2, 42.2 and 31.2% respectively. A better performance was also revealed in length estimations of secondary internodes and pedicels with percentage errors of 3.5 and 0.5%. This may largely facilitate related studies on disease control of grape bunches since internode numbers and lengths are closely related to bunch compactness which is the indicator of disease sensitivity. Especially, the proposed reconstruction pipeline shows an promising improvement in element coincidence with F1 scores of 0.90, 0.77, 0.45, 0.43 for respective element types. Knowing that element coincidence may influence the inner space utilisation of a bunch, the proposed work provides a better option for 3D grape bunch phenotyping.

@article{
 title = {An end-to-end lightweight model for grape and picking point simultaneous detection},
 type = {article},
 year = {2022},
 keywords = {zhao2022endtoendlightweightmodel},
 pages = {174-188},
 volume = {223},
 month = {11},
 publisher = {Academic Press},
 day = {1},
 id = {ca68fd98-ee1e-3da2-9c52-dec0c57e2b8b},
 created = {2023-10-27T06:48:43.273Z},
 accessed = {2023-10-27},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:25.583Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Grape clusters and their picking point detection (GCPPD) are pivotal in the visual tasks of automatic grape harvesting. In recent years, much progress has been made in increasing the accuracy of GCPPD based on deep learning models. However, GCPPD still has many problems. First, it is inevitable that grape cluster detection requires complex models with many parameters. Second, the prior work on picking point detection can be summarised as the image processing methods using predefined hand-crafted features. This leads to a lack of robustness in the proposed algorithms. To address this, a scheme for the simultaneous detection of grape clusters and their picking points is explored. Due to the superiority of simultaneous detection, the model is constructed as an end-to-end network. Thus, a lightweight end-to-end model called YOLO-GP (YOLO-Grape and Picking points) is proposed. Specifically, YOLO-GP utilises a ghost bottleneck to reduce model parameters. Additionally, this model adds the prediction of picking points using the novel idea, that the picking point follows the bounding box. The Grape-PP (Grape-Picking Point) dataset for model training is constructed, which contains 360 grape images with 4517 grape cluster bounding boxes and picking points. The experiments show that the mean Average Precision (mAP) of grape cluster detection by YOLO-GP is 93.27% with a decrease in the number of weight parameters by at least 10%. The distance error of picking point detection is less than 40 pixels. In summary, YOLO-GP achieves the simultaneous detection of grape clusters and their picking points, and its performance is comparable to that of baseline models.},
 bibtype = {article},
 author = {Zhao, Ruzhun and Zhu, Yuchang and Li, Yuanhong},
 doi = {10.1016/J.BIOSYSTEMSENG.2022.08.013},
 journal = {Biosystems Engineering}
}

Grape clusters and their picking point detection (GCPPD) are pivotal in the visual tasks of automatic grape harvesting. In recent years, much progress has been made in increasing the accuracy of GCPPD based on deep learning models. However, GCPPD still has many problems. First, it is inevitable that grape cluster detection requires complex models with many parameters. Second, the prior work on picking point detection can be summarised as the image processing methods using predefined hand-crafted features. This leads to a lack of robustness in the proposed algorithms. To address this, a scheme for the simultaneous detection of grape clusters and their picking points is explored. Due to the superiority of simultaneous detection, the model is constructed as an end-to-end network. Thus, a lightweight end-to-end model called YOLO-GP (YOLO-Grape and Picking points) is proposed. Specifically, YOLO-GP utilises a ghost bottleneck to reduce model parameters. Additionally, this model adds the prediction of picking points using the novel idea, that the picking point follows the bounding box. The Grape-PP (Grape-Picking Point) dataset for model training is constructed, which contains 360 grape images with 4517 grape cluster bounding boxes and picking points. The experiments show that the mean Average Precision (mAP) of grape cluster detection by YOLO-GP is 93.27% with a decrease in the number of weight parameters by at least 10%. The distance error of picking point detection is less than 40 pixels. In summary, YOLO-GP achieves the simultaneous detection of grape clusters and their picking points, and its performance is comparable to that of baseline models.

@article{
 title = {AUTOMATED LAG PHASE DETECTION IN WINE GRAPES},
 type = {article},
 year = {2022},
 keywords = {upadhyaya2022automatedlagphase},
 pages = {1-15},
 websites = {https://ageconsearch.umn.edu/record/322755?ln=en},
 id = {82bae107-c6b7-31d2-ad95-c8f83581031d},
 created = {2023-10-27T09:13:02.703Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:29.299Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 bibtype = {article},
 author = {Priyanka Upadhyaya, undefined and Karkee, Manoj and Kshetri, Safal and Zhang, Xin},
 number = {2018}
}

@article{
 title = {Pr rin t no t p ee r r ep rin t no t p ee r r},
 type = {article},
 year = {2022},
 id = {0ec236cc-8691-345c-b063-75856c45b6e3},
 created = {2023-10-27T09:37:09.872Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:29.694Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 bibtype = {article},
 author = {Ogayar, Carlos J and Feito, Francisco R and Sousa, Joaquim J}
}

@article{
 title = {Leaf Area Index Estimation of Pergola-Trained Vineyards in Arid Regions Based on UAV RGB and Multispectral Data Using Machine Learning Methods},
 type = {article},
 year = {2022},
 keywords = {ilniyaz2022leafareaindexestimation},
 pages = {415},
 volume = {14},
 websites = {https://www.mdpi.com/2072-4292/14/2/415/htm,https://www.mdpi.com/2072-4292/14/2/415},
 month = {1},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {17},
 id = {bb14a51a-f59c-31bb-a645-2f8dc73df657},
 created = {2023-10-28T17:34:20.354Z},
 accessed = {2023-10-28},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:32.042Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The leaf area index (LAI), a valuable variable for assessing vine vigor, reflects nutrient concentrations in vineyards and assists in precise management, including fertilization, improving yield, quality, and vineyard uniformity. Although some vegetation indices (VIs) have been successfully used to assess LAI variations, they are unsuitable for vineyards of different types and structures. By calibrating the light extinction coefficient of a digital photography algorithm for proximal LAI measurements, this study aimed to develop VI-LAI models for pergola-trained vineyards based on high-resolution RGB and multispectral images captured by an unmanned aerial vehicle (UAV). The models were developed by comparing five machine learning (ML) methods, and a robust ensemble model was proposed using the five models as base learners. The results showed that the ensemble model outperformed the base models. The highest R2 and lowest RMSE values that were obtained using the best combination of VIs with multispectral data were 0.899 and 0.434, respectively; those obtained using the RGB data were 0.825 and 0.547, respectively. By improving the results by feature selection, ML methods performed better with multispectral data than with RGB images, and better with higher spatial resolution data than with lower resolution data. LAI variations can be monitored efficiently and accurately for large areas of pergola-trained vineyards using this framework.},
 bibtype = {article},
 author = {Ilniyaz, Osman and Kurban, Alishir and Du, Qingyun},
 doi = {10.3390/RS14020415},
 journal = {Remote Sensing 2022, Vol. 14, Page 415},
 number = {2}
}

The leaf area index (LAI), a valuable variable for assessing vine vigor, reflects nutrient concentrations in vineyards and assists in precise management, including fertilization, improving yield, quality, and vineyard uniformity. Although some vegetation indices (VIs) have been successfully used to assess LAI variations, they are unsuitable for vineyards of different types and structures. By calibrating the light extinction coefficient of a digital photography algorithm for proximal LAI measurements, this study aimed to develop VI-LAI models for pergola-trained vineyards based on high-resolution RGB and multispectral images captured by an unmanned aerial vehicle (UAV). The models were developed by comparing five machine learning (ML) methods, and a robust ensemble model was proposed using the five models as base learners. The results showed that the ensemble model outperformed the base models. The highest R2 and lowest RMSE values that were obtained using the best combination of VIs with multispectral data were 0.899 and 0.434, respectively; those obtained using the RGB data were 0.825 and 0.547, respectively. By improving the results by feature selection, ML methods performed better with multispectral data than with RGB images, and better with higher spatial resolution data than with lower resolution data. LAI variations can be monitored efficiently and accurately for large areas of pergola-trained vineyards using this framework.

@article{
 title = {GRAPE QUALITY PREDICTION WITH PRE-POST HARVESTING USING FUSION DEEP LEARNING},
 type = {article},
 year = {2022},
 keywords = {patil2022grapequalityprediction},
 volume = {11},
 id = {e3fa5eee-b947-3f57-850c-8e093962086d},
 created = {2023-10-28T17:48:34.998Z},
 accessed = {2023-10-28},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-06T09:35:32.192Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {In modern society, the agriculture industry is crucial. Agriculture industry has to support the load of increasing population, year by year. Implementing robotics in agriculture can heavily rely on computer vision and perception techniques. Machine learning is very useful for object recognition and picture categorization. In this work, machine learning was used to identify grape bunches in vineyards at various phases of development, including the early stage right after bloom and the middle stage, when the grape bunches display an intermediate degree of development. The dataset is suggested in this investigation because the training inputs are not freely available. So we made a dataset. Separate metrics used to benchmark and explain the models. The results revealed that the built models successfully detect grape clusters from pictures. The suggested system's performance was adequate when considering the approach's minimal resource utilisation, cheap cost, and minimum power hardware device requirements, which allow for simpler models.},
 bibtype = {article},
 author = {Patil, Nisha and Bhise, Archana and Kumar Tiwari, Rajesh and Scholar, Research},
 journal = {I) Journal}
}

In modern society, the agriculture industry is crucial. Agriculture industry has to support the load of increasing population, year by year. Implementing robotics in agriculture can heavily rely on computer vision and perception techniques. Machine learning is very useful for object recognition and picture categorization. In this work, machine learning was used to identify grape bunches in vineyards at various phases of development, including the early stage right after bloom and the middle stage, when the grape bunches display an intermediate degree of development. The dataset is suggested in this investigation because the training inputs are not freely available. So we made a dataset. Separate metrics used to benchmark and explain the models. The results revealed that the built models successfully detect grape clusters from pictures. The suggested system's performance was adequate when considering the approach's minimal resource utilisation, cheap cost, and minimum power hardware device requirements, which allow for simpler models.

@article{
 title = {Grapevine Plant Image Dataset for Pruning},
 type = {article},
 year = {2022},
 keywords = {computer vision,grapevine pruning,image processing},
 pages = {110},
 volume = {7},
 websites = {https://www.mdpi.com/2306-5729/7/8/110/htm,https://www.mdpi.com/2306-5729/7/8/110},
 month = {8},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {9},
 id = {daf6bba0-d204-3b04-b403-11da4abb3952},
 created = {2023-11-06T10:19:37.777Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T06:17:49.955Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Grapevine pruning is conducted during winter, and it is a very important and expensive task for wine producers managing their vineyard. During grapevine pruning every year, the past year’s canes should be removed and should provide the possibility for new canes to grow and produce grapes. It is a difficult procedure, and it is not yet fully automated. However, some attempts have been made by the research community. Based on the literature, grapevine pruning automation is approximated with the help of computer vision and image processing methods. Despite the attempts that have been made to automate grapevine pruning, the task remains hard for the abovementioned domains. The reason for this is that several challenges such as cane overlapping or complex backgrounds appear. Additionally, there is no public image dataset for this problem which makes it difficult for the research community to approach it. Motivated by the above facts, an image dataset is proposed for grapevine canes’ segmentation for a pruning task. An experimental analysis is also conducted in the proposed dataset, achieving a 67% IoU and 78% F1 score in grapevine cane semantic segmentation with the U-net model.},
 bibtype = {article},
 author = {Apostolidis, Kyriakos D. and Kalampokas, Theofanis and Pachidis, Theodore P. and Kaburlasos, Vassilis G.},
 doi = {10.3390/DATA7080110},
 journal = {Data 2022, Vol. 7, Page 110},
 number = {8}
}

Grapevine pruning is conducted during winter, and it is a very important and expensive task for wine producers managing their vineyard. During grapevine pruning every year, the past year’s canes should be removed and should provide the possibility for new canes to grow and produce grapes. It is a difficult procedure, and it is not yet fully automated. However, some attempts have been made by the research community. Based on the literature, grapevine pruning automation is approximated with the help of computer vision and image processing methods. Despite the attempts that have been made to automate grapevine pruning, the task remains hard for the abovementioned domains. The reason for this is that several challenges such as cane overlapping or complex backgrounds appear. Additionally, there is no public image dataset for this problem which makes it difficult for the research community to approach it. Motivated by the above facts, an image dataset is proposed for grapevine canes’ segmentation for a pruning task. An experimental analysis is also conducted in the proposed dataset, achieving a 67% IoU and 78% F1 score in grapevine cane semantic segmentation with the U-net model.

@article{
 title = {Assessment of downy mildew in grapevine using computer vision and fuzzy logic. Development and validation of a new method},
 type = {article},
 year = {2022},
 keywords = {Plasmopara viticola,non-invasive sensing technologies,plant disease detection,precision viticulture},
 pages = {41-53},
 volume = {56},
 month = {6},
 publisher = {International Viticulture and Enology Society},
 day = {24},
 id = {7b0d9ad8-7540-35b3-b652-34d04c1a5770},
 created = {2023-11-06T10:33:26.499Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T06:17:49.802Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Downy mildew is a major disease of grapevine. Conventional methods for assessing crop diseases are time-consuming and require trained personnel. This work aimed to develop and validate a new method to automatically estimate the severity of downy mildew in grapevine leaves using fuzzy logic and computer vision techniques. Leaf discs of two grapevine varieties were inoculated with Plasmopara viticola and subsequently, RGB images were acquired under indoor conditions. Computer vision techniques were applied for leaf disc location in Petri dishes, image pre-processing and segmentation of pre-processed disc images to separate the pixels representing downy mildew sporulation from the rest of the leaf. Fuzzy logic was applied to improve the segmentation of disc images, rating pixels with a degree of infection according to the intensity of sporulation. To validate the new method, the downy mildew severity was visually evaluated by eleven experts and averaged score was used as the reference value. A coefficient of determination (R2) of 0.87 and a root mean squared error (RMSE) of 7.61 % was observed between the downy mildew severity obtained by the new method and the visual assessment values. Classification of the severity of the infection into three levels was also attempted, achieving an accuracy of 86 % and an F1 score of 0.78. These results indicate that computer vision and fuzzy logic can be used to automatically estimate the severity of downy mildew in grapevine leaves. A new method has been developed and validated to assess the severity of downy mildew in grapevine. The new method can be adapted to assess the severity of other diseases and crops in agriculture.},
 bibtype = {article},
 author = {Hernández, Inés and Gutiérrez, Salvador and Ceballos, Sara and Palacios, Fernando and Toffolatti, Silvia L. and Maddalena, Giuliana and Diago, María P. and Tardaguila, Javier},
 doi = {10.20870/oeno-one.2022.56.3.5359},
 journal = {Oeno One},
 number = {3}
}

Downy mildew is a major disease of grapevine. Conventional methods for assessing crop diseases are time-consuming and require trained personnel. This work aimed to develop and validate a new method to automatically estimate the severity of downy mildew in grapevine leaves using fuzzy logic and computer vision techniques. Leaf discs of two grapevine varieties were inoculated with Plasmopara viticola and subsequently, RGB images were acquired under indoor conditions. Computer vision techniques were applied for leaf disc location in Petri dishes, image pre-processing and segmentation of pre-processed disc images to separate the pixels representing downy mildew sporulation from the rest of the leaf. Fuzzy logic was applied to improve the segmentation of disc images, rating pixels with a degree of infection according to the intensity of sporulation. To validate the new method, the downy mildew severity was visually evaluated by eleven experts and averaged score was used as the reference value. A coefficient of determination (R2) of 0.87 and a root mean squared error (RMSE) of 7.61 % was observed between the downy mildew severity obtained by the new method and the visual assessment values. Classification of the severity of the infection into three levels was also attempted, achieving an accuracy of 86 % and an F1 score of 0.78. These results indicate that computer vision and fuzzy logic can be used to automatically estimate the severity of downy mildew in grapevine leaves. A new method has been developed and validated to assess the severity of downy mildew in grapevine. The new method can be adapted to assess the severity of other diseases and crops in agriculture.

@article{
 title = {Automatic detection of Flavescense Dorée grapevine disease in hyperspectral images using machine learning},
 type = {article},
 year = {2022},
 keywords = {Hyperspectral images,autoencoder,automatic detection,grapevine diseases,machine learning},
 pages = {125-132},
 volume = {196},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {fdd924cd-2629-3902-89e2-f162862dcc34},
 created = {2023-11-06T10:40:11.677Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T11:23:58.463Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The technological revolution that we have been witnessing recently has allowed components miniaturization and made electronic components accessible. Hyperspectral sensors benefited from these advances and could be mounted on unmanned aerial vehicles, which was unthinkable until recently. This fact significantly increased the applications of hyperspectral data, namely in agriculture, especially in the detection of diseases at an early stage. The vineyard is one of the agricultural sectors that has the most to gain from the use of this type of data, both by the economic value and by the number of diseases the plants are exposed to. The Flavescense dorée is a disease that attacks vineyards and may conduct to a significant loss. Nowadays, the detection of this disease is based on the visual identification of symptoms performed by experts who cover the entire area. However, this work remains tedious and relies only on the human eye, which is a problem since sometimes healthy plants are torn out, while diseased ones are left. If the experts think they have found symptoms, they take samples to send to the laboratory for further analysis. If the test is positive, then the whole vine is uprooted, to limit the spread of the disease. In this context, the use of hyperspectral data will allow the development of new disease detection methods. However, it will be necessary to reduce the volume of data used to make them usable by conventional resources. Fortunately, the advent of machine learning techniques empowered the development of systems that allow better decisions to be made, and consequently save time and money. In this article, a machine learning approach, which is based on an Autoencoder to automatically detect wine disease, is proposed.},
 bibtype = {article},
 author = {Silva, Diogo M. and Bernardin, Théo and Fanton, Kévin and Nepaul, Roshan and Pádua, Luís and Sousa, Joaquim J. and Cunha, António},
 doi = {10.1016/J.PROCS.2021.11.081},
 journal = {Procedia Computer Science}
}

The technological revolution that we have been witnessing recently has allowed components miniaturization and made electronic components accessible. Hyperspectral sensors benefited from these advances and could be mounted on unmanned aerial vehicles, which was unthinkable until recently. This fact significantly increased the applications of hyperspectral data, namely in agriculture, especially in the detection of diseases at an early stage. The vineyard is one of the agricultural sectors that has the most to gain from the use of this type of data, both by the economic value and by the number of diseases the plants are exposed to. The Flavescense dorée is a disease that attacks vineyards and may conduct to a significant loss. Nowadays, the detection of this disease is based on the visual identification of symptoms performed by experts who cover the entire area. However, this work remains tedious and relies only on the human eye, which is a problem since sometimes healthy plants are torn out, while diseased ones are left. If the experts think they have found symptoms, they take samples to send to the laboratory for further analysis. If the test is positive, then the whole vine is uprooted, to limit the spread of the disease. In this context, the use of hyperspectral data will allow the development of new disease detection methods. However, it will be necessary to reduce the volume of data used to make them usable by conventional resources. Fortunately, the advent of machine learning techniques empowered the development of systems that allow better decisions to be made, and consequently save time and money. In this article, a machine learning approach, which is based on an Autoencoder to automatically detect wine disease, is proposed.

@article{
 title = {Integrated IoT System for Prediction of Diseases in the Vineyards},
 type = {article},
 year = {2022},
 keywords = {jovanovska2022integratediotsystem},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {2c076ec3-f0b0-362b-b6d5-85117c503069},
 created = {2023-11-06T11:14:01.789Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T14:59:49.679Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Smart Agriculture is becoming more accepted by food industries, since they can see the difference in terms of quality of product, greater harvest, and lower cost expenses. Vine production has a great worldly role, so manufacturers are keen to experiment with new technologies that could improve their product. From a scientific perspective, we propose an integrated system for prediction of diseases in vineyards that provides services for data collection (from vineyard plantations), preprocessing, and integrated prediction of vineyard diseases using several models for prediction. Previous experiences showed that it is possible to develop a complete integrated system for prediction of diseases in viticulture and hence to provide a high efficient system for high quality vine production. Therefore, we will test the initial models when a decent set of data will be collected. The main contribution of this paper is comparing different methods based on experimental data from the prototype. This will allow us to build an integrated system for prediction of diseases in vineyards.},
 bibtype = {article},
 author = {Jovanovska, Elena M. and Chorbev, Ivan and Davcev, Danco and Mitreski, Kosta},
 doi = {10.1109/ICECCME55909.2022.9988415},
 journal = {International Conference on Electrical, Computer, Communications and Mechatronics Engineering, ICECCME 2022}
}

Smart Agriculture is becoming more accepted by food industries, since they can see the difference in terms of quality of product, greater harvest, and lower cost expenses. Vine production has a great worldly role, so manufacturers are keen to experiment with new technologies that could improve their product. From a scientific perspective, we propose an integrated system for prediction of diseases in vineyards that provides services for data collection (from vineyard plantations), preprocessing, and integrated prediction of vineyard diseases using several models for prediction. Previous experiences showed that it is possible to develop a complete integrated system for prediction of diseases in viticulture and hence to provide a high efficient system for high quality vine production. Therefore, we will test the initial models when a decent set of data will be collected. The main contribution of this paper is comparing different methods based on experimental data from the prototype. This will allow us to build an integrated system for prediction of diseases in vineyards.

@article{
 title = {Detection of Vineyard Diseases Using the Internet of Things Technology and Machine Learning Algorithms},
 type = {article},
 year = {2022},
 keywords = {roscuaneanu2022detectionvineyarddiseases},
 pages = {128-139},
 websites = {https://www.researchgate.net/publication/363472827},
 id = {ad0903d7-90d0-3844-b73d-9a07257404bf},
 created = {2023-11-06T11:21:57.175Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:04.532Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {In recent years, the Internet of Things concept has rapidly spread in most fields because of the benefits it offers, motivating viticulturists to implement new technologies that increase crop production and quality, as well as streamline production costs. The study's purpose is to monitor, using Internet of Things technology, two methods of identifying vine-specific diseases, which can be determined by environmental conditions (temperature, humidity, rainfall) or by analyzing diseased leaves from the vine. The first method is associated with a field study that involves placing Internet of Things sensors inside crops to measure environmental and plant parameters, which are then sent and stored in the Cloud. Based on these parameters, a correlation is made with the values that determine the occurrence of a specific vine disease (powdery mildew, downy mildew, and grey rot). The second method involves the use of Unmanned Aerial Vehicle imaging to take images containing healthy and diseased leaves from different parts of the vine. To analyze these images, a web page has been developed integrating a machine learning algorithm that detects the leaf state from the drone image footage. After the analysis all the values are stored in a database and the results are displayed as graphs and charts that are visualized by the viticulturist so that he can take the necessary actions. This study is an important step in the implementation of Internet of Things technology in viticulture, helping to monitor the main environmental and plant parameters, as well as detecting the occurrence of diseases among the vine cultures.},
 bibtype = {article},
 author = {Roscaneanu, Roxana and Balaceanu, Cristina Mihaela and Suciu, George and Maria, Ana and Dragulinescu, Claudia and Roșcăneanu, Roxana and Streche, Robert and Osiac, Filip and Bălăceanu, Cristina and Drăgulinescu, Ana Maria and Marcu, Ioana},
 doi = {10.24193/AWC2022_13},
 journal = {Air and Water – Components of the Environment” Conference Proceedings, Cluj-Napoca, Romania}
}

In recent years, the Internet of Things concept has rapidly spread in most fields because of the benefits it offers, motivating viticulturists to implement new technologies that increase crop production and quality, as well as streamline production costs. The study's purpose is to monitor, using Internet of Things technology, two methods of identifying vine-specific diseases, which can be determined by environmental conditions (temperature, humidity, rainfall) or by analyzing diseased leaves from the vine. The first method is associated with a field study that involves placing Internet of Things sensors inside crops to measure environmental and plant parameters, which are then sent and stored in the Cloud. Based on these parameters, a correlation is made with the values that determine the occurrence of a specific vine disease (powdery mildew, downy mildew, and grey rot). The second method involves the use of Unmanned Aerial Vehicle imaging to take images containing healthy and diseased leaves from different parts of the vine. To analyze these images, a web page has been developed integrating a machine learning algorithm that detects the leaf state from the drone image footage. After the analysis all the values are stored in a database and the results are displayed as graphs and charts that are visualized by the viticulturist so that he can take the necessary actions. This study is an important step in the implementation of Internet of Things technology in viticulture, helping to monitor the main environmental and plant parameters, as well as detecting the occurrence of diseases among the vine cultures.

@article{
 title = {Meteorological Data and UAV Images for the Detection and Identification of Grapevine Disease Using Deep Learning},
 type = {article},
 year = {2022},
 keywords = {ouhami2022meteorologicaldatauav},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {9b6bb1a0-04fb-3d64-a3fe-5d81b4ab86b4},
 created = {2023-11-06T11:27:14.937Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:04.333Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Vine disease is a serious issue in viticulture that causes an increase in the quantity and quantity of grapes. The early detection of vine disease can significantly improve their control and avoid the spread of pathogens, viruses, or fungi. One considerable solution is to use remote sensing as it is one of the important conventional eco-friendly methods for disease detection. Together with Deep learning models, remote sensing is used to improve the prediction quality and address effectively disease detection within semantic segmentation and disease identification based on meteorological history analysis. In this paper, we present our approach for downy mildew in vineyards identification and localization using a deep learning approach on multimodal decision fusion. We have reached an overall prediction segmentation accuracy and recall of 85,14% and 84,91% respectively.},
 bibtype = {article},
 author = {Ouhami, Maryam and Es-Saady, Youssef and El Hajj, Mohammed and Canals, Raphael and Hafiane, Adel},
 doi = {10.1109/EHB55594.2022.9991443},
 journal = {2022 10th E-Health and Bioengineering Conference, EHB 2022}
}

Vine disease is a serious issue in viticulture that causes an increase in the quantity and quantity of grapes. The early detection of vine disease can significantly improve their control and avoid the spread of pathogens, viruses, or fungi. One considerable solution is to use remote sensing as it is one of the important conventional eco-friendly methods for disease detection. Together with Deep learning models, remote sensing is used to improve the prediction quality and address effectively disease detection within semantic segmentation and disease identification based on meteorological history analysis. In this paper, we present our approach for downy mildew in vineyards identification and localization using a deep learning approach on multimodal decision fusion. We have reached an overall prediction segmentation accuracy and recall of 85,14% and 84,91% respectively.

@article{
 title = {Near Real-Time Vineyard Downy Mildew Detection and Severity Estimation},
 type = {article},
 year = {2022},
 pages = {9187-9194},
 volume = {2022-October},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {0cbaa6cb-e46b-34bc-b76b-e3065517f151},
 created = {2023-11-06T12:02:52.740Z},
 accessed = {2023-11-06},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T15:03:47.311Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The global grape and wine industry has been considerably impacted by diseases such as downy mildew (DM). Agricultural robots have demonstrated great potential to accurately and rapidly map DM infection for precision applications. Although the robots can autonomously acquire high-resolution images in the vineyard, data processing is mostly performed offline because of network infrastructure and onboard computing power constraints, limiting the use of agricultural robots for field operations. To address this issue, we developed a semantic segmentation model based on the modified DeepLabv3 network for near real time DM segmentation in high resolution images. Compared with state-of-the-art real time semantic segmentation models, the developed one achieved the best efficiency-accuracy balance on the DM dataset using embedded computing devices that can be easily integrated with commercial robotic platforms. DM severity estimation pipeline based on the model also showed a comparable measurement accuracy and statistical power in differentiation of fungicide treatments as the one based on offline semantic segmentation models. This enables the use of robotic perception systems for field operations.},
 bibtype = {article},
 author = {Liu, Ertai and Gold, Kaitlin and Cadle-Davidson, Lance and Combs, David and Jiang, Yu},
 doi = {10.1109/IROS47612.2022.9981404},
 journal = {IEEE International Conference on Intelligent Robots and Systems}
}

The global grape and wine industry has been considerably impacted by diseases such as downy mildew (DM). Agricultural robots have demonstrated great potential to accurately and rapidly map DM infection for precision applications. Although the robots can autonomously acquire high-resolution images in the vineyard, data processing is mostly performed offline because of network infrastructure and onboard computing power constraints, limiting the use of agricultural robots for field operations. To address this issue, we developed a semantic segmentation model based on the modified DeepLabv3 network for near real time DM segmentation in high resolution images. Compared with state-of-the-art real time semantic segmentation models, the developed one achieved the best efficiency-accuracy balance on the DM dataset using embedded computing devices that can be easily integrated with commercial robotic platforms. DM severity estimation pipeline based on the model also showed a comparable measurement accuracy and statistical power in differentiation of fungicide treatments as the one based on offline semantic segmentation models. This enables the use of robotic perception systems for field operations.

@article{
 title = {Identification of Grape Leaf Diseases Using Proposed Enhanced VGG16},
 type = {article},
 year = {2022},
 keywords = {farahani2022identificationgrapeleaf},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {d90271ab-5f27-37cf-8362-dcf5a2ac7efd},
 created = {2023-11-06T13:08:18.744Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T14:59:49.470Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Due to the damage that various tree diseases cause to agricultural fields, there are always different methods to deal with them. This paper proposes a new algorithm to identify three common grapevine diseases: Downey Mildew, Anthracnose, and Powdery Mildew. This new algorithm is based on the Faster Region-based Convolutional Neural Networks (R-CNN) with an Enhanced VGG16 model. The proposed Enhanced VGG16 model can diagnose better these three types of diseases. Experimental results show that the proposed algorithm has 0.53%, 0.912%, 2.759%, and 7.268% improvement in the mean Average Precision (mAP) criterion compared to ResNet50, VGG16, GoogLeNet, and AlexNet networks, respectively. The precision of the proposed method is better than other methods, and the number of layers used in it is acceptable compared to other methods.},
 bibtype = {article},
 author = {Farahani, Gholamreza and Mousavi, Seyedamirhossein and Farahani, Ameneh and Farahani, Hamidreza},
 doi = {10.1109/ICAC55051.2022.9911074},
 journal = {2022 27th International Conference on Automation and Computing: Smart Systems and Manufacturing, ICAC 2022}
}

Due to the damage that various tree diseases cause to agricultural fields, there are always different methods to deal with them. This paper proposes a new algorithm to identify three common grapevine diseases: Downey Mildew, Anthracnose, and Powdery Mildew. This new algorithm is based on the Faster Region-based Convolutional Neural Networks (R-CNN) with an Enhanced VGG16 model. The proposed Enhanced VGG16 model can diagnose better these three types of diseases. Experimental results show that the proposed algorithm has 0.53%, 0.912%, 2.759%, and 7.268% improvement in the mean Average Precision (mAP) criterion compared to ResNet50, VGG16, GoogLeNet, and AlexNet networks, respectively. The precision of the proposed method is better than other methods, and the number of layers used in it is acceptable compared to other methods.

@article{
 title = {Deep Learning Based Automatic Grape Downy Mildew Detection},
 type = {article},
 year = {2022},
 keywords = {zhang2022deeplearningbasedautomatic},
 pages = {872107},
 volume = {13},
 month = {6},
 publisher = {Frontiers Media S.A.},
 day = {9},
 id = {4b48ce58-fff8-318d-9553-30dcc463ed29},
 created = {2023-11-06T13:10:48.056Z},
 accessed = {2023-11-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:04.945Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Grape downy mildew (GDM) disease is a common plant leaf disease, and it causes serious damage to grape production, reducing yield and fruit quality. Traditional manual disease detection relies on farm experts and is often time-consuming. Computer vision technologies and artificial intelligence could provide automatic disease detection for real-time controlling the spread of disease on the grapevine in precision viticulture. To achieve the best trade-off between GDM detection accuracy and speed under natural environments, a deep learning based approach named YOLOv5-CA is proposed in this study. Here coordinate attention (CA) mechanism is integrated into YOLOv5, which highlights the downy mildew disease-related visual features to enhance the detection performance. A challenging GDM dataset was acquired in a vineyard under a nature scene (consisting of different illuminations, shadows, and backgrounds) to test the proposed approach. Experimental results show that the proposed YOLOv5-CA achieved a detection precision of 85.59%, a recall of 83.70%, and a mAP@0.5 of 89.55%, which is superior to the popular methods, including Faster R-CNN, YOLOv3, and YOLOv5. Furthermore, our proposed approach with inference occurring at 58.82 frames per second, could be deployed for the real-time disease control requirement. In addition, the proposed YOLOv5-CA based approach could effectively capture leaf disease related visual features resulting in higher GDE detection accuracy. Overall, this study provides a favorable deep learning based approach for the rapid and accurate diagnosis of grape leaf diseases in the field of automatic disease detection.},
 bibtype = {article},
 author = {Zhang, Zhao and Qiao, Yongliang and Guo, Yangyang and He, Dongjian},
 doi = {10.3389/FPLS.2022.872107/BIBTEX},
 journal = {Frontiers in Plant Science}
}

Grape downy mildew (GDM) disease is a common plant leaf disease, and it causes serious damage to grape production, reducing yield and fruit quality. Traditional manual disease detection relies on farm experts and is often time-consuming. Computer vision technologies and artificial intelligence could provide automatic disease detection for real-time controlling the spread of disease on the grapevine in precision viticulture. To achieve the best trade-off between GDM detection accuracy and speed under natural environments, a deep learning based approach named YOLOv5-CA is proposed in this study. Here coordinate attention (CA) mechanism is integrated into YOLOv5, which highlights the downy mildew disease-related visual features to enhance the detection performance. A challenging GDM dataset was acquired in a vineyard under a nature scene (consisting of different illuminations, shadows, and backgrounds) to test the proposed approach. Experimental results show that the proposed YOLOv5-CA achieved a detection precision of 85.59%, a recall of 83.70%, and a mAP@0.5 of 89.55%, which is superior to the popular methods, including Faster R-CNN, YOLOv3, and YOLOv5. Furthermore, our proposed approach with inference occurring at 58.82 frames per second, could be deployed for the real-time disease control requirement. In addition, the proposed YOLOv5-CA based approach could effectively capture leaf disease related visual features resulting in higher GDE detection accuracy. Overall, this study provides a favorable deep learning based approach for the rapid and accurate diagnosis of grape leaf diseases in the field of automatic disease detection.

@article{
 title = {Attention classification-and-segmentation network for micro-crack anomaly detection of photovoltaic module cells},
 type = {article},
 year = {2022},
 keywords = {Attention mechanism,Micro-crack anomaly detection,Photovoltaic module cell,Semantic segmentation,Transfer learning},
 pages = {291-304},
 volume = {238},
 month = {5},
 publisher = {Pergamon},
 day = {15},
 id = {ce19b742-0c9d-33be-af20-447407965970},
 created = {2023-11-07T08:25:21.186Z},
 accessed = {2023-11-07},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-07T08:25:25.429Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Micro-crack anomaly detection is a crucial part of the quality inspection of photovoltaic (PV) module cells. However, due to the complex background and the lack of sufficient anomaly samples, it is a challenging task to identify and locate micro-crack accurately. This paper presents a novel method for detecting micro-crack anomaly in PV module cells by designing an attention classification-and-segmentation network. Specifically, the proposed network consists of a classification network and a segmentation network. In the classification network, the real-time micro-crack anomaly discrimination task in the detection process is performed at first. The classification network introduces transfer learning and deep supervision mechanism to effectively extract and fuse multi-scale features to accurately predict the anomaly probability score. In the segmentation network, the pixel-level micro-crack detection is conducted on the sample determined as defects. The M-shaped structure is used to better extract and fuse shallow-level and deep-level features in the segmentation network, effectively solving the “All Black” issue in the training process. And the attention mechanism module is inserted into the M-shaped structure to more effectively extract micro-crack anomaly features and suppress background noise, thereby significantly improving the accuracy of segmentation. By the design of two-stage network architecture and the utilization of the attention module, the proposed network presents a strong capability for learning from a small set of labeled and annotated samples. Comprehensive experiments are conducted on the real PV electroluminescence (EL) images dataset. Experimental results show that, compared with InceptionV4 classification network and U-net segmentation network, the proposed network has superior performance with ACC of 100.0% and DICE of 0.541 in the micro-crack anomaly classification and segmentation task. Moreover, the experiments show the attention module inserted in the proposed network plays a significant role in improving the classification and segmentation accuracy.},
 bibtype = {article},
 author = {Jiang, Yu and Zhao, Chunhui},
 doi = {10.1016/J.SOLENER.2022.04.012},
 journal = {Solar Energy}
}

Micro-crack anomaly detection is a crucial part of the quality inspection of photovoltaic (PV) module cells. However, due to the complex background and the lack of sufficient anomaly samples, it is a challenging task to identify and locate micro-crack accurately. This paper presents a novel method for detecting micro-crack anomaly in PV module cells by designing an attention classification-and-segmentation network. Specifically, the proposed network consists of a classification network and a segmentation network. In the classification network, the real-time micro-crack anomaly discrimination task in the detection process is performed at first. The classification network introduces transfer learning and deep supervision mechanism to effectively extract and fuse multi-scale features to accurately predict the anomaly probability score. In the segmentation network, the pixel-level micro-crack detection is conducted on the sample determined as defects. The M-shaped structure is used to better extract and fuse shallow-level and deep-level features in the segmentation network, effectively solving the “All Black” issue in the training process. And the attention mechanism module is inserted into the M-shaped structure to more effectively extract micro-crack anomaly features and suppress background noise, thereby significantly improving the accuracy of segmentation. By the design of two-stage network architecture and the utilization of the attention module, the proposed network presents a strong capability for learning from a small set of labeled and annotated samples. Comprehensive experiments are conducted on the real PV electroluminescence (EL) images dataset. Experimental results show that, compared with InceptionV4 classification network and U-net segmentation network, the proposed network has superior performance with ACC of 100.0% and DICE of 0.541 in the micro-crack anomaly classification and segmentation task. Moreover, the experiments show the attention module inserted in the proposed network plays a significant role in improving the classification and segmentation accuracy.

@article{
 title = {Multitask Learning of Alfalfa Nutritive Value from UAV-Based Hyperspectral Images},
 type = {article},
 year = {2022},
 keywords = {Alfalfa,hyperspectral imagery,multitask learning,nutritive value,unmanned aerial vehicle (UAV)},
 volume = {19},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {77b05add-67cc-3ba5-ad99-c82dd94ad0b9},
 created = {2023-11-07T08:56:38.926Z},
 accessed = {2023-11-07},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-16T09:07:13.684Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Alfalfa is a valuable and widely adapted forage crop, and its nutritive value directly affects animal performance and ultimately affects the profitability of livestock production. Traditional nutritive value measurement method is labor-intensive and time-consuming and thus hinders the determination of alfalfa nutritive values over large fields. The adoption of unmanned aerial vehicles (UAVs) facilitates the generation of images with high spatial and temporal resolutions for field-level agricultural research. Additionally, compared with other imaging modalities, hyperspectral data usually consist of hundreds of narrow spectral bands and allow the accurate detection, identification, and quantification of crop quality. Although various machine-learning methods have been developed for alfalfa quality prediction, they were all single-task models that learned independently for each quality trait and failed to utilize the underlying relatedness between each task. Inspired by the idea of multitask learning (MTL), this study aims to develop an approach that simultaneously predicts multiple quality traits. The algorithm first extracts shared information through a long short-term memory (LSTM)-based common hidden layer. To enhance the model flexibility, it is then divided into multiple branches, each containing the same or different number of task-specific fully connected hidden layers. Through comparison with multiple mainstream single-task machine-learning models, the effectiveness of the model is illustrated based on the measured alfalfa quality data and multitemporal UAV-based hyperspectral imagery.},
 bibtype = {article},
 author = {Feng, Luwei and Zhang, Zhou and Ma, Yuchi and Sun, Yazhou and Du, Qingyun and Williams, Parker and Drewry, Jessica and Luck, Brian},
 doi = {10.1109/LGRS.2021.3079317},
 journal = {IEEE Geoscience and Remote Sensing Letters}
}

Alfalfa is a valuable and widely adapted forage crop, and its nutritive value directly affects animal performance and ultimately affects the profitability of livestock production. Traditional nutritive value measurement method is labor-intensive and time-consuming and thus hinders the determination of alfalfa nutritive values over large fields. The adoption of unmanned aerial vehicles (UAVs) facilitates the generation of images with high spatial and temporal resolutions for field-level agricultural research. Additionally, compared with other imaging modalities, hyperspectral data usually consist of hundreds of narrow spectral bands and allow the accurate detection, identification, and quantification of crop quality. Although various machine-learning methods have been developed for alfalfa quality prediction, they were all single-task models that learned independently for each quality trait and failed to utilize the underlying relatedness between each task. Inspired by the idea of multitask learning (MTL), this study aims to develop an approach that simultaneously predicts multiple quality traits. The algorithm first extracts shared information through a long short-term memory (LSTM)-based common hidden layer. To enhance the model flexibility, it is then divided into multiple branches, each containing the same or different number of task-specific fully connected hidden layers. Through comparison with multiple mainstream single-task machine-learning models, the effectiveness of the model is illustrated based on the measured alfalfa quality data and multitemporal UAV-based hyperspectral imagery.

@article{
 title = {Multitask Learning of Alfalfa Nutritive Value from UAV-Based Hyperspectral Images},
 type = {article},
 year = {2022},
 keywords = {Alfalfa,hyperspectral imagery,multitask learning,nutritive value,unmanned aerial vehicle (UAV)},
 volume = {19},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {e63eab8d-285d-3700-a469-a48b17041be1},
 created = {2023-11-08T09:41:21.722Z},
 accessed = {2023-11-08},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-08T09:41:30.654Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bc66e353-ef41-46d4-8108-778d5481c126,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Alfalfa is a valuable and widely adapted forage crop, and its nutritive value directly affects animal performance and ultimately affects the profitability of livestock production. Traditional nutritive value measurement method is labor-intensive and time-consuming and thus hinders the determination of alfalfa nutritive values over large fields. The adoption of unmanned aerial vehicles (UAVs) facilitates the generation of images with high spatial and temporal resolutions for field-level agricultural research. Additionally, compared with other imaging modalities, hyperspectral data usually consist of hundreds of narrow spectral bands and allow the accurate detection, identification, and quantification of crop quality. Although various machine-learning methods have been developed for alfalfa quality prediction, they were all single-task models that learned independently for each quality trait and failed to utilize the underlying relatedness between each task. Inspired by the idea of multitask learning (MTL), this study aims to develop an approach that simultaneously predicts multiple quality traits. The algorithm first extracts shared information through a long short-term memory (LSTM)-based common hidden layer. To enhance the model flexibility, it is then divided into multiple branches, each containing the same or different number of task-specific fully connected hidden layers. Through comparison with multiple mainstream single-task machine-learning models, the effectiveness of the model is illustrated based on the measured alfalfa quality data and multitemporal UAV-based hyperspectral imagery.},
 bibtype = {article},
 author = {Feng, Luwei and Zhang, Zhou and Ma, Yuchi and Sun, Yazhou and Du, Qingyun and Williams, Parker and Drewry, Jessica and Luck, Brian},
 doi = {10.1109/LGRS.2021.3079317},
 journal = {IEEE Geoscience and Remote Sensing Letters}
}

Alfalfa is a valuable and widely adapted forage crop, and its nutritive value directly affects animal performance and ultimately affects the profitability of livestock production. Traditional nutritive value measurement method is labor-intensive and time-consuming and thus hinders the determination of alfalfa nutritive values over large fields. The adoption of unmanned aerial vehicles (UAVs) facilitates the generation of images with high spatial and temporal resolutions for field-level agricultural research. Additionally, compared with other imaging modalities, hyperspectral data usually consist of hundreds of narrow spectral bands and allow the accurate detection, identification, and quantification of crop quality. Although various machine-learning methods have been developed for alfalfa quality prediction, they were all single-task models that learned independently for each quality trait and failed to utilize the underlying relatedness between each task. Inspired by the idea of multitask learning (MTL), this study aims to develop an approach that simultaneously predicts multiple quality traits. The algorithm first extracts shared information through a long short-term memory (LSTM)-based common hidden layer. To enhance the model flexibility, it is then divided into multiple branches, each containing the same or different number of task-specific fully connected hidden layers. Through comparison with multiple mainstream single-task machine-learning models, the effectiveness of the model is illustrated based on the measured alfalfa quality data and multitemporal UAV-based hyperspectral imagery.

@article{
 title = {Early detection of plant virus infection using multispectral imaging and spatial–spectral machine learning},
 type = {article},
 year = {2022},
 volume = {12},
 month = {12},
 publisher = {Nature Research},
 day = {1},
 id = {f21e3b65-06c3-36a3-a7b9-3367a70d7819},
 created = {2023-11-13T13:01:05.418Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-12T12:09:01.288Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {2bfb8d91-9fac-46f0-bd0c-93235d01dbed},
 private_publication = {false},
 abstract = {Cassava brown streak disease (CBSD) is an emerging viral disease that can greatly reduce cassava productivity, while causing only mild aerial symptoms that develop late in infection. Early detection of CBSD enables better crop management and intervention. Current techniques require laboratory equipment and are labour intensive and often inaccurate. We have developed a handheld active multispectral imaging (A-MSI) device combined with machine learning for early detection of CBSD in real-time. The principal benefits of A-MSI over passive MSI and conventional camera systems are improved spectral signal-to-noise ratio and temporal repeatability. Information fusion techniques further combine spectral and spatial information to reliably identify features that distinguish healthy cassava from plants with CBSD as early as 28 days post inoculation on a susceptible and a tolerant cultivar. Application of the device has the potential to increase farmers’ access to healthy planting materials and reduce losses due to CBSD in Africa. It can also be adapted for sensing other biotic and abiotic stresses in real-world situations where plants are exposed to multiple pest, pathogen and environmental stresses.},
 bibtype = {article},
 author = {Peng, Yao and Dallas, Mary M. and Ascencio-Ibáñez, José T. and Hoyer, J. Steen and Legg, James and Hanley-Bowdoin, Linda and Grieve, Bruce and Yin, Hujun},
 doi = {10.1038/s41598-022-06372-8},
 journal = {Scientific Reports},
 number = {1}
}

Cassava brown streak disease (CBSD) is an emerging viral disease that can greatly reduce cassava productivity, while causing only mild aerial symptoms that develop late in infection. Early detection of CBSD enables better crop management and intervention. Current techniques require laboratory equipment and are labour intensive and often inaccurate. We have developed a handheld active multispectral imaging (A-MSI) device combined with machine learning for early detection of CBSD in real-time. The principal benefits of A-MSI over passive MSI and conventional camera systems are improved spectral signal-to-noise ratio and temporal repeatability. Information fusion techniques further combine spectral and spatial information to reliably identify features that distinguish healthy cassava from plants with CBSD as early as 28 days post inoculation on a susceptible and a tolerant cultivar. Application of the device has the potential to increase farmers’ access to healthy planting materials and reduce losses due to CBSD in Africa. It can also be adapted for sensing other biotic and abiotic stresses in real-world situations where plants are exposed to multiple pest, pathogen and environmental stresses.

@article{
 title = {Plant Viral Disease Detection: From Molecular Diagnosis to Optical Sensing Technology—A Multidisciplinary Review},
 type = {article},
 year = {2022},
 keywords = {disease prediction modelling,hyperspectral imaging,machine learning,plant viruses,remote sensing},
 pages = {1542},
 volume = {14},
 websites = {https://www.mdpi.com/2072-4292/14/7/1542/htm,https://www.mdpi.com/2072-4292/14/7/1542},
 month = {3},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {23},
 id = {9e2ce3de-e45d-3d36-a756-f67bc4512e3f},
 created = {2023-11-14T10:11:48.495Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-05T13:07:33.585Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Plant viral diseases result in productivity and economic losses to agriculture, necessitating accurate detection for effective control. Lab-based molecular testing is the gold standard for providing reliable and accurate diagnostics; however, these tests are expensive, time-consuming, and labour-intensive, especially at the field-scale with a large number of samples. Recent advances in optical remote sensing offer tremendous potential for non-destructive diagnostics of plant viral diseases at large spatial scales. This review provides an overview of traditional diagnostic methods followed by a comprehensive description of optical sensing technology, including camera systems, platforms, and spectral data analysis to detect plant viral diseases. The paper is organized along six multidisciplinary sections: (1) Impact of plant viral disease on plant physiology and consequent phenotypic changes, (2) direct diagnostic methods, (3) traditional indirect detection methods, (4) optical sensing technologies, (5) data processing techniques and modelling for disease detection, and (6) comparison of the costs. Finally, the current challenges and novel ideas of optical sensing for detecting plant viruses are discussed.},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Gautam, Deepak and Habili, Nuredin and Pagay, Vinay},
 doi = {10.3390/RS14071542},
 journal = {Remote Sensing 2022, Vol. 14, Page 1542},
 number = {7}
}

Plant viral diseases result in productivity and economic losses to agriculture, necessitating accurate detection for effective control. Lab-based molecular testing is the gold standard for providing reliable and accurate diagnostics; however, these tests are expensive, time-consuming, and labour-intensive, especially at the field-scale with a large number of samples. Recent advances in optical remote sensing offer tremendous potential for non-destructive diagnostics of plant viral diseases at large spatial scales. This review provides an overview of traditional diagnostic methods followed by a comprehensive description of optical sensing technology, including camera systems, platforms, and spectral data analysis to detect plant viral diseases. The paper is organized along six multidisciplinary sections: (1) Impact of plant viral disease on plant physiology and consequent phenotypic changes, (2) direct diagnostic methods, (3) traditional indirect detection methods, (4) optical sensing technologies, (5) data processing techniques and modelling for disease detection, and (6) comparison of the costs. Finally, the current challenges and novel ideas of optical sensing for detecting plant viruses are discussed.

@article{
 title = {A grape leaves disease recognition using Amazon Sage Maker},
 type = {article},
 year = {2022},
 keywords = {tockova2022grapeleavesdisease},
 pages = {45-56},
 volume = {5},
 id = {2c25adbe-4b99-34b3-b0e4-b00546b07fc6},
 created = {2023-11-14T11:15:42.825Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:04.753Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {article},
 author = {Angela Tockova, Zoran Zlatev, Saso Koceski},
 journal = {BALKAN JOURNAL OF APPLIED MATHEMATICS AND INFORMATICS (BJAMI)},
 number = {2}
}

@article{
 title = {How important is UAVs RTK accuracy for the identification of certain vine diseases?},
 type = {article},
 year = {2022},
 keywords = {zottele2022howimportantuav},
 pages = {239-243},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {e1cddb09-19c2-3678-92e3-1be5930e187a},
 created = {2023-11-14T12:25:29.965Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:05.139Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The recent deployment of GNSS-RTK positioning on remotely piloted vehicles has increased real-time positioning accuracy by almost three orders of magnitude. This not only provides a significant geometric improvement but, in practice, makes possible some applications that were simply not possible before. For example, positioning crops with centimetre accuracy makes it possible to distinguish a single plant and to detect or treat that very plant without any possibility of misunderstanding. This is obviously not possible with 'traditional' drones that work in 'point positioning' with indeterminacies of even tens of metres. In this paper we will illustrate how the possibilities of RTK can be applied to a specific vinepathology. The symptoms of the flavescence dorée and bois noir are grouped into the so-called Grapevine Yellows (GY). These diseases are affecting the viticultural regions worldwide and all varieties and rootstocks seem susceptible but with varying degrees of severity. Typical symptoms include discolouration and necrosis of leaf veins and leaf blades, downward curling of leaves, lack or incomplete lignification of shoots, stunting and necrosis of shoots, abortion of inflorescences and shrivelling of berries. The compulsory control plan for the fight of these diseases includes both the use of insecticides and the eradication of the vines. This latter is part of a monitoring plan of the grapevine yellows that aims to identify outbreaks of the disease and its progression and limit the compulsory phytosanitary control only in the truly affected areas. The identification of the GY is a very time-consuming technical work because each vineyard must be visually inspected plant by plant. This type of monitoring is made even more difficult in the case of steeply sloping vineyards and where the vineyard landscape is fragmented. So, we raised the following question: is it possible to use Unmanned Aerial Vehicles (UAVs or drones) to remotely monitor the vines that are difficult to reach and identify the grapevine yellows? We present here the results of our field tests made in Trentino (IT) with different drone models (prosumer and professional) and with different types of image acquisition sensors (RGB and multi-spectral).},
 bibtype = {article},
 author = {Zottele, Fabio and Crocetta, Paolo and Baiocchi, Valerio},
 doi = {10.1109/METROAGRIFOR55389.2022.9964928},
 journal = {2022 IEEE Workshop on Metrology for Agriculture and Forestry, MetroAgriFor 2022 - Proceedings}
}

The recent deployment of GNSS-RTK positioning on remotely piloted vehicles has increased real-time positioning accuracy by almost three orders of magnitude. This not only provides a significant geometric improvement but, in practice, makes possible some applications that were simply not possible before. For example, positioning crops with centimetre accuracy makes it possible to distinguish a single plant and to detect or treat that very plant without any possibility of misunderstanding. This is obviously not possible with 'traditional' drones that work in 'point positioning' with indeterminacies of even tens of metres. In this paper we will illustrate how the possibilities of RTK can be applied to a specific vinepathology. The symptoms of the flavescence dorée and bois noir are grouped into the so-called Grapevine Yellows (GY). These diseases are affecting the viticultural regions worldwide and all varieties and rootstocks seem susceptible but with varying degrees of severity. Typical symptoms include discolouration and necrosis of leaf veins and leaf blades, downward curling of leaves, lack or incomplete lignification of shoots, stunting and necrosis of shoots, abortion of inflorescences and shrivelling of berries. The compulsory control plan for the fight of these diseases includes both the use of insecticides and the eradication of the vines. This latter is part of a monitoring plan of the grapevine yellows that aims to identify outbreaks of the disease and its progression and limit the compulsory phytosanitary control only in the truly affected areas. The identification of the GY is a very time-consuming technical work because each vineyard must be visually inspected plant by plant. This type of monitoring is made even more difficult in the case of steeply sloping vineyards and where the vineyard landscape is fragmented. So, we raised the following question: is it possible to use Unmanned Aerial Vehicles (UAVs or drones) to remotely monitor the vines that are difficult to reach and identify the grapevine yellows? We present here the results of our field tests made in Trentino (IT) with different drone models (prosumer and professional) and with different types of image acquisition sensors (RGB and multi-spectral).

@article{
 title = {A multispectral acquisition system for potential detection of Flavescence dorée},
 type = {article},
 year = {2022},
 keywords = {barjaktarovic2022multispectralacquisition},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {e60e55e6-0070-3975-9dd0-89f857ba7828},
 created = {2023-11-14T12:42:05.511Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:12:01.290Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Agriculture is under constant pressure to increase the production rate and to provide more food or resources for other industries. Without precision agriculture it is not possible to fulfil these requirements. Medium and large-size farms already adopted this technology, but small farms are far from precision agriculture due to the high initial costs. In this paper we present a bespoke and affordable multispectral camera for precision farming and illustrate its application in the detection of Flavescence dorée. This is a grapevine disease that makes a great concern to grapevine producers in the whole Mediterranean region. Flavescence dorée is the only quarantine disease in the European region. The related mandatory control procedures include uprooting every infected plant, with the obligation to uproot the vineyard when the infection exceed the 20% of threshold infection, hence resulting in a high economical loss. Thus, it is highly important to detect even a single infected plant over an entire vineyard plot, preventing the spreading of Flavescence dorée at earliest stages. We used the in-house developed multispectral camera together with a hyperspectral camera to acquire data from two vineyards near Riva del Garda, Trentino, Italy, during the summer of 2022. These data are the starting point for selecting the optimal spectral bands for the detection of Flavescence dorée using affordable multispectral approaches and developing an appropriate classification algorithm.},
 bibtype = {article},
 author = {Barjaktarovic, Marko and Faralli, Michele and Bertamini, Massimo and Bruzzone, Lorenzo and Santoni, Massimo},
 doi = {10.1109/TELFOR56187.2022.9983685},
 journal = {2022 30th Telecommunications Forum, TELFOR 2022 - Proceedings}
}

Agriculture is under constant pressure to increase the production rate and to provide more food or resources for other industries. Without precision agriculture it is not possible to fulfil these requirements. Medium and large-size farms already adopted this technology, but small farms are far from precision agriculture due to the high initial costs. In this paper we present a bespoke and affordable multispectral camera for precision farming and illustrate its application in the detection of Flavescence dorée. This is a grapevine disease that makes a great concern to grapevine producers in the whole Mediterranean region. Flavescence dorée is the only quarantine disease in the European region. The related mandatory control procedures include uprooting every infected plant, with the obligation to uproot the vineyard when the infection exceed the 20% of threshold infection, hence resulting in a high economical loss. Thus, it is highly important to detect even a single infected plant over an entire vineyard plot, preventing the spreading of Flavescence dorée at earliest stages. We used the in-house developed multispectral camera together with a hyperspectral camera to acquire data from two vineyards near Riva del Garda, Trentino, Italy, during the summer of 2022. These data are the starting point for selecting the optimal spectral bands for the detection of Flavescence dorée using affordable multispectral approaches and developing an appropriate classification algorithm.

@article{
 title = {Lite CNN Models for Real-Time Post-Harvest Grape Disease Detection},
 type = {article},
 year = {2022},
 keywords = {mohimont2022litecnnmodels},
 pages = {116-125},
 websites = {https://ebooks.iospress.nl/doi/10.3233/AISE220029},
 month = {6},
 publisher = {IOS Press},
 day = {16},
 id = {f4503988-91d5-3ca8-8f23-d32103d778d3},
 created = {2023-11-14T12:49:57.475Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T11:23:57.812Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {… with an object detection model based on Faster R-CNN and it reached 81% mean Average-… is labeled for binary segmentation with LabelMe polygonal mask [15]. Most images contain …},
 bibtype = {article},
 author = {Mohimont, Lucas and Alin, François and Gaveau, Nathalie and Steffenel, Luiz Angelo},
 doi = {10.3233/aise220029}
}

… with an object detection model based on Faster R-CNN and it reached 81% mean Average-… is labeled for binary segmentation with LabelMe polygonal mask [15]. Most images contain …

@article{
 title = {Design of Intelligent Detection Platform for Wine Grape Pests and Diseases in Ningxia},
 type = {article},
 year = {2022},
 keywords = {wang2022designintelligentdetection},
 pages = {106},
 volume = {12},
 websites = {https://www.mdpi.com/2223-7747/12/1/106/htm,https://www.mdpi.com/2223-7747/12/1/106},
 month = {12},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {26},
 id = {1b9a1d1b-8702-3b22-9941-1d4c4ed6dae3},
 created = {2023-11-14T13:14:14.985Z},
 accessed = {2023-11-14},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T18:53:05.375Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {In order to reduce the impact of pests and diseases on the yield and quality of Ningxia wine grapes and to improve the efficiency and intelligence of detection, this paper designs an intelligent detection platform for pests and diseases. The optimal underlying network is selected by comparing the recognition accuracy of both MobileNet V2 and YOLOX_s networks trained on the Public Dataset. Based on this network, the effect of adding attention mechanism and replacing loss function on recognition effect is investigated by permutation in the Custom Dataset, resulting in the improved network YOLOX_s + CBAM. The improved network was trained on the Overall Dataset, and finally a recognition model capable of identifying nine types of pests was obtained, with a recognition accuracy of 93.35% in the validation set, an improvement of 1.35% over the original network. The recognition model is deployed on the Web side and Raspberry Pi to achieve independent detection functions; the channel between the two platforms is built through Ngrok, and remote interconnection is achieved through VNC desktop. Users can choose to upload local images on the Web side for detection, handheld Raspberry Pi for field detection, or Raspberry Pi and Web interconnection for remote detection.},
 bibtype = {article},
 author = {Wang, Yutan and Wei, Chi and Sun, Haowei and Qu, Aili},
 doi = {10.3390/PLANTS12010106},
 journal = {Plants 2023, Vol. 12, Page 106},
 number = {1}
}

In order to reduce the impact of pests and diseases on the yield and quality of Ningxia wine grapes and to improve the efficiency and intelligence of detection, this paper designs an intelligent detection platform for pests and diseases. The optimal underlying network is selected by comparing the recognition accuracy of both MobileNet V2 and YOLOX_s networks trained on the Public Dataset. Based on this network, the effect of adding attention mechanism and replacing loss function on recognition effect is investigated by permutation in the Custom Dataset, resulting in the improved network YOLOX_s + CBAM. The improved network was trained on the Overall Dataset, and finally a recognition model capable of identifying nine types of pests was obtained, with a recognition accuracy of 93.35% in the validation set, an improvement of 1.35% over the original network. The recognition model is deployed on the Web side and Raspberry Pi to achieve independent detection functions; the channel between the two platforms is built through Ngrok, and remote interconnection is achieved through VNC desktop. Users can choose to upload local images on the Web side for detection, handheld Raspberry Pi for field detection, or Raspberry Pi and Web interconnection for remote detection.

@article{
 title = {Near Real-Time Vineyard Downy Mildew Detection and Severity Estimation},
 type = {article},
 year = {2022},
 pages = {9187-9194},
 volume = {2022-Octob},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {b4da8d4f-309a-3b1e-b5f6-17ca1a182908},
 created = {2023-11-21T11:23:29.534Z},
 accessed = {2023-11-21},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T14:59:50.134Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The global grape and wine industry has been considerably impacted by diseases such as downy mildew (DM). Agricultural robots have demonstrated great potential to accurately and rapidly map DM infection for precision applications. Although the robots can autonomously acquire high-resolution images in the vineyard, data processing is mostly performed offline because of network infrastructure and onboard computing power constraints, limiting the use of agricultural robots for field operations. To address this issue, we developed a semantic segmentation model based on the modified DeepLabv3 network for near real time DM segmentation in high resolution images. Compared with state-of-the-art real time semantic segmentation models, the developed one achieved the best efficiency-accuracy balance on the DM dataset using embedded computing devices that can be easily integrated with commercial robotic platforms. DM severity estimation pipeline based on the model also showed a comparable measurement accuracy and statistical power in differentiation of fungicide treatments as the one based on offline semantic segmentation models. This enables the use of robotic perception systems for field operations.},
 bibtype = {article},
 author = {Liu, Ertai and Gold, Kaitlin and Cadle-Davidson, Lance and Combs, David and Jiang, Yu},
 doi = {10.1109/IROS47612.2022.9981404},
 journal = {IEEE International Conference on Intelligent Robots and Systems},
 keywords = {liu2022nearrealtimevieyard}
}

The global grape and wine industry has been considerably impacted by diseases such as downy mildew (DM). Agricultural robots have demonstrated great potential to accurately and rapidly map DM infection for precision applications. Although the robots can autonomously acquire high-resolution images in the vineyard, data processing is mostly performed offline because of network infrastructure and onboard computing power constraints, limiting the use of agricultural robots for field operations. To address this issue, we developed a semantic segmentation model based on the modified DeepLabv3 network for near real time DM segmentation in high resolution images. Compared with state-of-the-art real time semantic segmentation models, the developed one achieved the best efficiency-accuracy balance on the DM dataset using embedded computing devices that can be easily integrated with commercial robotic platforms. DM severity estimation pipeline based on the model also showed a comparable measurement accuracy and statistical power in differentiation of fungicide treatments as the one based on offline semantic segmentation models. This enables the use of robotic perception systems for field operations.

@article{
 title = {wGrapeUNIPD-DL: An open dataset for white grape bunch detection},
 type = {article},
 year = {2022},
 keywords = {Cluster detection,Crop load mapping,Digital agriculture,Early yield estimation,Grape yield,Object detection,Precision viticulture},
 pages = {108466},
 volume = {43},
 month = {8},
 publisher = {Elsevier},
 day = {1},
 id = {0df3a2f8-44ee-3c98-b4bb-164d98d3c8a0},
 created = {2023-11-21T15:01:43.623Z},
 accessed = {2023-11-21},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-21T15:32:10.022Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {National and international Vitis variety catalogues can be used as image datasets for computer vision in viticulture. These databases archive ampelographic features and phenology of several grape varieties and plant structures images (e.g. leaf, bunch, shoots). Although these archives represent a potential database for computer vision in viticulture, plant structure images are acquired singularly and mostly not directly in the vineyard. Localization computer vision models would take advantage of multiple objects in the same image, allowing more efficient training. The present images and labels dataset was designed to overcome such limitations and provide suitable images for multiple cluster identification in white grape varieties. A group of 373 images were acquired from later view in vertical shoot position vineyards in six different Italian locations at different phenological stages. Images were then labelled in YOLO labelling format. The dataset was made available both in terms of images and labels. The real number of bunches counted in the field, and the number of bunches visible in the image (not covered by other vine structures) was recorded for a group of images in this dataset.},
 bibtype = {article},
 author = {Sozzi, Marco and Cantalamessa, Silvia and Cogato, Alessia and Kayad, Ahmed and Marinello, Francesco},
 doi = {10.1016/J.DIB.2022.108466},
 journal = {Data in Brief}
}

National and international Vitis variety catalogues can be used as image datasets for computer vision in viticulture. These databases archive ampelographic features and phenology of several grape varieties and plant structures images (e.g. leaf, bunch, shoots). Although these archives represent a potential database for computer vision in viticulture, plant structure images are acquired singularly and mostly not directly in the vineyard. Localization computer vision models would take advantage of multiple objects in the same image, allowing more efficient training. The present images and labels dataset was designed to overcome such limitations and provide suitable images for multiple cluster identification in white grape varieties. A group of 373 images were acquired from later view in vertical shoot position vineyards in six different Italian locations at different phenological stages. Images were then labelled in YOLO labelling format. The dataset was made available both in terms of images and labels. The real number of bunches counted in the field, and the number of bunches visible in the image (not covered by other vine structures) was recorded for a group of images in this dataset.

@article{
 title = {Semi-supervised deep learning and low-cost cameras for the semantic segmentation of natural images in viticulture},
 type = {article},
 year = {2022},
 keywords = {Agricultural robot sensing,Grape bunches,Natural images,Semantic segmentation,Semi-supervised learning},
 pages = {2001-2026},
 volume = {23},
 websites = {https://link.springer.com/article/10.1007/s11119-022-09929-9},
 month = {12},
 publisher = {Springer},
 day = {1},
 id = {3f414c31-81e1-3ad4-aa77-de965f3de605},
 created = {2023-12-13T07:02:54.917Z},
 accessed = {2023-12-13},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-12-13T07:45:25.015Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Automatic yield monitoring and in-field robotic harvesting by low-cost cameras require object detection and segmentation solutions to tackle the poor quality of natural images and the lack of exactly-labeled datasets of consistent sizes. This work proposed the application of deep learning for semantic segmentation of natural images acquired by a low-cost RGB-D camera in a commercial vineyard. Several deep architectures were trained and compared on 85 labeled images. Three semi-supervised learning methods (PseudoLabeling, Distillation and Model Distillation) were proposed to take advantage of 320 non-annotated images. In these experiments, the DeepLabV3+ architecture with a ResNext50 backbone, trained with the set of labeled images, achieved the best overall accuracy of 84.78%. In contrast, the Manet architecture combined with the EfficientnetB3 backbone reached the highest accuracy for the bunch class (85.69%). The application of semi-supervised learning methods boosted the segmentation accuracy between 5.62 and 6.01%, on average. Further discussions are presented to show the effects of a fine-grained manual image annotation on the accuracy of the proposed methods and to compare time requirements.},
 bibtype = {article},
 author = {Casado-García, A. and Heras, J. and Milella, A. and Marani, R.},
 doi = {10.1007/S11119-022-09929-9/TABLES/9},
 journal = {Precision Agriculture},
 number = {6}
}

Automatic yield monitoring and in-field robotic harvesting by low-cost cameras require object detection and segmentation solutions to tackle the poor quality of natural images and the lack of exactly-labeled datasets of consistent sizes. This work proposed the application of deep learning for semantic segmentation of natural images acquired by a low-cost RGB-D camera in a commercial vineyard. Several deep architectures were trained and compared on 85 labeled images. Three semi-supervised learning methods (PseudoLabeling, Distillation and Model Distillation) were proposed to take advantage of 320 non-annotated images. In these experiments, the DeepLabV3+ architecture with a ResNext50 backbone, trained with the set of labeled images, achieved the best overall accuracy of 84.78%. In contrast, the Manet architecture combined with the EfficientnetB3 backbone reached the highest accuracy for the bunch class (85.69%). The application of semi-supervised learning methods boosted the segmentation accuracy between 5.62 and 6.01%, on average. Further discussions are presented to show the effects of a fine-grained manual image annotation on the accuracy of the proposed methods and to compare time requirements.

@article{
 title = {Cross-Task Knowledge Distillation in Multi-Task Recommendation},
 type = {article},
 year = {2022},
 keywords = {Data Mining & Knowledge Management (DMKM)},
 pages = {4318-4326},
 volume = {36},
 websites = {https://ojs.aaai.org/index.php/AAAI/article/view/20352},
 month = {6},
 publisher = {Association for the Advancement of Artificial Intelligence},
 day = {28},
 id = {918c3ff1-8e41-30b2-9ec3-bc5095d4b729},
 created = {2023-12-13T07:21:41.686Z},
 accessed = {2023-12-13},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-12-13T07:21:45.963Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {d25a2be2-b54f-400b-918b-b254e8044e39},
 private_publication = {false},
 abstract = {Multi-task learning (MTL) has been widely used in recommender systems, wherein predicting each type of user feedback on items (e.g, click, purchase) are treated as individual tasks and jointly trained with a unified model. Our key observation is that the prediction results of each task may contain task-specific knowledge about user’s fine-grained preference towards items. While such knowledge could be transferred to benefit other tasks, it is being overlooked under the current MTL paradigm. This paper, instead, proposes a Cross-Task Knowledge Distillation framework that attempts to leverage prediction results of one task as supervised signals to teach another task. However, integrating MTL and KD in a proper manner is non-trivial due to several challenges including task conflicts, inconsistent magnitude and requirement of synchronous optimization. As countermeasures, we 1) introduce auxiliary tasks with quadruplet loss functions to capture cross-task fine-grained ranking information and avoid task conflicts, 2) design a calibrated distillation approach to align and distill knowledge from auxiliary tasks, and 3) propose a novel error correction mechanism to enable and facilitate synchronous training of teacher and student models. Comprehensive experiments are conducted to verify the effectiveness of our framework in real-world datasets.},
 bibtype = {article},
 author = {Yang, Chenxiao and Pan, Junwei and Gao, Xiaofeng and Jiang, Tingyu and Liu, Dapeng and Chen, Guihai},
 doi = {10.1609/AAAI.V36I4.20352},
 journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
 number = {4}
}

Multi-task learning (MTL) has been widely used in recommender systems, wherein predicting each type of user feedback on items (e.g, click, purchase) are treated as individual tasks and jointly trained with a unified model. Our key observation is that the prediction results of each task may contain task-specific knowledge about user’s fine-grained preference towards items. While such knowledge could be transferred to benefit other tasks, it is being overlooked under the current MTL paradigm. This paper, instead, proposes a Cross-Task Knowledge Distillation framework that attempts to leverage prediction results of one task as supervised signals to teach another task. However, integrating MTL and KD in a proper manner is non-trivial due to several challenges including task conflicts, inconsistent magnitude and requirement of synchronous optimization. As countermeasures, we 1) introduce auxiliary tasks with quadruplet loss functions to capture cross-task fine-grained ranking information and avoid task conflicts, 2) design a calibrated distillation approach to align and distill knowledge from auxiliary tasks, and 3) propose a novel error correction mechanism to enable and facilitate synchronous training of teacher and student models. Comprehensive experiments are conducted to verify the effectiveness of our framework in real-world datasets.

@article{
 title = {Design and Implementation of an Urban Farming Robot},
 type = {article},
 year = {2022},
 keywords = {Plant detection,Precision irrigation,Robot,Urban agriculture,Vegetables},
 volume = {13},
 month = {2},
 publisher = {MDPI},
 day = {1},
 id = {22952c90-f177-3cde-9ba4-2810ce83f459},
 created = {2024-01-11T17:03:37.321Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-11T17:03:37.911Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {2bfb8d91-9fac-46f0-bd0c-93235d01dbed},
 private_publication = {false},
 abstract = {Urban agriculture can be shortly defined as the growing of plants and/or the livestock husbandry in and around cities. Although it has been a common occupation for the urban population all along, recently there is a growing interest in it both from public bodies and researchers, as well as from ordinary citizens who want to engage in self-cultivation. The modern citizen, though, will hardly find the free time to grow his own vegetables as it is a process that requires, in addition to knowledge and disposition, consistency. Given the above considerations, the purpose of this work was to develop an economic robotic system for the automatic monitoring and management of an urban garden. The robotic system was designed and built entirely from scratch. It had to have suitable dimensions so that it could be placed in a balcony or a terrace, and be able to scout vegetables from planting to harvest and primarily conduct precision irrigation based on the growth stage of each plant. Fertigation and weed control will also follow. For its development, a number of technologies were combined, such as Cartesian robots’ motion, machine vision, deep learning for the identification and detection of plants, irrigation dosage and scheduling based on plants’ growth stage, and cloud storage. The complete process of software and hardware development to a robust robotic platform is described in detail in the respective sections. The experimental procedure was performed for lettuce plants, with the robotic system providing precise movement of its actuator and applying precision irrigation based on the specific needs of the plants.},
 bibtype = {article},
 author = {Moraitis, Michail and Vaiopoulos, Konstantinos and Balafoutis, Athanasios T.},
 doi = {10.3390/mi13020250},
 journal = {Micromachines},
 number = {2}
}

Urban agriculture can be shortly defined as the growing of plants and/or the livestock husbandry in and around cities. Although it has been a common occupation for the urban population all along, recently there is a growing interest in it both from public bodies and researchers, as well as from ordinary citizens who want to engage in self-cultivation. The modern citizen, though, will hardly find the free time to grow his own vegetables as it is a process that requires, in addition to knowledge and disposition, consistency. Given the above considerations, the purpose of this work was to develop an economic robotic system for the automatic monitoring and management of an urban garden. The robotic system was designed and built entirely from scratch. It had to have suitable dimensions so that it could be placed in a balcony or a terrace, and be able to scout vegetables from planting to harvest and primarily conduct precision irrigation based on the growth stage of each plant. Fertigation and weed control will also follow. For its development, a number of technologies were combined, such as Cartesian robots’ motion, machine vision, deep learning for the identification and detection of plants, irrigation dosage and scheduling based on plants’ growth stage, and cloud storage. The complete process of software and hardware development to a robust robotic platform is described in detail in the respective sections. The experimental procedure was performed for lettuce plants, with the robotic system providing precise movement of its actuator and applying precision irrigation based on the specific needs of the plants.

@article{
 title = {Heatmap-based Explanation of YOLOv5 Object Detection with Layer-wise Relevance Propagation},
 type = {article},
 year = {2022},
 keywords = {Explainable AI,Heatmaps,Layer-wise Relevance Propagation,YOLO},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {c59d3b3d-7d27-339b-9b9d-1eaeb0adb381},
 created = {2024-01-12T09:00:12.721Z},
 accessed = {2024-01-12},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-31T12:52:29.418Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c},
 private_publication = {false},
 abstract = {Deep Neural Networks (DNNs) have been effective in providing solutions to most prevailing problems in the field Computer Vision. Modern Object Detector architecture families such as Single Shot Detector, You Only Look Once (YOLO), Region Based Convolutional Neural Networks (RCNN) and others are powerful models with strong capabilities of distinguishing many common object classes in images with unprecedented speed and accuracy. However, in most of their applications the use of deep networks conceals the inference of object class and location from the input data by introducing many layers of complex black-box functionality. This leads to sometimes bizarre and inexplicable detection outputs. A potential use of these networks in risk-prone applications motivates the development of techniques for visually interpreting their outputs, which could lead to more controlled training and deployment. Layer-wise Relevance Prop-agation (LRP) is a popular 'eXplainable AI' (XAI) technique for constructing such explanations, frequently used in investigating the output of DNNs. For object detection, LRP could provide us with a useful heatmap of the input's relevance to the output class, localized within an estimate of a bounding box. In this work, we propose a new and efficient method for conducting relevance propagation in object detector DNNs that leads to visually informative results. In our experiments, we make use of the YOLOv5 model showing meaningful visual explanations of its output. Our implementation philosophy exemplifies how relevance propagation can be regulated for complex modular architectures.},
 bibtype = {article},
 author = {Karasmanoglou, Apostolos and Antonakakis, Marios and Zervakis, Michalis},
 doi = {10.1109/IST55454.2022.9827744},
 journal = {IST 2022 - IEEE International Conference on Imaging Systems and Techniques, Proceedings}
}

Deep Neural Networks (DNNs) have been effective in providing solutions to most prevailing problems in the field Computer Vision. Modern Object Detector architecture families such as Single Shot Detector, You Only Look Once (YOLO), Region Based Convolutional Neural Networks (RCNN) and others are powerful models with strong capabilities of distinguishing many common object classes in images with unprecedented speed and accuracy. However, in most of their applications the use of deep networks conceals the inference of object class and location from the input data by introducing many layers of complex black-box functionality. This leads to sometimes bizarre and inexplicable detection outputs. A potential use of these networks in risk-prone applications motivates the development of techniques for visually interpreting their outputs, which could lead to more controlled training and deployment. Layer-wise Relevance Prop-agation (LRP) is a popular 'eXplainable AI' (XAI) technique for constructing such explanations, frequently used in investigating the output of DNNs. For object detection, LRP could provide us with a useful heatmap of the input's relevance to the output class, localized within an estimate of a bounding box. In this work, we propose a new and efficient method for conducting relevance propagation in object detector DNNs that leads to visually informative results. In our experiments, we make use of the YOLOv5 model showing meaningful visual explanations of its output. Our implementation philosophy exemplifies how relevance propagation can be regulated for complex modular architectures.

@techreport{
 title = {Industry Best Practices in Robotics Software Engineering},
 type = {techreport},
 year = {2022},
 websites = {https://rose-workshops.github.io},
 id = {93beb270-ffb1-3587-b39f-e1ec270fc816},
 created = {2024-01-29T10:34:29.025Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-01-29T10:34:29.539Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {9e28b3ab-dc65-4cb8-8373-439f2758a335},
 private_publication = {false},
 abstract = {Robotics software is pushing the limits of software engineering practice. The 3rd International Workshop on Robotics Software Engineering held a panel on "the best practices for robotic software engineering". This article shares the key takeaways that emerged from the discussion among the panelists and the workshop, ranging from architecting practices at the NASA/Caltech Jet Propulsion Laboratory, model-driven development at Bosch, development and testing of autonomous driving systems at Waymo, and testing of robotics software at XITASO. Researchers and practitioners can build on the contents of this paper to gain a fresh perspective on their activities and focus on the most pressing practices and challenges in developing robotics software today.},
 bibtype = {techreport},
 author = {Bocchino, Robert and Thackston Waymo, Allison and Angerer XITASO GmbH, Andreas and Ciccozzi, Federico}
}

Robotics software is pushing the limits of software engineering practice. The 3rd International Workshop on Robotics Software Engineering held a panel on "the best practices for robotic software engineering". This article shares the key takeaways that emerged from the discussion among the panelists and the workshop, ranging from architecting practices at the NASA/Caltech Jet Propulsion Laboratory, model-driven development at Bosch, development and testing of autonomous driving systems at Waymo, and testing of robotics software at XITASO. Researchers and practitioners can build on the contents of this paper to gain a fresh perspective on their activities and focus on the most pressing practices and challenges in developing robotics software today.

@article{
 title = {FlexBRDF: A flexible BRDF correction for grouped processing of airborne imaging spectroscopy flightlines},
 type = {article},
 year = {2022},
 pages = {e2021JG006622},
 volume = {127},
 id = {24cf56f2-84e0-3cd8-952b-2087b2dc6feb},
 created = {2024-02-14T13:08:56.273Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.273Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {queally2022flexbrdfflexiblebrdf},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Queally, Natalie and Ye, Zhiwei and Zheng, Ting and Chlus, Adam and Schneider, Fabian and Pavlick, Ryan P and Townsend, Philip A},
 journal = {Journal of Geophysical Research: Biogeosciences},
 number = {1}
}

@article{
 title = {A grape leaves disease recognition using Amazon Sage Maker},
 type = {article},
 year = {2022},
 pages = {45-55},
 volume = {5},
 id = {60243204-20d7-3411-b2f4-de931d15124f},
 created = {2024-02-14T13:08:56.420Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.420Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {tockova2022grapeleavesdisease},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Tockova, Angela and Zlatev, Zoran and Koceski, Saso},
 journal = {Balkan Journal of Applied Mathematics and Informatics},
 number = {2}
}

@article{
 title = {A CNN-SVM study based on selected deep features for grapevine leaves classification},
 type = {article},
 year = {2022},
 pages = {110425},
 volume = {188},
 id = {da46bf5b-da30-3784-b72b-a4f9ab5404e1},
 created = {2024-02-14T13:08:56.447Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.447Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {koklu2022cnnscmstudy},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Koklu, Murat and Unlersen, M Fahri and Ozkan, Ilker Ali and Aslan, M Fatih and Sabanci, Kadir},
 journal = {Measurement}
}

@inproceedings{
 title = {How important is UAVs RTK accuracy for the identification of certain vine diseases?},
 type = {inproceedings},
 year = {2022},
 pages = {239-243},
 id = {a2d55eba-d7f8-3ff9-96d2-5fda028b2a3e},
 created = {2024-02-14T13:08:56.451Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.451Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {zottele2022howimportantuav},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Zottele, Fabio and Crocetta, Paolo and Baiocchi, Valerio},
 booktitle = {2022 IEEE Workshop on Metrology for Agriculture and Forestry (MetroAgriFor)}
}

@inproceedings{
 title = {Meteorological Data and UAV Images for the Detection and Identification of Grapevine Disease Using Deep Learning},
 type = {inproceedings},
 year = {2022},
 pages = {1-4},
 id = {3d9b9186-d97c-3400-b4bf-c1f26a201a80},
 created = {2024-02-14T13:08:56.821Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:56.821Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {ouhami2022meteorologicaldatauav},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Ouhami, Maryam and Es-saady, Youssef and Hajj, Mohammed El and Canals, Raphael and Hafiane, Adel},
 booktitle = {2022 E-Health and Bioengineering Conference (EHB)}
}

@article{
 title = {Deep semantic segmentation for the quantification of grape foliar diseases in the vineyard},
 type = {article},
 year = {2022},
 pages = {978761},
 volume = {13},
 id = {066c8f6f-17ec-34ee-b128-1283982a1648},
 created = {2024-02-14T13:08:57.468Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:57.468Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {liu2022deepsemanticsegmentation},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Liu, Ertai and Gold, Kaitlin M and Combs, David and Cadle-Davidson, Lance and Jiang, Yu},
 journal = {Frontiers in Plant Science}
}

@article{
 title = {The spectral mixture residual: A source of low-variance information to enhance the explainability and accuracy of surface biology and geology retrievals},
 type = {article},
 year = {2022},
 pages = {e2021JG006672},
 volume = {127},
 id = {0ee1f1ca-f078-3684-83b7-41f32c3660aa},
 created = {2024-02-14T13:08:57.892Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:57.892Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {sousa2022spectralmixtureresidual},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Sousa, Daniel and Brodrick, P and Cawse-Nicholson, Kerry and Fisher, Joshua B and Pavlick, Ryan and Small, Christopher and Thompson, David Ray},
 journal = {Journal of Geophysical Research: Biogeosciences},
 number = {2}
}

@article{
 title = {Comparison of Deep Neural Networks in Detecting Field Grapevine Diseases Using Transfer Learning},
 type = {article},
 year = {2022},
 pages = {4648},
 volume = {14},
 id = {e1d7ff46-8d87-3b42-a70c-ab82c24c23a8},
 created = {2024-02-14T13:08:58.933Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:58.933Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {morellos2022comparisondeepneural},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Morellos, Antonios and Pantazi, Xanthoula Eirini and Paraskevas, Charalampos and Moshou, Dimitrios},
 journal = {Remote Sensing},
 number = {18}
}

@article{
 title = {A Novel Enhanced VGG16 Model to Tackle Grapevine Leaves Diseases With Automatic Method},
 type = {article},
 year = {2022},
 pages = {111564-111578},
 volume = {10},
 id = {0e76473b-7eea-3a58-a9db-d27842e5f5d3},
 created = {2024-02-14T13:08:59.543Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.543Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {mousavi2022novelenhancedvgg16},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Mousavi, Seyed Amirhossein and Farahani, Gholamreza},
 journal = {IEEE Access}
}

@article{
 title = {Grapevine Plant Image Dataset for Pruning},
 type = {article},
 year = {2022},
 pages = {110},
 volume = {7},
 id = {9d160e1a-9090-3cf1-b945-2bd7504914aa},
 created = {2024-02-14T13:08:59.833Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.833Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {apostolidis2022grapevineplantimagedataset},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Apostolidis, Kyriakos D and Kalampokas, Theofanis and Pachidis, Theodore P and Kaburlasos, Vassilis G},
 journal = {Data},
 number = {8}
}

@article{
 title = {VineInspector: The Vineyard Assistant},
 type = {article},
 year = {2022},
 pages = {730},
 volume = {12},
 id = {ae50194d-91fe-3b55-a662-f797194da7ba},
 created = {2024-02-14T13:08:59.845Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.845Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {mendes2022vineinspector},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Mendes, Jorge and Peres, Emanuel and dos Santos, Filipe and Silva, Nuno and Silva, Renato and Sousa, Joaquim João and Cortez, Isabel and Morais, Raul},
 journal = {Agriculture},
 number = {5}
}

@article{
 title = {wGrapeUNIPD-DL: An open dataset for white grape bunch detection},
 type = {article},
 year = {2022},
 pages = {108466},
 volume = {43},
 id = {1cb99ee8-bce4-30b2-9fa9-b888f74185a6},
 created = {2024-02-14T13:08:59.981Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:08:59.981Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {sozzi2022wgrapeunipddlopendataset},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Sozzi, Marco and Cantalamessa, Silvia and Cogato, Alessia and Kayad, Ahmed and Marinello, Francesco},
 journal = {Data in Brief}
}

@article{
 title = {Detection of Vineyard Diseases Using the Internet of Things Technology and Machine Learning Algorithms},
 type = {article},
 year = {2022},
 pages = {128-139},
 id = {cda9b581-ed99-34bd-b161-2394d1d8313f},
 created = {2024-02-14T13:09:00.333Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:00.333Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {roscuaneanu2022detectionvineyarddiseases},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Ro\csc\uaneanu, Roxana and Streche, Robert and Osiac, Filip and B\ual\uaceanu, Cristina and Suciu, George and Dr\uagulinescu, Ana Maria and Marcu, Ioana},
 journal = {Aerul si Apa. Componente ale Mediului}
}

@article{
 title = {Deep Learning Based Automatic Grape Downy Mildew Detection},
 type = {article},
 year = {2022},
 volume = {13},
 id = {8aab7577-2a38-307b-a02e-1da31a8d6f95},
 created = {2024-02-14T13:09:00.402Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:00.402Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {zhang2022deeplearningbasedautomatic},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Zhang, Zhao and Qiao, Yongliang and Guo, Yangyang and He, Dongjian},
 journal = {Frontiers in Plant Science}
}

@article{
 title = {Fruit yield prediction and estimation in orchards: a state-of-the-art comprehensive review for both direct and indirect methods},
 type = {article},
 year = {2022},
 pages = {106812},
 volume = {195},
 id = {722daf09-9864-3ec2-9c94-41d9610a95cf},
 created = {2024-02-14T13:09:00.743Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:00.743Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {he2022fruityieldprediction},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {He, Leilei and Fang, Wentai and Zhao, Guanao and Wu, Zhenchao and Fu, Longsheng and Li, Rui and Majeed, Yaqoob and Dhupia, Jaspreet Singh},
 journal = {Computers and Electronics in Agriculture}
}

@article{
 title = {Detection of Anomalous Grapevine Berries Using Variational Autoencoders},
 type = {article},
 year = {2022},
 pages = {729097},
 volume = {13},
 id = {bd9d2bb9-2a29-3789-ac6d-191045c1ce68},
 created = {2024-02-14T13:09:01.124Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:01.124Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {miranda2022detectionanomalousgrapevine},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Miranda, Miro and Zabawa, Laura and Kicherer, Anna and Strothmann, Laurenz and Rascher, Uwe and Roscher, Ribana},
 journal = {Frontiers in Plant Science}
}

@inproceedings{
 title = {Integrated IoT System for Prediction of Diseases in the Vineyards},
 type = {inproceedings},
 year = {2022},
 pages = {1-6},
 id = {573b48db-542a-3365-9ee2-e3d64f89fa66},
 created = {2024-02-14T13:09:01.127Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:01.127Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {jovanovska2022integratediotsystem},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Jovanovska, Elena M and Chorbev, Ivan and Davcev, Danco and Mitreski, Kosta},
 booktitle = {2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME)}
}

@article{
 title = {Computer Vision and Deep Learning for Precision Viticulture},
 type = {article},
 year = {2022},
 pages = {2463},
 volume = {12},
 id = {c971d344-7b4d-39ca-8e2d-94af6dae4635},
 created = {2024-02-14T13:09:01.670Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:01.670Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {mohimont2022computervisiondeep},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Mohimont, Lucas and Alin, François and Rondeau, Marine and Gaveau, Nathalie and Steffenel, Luiz Angelo},
 journal = {Agronomy},
 number = {10}
}

@inproceedings{
 title = {A multispectral acquisition system for potential detection of Flavescence dorée},
 type = {inproceedings},
 year = {2022},
 pages = {1-4},
 id = {8d2d9f65-d879-35c8-9206-a28f4e99988e},
 created = {2024-02-14T13:09:02.115Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:02.115Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {barjaktarovic2022multispectralacquisition},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Barjaktarović, Marko and Santoni, Massimo and Faralli, Michele and Bertamini, Massimo and Bruzzone, Lorenzo},
 booktitle = {2022 30th Telecommunications Forum (TELFOR)}
}

@article{
 title = {Design of Intelligent Detection Platform for Wine Grape Pests and Diseases in Ningxia},
 type = {article},
 year = {2022},
 pages = {106},
 volume = {12},
 id = {638f9154-e36b-3f70-bde7-3608b82120e5},
 created = {2024-02-14T13:09:02.502Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:02.502Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {wang2022designintelligentdetection},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Wang, Yutan and Wei, Chi and Sun, Haowei and Qu, Aili},
 journal = {Plants},
 number = {1}
}

@misc{
 title = {VitiVisor Vineyard Datasets},
 type = {misc},
 year = {2022},
 websites = {https://www.kaggle.com/ds/1248678},
 publisher = {The University of Adelaide},
 id = {625a018c-f923-398d-b8a2-158cc9192fa0},
 created = {2024-02-14T13:09:03.746Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:03.746Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {collins2022vitivisor},
 source_type = {misc},
 private_publication = {false},
 bibtype = {misc},
 author = {Collins, Cassandra and Kiely, Wayne and Grigg, Dylan}
}

@inproceedings{
 title = {Identification of Grape Leaf Diseases Using Proposed Enhanced VGG16},
 type = {inproceedings},
 year = {2022},
 pages = {1-6},
 id = {09690156-40e9-338c-84f9-92a1a861f92c},
 created = {2024-02-14T13:09:04.406Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:04.406Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {farahani2022identificationgrapeleaf},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Farahani, Gholamreza and Mousavi, Seyedamirhossein and Farahani, Ameneh and Farahani, Hamidreza},
 booktitle = {2022 27th International Conference on Automation and Computing (ICAC)}
}

@article{
 title = {Semi-supervised deep learning and low-cost cameras for the semantic segmentation of natural images in viticulture},
 type = {article},
 year = {2022},
 pages = {1-26},
 id = {a0ac9b9a-fffc-3dd9-8b54-a32d5bae286c},
 created = {2024-02-14T13:09:04.487Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:04.487Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {casado2022semisuperviseddeep},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Casado-García, A and Heras, J and Milella, A and Marani, R},
 journal = {Precision Agriculture}
}

@article{
 title = {Multispectral vineyard segmentation: a deep learning comparison study},
 type = {article},
 year = {2022},
 pages = {106782},
 volume = {195},
 id = {f4618b7f-c438-34eb-a913-8084e0b137e2},
 created = {2024-02-14T13:09:04.698Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:04.698Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {barros2022multispectralvineyardsegmentation},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Barros, Tiago and Conde, Pedro and Gonçalves, Gil and Premebida, Cristiano and Monteiro, Miguel and Ferreira, Carla Sofia Santos and Nunes, Urbano J},
 journal = {Computers and Electronics in Agriculture}
}

@article{
 title = {Plant Viral Disease Detection: From Molecular Diagnosis to Optical Sensing Technology - A Multidisciplinary Review},
 type = {article},
 year = {2022},
 pages = {1542},
 volume = {14},
 id = {6a8ea89c-5f7a-39b3-be07-cdda04e63276},
 created = {2024-02-14T13:09:05.001Z},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-02-14T13:09:05.001Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {wang2022plantviraldisease},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Wang, Yeniu Mickey and Ostendorf, Bertram and Gautam, Deepak and Habili, Nuredin and Pagay, Vinay},
 journal = {Remote Sensing},
 number = {7}
}

@article{
 title = {A General Gaussian Heatmap Label Assignment for Arbitrary-Oriented Object Detection},
 type = {article},
 year = {2022},
 keywords = {Arbitrary-oriented object,convolutional neural network,gaussian heatmap,label assignment,object detection},
 pages = {1895-1910},
 volume = {31},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {865832c1-dde7-3f65-8514-377700a136be},
 created = {2024-02-16T10:27:33.307Z},
 accessed = {2024-02-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-04-08T06:04:34.463Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Recently, many arbitrary-oriented object detection (AOOD) methods have been proposed and attracted widespread attention in many fields. However, most of them are based on anchor-boxes or standard Gaussian heatmaps. Such label assignment strategy may not only fail to reflect the shape and direction characteristics of arbitrary-oriented objects, but also have high parameter-tuning efforts. In this paper, a novel AOOD method called General Gaussian Heatmap Label Assignment (GGHL) is proposed. Specifically, an anchor-free object-adaptation label assignment (OLA) strategy is presented to define the positive candidates based on two-dimensional (2D) oriented Gaussian heatmaps, which reflect the shape and direction features of arbitrary-oriented objects. Based on OLA, an oriented-bounding-box (OBB) representation component (ORC) is developed to indicate OBBs and adjust the Gaussian center prior weights to fit the characteristics of different objects adaptively through neural network learning. Moreover, a joint-optimization loss (JOL) with area normalization and dynamic confidence weighting is designed to refine the misalign optimal results of different subtasks. Extensive experiments on public datasets demonstrate that the proposed GGHL improves the AOOD performance with low parameter-tuning and time costs. Furthermore, it is generally applicable to most AOOD methods to improve their performance including lightweight models on embedded platforms.},
 bibtype = {article},
 author = {Huang, Zhanchao and Li, Wei and Xia, Xiang Gen and Tao, Ran},
 doi = {10.1109/TIP.2022.3148874},
 journal = {IEEE Transactions on Image Processing}
}

Recently, many arbitrary-oriented object detection (AOOD) methods have been proposed and attracted widespread attention in many fields. However, most of them are based on anchor-boxes or standard Gaussian heatmaps. Such label assignment strategy may not only fail to reflect the shape and direction characteristics of arbitrary-oriented objects, but also have high parameter-tuning efforts. In this paper, a novel AOOD method called General Gaussian Heatmap Label Assignment (GGHL) is proposed. Specifically, an anchor-free object-adaptation label assignment (OLA) strategy is presented to define the positive candidates based on two-dimensional (2D) oriented Gaussian heatmaps, which reflect the shape and direction features of arbitrary-oriented objects. Based on OLA, an oriented-bounding-box (OBB) representation component (ORC) is developed to indicate OBBs and adjust the Gaussian center prior weights to fit the characteristics of different objects adaptively through neural network learning. Moreover, a joint-optimization loss (JOL) with area normalization and dynamic confidence weighting is designed to refine the misalign optimal results of different subtasks. Extensive experiments on public datasets demonstrate that the proposed GGHL improves the AOOD performance with low parameter-tuning and time costs. Furthermore, it is generally applicable to most AOOD methods to improve their performance including lightweight models on embedded platforms.

@article{
 title = {Align Deep Features for Oriented Object Detection},
 type = {article},
 year = {2022},
 keywords = {Aerial images,deep learning,feature alignment,object detection},
 volume = {60},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {727ad0ce-566d-3250-98a6-dec0ab03c05d},
 created = {2024-02-16T10:35:07.613Z},
 accessed = {2024-02-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-27T10:37:23.542Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.},
 bibtype = {article},
 author = {Han, Jiaming and Ding, Jian and Li, Jie and Xia, Gui Song},
 doi = {10.1109/TGRS.2021.3062048},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.

@article{
 title = {Align Deep Features for Oriented Object Detection},
 type = {article},
 year = {2022},
 keywords = {Aerial images,deep learning,feature alignment,object detection},
 volume = {60},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {5769d2d3-5629-3188-bc61-c0fd0c0c0b94},
 created = {2024-02-16T10:35:49.579Z},
 accessed = {2024-02-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-14T10:29:17.441Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 citation_key = {Han2022},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0,5a010301-acb6-4642-a6b2-8afaee1b741c},
 private_publication = {false},
 abstract = {The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.},
 bibtype = {article},
 author = {Han, Jiaming and Ding, Jian and Li, Jie and Xia, Gui Song},
 doi = {10.1109/TGRS.2021.3062048},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.

@article{
 title = {Anchor-Free Oriented Proposal Generator for Object Detection},
 type = {article},
 year = {2022},
 keywords = {Anchor-free oriented proposal generator (AOPG),oriented object detection,oriented proposal generation},
 volume = {60},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {75eef121-8aa3-3414-81a0-0782295e9deb},
 created = {2024-03-04T07:49:55.970Z},
 accessed = {2024-03-04},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-04T11:58:10.399Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Oriented object detection is a practical and challenging task in remote sensing image interpretation. Nowadays, oriented detectors mostly use horizontal boxes as intermedium to derive oriented boxes from them. However, the horizontal boxes are inclined to get small Intersection-over-Unions (IoUs) with ground truths, which may have some undesirable effects, such as introducing redundant noise, mismatching with ground truths, and detracting from the robustness of detectors. In this article, we propose a novel anchor-free oriented proposal generator (AOPG) that abandons horizontal box-related operations from the network architecture. AOPG first produces coarse oriented boxes by a coarse location module (CLM) in an anchor-free manner and then refines them into high-quality oriented proposals. After AOPG, we apply a Fast Region-based Convolutional Neural Network (R-CNN) head to produce the final detection results. Furthermore, the shortage of large-scale datasets is also a hindrance to the development of oriented object detection. To alleviate the data insufficiency, we release a new dataset on the basis of our DIOR dataset and name it DIOR-R. Massive experiments demonstrate the effectiveness of AOPG. Particularly, without bells and whistles, we achieve the accuracy of 64.41%, 75.24%, and 96.22% mAP on the DIOR-R, DOTA, and HRSC2016 datasets, respectively. Code and models are available at https://github.com/jbwang1997/AOPG.},
 bibtype = {article},
 author = {Cheng, Gong and Wang, Jiabao and Li, Ke and Xie, Xingxing and Lang, Chunbo and Yao, Yanqing and Han, Junwei},
 doi = {10.1109/TGRS.2022.3183022},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

Oriented object detection is a practical and challenging task in remote sensing image interpretation. Nowadays, oriented detectors mostly use horizontal boxes as intermedium to derive oriented boxes from them. However, the horizontal boxes are inclined to get small Intersection-over-Unions (IoUs) with ground truths, which may have some undesirable effects, such as introducing redundant noise, mismatching with ground truths, and detracting from the robustness of detectors. In this article, we propose a novel anchor-free oriented proposal generator (AOPG) that abandons horizontal box-related operations from the network architecture. AOPG first produces coarse oriented boxes by a coarse location module (CLM) in an anchor-free manner and then refines them into high-quality oriented proposals. After AOPG, we apply a Fast Region-based Convolutional Neural Network (R-CNN) head to produce the final detection results. Furthermore, the shortage of large-scale datasets is also a hindrance to the development of oriented object detection. To alleviate the data insufficiency, we release a new dataset on the basis of our DIOR dataset and name it DIOR-R. Massive experiments demonstrate the effectiveness of AOPG. Particularly, without bells and whistles, we achieve the accuracy of 64.41%, 75.24%, and 96.22% mAP on the DIOR-R, DOTA, and HRSC2016 datasets, respectively. Code and models are available at https://github.com/jbwang1997/AOPG.

@article{
 title = {Anchor-Free Oriented Proposal Generator for Object Detection},
 type = {article},
 year = {2022},
 keywords = {Anchor-free oriented proposal generator (AOPG),oriented object detection,oriented proposal generation},
 volume = {60},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {2867128e-7a00-3436-8ab1-40dd7539a140},
 created = {2024-03-04T07:54:14.986Z},
 accessed = {2024-03-04},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-14T10:29:17.637Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 citation_key = {Cheng2022},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Oriented object detection is a practical and challenging task in remote sensing image interpretation. Nowadays, oriented detectors mostly use horizontal boxes as intermedium to derive oriented boxes from them. However, the horizontal boxes are inclined to get small Intersection-over-Unions (IoUs) with ground truths, which may have some undesirable effects, such as introducing redundant noise, mismatching with ground truths, and detracting from the robustness of detectors. In this article, we propose a novel anchor-free oriented proposal generator (AOPG) that abandons horizontal box-related operations from the network architecture. AOPG first produces coarse oriented boxes by a coarse location module (CLM) in an anchor-free manner and then refines them into high-quality oriented proposals. After AOPG, we apply a Fast Region-based Convolutional Neural Network (R-CNN) head to produce the final detection results. Furthermore, the shortage of large-scale datasets is also a hindrance to the development of oriented object detection. To alleviate the data insufficiency, we release a new dataset on the basis of our DIOR dataset and name it DIOR-R. Massive experiments demonstrate the effectiveness of AOPG. Particularly, without bells and whistles, we achieve the accuracy of 64.41%, 75.24%, and 96.22% mAP on the DIOR-R, DOTA, and HRSC2016 datasets, respectively. Code and models are available at https://github.com/jbwang1997/AOPG.},
 bibtype = {article},
 author = {Cheng, Gong and Wang, Jiabao and Li, Ke and Xie, Xingxing and Lang, Chunbo and Yao, Yanqing and Han, Junwei},
 doi = {10.1109/TGRS.2022.3183022},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

Oriented object detection is a practical and challenging task in remote sensing image interpretation. Nowadays, oriented detectors mostly use horizontal boxes as intermedium to derive oriented boxes from them. However, the horizontal boxes are inclined to get small Intersection-over-Unions (IoUs) with ground truths, which may have some undesirable effects, such as introducing redundant noise, mismatching with ground truths, and detracting from the robustness of detectors. In this article, we propose a novel anchor-free oriented proposal generator (AOPG) that abandons horizontal box-related operations from the network architecture. AOPG first produces coarse oriented boxes by a coarse location module (CLM) in an anchor-free manner and then refines them into high-quality oriented proposals. After AOPG, we apply a Fast Region-based Convolutional Neural Network (R-CNN) head to produce the final detection results. Furthermore, the shortage of large-scale datasets is also a hindrance to the development of oriented object detection. To alleviate the data insufficiency, we release a new dataset on the basis of our DIOR dataset and name it DIOR-R. Massive experiments demonstrate the effectiveness of AOPG. Particularly, without bells and whistles, we achieve the accuracy of 64.41%, 75.24%, and 96.22% mAP on the DIOR-R, DOTA, and HRSC2016 datasets, respectively. Code and models are available at https://github.com/jbwang1997/AOPG.

@article{
 title = {A General Gaussian Heatmap Label Assignment for Arbitrary-Oriented Object Detection},
 type = {article},
 year = {2022},
 keywords = {Arbitrary-oriented object,convolutional neural network,gaussian heatmap,label assignment,object detection},
 pages = {1895-1910},
 volume = {31},
 id = {ee5720ae-2fa4-348c-aa41-555c98893e74},
 created = {2024-03-04T11:52:40.167Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-04-08T06:04:35.107Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {Recently, many arbitrary-oriented object detection (AOOD) methods have been proposed and attracted widespread attention in many fields. However, most of them are based on anchor-boxes or standard Gaussian heatmaps. Such label assignment strategy may not only fail to reflect the shape and direction characteristics of arbitrary-oriented objects, but also have high parameter-tuning efforts. In this paper, a novel AOOD method called General Gaussian Heatmap Label Assignment (GGHL) is proposed. Specifically, an anchor-free object-adaptation label assignment (OLA) strategy is presented to define the positive candidates based on two-dimensional (2D) oriented Gaussian heatmaps, which reflect the shape and direction features of arbitrary-oriented objects. Based on OLA, an oriented-bounding-box (OBB) representation component (ORC) is developed to indicate OBBs and adjust the Gaussian center prior weights to fit the characteristics of different objects adaptively through neural network learning. Moreover, a joint-optimization loss (JOL) with area normalization and dynamic confidence weighting is designed to refine the misalign optimal results of different subtasks. Extensive experiments on public datasets demonstrate that the proposed GGHL improves the AOOD performance with low parameter-tuning and time costs. Furthermore, it is generally applicable to most AOOD methods to improve their performance including lightweight models on embedded platforms.},
 bibtype = {article},
 author = {Huang, Zhanchao and Li, Wei and Xia, Xiang Gen and Tao, Ran},
 doi = {10.1109/TIP.2022.3148874},
 journal = {IEEE Transactions on Image Processing}
}

Recently, many arbitrary-oriented object detection (AOOD) methods have been proposed and attracted widespread attention in many fields. However, most of them are based on anchor-boxes or standard Gaussian heatmaps. Such label assignment strategy may not only fail to reflect the shape and direction characteristics of arbitrary-oriented objects, but also have high parameter-tuning efforts. In this paper, a novel AOOD method called General Gaussian Heatmap Label Assignment (GGHL) is proposed. Specifically, an anchor-free object-adaptation label assignment (OLA) strategy is presented to define the positive candidates based on two-dimensional (2D) oriented Gaussian heatmaps, which reflect the shape and direction features of arbitrary-oriented objects. Based on OLA, an oriented-bounding-box (OBB) representation component (ORC) is developed to indicate OBBs and adjust the Gaussian center prior weights to fit the characteristics of different objects adaptively through neural network learning. Moreover, a joint-optimization loss (JOL) with area normalization and dynamic confidence weighting is designed to refine the misalign optimal results of different subtasks. Extensive experiments on public datasets demonstrate that the proposed GGHL improves the AOOD performance with low parameter-tuning and time costs. Furthermore, it is generally applicable to most AOOD methods to improve their performance including lightweight models on embedded platforms.

@article{
 title = {RAOD: refined oriented detector with augmented feature in remote sensing images object detection},
 type = {article},
 year = {2022},
 keywords = {Augmented feature pyramid,Deformable RoI pooling,Oriented object detection,Remote sensing image,Rotated RoI align},
 pages = {15278-15294},
 volume = {52},
 publisher = {Applied Intelligence},
 id = {2190af1b-a4c7-313d-85e4-335e52203ac6},
 created = {2024-03-04T11:52:40.168Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-04T11:52:58.344Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Object detection is a challenging task in remote sensing. Aerial images are distinguished by complex backgrounds, arbitrary orientations, and dense distributions. Considering those difficulties, this paper proposes a two-stage refined oriented detector with augmented features named RAOD. First, a novel Augmented Feature Pyramid Network (A-FPN) is built to enhance fusion both in spatial and channel dimensions. Specifically, it mainly consists of three modules: Scale Transfer Module (STM), Feature Aggregate Module (FAM) and Feature Refinement Module (FRM). STM reduces information loss when fusing features in the top-down pathway. FAM aggregates features from different scales. FRM aims to refine the integrated features using a lightweight attention module. Then, we adopt a two-step processing, which consists of a coarse stage and a refinement stage. In the coarse stage, deformable RoI pooling is adopted to improve the network’s ability of modeling spatial transformations and then horizontal proposals are transformed into oriented ones. In the refinement stage, Rotated RoI align (RRoI align) is used to extract rotation-invariant features from rotated RoIs and further optimize the localization. To enhance stability and robustness during training, smooth Ln is chosen as regression loss as it has better ability in terms of robustness and stability than smooth L1 loss. Extensive experiments on several rotation detection datasets demonstrate the effectiveness of our method. Results show that our method is able to achieve 79.78%, 74.7% and 94.82% on DOTA-v1.0, DOTA-v1.5 and HRSC2016, respectively.},
 bibtype = {article},
 author = {Shi, Qin and Zhu, Yu and Fang, Chuantao and Wang, Nan and Lin, Jiajun},
 doi = {10.1007/s10489-022-03393-8},
 journal = {Applied Intelligence},
 number = {13}
}

Object detection is a challenging task in remote sensing. Aerial images are distinguished by complex backgrounds, arbitrary orientations, and dense distributions. Considering those difficulties, this paper proposes a two-stage refined oriented detector with augmented features named RAOD. First, a novel Augmented Feature Pyramid Network (A-FPN) is built to enhance fusion both in spatial and channel dimensions. Specifically, it mainly consists of three modules: Scale Transfer Module (STM), Feature Aggregate Module (FAM) and Feature Refinement Module (FRM). STM reduces information loss when fusing features in the top-down pathway. FAM aggregates features from different scales. FRM aims to refine the integrated features using a lightweight attention module. Then, we adopt a two-step processing, which consists of a coarse stage and a refinement stage. In the coarse stage, deformable RoI pooling is adopted to improve the network’s ability of modeling spatial transformations and then horizontal proposals are transformed into oriented ones. In the refinement stage, Rotated RoI align (RRoI align) is used to extract rotation-invariant features from rotated RoIs and further optimize the localization. To enhance stability and robustness during training, smooth Ln is chosen as regression loss as it has better ability in terms of robustness and stability than smooth L1 loss. Extensive experiments on several rotation detection datasets demonstrate the effectiveness of our method. Results show that our method is able to achieve 79.78%, 74.7% and 94.82% on DOTA-v1.0, DOTA-v1.5 and HRSC2016, respectively.

@article{
 title = {Align Deep Features for Oriented Object Detection},
 type = {article},
 year = {2022},
 keywords = {Aerial images,deep learning,feature alignment,object detection},
 pages = {1-10},
 volume = {60},
 id = {ba6421e9-aa4f-3535-9594-e1f74cbef1e1},
 created = {2024-03-04T11:52:40.171Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-04T11:52:51.381Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.},
 bibtype = {article},
 author = {Han, Jiaming and Ding, Jian and Li, Jie and Xia, Gui Song},
 doi = {10.1109/TGRS.2021.3062048},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.

@article{
 title = {An Elliptic Centerness for Object Instance Segmentation in Aerial Images},
 type = {article},
 year = {2022},
 volume = {2022},
 id = {fa9bfd49-a3db-33cd-ac4d-f861e599aa93},
 created = {2024-03-04T11:52:40.358Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-04T11:52:54.950Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Instance segmentation in aerial images is an important and challenging task. Most of the existing methods have adapted instance segmentation algorithms developed for natural images to aerial images. However, these methods easily suffer from performance degradation in aerial images, due to the scale variations, large aspect ratios, and arbitrary orientations of instances caused by the bird’s-eye view of aerial images. To address this issue, we propose an elliptic centerness (EC) for instance segmentation in aerial images, which can assign the proper centerness values to the intricate aerial instances and thus mitigate the performance degradation. Specifically, we introduce ellipses to fit the various contours of aerial instances and measure these fitted ellipses by two-dimensional anisotropic Gaussian distribution. Armed with EC, we develop a one-stage aerial instance segmentation network. Extensive experiments on a commonly used dataset, the instance segmentation in aerial images dataset (iSAID), demonstrate that our proposed method can achieve a remarkable performance of instance segmentation while introducing negligible computational cost.},
 bibtype = {article},
 author = {Luo, Yixin and Han, Jiaming and Liu, Zhou and Wang, Mi and Xia, Gui Song},
 doi = {10.34133/2022/9809505},
 journal = {Journal of Remote Sensing (United States)}
}

Instance segmentation in aerial images is an important and challenging task. Most of the existing methods have adapted instance segmentation algorithms developed for natural images to aerial images. However, these methods easily suffer from performance degradation in aerial images, due to the scale variations, large aspect ratios, and arbitrary orientations of instances caused by the bird’s-eye view of aerial images. To address this issue, we propose an elliptic centerness (EC) for instance segmentation in aerial images, which can assign the proper centerness values to the intricate aerial instances and thus mitigate the performance degradation. Specifically, we introduce ellipses to fit the various contours of aerial instances and measure these fitted ellipses by two-dimensional anisotropic Gaussian distribution. Armed with EC, we develop a one-stage aerial instance segmentation network. Extensive experiments on a commonly used dataset, the instance segmentation in aerial images dataset (iSAID), demonstrate that our proposed method can achieve a remarkable performance of instance segmentation while introducing negligible computational cost.

@article{
 title = {On the Arbitrary-Oriented Object Detection: Classification Based Approaches Revisited},
 type = {article},
 year = {2022},
 keywords = {Arbitrary-oriented object detection,Boundary problem,Circular smooth label,Densely coded labels,Object heading detection},
 pages = {1340-1365},
 volume = {130},
 id = {991f6494-8bda-35ec-b416-58ec9e58320d},
 created = {2024-03-15T07:41:30.928Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2025-01-27T13:24:54.891Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Arbitrary-oriented object detection has been a building block for rotation sensitive tasks. We first show that the boundary problem suffered in existing dominant regression-based rotation detectors, is caused by angular periodicity or corner ordering, according to the parameterization protocol. We also show that the root cause is that the ideal predictions can be out of the defined range. Accordingly, we transform the angular prediction task from a regression problem to a classification one. For the resulting circularly distributed angle classification problem, we first devise a Circular Smooth Label technique to handle the periodicity of angle and increase the error tolerance to adjacent angles. To reduce the excessive model parameters by Circular Smooth Label, we further design a Densely Coded Labels, which greatly reduces the length of the encoding. Finally, we further develop an object heading detection module, which can be useful when the exact heading orientation information is needed e.g. for ship and plane heading detection. We release our OHD-SJTU dataset and OHDet detector for heading detection. Extensive experimental results on three large-scale public datasets for aerial images i.e. DOTA, HRSC2016, OHD-SJTU, and face dataset FDDB, as well as scene text dataset ICDAR2015 and MLT, show the effectiveness of our approach.},
 bibtype = {article},
 author = {Yang, Xue and Yan, Junchi},
 doi = {10.1007/s11263-022-01593-w},
 journal = {International Journal of Computer Vision},
 number = {5}
}

Arbitrary-oriented object detection has been a building block for rotation sensitive tasks. We first show that the boundary problem suffered in existing dominant regression-based rotation detectors, is caused by angular periodicity or corner ordering, according to the parameterization protocol. We also show that the root cause is that the ideal predictions can be out of the defined range. Accordingly, we transform the angular prediction task from a regression problem to a classification one. For the resulting circularly distributed angle classification problem, we first devise a Circular Smooth Label technique to handle the periodicity of angle and increase the error tolerance to adjacent angles. To reduce the excessive model parameters by Circular Smooth Label, we further design a Densely Coded Labels, which greatly reduces the length of the encoding. Finally, we further develop an object heading detection module, which can be useful when the exact heading orientation information is needed e.g. for ship and plane heading detection. We release our OHD-SJTU dataset and OHDet detector for heading detection. Extensive experimental results on three large-scale public datasets for aerial images i.e. DOTA, HRSC2016, OHD-SJTU, and face dataset FDDB, as well as scene text dataset ICDAR2015 and MLT, show the effectiveness of our approach.

@article{
 title = {RTMDet: An Empirical Study of Designing Real-Time Object Detectors},
 type = {article},
 year = {2022},
 pages = {63-65},
 websites = {http://arxiv.org/abs/2212.07784},
 id = {ccb3a7ac-19a9-3f6e-a3bd-01e09efc2845},
 created = {2024-03-15T07:41:30.933Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-15T07:41:41.615Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {In this paper, we aim to design an efficient real-time object detector that exceeds the YOLO series and is easily extensible for many object recognition tasks such as instance segmentation and rotated object detection. To obtain a more efficient model architecture, we explore an architecture that has compatible capacities in the backbone and neck, constructed by a basic building block that consists of large-kernel depth-wise convolutions. We further introduce soft labels when calculating matching costs in the dynamic label assignment to improve accuracy. Together with better training techniques, the resulting object detector, named RTMDet, achieves 52.8% AP on COCO with 300+ FPS on an NVIDIA 3090 GPU, outperforming the current mainstream industrial detectors. RTMDet achieves the best parameter-accuracy trade-off with tiny/small/medium/large/extra-large model sizes for various application scenarios, and obtains new state-of-the-art performance on real-time instance segmentation and rotated object detection. We hope the experimental results can provide new insights into designing versatile real-time object detectors for many object recognition tasks. Code and models are released at https://github.com/open-mmlab/mmdetection/tree/3.x/configs/rtmdet.},
 bibtype = {article},
 author = {Lyu, Chengqi and Zhang, Wenwei and Huang, Haian and Zhou, Yue and Wang, Yudong and Liu, Yanyi and Zhang, Shilong and Chen, Kai}
}

In this paper, we aim to design an efficient real-time object detector that exceeds the YOLO series and is easily extensible for many object recognition tasks such as instance segmentation and rotated object detection. To obtain a more efficient model architecture, we explore an architecture that has compatible capacities in the backbone and neck, constructed by a basic building block that consists of large-kernel depth-wise convolutions. We further introduce soft labels when calculating matching costs in the dynamic label assignment to improve accuracy. Together with better training techniques, the resulting object detector, named RTMDet, achieves 52.8% AP on COCO with 300+ FPS on an NVIDIA 3090 GPU, outperforming the current mainstream industrial detectors. RTMDet achieves the best parameter-accuracy trade-off with tiny/small/medium/large/extra-large model sizes for various application scenarios, and obtains new state-of-the-art performance on real-time instance segmentation and rotated object detection. We hope the experimental results can provide new insights into designing versatile real-time object detectors for many object recognition tasks. Code and models are released at https://github.com/open-mmlab/mmdetection/tree/3.x/configs/rtmdet.

@article{
 title = {Object Detection using YOLO: A Survey},
 type = {article},
 year = {2022},
 keywords = {Computer Vision,Object Detection,Regression,State-of-the-art,YOLO},
 pages = {747-752},
 publisher = {IEEE},
 id = {b7d883ec-f010-35a4-b46e-8810989cedcc},
 created = {2024-03-16T12:43:16.030Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-16T12:43:19.325Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {In recent years, object detection is becoming very popular field in computer vision developments. Object detection has many applications viz. vehicle detection, pedestrian detection, blood cell counting etc. Various studies have been conducted in order to improve object detecting accuracy and speed. The latest technique is You Only Look Once object detection. It is state-of-the-art detection technique and considered as a regression problem. YOLO has the ability to predict various objects present in an image in a single run. This paper presents a survey of various detections based on YOLO which aims to improve the accuracy of existing system. This paper presents various modifications done on basic YOLO method and shows their analysis.},
 bibtype = {article},
 author = {Tripathi, Abhinandan and Gupta, Manish Kumar and Srivastava, Chaynika and Dixit, Pallavi and Pandey, Shrawan Kumar},
 doi = {10.1109/IC3I56241.2022.10073281},
 journal = {Proceedings of 5th International Conference on Contemporary Computing and Informatics, IC3I 2022},
 number = {December 2022}
}

In recent years, object detection is becoming very popular field in computer vision developments. Object detection has many applications viz. vehicle detection, pedestrian detection, blood cell counting etc. Various studies have been conducted in order to improve object detecting accuracy and speed. The latest technique is You Only Look Once object detection. It is state-of-the-art detection technique and considered as a regression problem. YOLO has the ability to predict various objects present in an image in a single run. This paper presents a survey of various detections based on YOLO which aims to improve the accuracy of existing system. This paper presents various modifications done on basic YOLO method and shows their analysis.

@article{
 title = {A Review of Yolo Algorithm Developments},
 type = {article},
 year = {2022},
 keywords = {Object detection,Public data analysis,Review,Yolo},
 pages = {1066-1073},
 volume = {199},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {a19843d7-8369-3328-b2f5-9cfd37e1458a},
 created = {2024-03-16T12:58:17.064Z},
 accessed = {2024-03-16},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-16T12:58:19.655Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Object detection techniques are the foundation for the artificial intelligence field. This research paper gives a brief overview of the You Only Look Once (YOLO) algorithm and its subsequent advanced versions. Through the analysis, we reach many remarks and insightful results. The results show the differences and similarities among the YOLO versions and between YOLO and Convolutional Neural Networks (CNNs). The central insight is the YOLO algorithm improvement is still ongoing.This article briefly describes the development process of the YOLO algorithm, summarizes the methods of target recognition and feature selection, and provides literature support for the targeted picture news and feature extraction in the financial and other fields. Besides, this paper contributes a lot to YOLO and other object detection literature.},
 bibtype = {article},
 author = {Jiang, Peiyuan and Ergu, Daji and Liu, Fangyao and Cai, Ying and Ma, Bo},
 doi = {10.1016/J.PROCS.2022.01.135},
 journal = {Procedia Computer Science}
}

Object detection techniques are the foundation for the artificial intelligence field. This research paper gives a brief overview of the You Only Look Once (YOLO) algorithm and its subsequent advanced versions. Through the analysis, we reach many remarks and insightful results. The results show the differences and similarities among the YOLO versions and between YOLO and Convolutional Neural Networks (CNNs). The central insight is the YOLO algorithm improvement is still ongoing.This article briefly describes the development process of the YOLO algorithm, summarizes the methods of target recognition and feature selection, and provides literature support for the targeted picture news and feature extraction in the financial and other fields. Besides, this paper contributes a lot to YOLO and other object detection literature.

@article{
 title = {Anchor-Free Oriented Proposal Generator for Object Detection},
 type = {article},
 year = {2022},
 keywords = {Anchor-free oriented proposal generator (AOPG),oriented object detection,oriented proposal generation},
 volume = {60},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {930a6ae0-2d04-35e0-bbb2-f2050260958f},
 created = {2024-03-27T10:37:22.134Z},
 file_attached = {false},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-27T10:37:24.653Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {Cheng2022},
 source_type = {article},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Oriented object detection is a practical and challenging task in remote sensing image interpretation. Nowadays, oriented detectors mostly use horizontal boxes as intermedium to derive oriented boxes from them. However, the horizontal boxes are inclined to get small Intersection-over-Unions (IoUs) with ground truths, which may have some undesirable effects, such as introducing redundant noise, mismatching with ground truths, and detracting from the robustness of detectors. In this article, we propose a novel anchor-free oriented proposal generator (AOPG) that abandons horizontal box-related operations from the network architecture. AOPG first produces coarse oriented boxes by a coarse location module (CLM) in an anchor-free manner and then refines them into high-quality oriented proposals. After AOPG, we apply a Fast Region-based Convolutional Neural Network (R-CNN) head to produce the final detection results. Furthermore, the shortage of large-scale datasets is also a hindrance to the development of oriented object detection. To alleviate the data insufficiency, we release a new dataset on the basis of our DIOR dataset and name it DIOR-R. Massive experiments demonstrate the effectiveness of AOPG. Particularly, without bells and whistles, we achieve the accuracy of 64.41%, 75.24%, and 96.22% mAP on the DIOR-R, DOTA, and HRSC2016 datasets, respectively. Code and models are available at https://github.com/jbwang1997/AOPG.},
 bibtype = {article},
 author = {Cheng, Gong and Wang, Jiabao and Li, Ke and Xie, Xingxing and Lang, Chunbo and Yao, Yanqing and Han, Junwei},
 doi = {10.1109/TGRS.2022.3183022},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

Oriented object detection is a practical and challenging task in remote sensing image interpretation. Nowadays, oriented detectors mostly use horizontal boxes as intermedium to derive oriented boxes from them. However, the horizontal boxes are inclined to get small Intersection-over-Unions (IoUs) with ground truths, which may have some undesirable effects, such as introducing redundant noise, mismatching with ground truths, and detracting from the robustness of detectors. In this article, we propose a novel anchor-free oriented proposal generator (AOPG) that abandons horizontal box-related operations from the network architecture. AOPG first produces coarse oriented boxes by a coarse location module (CLM) in an anchor-free manner and then refines them into high-quality oriented proposals. After AOPG, we apply a Fast Region-based Convolutional Neural Network (R-CNN) head to produce the final detection results. Furthermore, the shortage of large-scale datasets is also a hindrance to the development of oriented object detection. To alleviate the data insufficiency, we release a new dataset on the basis of our DIOR dataset and name it DIOR-R. Massive experiments demonstrate the effectiveness of AOPG. Particularly, without bells and whistles, we achieve the accuracy of 64.41%, 75.24%, and 96.22% mAP on the DIOR-R, DOTA, and HRSC2016 datasets, respectively. Code and models are available at https://github.com/jbwang1997/AOPG.

@article{
 title = {Align Deep Features for Oriented Object Detection},
 type = {article},
 year = {2022},
 keywords = {Aerial images,deep learning,feature alignment,object detection},
 volume = {60},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {efdd13af-e2bb-3a13-a599-189565a81596},
 created = {2024-03-27T10:37:22.252Z},
 file_attached = {false},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-03-27T10:37:24.436Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {Han2022},
 source_type = {article},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.},
 bibtype = {article},
 author = {Han, Jiaming and Ding, Jian and Li, Jie and Xia, Gui Song},
 doi = {10.1109/TGRS.2021.3062048},
 journal = {IEEE Transactions on Geoscience and Remote Sensing}
}

The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large-scale variations and arbitrary orientations. However, most of existing methods rely on heuristically defined anchors with different scales, angles, and aspect ratios, and usually suffer from severe misalignment between anchor boxes (ABs) and axis-aligned convolutional features, which lead to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a single-shot alignment network (S2A-Net) consisting of two modules: a feature alignment module (FAM) and an oriented detection module (ODM). The FAM can generate high-quality anchors with an anchor refinement network and adaptively align the convolutional features according to the ABs with a novel alignment convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve the state-of-the-art performance on two commonly used aerial objects' data sets (i.e., DOTA and HRSC2016) while keeping high efficiency.

@article{
 title = {H2RBox: Horizontal Box Annotation is All You Need for Oriented Object Detection},
 type = {article},
 year = {2022},
 websites = {https://arxiv.org/abs/2210.06742v5},
 month = {10},
 day = {13},
 id = {129b08ff-419d-3dd7-8014-3f6945153d78},
 created = {2024-07-02T13:11:08.648Z},
 accessed = {2024-07-02},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-07-02T13:11:20.211Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {df28411a-ed7f-4991-8358-d39685eb4bf0},
 private_publication = {false},
 abstract = {Oriented object detection emerges in many applications from aerial images to autonomous driving, while many existing detection benchmarks are annotated with horizontal bounding box only which is also less costive than fine-grained rotated box, leading to a gap between the readily available training corpus and the rising demand for oriented object detection. This paper proposes a simple yet effective oriented object detection approach called H2RBox merely using horizontal box annotation for weakly-supervised training, which closes the above gap and shows competitive performance even against those trained with rotated boxes. The cores of our method are weakly- and self-supervised learning, which predicts the angle of the object by learning the consistency of two different views. To our best knowledge, H2RBox is the first horizontal box annotation-based oriented object detector. Compared to an alternative i.e. horizontal box-supervised instance segmentation with our post adaption to oriented object detection, our approach is not susceptible to the prediction quality of mask and can perform more robustly in complex scenes containing a large number of dense objects and outliers. Experimental results show that H2RBox has significant performance and speed advantages over horizontal box-supervised instance segmentation methods, as well as lower memory requirements. While compared to rotated box-supervised oriented object detectors, our method shows very close performance and speed. The source code is available at PyTorch-based \hrefhttps://github.com/yangxue0827/h2rbox-mmrotateMMRotate and Jittor-based \hrefhttps://github.com/yangxue0827/h2rbox-jittorJDet.},
 bibtype = {article},
 author = {Yang, Xue and Zhang, Gefan and Li, Wentong and Wang, Xuehui and Zhou, Yue and Yan, Junchi and Key, Moe}
}

Oriented object detection emerges in many applications from aerial images to autonomous driving, while many existing detection benchmarks are annotated with horizontal bounding box only which is also less costive than fine-grained rotated box, leading to a gap between the readily available training corpus and the rising demand for oriented object detection. This paper proposes a simple yet effective oriented object detection approach called H2RBox merely using horizontal box annotation for weakly-supervised training, which closes the above gap and shows competitive performance even against those trained with rotated boxes. The cores of our method are weakly- and self-supervised learning, which predicts the angle of the object by learning the consistency of two different views. To our best knowledge, H2RBox is the first horizontal box annotation-based oriented object detector. Compared to an alternative i.e. horizontal box-supervised instance segmentation with our post adaption to oriented object detection, our approach is not susceptible to the prediction quality of mask and can perform more robustly in complex scenes containing a large number of dense objects and outliers. Experimental results show that H2RBox has significant performance and speed advantages over horizontal box-supervised instance segmentation methods, as well as lower memory requirements. While compared to rotated box-supervised oriented object detectors, our method shows very close performance and speed. The source code is available at PyTorch-based \hrefhttps://github.com/yangxue0827/h2rbox-mmrotateMMRotate and Jittor-based \hrefhttps://github.com/yangxue0827/h2rbox-jittorJDet.

@article{
 title = {Disease detection, severity prediction, and crop loss estimation in MaizeCrop using deep learning},
 type = {article},
 year = {2022},
 keywords = {Crop loss,Deep learning,Disease detection,Maize,Severity},
 pages = {276-291},
 volume = {6},
 month = {1},
 publisher = {Elsevier},
 day = {1},
 id = {3cc36f2b-d0ff-31da-9bcd-bba8017896b7},
 created = {2024-09-20T11:57:43.897Z},
 accessed = {2024-09-20},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-09-23T09:20:49.724Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {b2e80568-e430-43c3-9f7c-33cc6051f581},
 private_publication = {false},
 abstract = {The increasing gap between the demand and productivity of maize crop is a point of concern for the food industry, and farmers. Its' susceptibility to diseases such as Turcicum Leaf Blight, and Rust is a major cause for reducing its production. Manual detection, and classification of these diseases, calculation of disease severity, and crop loss estimation is a time-consuming task. Also, it requires expertise in disease detection. Thus, there is a need to find an alternative for automatic disease detection, severity prediction, and crop loss estimation. The promising results of machine learning, and deep learning algorithms in pattern recognition, object detection, and data analysis motivate researchers to employ these techniques for disease detection, classification, and crop loss estimation in maize crop. The research works available in literature, have proven their potential in automatic disease detection using machine learning, and deep learning models. But, there is a lack none of these works a reliable and real-life labelled dataset for training these models. Also, none of the existing works focus on severity prediction, and crop loss estimation. The authors in this manuscript collect the real-life dataset labelled by plant pathologists. They propose a deep learning-based framework for pre-processing of dataset, automatic disease detection, severity prediction, and crop loss estimation. It uses the K-Means clustering algorithm for extracting the region of interest. Next, they employ the customized deep learning model ‘MaizeNet’ for disease detection, severity prediction, and crop loss estimation. The model reports the highest accuracy of 98.50%. Also, the authors perform the feature visualization using the Grad-CAM. Now, the proposed model is integrated with a web application to provide a user-friendly interface. The efficacy of the model in extracting the relevant features, a smaller number of parameters, low training time, high accuracy favors its importance as an assisting tool for plant pathology experts.The copyright for the associated web application ‘Maize-Disease-Detector’ is filed with diary number: 17006/2021-CO/SW.},
 bibtype = {article},
 author = {Kundu, Nidhi and Rani, Geeta and Dhaka, Vijaypal Singh and Gupta, Kalpit and Nayaka, Siddaiah Chandra and Vocaturo, Eugenio and Zumpano, Ester},
 doi = {10.1016/J.AIIA.2022.11.002},
 journal = {Artificial Intelligence in Agriculture}
}

The increasing gap between the demand and productivity of maize crop is a point of concern for the food industry, and farmers. Its' susceptibility to diseases such as Turcicum Leaf Blight, and Rust is a major cause for reducing its production. Manual detection, and classification of these diseases, calculation of disease severity, and crop loss estimation is a time-consuming task. Also, it requires expertise in disease detection. Thus, there is a need to find an alternative for automatic disease detection, severity prediction, and crop loss estimation. The promising results of machine learning, and deep learning algorithms in pattern recognition, object detection, and data analysis motivate researchers to employ these techniques for disease detection, classification, and crop loss estimation in maize crop. The research works available in literature, have proven their potential in automatic disease detection using machine learning, and deep learning models. But, there is a lack none of these works a reliable and real-life labelled dataset for training these models. Also, none of the existing works focus on severity prediction, and crop loss estimation. The authors in this manuscript collect the real-life dataset labelled by plant pathologists. They propose a deep learning-based framework for pre-processing of dataset, automatic disease detection, severity prediction, and crop loss estimation. It uses the K-Means clustering algorithm for extracting the region of interest. Next, they employ the customized deep learning model ‘MaizeNet’ for disease detection, severity prediction, and crop loss estimation. The model reports the highest accuracy of 98.50%. Also, the authors perform the feature visualization using the Grad-CAM. Now, the proposed model is integrated with a web application to provide a user-friendly interface. The efficacy of the model in extracting the relevant features, a smaller number of parameters, low training time, high accuracy favors its importance as an assisting tool for plant pathology experts.The copyright for the associated web application ‘Maize-Disease-Detector’ is filed with diary number: 17006/2021-CO/SW.

@article{
 title = {Driver and Pedestrian Mutual Awareness for Path Prediction and Collision Risk Estimation},
 type = {article},
 year = {2022},
 volume = {7},
 websites = {https://doi.org/10.1109/TIV.},
 id = {b130bddb-706f-3de8-84bf-3c2d16089222},
 created = {2024-11-06T13:04:00.272Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-07T10:52:46.769Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4cda9297-f98e-4246-88d6-ffeeade205c3},
 private_publication = {false},
 abstract = {We present a novel method for vehicle-pedestrian path prediction that takes into account the awareness of the driver and the pedestrian towards each other. The method jointly models the paths of vehicle and pedestrian within a single Dynamic Bayesian Network (DBN). In this DBN, sub-graphs model the environment and entity-specific context cues of the vehicle and pedestrian (incl. awareness), which affect their future motion and allow to increase the prediction horizon. These sub-graphs share a latent state which models whether vehicle and pedestrian are on collision course; this accounts for a certain degree of motion coupling. The method was validated with real-world data obtained by on-board vehicle sensing (stereo vision, GNSS and proprioceptive). Data consist of 93 vehicle and pedestrian encounters, spanning various awareness conditions and dynamic characteristics of the participants. In ablation studies, we quantify the benefits of various components of our proposed DBN model for path prediction and collision risk estimation. Results show that at a prediction horizon of 1.5 s, context-aware models outperform context-agnostic models in path prediction for scenarios with a dynamics change, while performing similarly otherwise. Results further indicate that driver attention-aware models improve collision risk estimation compared to driver-agnostic models. Index Terms-Collision risk estimation, driver awareness, path prediction, pedestrian awareness.},
 bibtype = {article},
 author = {Roth, Markus and Stapel, Jork and Happee, Riender and Gavrila, Dariu M},
 doi = {10.1109/TIV},
 journal = {IEEE TRANSACTIONS ON INTELLIGENT VEHICLES},
 number = {4}
}

We present a novel method for vehicle-pedestrian path prediction that takes into account the awareness of the driver and the pedestrian towards each other. The method jointly models the paths of vehicle and pedestrian within a single Dynamic Bayesian Network (DBN). In this DBN, sub-graphs model the environment and entity-specific context cues of the vehicle and pedestrian (incl. awareness), which affect their future motion and allow to increase the prediction horizon. These sub-graphs share a latent state which models whether vehicle and pedestrian are on collision course; this accounts for a certain degree of motion coupling. The method was validated with real-world data obtained by on-board vehicle sensing (stereo vision, GNSS and proprioceptive). Data consist of 93 vehicle and pedestrian encounters, spanning various awareness conditions and dynamic characteristics of the participants. In ablation studies, we quantify the benefits of various components of our proposed DBN model for path prediction and collision risk estimation. Results show that at a prediction horizon of 1.5 s, context-aware models outperform context-agnostic models in path prediction for scenarios with a dynamics change, while performing similarly otherwise. Results further indicate that driver attention-aware models improve collision risk estimation compared to driver-agnostic models. Index Terms-Collision risk estimation, driver awareness, path prediction, pedestrian awareness.

@inproceedings{
 title = {On-Board Pedestrian Trajectory Prediction Using Behavioral Features},
 type = {inproceedings},
 year = {2022},
 pages = {437-443},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {9b3724d1-625a-3625-8129-1794d8f44a5e},
 created = {2024-11-06T13:04:00.806Z},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-07T10:52:44.352Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4cda9297-f98e-4246-88d6-ffeeade205c3},
 private_publication = {false},
 abstract = {This paper presents a novel approach to pedestrian trajectory prediction for on-board camera systems, which utilizes behavioral features of pedestrians that can be inferred from visual observations. Our proposed method, called Behavior-Aware Pedestrian Trajectory Prediction (BA-PTP), processes multiple input modalities, i.e. bounding boxes, body and head orientation of pedestrians as well as their pose, with independent encoding streams. The encodings of each stream are fused using a modality attention mechanism, resulting in a final embedding that is used to predict future bounding boxes in the image.In experiments on two datasets for pedestrian behavior prediction, we demonstrate the benefit of using behavioral features for pedestrian trajectory prediction and evaluate the effectiveness of the proposed encoding strategy. Additionally, we investigate the relevance of different behavioral features on the prediction performance based on an ablation study.},
 bibtype = {inproceedings},
 author = {Czech, Phillip and Braun, Markus and Kreßel, Ulrich and Yang, Bin},
 doi = {10.1109/ICMLA55696.2022.00070},
 booktitle = {Proceedings - 21st IEEE International Conference on Machine Learning and Applications, ICMLA 2022}
}

This paper presents a novel approach to pedestrian trajectory prediction for on-board camera systems, which utilizes behavioral features of pedestrians that can be inferred from visual observations. Our proposed method, called Behavior-Aware Pedestrian Trajectory Prediction (BA-PTP), processes multiple input modalities, i.e. bounding boxes, body and head orientation of pedestrians as well as their pose, with independent encoding streams. The encodings of each stream are fused using a modality attention mechanism, resulting in a final embedding that is used to predict future bounding boxes in the image.In experiments on two datasets for pedestrian behavior prediction, we demonstrate the benefit of using behavioral features for pedestrian trajectory prediction and evaluate the effectiveness of the proposed encoding strategy. Additionally, we investigate the relevance of different behavioral features on the prediction performance based on an ablation study.

@article{
 title = {Drones in agriculture: A review and bibliometric analysis},
 type = {article},
 year = {2022},
 keywords = {Bibliometrics,Drones,Internet of Things,Precision agriculture,UAV},
 pages = {107017},
 volume = {198},
 month = {7},
 publisher = {Elsevier},
 day = {1},
 id = {836648aa-2cf4-30cf-9cae-ce9b86c25c01},
 created = {2024-12-06T10:30:28.906Z},
 accessed = {2024-12-06},
 file_attached = {true},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-06T10:30:33.016Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e},
 private_publication = {false},
 abstract = {Drones, also called Unmanned Aerial Vehicles (UAV), have witnessed a remarkable development in recent decades. In agriculture, they have changed farming practices by offering farmers substantial cost savings, increased operational efficiency, and better profitability. Over the past decades, the topic of agricultural drones has attracted remarkable academic attention. We therefore conduct a comprehensive review based on bibliometrics to summarize and structure existing academic literature and reveal current research trends and hotspots. We apply bibliometric techniques and analyze the literature surrounding agricultural drones to summarize and assess previous research. Our analysis indicates that remote sensing, precision agriculture, deep learning, machine learning, and the Internet of Things are critical topics related to agricultural drones. The co-citation analysis reveals six broad research clusters in the literature. This study is one of the first attempts to summarize drone research in agriculture and suggest future research directions.},
 bibtype = {article},
 author = {Rejeb, Abderahman and Abdollahi, Alireza and Rejeb, Karim and Treiblmaier, Horst},
 doi = {10.1016/J.COMPAG.2022.107017},
 journal = {Computers and Electronics in Agriculture}
}

Drones, also called Unmanned Aerial Vehicles (UAV), have witnessed a remarkable development in recent decades. In agriculture, they have changed farming practices by offering farmers substantial cost savings, increased operational efficiency, and better profitability. Over the past decades, the topic of agricultural drones has attracted remarkable academic attention. We therefore conduct a comprehensive review based on bibliometrics to summarize and structure existing academic literature and reveal current research trends and hotspots. We apply bibliometric techniques and analyze the literature surrounding agricultural drones to summarize and assess previous research. Our analysis indicates that remote sensing, precision agriculture, deep learning, machine learning, and the Internet of Things are critical topics related to agricultural drones. The co-citation analysis reveals six broad research clusters in the literature. This study is one of the first attempts to summarize drone research in agriculture and suggest future research directions.

@article{
 title = {An annotated image dataset of downy mildew symptoms on Merlot grape variety},
 type = {article},
 year = {2021},
 pages = {107250},
 volume = {37},
 month = {8},
 publisher = {Elsevier BV},
 id = {c1791b5c-2d97-35b6-b647-3ea7a8642243},
 created = {2022-07-15T12:46:39.764Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:00.411Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {This article introduces a dataset of high-resolution colour images of grapevines. It contains 99 images acquired in the vineyard from a cruising tractor. Each image includes the full foliage of a grapevine plant. These images display a diverse range of symptoms caused by the grapevine downy mildew (Plasmopara viticola), a major fungal disease. The dataset also includes various confounding factors, i.e. anomalies that are not related to the disease. These anomalies are the natural and common phenomena affecting vineyards such as results of mechanical wounds, necroses, chemical burns or yellowing and discolorations due to nutritional or hydric deficiencies. Images were acquired in-situ on "Le Domaine de la Grande Ferrade" a public experimental facility of INRAE, in the area of Bordeaux. Acquisitions took place at early fruiting stages (BBCH 75-79) corresponding to the main sanitary pressure during growth. The acquisition device, embedded on a vine tractor, is composed of an industrial colour camera synchronised with powerful flashes. The purpose of this device is to produce a "day for night" effect that mitigates the variation of sunlight. It enables to homogenise images acquired during different weathers and time of the day and to ensure that the foreground (containing foliage) displays appropriate brightness, with minimum shadows while the background is darker. The images of the dataset were annotated manually by photo-interpretation with a careful review of expertise regarding phytopathology and physiological disorders. The annotation process consists in associating pixels with a class that defines its membership to a type of organ and its physiological state. Pixels from healthy, symptomatic or abnormal grapevine tissues were labelled into seven classes: "Limbus", "Leaf edges", "Berries", "Stems", "Foliar mildew", "Berries mildew" and "Anomalies". The annotation is achieved with the GIMP2 software as mask images where the value attributed to each pixel corresponds to one of the seven considered classes.},
 bibtype = {article},
 author = {Abdelghafour, Florent and Keresztes, Barna and Deshayes, Aymeric and Germain, Christian and Da Costa, Jean-Pierre},
 doi = {10.1016/j.dib.2021.107250},
 journal = {Data in Brief},
 keywords = {abdelghafour2021annotatedimagedataset}
}

This article introduces a dataset of high-resolution colour images of grapevines. It contains 99 images acquired in the vineyard from a cruising tractor. Each image includes the full foliage of a grapevine plant. These images display a diverse range of symptoms caused by the grapevine downy mildew (Plasmopara viticola), a major fungal disease. The dataset also includes various confounding factors, i.e. anomalies that are not related to the disease. These anomalies are the natural and common phenomena affecting vineyards such as results of mechanical wounds, necroses, chemical burns or yellowing and discolorations due to nutritional or hydric deficiencies. Images were acquired in-situ on "Le Domaine de la Grande Ferrade" a public experimental facility of INRAE, in the area of Bordeaux. Acquisitions took place at early fruiting stages (BBCH 75-79) corresponding to the main sanitary pressure during growth. The acquisition device, embedded on a vine tractor, is composed of an industrial colour camera synchronised with powerful flashes. The purpose of this device is to produce a "day for night" effect that mitigates the variation of sunlight. It enables to homogenise images acquired during different weathers and time of the day and to ensure that the foreground (containing foliage) displays appropriate brightness, with minimum shadows while the background is darker. The images of the dataset were annotated manually by photo-interpretation with a careful review of expertise regarding phytopathology and physiological disorders. The annotation process consists in associating pixels with a class that defines its membership to a type of organ and its physiological state. Pixels from healthy, symptomatic or abnormal grapevine tissues were labelled into seven classes: "Limbus", "Leaf edges", "Berries", "Stems", "Foliar mildew", "Berries mildew" and "Anomalies". The annotation is achieved with the GIMP2 software as mask images where the value attributed to each pixel corresponds to one of the seven considered classes.

@article{
 title = {Knowledge Distillation: A Survey},
 type = {article},
 year = {2021},
 pages = {1789--1819},
 volume = {129},
 websites = {http://arxiv.org/abs/2006.05525,http://dx.doi.org/10.1007/s11263-021-01453-z},
 month = {6},
 day = {9},
 id = {236c67e5-319f-3581-a336-dcbf47c37322},
 created = {2022-08-17T13:07:57.391Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-29T07:46:17.694Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {71c17bf5-4635-46a8-b746-49e98883b113},
 private_publication = {false},
 abstract = {In recent years, deep neural networks have been successful in both industry and academia, especially for computer vision tasks. The great success of deep learning is mainly due to its scalability to encode large-scale data and to maneuver billions of model parameters. However, it is a challenge to deploy these cumbersome deep models on devices with limited resources, e.g., mobile phones and embedded devices, not only because of the high computational complexity but also the large storage requirements. To this end, a variety of model compression and acceleration techniques have been developed. As a representative type of model compression and acceleration, knowledge distillation effectively learns a small student model from a large teacher model. It has received rapid increasing attention from the community. This paper provides a comprehensive survey of knowledge distillation from the perspectives of knowledge categories, training schemes, teacher-student architecture, distillation algorithms, performance comparison and applications. Furthermore, challenges in knowledge distillation are briefly reviewed and comments on future research are discussed and forwarded.},
 bibtype = {article},
 author = {Gou, Jianping and Yu, Baosheng and Maybank, Stephen John and Tao, Dacheng},
 doi = {10.1007/s11263-021-01453-z},
 journal = {Int. J. Comput. Vis.},
 number = {6}
}

In recent years, deep neural networks have been successful in both industry and academia, especially for computer vision tasks. The great success of deep learning is mainly due to its scalability to encode large-scale data and to maneuver billions of model parameters. However, it is a challenge to deploy these cumbersome deep models on devices with limited resources, e.g., mobile phones and embedded devices, not only because of the high computational complexity but also the large storage requirements. To this end, a variety of model compression and acceleration techniques have been developed. As a representative type of model compression and acceleration, knowledge distillation effectively learns a small student model from a large teacher model. It has received rapid increasing attention from the community. This paper provides a comprehensive survey of knowledge distillation from the perspectives of knowledge categories, training schemes, teacher-student architecture, distillation algorithms, performance comparison and applications. Furthermore, challenges in knowledge distillation are briefly reviewed and comments on future research are discussed and forwarded.

@article{
 title = {Visual and Visual-Inertial SLAM: State of the Art, Classification, and Experimental Benchmarking},
 type = {article},
 year = {2021},
 volume = {2021},
 publisher = {Hindawi Limited},
 id = {b9c4cfe4-423b-32a0-bdf6-5554679fe666},
 created = {2022-08-19T08:58:04.657Z},
 accessed = {2022-08-19},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-06T13:33:38.032Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {Simultaneous Localization and Mapping is now widely adopted by many applications, and researchers have produced very dense literature on this topic. With the advent of smart devices, embedding cameras, inertial measurement units, visual SLAM (vSLAM), and visual-inertial SLAM (viSLAM) are enabling novel general public applications. In this context, this paper conducts a review of popular SLAM approaches with a focus on vSLAM/viSLAM, both at fundamental and experimental levels. It starts with a structured overview of existing vSLAM and viSLAM designs and continues with a new classification of a dozen main state-of-the-art methods. A chronological survey of viSLAM's development highlights the historical milestones and presents more recent methods into a classification. Finally, the performance of vSLAM is experimentally assessed for the use case of pedestrian pose estimation with a handheld device in urban environments. The performance of five open-source methods Vins-Mono, ROVIO, ORB-SLAM2, DSO, and LSD-SLAM is compared using the EuRoC MAV dataset and a new visual-inertial dataset corresponding to urban pedestrian navigation. A detailed analysis of the computation results identifies the strengths and weaknesses for each method. Globally, ORB-SLAM2 appears to be the most promising algorithm to address the challenges of urban pedestrian navigation, tested with two datasets.},
 bibtype = {article},
 author = {Servières, Myriam and Renaudin, Valérie and Dupuis, Alexis and Antigny, Nicolas},
 doi = {10.1155/2021/2054828},
 journal = {Journal of Sensors}
}

Simultaneous Localization and Mapping is now widely adopted by many applications, and researchers have produced very dense literature on this topic. With the advent of smart devices, embedding cameras, inertial measurement units, visual SLAM (vSLAM), and visual-inertial SLAM (viSLAM) are enabling novel general public applications. In this context, this paper conducts a review of popular SLAM approaches with a focus on vSLAM/viSLAM, both at fundamental and experimental levels. It starts with a structured overview of existing vSLAM and viSLAM designs and continues with a new classification of a dozen main state-of-the-art methods. A chronological survey of viSLAM's development highlights the historical milestones and presents more recent methods into a classification. Finally, the performance of vSLAM is experimentally assessed for the use case of pedestrian pose estimation with a handheld device in urban environments. The performance of five open-source methods Vins-Mono, ROVIO, ORB-SLAM2, DSO, and LSD-SLAM is compared using the EuRoC MAV dataset and a new visual-inertial dataset corresponding to urban pedestrian navigation. A detailed analysis of the computation results identifies the strengths and weaknesses for each method. Globally, ORB-SLAM2 appears to be the most promising algorithm to address the challenges of urban pedestrian navigation, tested with two datasets.

@article{
 title = {Survey and Evaluation of RGB-D SLAM},
 type = {article},
 year = {2021},
 keywords = {Computer vision,RGB-D SLAM,evaluation,robotics,survey},
 pages = {21367-21387},
 volume = {9},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {9eb9ee0e-d51c-38b1-8478-f0fceb2de606},
 created = {2022-08-19T08:59:53.362Z},
 accessed = {2022-08-19},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-08-19T09:00:13.289Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {The traditional visual SLAM systems take the monocular or stereo camera as input sensor, with complex map initialization and map point triangulation steps needed for 3D map reconstruction, which are easy to fail, computationally complex and can cause noisy measurements. The emergence of RGB-D camera which provides RGB image together with depth information breaks this situation. While a number of RGB-D SLAM systems have been proposed in recent years, the current classification research on RGB-D SLAM is very lacking, and their advantages and shortcomings remain unclear regarding different applications and perturbations, such as illumination transformation, noise and rolling shutter effect of sensors. In this paper, we mainly introduced the basic concept and structure of the RGB-D SLAM system, and then introduced the differences between the various RGB-D SLAM systems in the three aspects of tracking, mapping, and loop detection, and we make a classification study on different RGB-D SLAM algorithms according to the three aspect. Furthermore, we discuss some advanced topics and open problems of RGB-D SLAM, hoping that it will help for future exploring. In the end, we conducted a large number of evaluation experiments on multiple RGB-D SLAM systems, and analyzed their advantages and disadvantages, as well as performance differences in different application scenarios, and provided references for researchers and developers.},
 bibtype = {article},
 author = {Zhang, Shishun and Zheng, Longyu and Tao, Wenbing},
 doi = {10.1109/ACCESS.2021.3053188},
 journal = {IEEE Access}
}

The traditional visual SLAM systems take the monocular or stereo camera as input sensor, with complex map initialization and map point triangulation steps needed for 3D map reconstruction, which are easy to fail, computationally complex and can cause noisy measurements. The emergence of RGB-D camera which provides RGB image together with depth information breaks this situation. While a number of RGB-D SLAM systems have been proposed in recent years, the current classification research on RGB-D SLAM is very lacking, and their advantages and shortcomings remain unclear regarding different applications and perturbations, such as illumination transformation, noise and rolling shutter effect of sensors. In this paper, we mainly introduced the basic concept and structure of the RGB-D SLAM system, and then introduced the differences between the various RGB-D SLAM systems in the three aspects of tracking, mapping, and loop detection, and we make a classification study on different RGB-D SLAM algorithms according to the three aspect. Furthermore, we discuss some advanced topics and open problems of RGB-D SLAM, hoping that it will help for future exploring. In the end, we conducted a large number of evaluation experiments on multiple RGB-D SLAM systems, and analyzed their advantages and disadvantages, as well as performance differences in different application scenarios, and provided references for researchers and developers.

@article{
 title = {ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial, and Multimap SLAM},
 type = {article},
 year = {2021},
 keywords = {Computer vision,inertial navigation,simult- aneous localization and mapping},
 pages = {1874-1890},
 volume = {37},
 month = {12},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 day = {1},
 id = {0f646c01-c43f-3870-ba77-f856f1b0a26a},
 created = {2022-08-19T09:00:26.344Z},
 accessed = {2022-08-19},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-07T08:31:25.573Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {This article presents ORB-SLAM3, the first system able to perform visual, visual-inertial and multimap SLAM with monocular, stereo and RGB-D cameras, using pin-hole and fisheye lens models. The first main novelty is a tightly integrated visual-inertial SLAM system that fully relies on maximum a posteriori (MAP) estimation, even during IMU initialization, resulting in real-time robust operation in small and large, indoor and outdoor environments, being two to ten times more accurate than previous approaches. The second main novelty is a multiple map system relying on a new place recognition method with improved recall that lets ORB-SLAM3 survive to long periods of poor visual information: when it gets lost, it starts a new map that will be seamlessly merged with previous maps when revisiting them. Compared with visual odometry systems that only use information from the last few seconds, ORB-SLAM3 is the first system able to reuse in all the algorithm stages all previous information from high parallax co-visible keyframes, even if they are widely separated in time or come from previous mapping sessions, boosting accuracy. Our experiments show that, in all sensor configurations, ORB-SLAM3 is as robust as the best systems available in the literature and significantly more accurate. Notably, our stereo-inertial SLAM achieves an average accuracy of 3.5 cm in the EuRoC drone and 9 mm under quick hand-held motions in the room of TUM-VI dataset, representative of AR/VR scenarios. For the benefit of the community we make public the source code.},
 bibtype = {article},
 author = {Campos, Carlos and Elvira, Richard and Rodriguez, Juan J.Gomez and Montiel, Jose M.M. and Tardos, Juan D.},
 doi = {10.1109/TRO.2021.3075644},
 journal = {IEEE Transactions on Robotics},
 number = {6}
}

This article presents ORB-SLAM3, the first system able to perform visual, visual-inertial and multimap SLAM with monocular, stereo and RGB-D cameras, using pin-hole and fisheye lens models. The first main novelty is a tightly integrated visual-inertial SLAM system that fully relies on maximum a posteriori (MAP) estimation, even during IMU initialization, resulting in real-time robust operation in small and large, indoor and outdoor environments, being two to ten times more accurate than previous approaches. The second main novelty is a multiple map system relying on a new place recognition method with improved recall that lets ORB-SLAM3 survive to long periods of poor visual information: when it gets lost, it starts a new map that will be seamlessly merged with previous maps when revisiting them. Compared with visual odometry systems that only use information from the last few seconds, ORB-SLAM3 is the first system able to reuse in all the algorithm stages all previous information from high parallax co-visible keyframes, even if they are widely separated in time or come from previous mapping sessions, boosting accuracy. Our experiments show that, in all sensor configurations, ORB-SLAM3 is as robust as the best systems available in the literature and significantly more accurate. Notably, our stereo-inertial SLAM achieves an average accuracy of 3.5 cm in the EuRoC drone and 9 mm under quick hand-held motions in the room of TUM-VI dataset, representative of AR/VR scenarios. For the benefit of the community we make public the source code.

@article{
 title = {Research on the Application of Visual SLAM in Embedded GPU},
 type = {article},
 year = {2021},
 volume = {2021},
 publisher = {Hindawi Limited},
 id = {0b16026a-adfe-3842-afc7-adf285ffab15},
 created = {2022-08-19T09:31:09.117Z},
 accessed = {2022-08-19},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-07T11:17:52.576Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {In the automatic navigation robot field, robotic autonomous positioning is one of the most difficult challenges. Simultaneous localization and mapping (SLAM) technology can incrementally construct a map of the robot's moving path in an unknown environment while estimating the position of the robot in the map, providing an effective solution for robots to fully navigate autonomously. The camera can obtain corresponding two-dimensional digital images from the real three-dimensional world. These images contain very rich colour, texture information, and highly recognizable features, which provide indispensable information for robots to understand and recognize the environment based on the ability to autonomously explore the unknown environment. Therefore, more and more researchers use cameras to solve SLAM problems, also known as visual SLAM. Visual SLAM needs to process a large number of image data collected by the camera, which has high performance requirements for computing hardware, and thus, its application on embedded mobile platforms is greatly limited. This paper presents a parallelization method based on embedded hardware equipped with embedded GPU. Use CUDA, a parallel computing platform, to accelerate the visual front-end processing of the visual SLAM algorithm. Extensive experiments are done to verify the effectiveness of the method. The results show that the presented method effectively improves the operating efficiency of the visual SLAM algorithm and ensures the original accuracy of the algorithm.},
 bibtype = {article},
 author = {Ma, Tianji and Bai, Nanyang and Shi, Wentao and Wu, Xi and Wang, Lutao and Wu, Tao and Zhao, Changming},
 doi = {10.1155/2021/6691262},
 journal = {Wireless Communications and Mobile Computing}
}

In the automatic navigation robot field, robotic autonomous positioning is one of the most difficult challenges. Simultaneous localization and mapping (SLAM) technology can incrementally construct a map of the robot's moving path in an unknown environment while estimating the position of the robot in the map, providing an effective solution for robots to fully navigate autonomously. The camera can obtain corresponding two-dimensional digital images from the real three-dimensional world. These images contain very rich colour, texture information, and highly recognizable features, which provide indispensable information for robots to understand and recognize the environment based on the ability to autonomously explore the unknown environment. Therefore, more and more researchers use cameras to solve SLAM problems, also known as visual SLAM. Visual SLAM needs to process a large number of image data collected by the camera, which has high performance requirements for computing hardware, and thus, its application on embedded mobile platforms is greatly limited. This paper presents a parallelization method based on embedded hardware equipped with embedded GPU. Use CUDA, a parallel computing platform, to accelerate the visual front-end processing of the visual SLAM algorithm. Extensive experiments are done to verify the effectiveness of the method. The results show that the presented method effectively improves the operating efficiency of the visual SLAM algorithm and ensures the original accuracy of the algorithm.

@inproceedings{
 title = {Kimera-Multi: a System for Distributed Multi-Robot Metric-Semantic Simultaneous Localization and Mapping},
 type = {inproceedings},
 year = {2021},
 id = {e94e2419-9f25-3beb-8b00-d4e6bce2974f},
 created = {2022-09-06T14:03:31.581Z},
 accessed = {2022-09-06},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-12T10:25:30.493Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {We present the first fully distributed multi-robot system for dense metric-semantic Simultaneous Localization and Mapping (SLAM). Our system, dubbed Kimera-Multi, is implemented by a team of robots equipped with visual-inertial sensors, and builds a 3D mesh model of the environment in real-time, where each face of the mesh is annotated with a semantic label (e.g., building, road, objects). In Kimera-Multi, each robot builds a local trajectory estimate and a local mesh using Kimera [1]. Then, when two robots are within communication range, they initiate a distributed place recognition and robust pose graph optimization protocol with a novel incremental maximum clique outlier rejection; the protocol allows the robots to improve their local trajectory estimates by leveraging inter-robot loop closures. Finally, each robot uses its improved trajectory estimate to correct the local mesh using mesh deformation techniques. We demonstrate Kimera-Multi in photo-realistic simulations and real data. Kimera-Multi (i) is able to build accurate 3D metric-semantic meshes, (ii) is robust to incorrect loop closures while requiring less computation than state-of-the-art distributed SLAM back-ends, and (iii) is efficient, both in terms of computation at each robot as well as communication bandwidth.},
 bibtype = {inproceedings},
 author = {Chang, Yun and Tian, Yulun and How, Jonathan P and Carlone, Luca},
 booktitle = {IEEE International Conference on Robotics and Automation, ICRA 2021, Xi'an, China, May 30 - June 5, 2021}
}

We present the first fully distributed multi-robot system for dense metric-semantic Simultaneous Localization and Mapping (SLAM). Our system, dubbed Kimera-Multi, is implemented by a team of robots equipped with visual-inertial sensors, and builds a 3D mesh model of the environment in real-time, where each face of the mesh is annotated with a semantic label (e.g., building, road, objects). In Kimera-Multi, each robot builds a local trajectory estimate and a local mesh using Kimera [1]. Then, when two robots are within communication range, they initiate a distributed place recognition and robust pose graph optimization protocol with a novel incremental maximum clique outlier rejection; the protocol allows the robots to improve their local trajectory estimates by leveraging inter-robot loop closures. Finally, each robot uses its improved trajectory estimate to correct the local mesh using mesh deformation techniques. We demonstrate Kimera-Multi in photo-realistic simulations and real data. Kimera-Multi (i) is able to build accurate 3D metric-semantic meshes, (ii) is robust to incorrect loop closures while requiring less computation than state-of-the-art distributed SLAM back-ends, and (iii) is efficient, both in terms of computation at each robot as well as communication bandwidth.

@article{
 title = {NICE-SLAM: Neural Implicit Scalable Encoding for SLAM},
 type = {article},
 year = {2021},
 websites = {http://arxiv.org/abs/2112.12130},
 id = {6f0076ca-2e56-3add-83b5-676cf7996fb9},
 created = {2022-09-08T06:17:30.722Z},
 accessed = {2022-09-08},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-12T10:25:30.461Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {Neural implicit representations have recently shown encouraging results in various domains, including promising progress in simultaneous localization and mapping (SLAM). Nevertheless, existing methods produce over-smoothed scene reconstructions and have difficulty scaling up to large scenes. These limitations are mainly due to their simple fully-connected network architecture that does not incorporate local information in the observations. In this paper, we present NICE-SLAM, a dense SLAM system that incorporates multi-level local information by introducing a hierarchical scene representation. Optimizing this representation with pre-trained geometric priors enables detailed reconstruction on large indoor scenes. Compared to recent neural implicit SLAM systems, our approach is more scalable, efficient, and robust. Experiments on five challenging datasets demonstrate competitive results of NICE-SLAM in both mapping and tracking quality. Project page: https://pengsongyou.github.io/nice-slam},
 bibtype = {article},
 author = {Zhu, Zihan and Peng, Songyou and Larsson, Viktor and Xu, Weiwei and Bao, Hujun and Cui, Zhaopeng and Oswald, Martin R. and Pollefeys, Marc},
 journal = {CoRR}
}

Neural implicit representations have recently shown encouraging results in various domains, including promising progress in simultaneous localization and mapping (SLAM). Nevertheless, existing methods produce over-smoothed scene reconstructions and have difficulty scaling up to large scenes. These limitations are mainly due to their simple fully-connected network architecture that does not incorporate local information in the observations. In this paper, we present NICE-SLAM, a dense SLAM system that incorporates multi-level local information by introducing a hierarchical scene representation. Optimizing this representation with pre-trained geometric priors enables detailed reconstruction on large indoor scenes. Compared to recent neural implicit SLAM systems, our approach is more scalable, efficient, and robust. Experiments on five challenging datasets demonstrate competitive results of NICE-SLAM in both mapping and tracking quality. Project page: https://pengsongyou.github.io/nice-slam

@inproceedings{
 title = {iMAP: Implicit Mapping and Positioning in Real-Time},
 type = {inproceedings},
 year = {2021},
 pages = {6209--6218},
 id = {93677656-fb3d-3224-b05c-da29923e8bff},
 created = {2022-09-08T06:40:04.220Z},
 accessed = {2022-09-08},
 file_attached = {true},
 profile_id = {48fc0258-023d-3602-860e-824092d62c56},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-11-13T10:32:39.242Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {We show for the first time that a multilayer perceptron (MLP) can serve as the only scene representation in a real-time SLAM system for a handheld RGB-D camera. Our network is trained in live operation without prior data, building a dense, scene-specific implicit 3D model of occupancy and colour which is also immediately used for tracking. Achieving real-time SLAM via continual training of a neural network against a live image stream requires significant innovation. Our iMAP algorithm uses a keyframe structure and multi-processing computation flow, with dynamic information-guided pixel sampling for speed, with tracking at 10 Hz and global map updating at 2 Hz. The advantages of an implicit MLP over standard dense SLAM techniques include efficient geometry representation with automatic detail control and smooth, plausible filling-in of unobserved regions such as the back surfaces of objects.},
 bibtype = {inproceedings},
 author = {Sucar, Edgar and Liu, Shikun and Ortiz, Joseph and Davison, Andrew J},
 booktitle = {2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021, Montreal, QC, Canada, October 10-17, 2021}
}

We show for the first time that a multilayer perceptron (MLP) can serve as the only scene representation in a real-time SLAM system for a handheld RGB-D camera. Our network is trained in live operation without prior data, building a dense, scene-specific implicit 3D model of occupancy and colour which is also immediately used for tracking. Achieving real-time SLAM via continual training of a neural network against a live image stream requires significant innovation. Our iMAP algorithm uses a keyframe structure and multi-processing computation flow, with dynamic information-guided pixel sampling for speed, with tracking at 10 Hz and global map updating at 2 Hz. The advantages of an implicit MLP over standard dense SLAM techniques include efficient geometry representation with automatic detail control and smooth, plausible filling-in of unobserved regions such as the back surfaces of objects.

@article{
 title = {Neural Reprojection Error: Merging Feature Learning and Camera Pose Estimation},
 type = {article},
 year = {2021},
 pages = {414-423},
 id = {b3e876f4-b821-3375-9d7e-daf24b855407},
 created = {2022-09-12T09:39:09.793Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-10-12T09:47:44.968Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {97693603-b330-4e3e-8cf5-d549e6474921},
 private_publication = {false},
 abstract = {Absolute camera pose estimation is usually addressed by sequentially solving two distinct subproblems: First a feature matching problem that seeks to establish putative 2D-3D correspondences, and then a Perspective-n-Point problem that minimizes, w.r.t. the camera pose, the sum of so-called Reprojection Errors (RE). We argue that generating putative 2D-3D correspondences 1) leads to an important loss of information that needs to be compensated as far as possible, within RE, through the choice of a robust loss and the tuning of its hyperparameters and 2) may lead to an RE that conveys erroneous data to the pose estimator. In this paper, we introduce the Neural Reprojection Error (NRE) as a substitute for RE. NRE allows to rethink the camera pose estimation problem by merging it with the feature learning problem, hence leveraging richer information than 2D-3D correspondences and eliminating the need for choosing a robust loss and its hyperparameters. Thus NRE can be used as training loss to learn image descriptors tailored for pose estimation. We also propose a coarse-to-fine optimization method able to very efficiently minimize a sum of NRE terms w.r.t. the camera pose. We experimentally demonstrate that NRE is a good substitute for RE as it significantly improves both the robustness and the accuracy of the camera pose estimate while being computationally and memory highly efficient. From a broader point of view, we believe this new way of merging deep learning and 3D geometry may be useful in other computer vision applications. Source code and model weights will be made available at hugogermain.com/nre.},
 bibtype = {article},
 author = {Germain, Hugo and Lepetit, Vincent and Bourmaud, Guillaume},
 doi = {10.1109/CVPR46437.2021.00048},
 journal = {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition},
 keywords = {Benji}
}

Absolute camera pose estimation is usually addressed by sequentially solving two distinct subproblems: First a feature matching problem that seeks to establish putative 2D-3D correspondences, and then a Perspective-n-Point problem that minimizes, w.r.t. the camera pose, the sum of so-called Reprojection Errors (RE). We argue that generating putative 2D-3D correspondences 1) leads to an important loss of information that needs to be compensated as far as possible, within RE, through the choice of a robust loss and the tuning of its hyperparameters and 2) may lead to an RE that conveys erroneous data to the pose estimator. In this paper, we introduce the Neural Reprojection Error (NRE) as a substitute for RE. NRE allows to rethink the camera pose estimation problem by merging it with the feature learning problem, hence leveraging richer information than 2D-3D correspondences and eliminating the need for choosing a robust loss and its hyperparameters. Thus NRE can be used as training loss to learn image descriptors tailored for pose estimation. We also propose a coarse-to-fine optimization method able to very efficiently minimize a sum of NRE terms w.r.t. the camera pose. We experimentally demonstrate that NRE is a good substitute for RE as it significantly improves both the robustness and the accuracy of the camera pose estimate while being computationally and memory highly efficient. From a broader point of view, we believe this new way of merging deep learning and 3D geometry may be useful in other computer vision applications. Source code and model weights will be made available at hugogermain.com/nre.

@article{
 title = {Deep Semantic Segmentation at the Edge for Autonomous Navigation in Vineyard Rows},
 type = {article},
 year = {2021},
 pages = {3421-3428},
 id = {28737b74-3d7d-3311-a18b-ba557fe703f4},
 created = {2022-09-14T06:35:12.952Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:01.188Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4c7c81ce-f24b-44ae-bc2a-bf60600a3a24,35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Precision agriculture is a fast-growing field that aims at introducing affordable and effective automation into agricultural processes. Nowadays, algorithmic solutions for navigation in vineyards require expensive sensors and high computational workloads that preclude large-scale applicability of autonomous robotic platforms in real business case scenarios. From this perspective, our novel proposed control leverages the latest advancement in machine perception and edge AI techniques to achieve highly affordable and reliable navigation inside vineyard rows with low computational and power consumption. Indeed, using a custom-trained segmentation network and a low-range RGB-D camera, we are able to take advantage of the semantic information of the environment to produce smooth trajectories and stable control in different vineyards scenarios. Moreover, the segmentation maps generated by the control algorithm itself could be directly exploited as filters for a vegetative assessment of the crop status. Extensive experimentations and evaluations against real-world data and simulated environments demonstrated the effectiveness and intrinsic robustness of our methodology.},
 bibtype = {article},
 author = {Aghi, Diego and Cerrato, Simone and Mazzia, Vittorio and Chiaberge, Marcello},
 doi = {10.1109/IROS51168.2021.9635969},
 journal = {IEEE International Conference on Intelligent Robots and Systems},
 keywords = {aghi2021deepsemanticsegmentation}
}

Precision agriculture is a fast-growing field that aims at introducing affordable and effective automation into agricultural processes. Nowadays, algorithmic solutions for navigation in vineyards require expensive sensors and high computational workloads that preclude large-scale applicability of autonomous robotic platforms in real business case scenarios. From this perspective, our novel proposed control leverages the latest advancement in machine perception and edge AI techniques to achieve highly affordable and reliable navigation inside vineyard rows with low computational and power consumption. Indeed, using a custom-trained segmentation network and a low-range RGB-D camera, we are able to take advantage of the semantic information of the environment to produce smooth trajectories and stable control in different vineyards scenarios. Moreover, the segmentation maps generated by the control algorithm itself could be directly exploited as filters for a vegetative assessment of the crop status. Extensive experimentations and evaluations against real-world data and simulated environments demonstrated the effectiveness and intrinsic robustness of our methodology.

@article{
 title = {Unmanned vehicles in smart farming: A survey and a glance at future horizons},
 type = {article},
 year = {2021},
 pages = {1-8},
 id = {38e785d2-71c8-3b67-b825-e485207b1d91},
 created = {2022-09-14T06:35:12.958Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:01.173Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Smart and Precision Agriculture is nowadays already exploiting advanced drones and machinery, but it is foreseen that in the near future more complex and intelligent applications will be required to be brought on-board to improve qualitatively and quantitative production. In this paper, we present an overview on the current usage of autonomous drones in this field and on the augmented computing capabilities that they could count on when companion computers are coupled to flight controllers. The paper also present a novel architecture for companion computers that is under development within the Comp4Drones ECSEL-JU project.},
 bibtype = {article},
 author = {Madroñal, Daniel and Palumbo, Francesca and Capotondi, Alessandro and Marongiu, Andrea},
 doi = {10.1145/3444950.3444958},
 journal = {ACM International Conference Proceeding Series},
 keywords = {madronal2021unmannedvechiclessmart}
}

Smart and Precision Agriculture is nowadays already exploiting advanced drones and machinery, but it is foreseen that in the near future more complex and intelligent applications will be required to be brought on-board to improve qualitatively and quantitative production. In this paper, we present an overview on the current usage of autonomous drones in this field and on the augmented computing capabilities that they could count on when companion computers are coupled to flight controllers. The paper also present a novel architecture for companion computers that is under development within the Comp4Drones ECSEL-JU project.

@article{
 title = {In-field automatic detection of grape bunches under a totally uncontrolled environment},
 type = {article},
 year = {2021},
 keywords = {ghiani2021infieldautomaticdetection},
 volume = {21},
 id = {7b6301f5-c401-3e1a-9551-1f8304d9999d},
 created = {2022-09-14T06:35:13.079Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:15.039Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {An early estimation of the exact number of fruits, flowers, and trees helps farmers to make better decisions on cultivation practices, plant disease prevention, and the size of harvest labor force. The current practice of yield estimation based on manual counting of fruits or flowers by workers is a time consuming and expensive process and it is not feasible for large fields. Automatic yield estimation based on robotic agriculture provides a viable solution in this regard. In a typical image classification process, the task is not only to specify the presence or absence of a given object on a specific location, while counting how many objects are present in the scene. The success of these tasks largely depends on the availability of a large amount of training samples. This paper presents a detector of bunches of one fruit, grape, based on a deep convolutional neural network trained to detect vine bunches directly on the field. Experimental results show a 91% mean Average Precision.},
 bibtype = {article},
 author = {Ghiani, Luca and Sassu, Alberto and Palumbo, Francesca and Mercenaro, Luca and Gambella, Filippo},
 doi = {10.3390/s21113908},
 journal = {Sensors},
 number = {11}
}

An early estimation of the exact number of fruits, flowers, and trees helps farmers to make better decisions on cultivation practices, plant disease prevention, and the size of harvest labor force. The current practice of yield estimation based on manual counting of fruits or flowers by workers is a time consuming and expensive process and it is not feasible for large fields. Automatic yield estimation based on robotic agriculture provides a viable solution in this regard. In a typical image classification process, the task is not only to specify the presence or absence of a given object on a specific location, while counting how many objects are present in the scene. The success of these tasks largely depends on the availability of a large amount of training samples. This paper presents a detector of bunches of one fruit, grape, based on a deep convolutional neural network trained to detect vine bunches directly on the field. Experimental results show a 91% mean Average Precision.

@article{
 title = {InLoc: Indoor Visual Localization with Dense Matching and View Synthesis},
 type = {article},
 year = {2021},
 keywords = {Visual localization,feature matching,image retrieval,place recognition,pose estimation,view synthesis},
 pages = {1293-1307},
 volume = {43},
 id = {deb9b064-97d7-328f-89d6-7366aef124ca},
 created = {2022-09-19T10:49:12.129Z},
 file_attached = {true},
 profile_id = {276016a7-2c9d-3507-8888-093db7c54774},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-09-26T08:15:25.640Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {02fb5526-03ff-44ad-8d5c-42bd496c3100},
 private_publication = {false},
 abstract = {We seek to predict the 6 degree-of-freedom (6DoF) pose of a query photograph with respect to a large indoor 3D map. The contributions of this work are three-fold. First, we develop a new large-scale visual localization method targeted for indoor spaces. The method proceeds along three steps: (i) efficient retrieval of candidate poses that scales to large-scale environments, (ii) pose estimation using dense matching rather than sparse local features to deal with weakly textured indoor scenes, and (iii) pose verification by virtual view synthesis that is robust to significant changes in viewpoint, scene layout, and occlusion. Second, we release a new dataset with reference 6DoF poses for large-scale indoor localization. Query photographs are captured by mobile phones at a different time than the reference 3D map, thus presenting a realistic indoor localization scenario. Third, we demonstrate that our method significantly outperforms current state-of-the-art indoor localization approaches on this new challenging data. Code and data are publicly available.},
 bibtype = {article},
 author = {Taira, Hajime and Okutomi, Masatoshi and Sattler, Torsten and Cimpoi, Mircea and Pollefeys, Marc and Sivic, Josef and Pajdla, Tomas and Torii, Akihiko},
 doi = {10.1109/TPAMI.2019.2952114},
 journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
 number = {4}
}

We seek to predict the 6 degree-of-freedom (6DoF) pose of a query photograph with respect to a large indoor 3D map. The contributions of this work are three-fold. First, we develop a new large-scale visual localization method targeted for indoor spaces. The method proceeds along three steps: (i) efficient retrieval of candidate poses that scales to large-scale environments, (ii) pose estimation using dense matching rather than sparse local features to deal with weakly textured indoor scenes, and (iii) pose verification by virtual view synthesis that is robust to significant changes in viewpoint, scene layout, and occlusion. Second, we release a new dataset with reference 6DoF poses for large-scale indoor localization. Query photographs are captured by mobile phones at a different time than the reference 3D map, thus presenting a realistic indoor localization scenario. Third, we demonstrate that our method significantly outperforms current state-of-the-art indoor localization approaches on this new challenging data. Code and data are publicly available.

@article{
 title = {Technological support for detection and prediction of plant diseases: A systematic mapping study},
 type = {article},
 year = {2021},
 keywords = {bischoff2021technologicalsupportdetection},
 pages = {105922},
 volume = {181},
 websites = {https://doi.org/10.1016/j.compag.2020.105922},
 publisher = {Elsevier B.V.},
 id = {7d57ad94-a67b-39f2-8fd6-b8c452397043},
 created = {2022-09-21T05:45:27.408Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T11:58:00.578Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The field of plant disease diagnosis and epidemiology seeks to assess symptoms caused by pathogens. Different infectious and non-infectious agents can cause similar symptoms in plant organs. Diagnosing diseases is crucial, but it remains an inherently manual and error-prone task. Many works have been proposed to diagnose plant diseases, mainly using machine learning approaches. Even though this field affects agribusiness areas, little has been done to classify and map the current literature. This article presents a comprehensive overview of the current literature, and draw some research gaps, trends, and challenges that are worth investigating. A systematic mapping of the literature was carried out in pairs, following well-established practice guidelines. In total, 56 primary studies were carefully selected from a sample of 668 papers, which were retrieved from 9 widely recognized electronic databases. They were analyzed and categorized to answer seven research questions. The results show that 41% of primary studies applied machine learning techniques to detect diseases, 32% used image sensors to identify symptoms related to plant diseases, 30% focused on proposing new models of machine learning to detect diseases 34% were evaluation studies, and 71% were published in scientific journals. The association between computer vision and neural networks appears as a promising field of research for the detection of diseases. Finally, this article can serve as a starting point for upcoming studies, providing insights from a systematic map of the literature.},
 bibtype = {article},
 author = {Bischoff, Vinicius and Farias, Kleinner and Menzen, Juliano Paulo and Pessin, Gustavo},
 doi = {10.1016/j.compag.2020.105922},
 journal = {Computers and Electronics in Agriculture},
 number = {June 2019}
}

The field of plant disease diagnosis and epidemiology seeks to assess symptoms caused by pathogens. Different infectious and non-infectious agents can cause similar symptoms in plant organs. Diagnosing diseases is crucial, but it remains an inherently manual and error-prone task. Many works have been proposed to diagnose plant diseases, mainly using machine learning approaches. Even though this field affects agribusiness areas, little has been done to classify and map the current literature. This article presents a comprehensive overview of the current literature, and draw some research gaps, trends, and challenges that are worth investigating. A systematic mapping of the literature was carried out in pairs, following well-established practice guidelines. In total, 56 primary studies were carefully selected from a sample of 668 papers, which were retrieved from 9 widely recognized electronic databases. They were analyzed and categorized to answer seven research questions. The results show that 41% of primary studies applied machine learning techniques to detect diseases, 32% used image sensors to identify symptoms related to plant diseases, 30% focused on proposing new models of machine learning to detect diseases 34% were evaluation studies, and 71% were published in scientific journals. The association between computer vision and neural networks appears as a promising field of research for the detection of diseases. Finally, this article can serve as a starting point for upcoming studies, providing insights from a systematic map of the literature.

@article{
 title = {A grapevine leaves dataset for early detection and classification of esca disease in vineyards through machine learning},
 type = {article},
 year = {2021},
 keywords = {alessandrini2021grapevineleavesdataset},
 pages = {106809},
 volume = {35},
 websites = {https://doi.org/10.1016/j.dib.2021.106809},
 publisher = {Elsevier Inc.},
 id = {6d02e585-b33a-38ac-a47a-12a22026b0aa},
 created = {2022-09-21T05:45:27.534Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:00.677Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {e09e8a73-297c-40aa-8415-81ec386a90b0,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Esca is one of the most common disease that can severely damage grapevine. This disease, if not properly treated in time, is the cause of vegetative stress or death of the attacked plant, with the consequence of losses in production as well as a rising risk of propagation to the closer grapevines. Nowadays, the detection of Esca is carried out manually through visual surveys usually done by agronomists, requiring enormous amount of time. Recently, image processing, computer vision and machine learning methods have been widely adopted for plant diseases classification. These methods can minimize the time spent for anomaly detection ensuring an early detection of Esca disease in grapevine plants that helps in preventing it to spread in the vineyards and in minimizing the financial loss to the wine producers. In this article, an image dataset of grapevine leaves is presented. The dataset holds grapevine leaves images belonging to two classes: unhealthy leaves acquired from plants affected by Esca disease and healthy leaves. The data presented has been collected to be used in a research project jointly developed by the Department of Information Engineering, Polytechnic University of Marche, Ancona, Italy and the STMicroelectronics, Italy, under the cooperation of the Umani Ronchi SPA winery, Osimo, Ancona, Marche, Italy. The dataset could be helpful to researchers who use machine learning and computer vision algorithms to develop applications that help agronomists in early detection of grapevine plant diseases. The dataset is freely available at http://dx.doi.org/10.17632/89cnxc58kj.1},
 bibtype = {article},
 author = {Alessandrini, M. and Calero Fuentes Rivera, R. and Falaschetti, L. and Pau, D. and Tomaselli, V. and Turchetti, C.},
 doi = {10.1016/j.dib.2021.106809},
 journal = {Data in Brief}
}

Esca is one of the most common disease that can severely damage grapevine. This disease, if not properly treated in time, is the cause of vegetative stress or death of the attacked plant, with the consequence of losses in production as well as a rising risk of propagation to the closer grapevines. Nowadays, the detection of Esca is carried out manually through visual surveys usually done by agronomists, requiring enormous amount of time. Recently, image processing, computer vision and machine learning methods have been widely adopted for plant diseases classification. These methods can minimize the time spent for anomaly detection ensuring an early detection of Esca disease in grapevine plants that helps in preventing it to spread in the vineyards and in minimizing the financial loss to the wine producers. In this article, an image dataset of grapevine leaves is presented. The dataset holds grapevine leaves images belonging to two classes: unhealthy leaves acquired from plants affected by Esca disease and healthy leaves. The data presented has been collected to be used in a research project jointly developed by the Department of Information Engineering, Polytechnic University of Marche, Ancona, Italy and the STMicroelectronics, Italy, under the cooperation of the Umani Ronchi SPA winery, Osimo, Ancona, Marche, Italy. The dataset could be helpful to researchers who use machine learning and computer vision algorithms to develop applications that help agronomists in early detection of grapevine plant diseases. The dataset is freely available at http://dx.doi.org/10.17632/89cnxc58kj.1

@article{
 title = {Computer vision, IoT and data fusion for crop disease detection using machine learning: A survey and ongoing research},
 type = {article},
 year = {2021},
 keywords = {ouhami2021computervisioniot},
 volume = {13},
 id = {2b6de8be-0e82-33ac-9144-1805b8b59364},
 created = {2022-09-21T05:45:27.588Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T12:03:51.221Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Crop diseases constitute a serious issue in agriculture, affecting both quality and quantity of agriculture production. Disease control has been a research object in many scientific and technologic domains. Technological advances in sensors, data storage, computing resources and artificial intelligence have shown enormous potential to control diseases effectively. A growing body of literature recognizes the importance of using data from different types of sensors and machine learning approaches to build models for detection, prediction, analysis, assessment, etc. However, the increasing number and diversity of research studies requires a literature review for further developments and contributions in this area. This paper reviews state-of-the-art machine learning methods that use different data sources, applied to plant disease detection. It lists traditional and deep learning methods associated with the main data acquisition modalities, namely IoT, ground imaging, unmanned aerial vehicle imaging and satellite imaging. In addition, this study examines the role of data fusion for ongoing research in the context of disease detection. It highlights the advantage of intelligent data fusion techniques, from heterogeneous data sources, to improve plant health status prediction and presents the main challenges facing this field. The study concludes with a discussion of several current issues and research trends.},
 bibtype = {article},
 author = {Ouhami, Maryam and Hafiane, Adel and Es-Saady, Youssef and El Hajji, Mohamed and Canals, Raphael},
 doi = {10.3390/rs13132486},
 journal = {Remote Sensing},
 number = {13}
}

Crop diseases constitute a serious issue in agriculture, affecting both quality and quantity of agriculture production. Disease control has been a research object in many scientific and technologic domains. Technological advances in sensors, data storage, computing resources and artificial intelligence have shown enormous potential to control diseases effectively. A growing body of literature recognizes the importance of using data from different types of sensors and machine learning approaches to build models for detection, prediction, analysis, assessment, etc. However, the increasing number and diversity of research studies requires a literature review for further developments and contributions in this area. This paper reviews state-of-the-art machine learning methods that use different data sources, applied to plant disease detection. It lists traditional and deep learning methods associated with the main data acquisition modalities, namely IoT, ground imaging, unmanned aerial vehicle imaging and satellite imaging. In addition, this study examines the role of data fusion for ongoing research in the context of disease detection. It highlights the advantage of intelligent data fusion techniques, from heterogeneous data sources, to improve plant health status prediction and presents the main challenges facing this field. The study concludes with a discussion of several current issues and research trends.

@article{
 title = {Deep learning for the differentiation of downy mildew and spider mite in grapevine under field conditions},
 type = {article},
 year = {2021},
 keywords = {gutierrez2021deeplearningdifferentiation},
 pages = {1-9},
 volume = {182},
 id = {e8def660-d90f-33b5-bbc8-fb23f7aa2020},
 created = {2022-09-21T05:45:27.656Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-23T14:59:48.223Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {6b565182-74c4-44fd-98cc-10618152e2ae,4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Diseases and pests cause serious damage in crop production, reducing yield and fruit quality. Their identification is often time-consuming and requires trained personnel. New sensing technologies and artificial intelligence could be used for automatic identification of disease and pest symptoms on grapevine in precision viticulture. The aim of this work was to apply deep learning modelling and computer vision for the detection and differentiation of downy mildew and spider mite symptoms in grapevine leaves under field conditions. RGB images of grapevine canopy leaves with downy mildew symptoms, with spider mite symptoms and without symptoms were taken under field conditions in a commercial vineyard. The images were prepared using computer vision techniques to increase disease visual features. Finally, deep learning was used to train a model capable of differentiating leaf images of the three classes. An accuracy up to 0.94 (F1-score of 0.94) was obtained by classifying leaves with downy mildew, spider mite and without symptoms at the same time, using a hold-out validation. Additionally, accuracies between 0.89 and 0.91 (F1-scores between 0.89 and 0.91) were obtained in the binary classification of the disease and pest, obtaining the best results in differentiating downy mildew from spider mite symptoms. This high accuracy demonstrates the effectiveness of deep learning and computer vision techniques for the classification of grapevine leaf images taken under field conditions, automatically finding complex features capable of differentiating leaves with spider mite symptoms, with downy mildew symptoms and without any. These results prove the potential of these non-invasive techniques in the detection and differentiation of pests and diseases in commercial crop production.},
 bibtype = {article},
 author = {Gutiérrez, Salvador and Hernández, Inés and Ceballos, Sara and Barrio, Ignacio and Díez-Navajas, Ana M. and Tardaguila, Javier},
 doi = {10.1016/j.compag.2021.105991},
 journal = {Computers and Electronics in Agriculture},
 number = {February}
}

Diseases and pests cause serious damage in crop production, reducing yield and fruit quality. Their identification is often time-consuming and requires trained personnel. New sensing technologies and artificial intelligence could be used for automatic identification of disease and pest symptoms on grapevine in precision viticulture. The aim of this work was to apply deep learning modelling and computer vision for the detection and differentiation of downy mildew and spider mite symptoms in grapevine leaves under field conditions. RGB images of grapevine canopy leaves with downy mildew symptoms, with spider mite symptoms and without symptoms were taken under field conditions in a commercial vineyard. The images were prepared using computer vision techniques to increase disease visual features. Finally, deep learning was used to train a model capable of differentiating leaf images of the three classes. An accuracy up to 0.94 (F1-score of 0.94) was obtained by classifying leaves with downy mildew, spider mite and without symptoms at the same time, using a hold-out validation. Additionally, accuracies between 0.89 and 0.91 (F1-scores between 0.89 and 0.91) were obtained in the binary classification of the disease and pest, obtaining the best results in differentiating downy mildew from spider mite symptoms. This high accuracy demonstrates the effectiveness of deep learning and computer vision techniques for the classification of grapevine leaf images taken under field conditions, automatically finding complex features capable of differentiating leaves with spider mite symptoms, with downy mildew symptoms and without any. These results prove the potential of these non-invasive techniques in the detection and differentiation of pests and diseases in commercial crop production.

@article{
 title = {Advances in agriculture robotics: A state-of-the-art review and challenges ahead},
 type = {article},
 year = {2021},
 keywords = {Agricultural robots,Agriculture 4.0,Precision agriculture},
 pages = {1-31},
 volume = {10},
 id = {dfb8e567-50a5-3f59-80d4-79c8f5eba7f3},
 created = {2022-09-21T05:45:27.727Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T12:02:27.699Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {Oliveira2021},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The constant advances in agricultural robotics aim to overcome the challenges imposed by population growth, accelerated urbanization, high competitiveness of high-quality products, environmental preservation and a lack of qualified labor. In this sense, this review paper surveys the main existing applications of agricultural robotic systems for the execution of land preparation before planting, sowing, planting, plant treatment, harvesting, yield estimation and phenotyping. In general, all robots were evaluated according to the following criteria: its locomotion system, what is the final application, if it has sensors, robotic arm and/or computer vision algorithm, what is its development stage and which country and continent they belong. After evaluating all similar characteristics, to expose the research trends, common pitfalls and the characteristics that hinder commercial development, and discover which countries are investing into Research and Development (R&D) in these technologies for the future, four major areas that need future research work for enhancing the state of the art in smart agriculture were highlighted: locomotion systems, sensors, computer vision algorithms and communication technologies. The results of this research suggest that the investment in agricultural robotic systems allows to achieve short—harvest monitoring—and long-term objectives—yield estimation.},
 bibtype = {article},
 author = {Oliveira, Luiz F.P. and Moreira, António P. and Silva, Manuel F.},
 doi = {10.3390/robotics10020052},
 journal = {Robotics},
 number = {2}
}

The constant advances in agricultural robotics aim to overcome the challenges imposed by population growth, accelerated urbanization, high competitiveness of high-quality products, environmental preservation and a lack of qualified labor. In this sense, this review paper surveys the main existing applications of agricultural robotic systems for the execution of land preparation before planting, sowing, planting, plant treatment, harvesting, yield estimation and phenotyping. In general, all robots were evaluated according to the following criteria: its locomotion system, what is the final application, if it has sensors, robotic arm and/or computer vision algorithm, what is its development stage and which country and continent they belong. After evaluating all similar characteristics, to expose the research trends, common pitfalls and the characteristics that hinder commercial development, and discover which countries are investing into Research and Development (R&D) in these technologies for the future, four major areas that need future research work for enhancing the state of the art in smart agriculture were highlighted: locomotion systems, sensors, computer vision algorithms and communication technologies. The results of this research suggest that the investment in agricultural robotic systems allows to achieve short—harvest monitoring—and long-term objectives—yield estimation.

@article{
 title = {Real-time multispectral image processing and registration on 3D point Cloud for Vineyard Analysis},
 type = {article},
 year = {2021},
 keywords = {clamens2021realtimemultispectral},
 pages = {388-398},
 volume = {4},
 id = {69e44fe6-4fc5-324c-a588-468ebc8466ad},
 created = {2022-09-21T05:45:27.783Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:01.403Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Nowadays, precision agriculture and precision viticulture are under strong development. In order to accomplish effective actions, robots require robust perception of the culture and the surrounding environment. Computer vision systems have to identify plant parts (branches, stems, leaves, flowers, fruits, vegetables, etc.) and their respective health status. Moreover, they must merge various plant information, to measure agronomic indices, to classify them and finally to extract data to enable the agriculturist or expert to make a relevant decision. We propose a real-time method to acquire, process and register multispectral images fused to 3D. The sensors system, consisting of a multispectral camera and a RGB-D sensor, can be embedded on a ground robot or other terrestrial vehicles. Experiments conducted in the vineyard field demonstrate that agronomic analyses are allowed.},
 bibtype = {article},
 author = {Clamens, Thibault and Alexakis, Georgios and Duverne, Raphaël and Seulin, Ralph and Fauvet, Eric and Fofi, David},
 doi = {10.5220/0010266203880398},
 journal = {VISIGRAPP 2021 - Proceedings of the 16th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications},
 number = {Visigrapp}
}

Nowadays, precision agriculture and precision viticulture are under strong development. In order to accomplish effective actions, robots require robust perception of the culture and the surrounding environment. Computer vision systems have to identify plant parts (branches, stems, leaves, flowers, fruits, vegetables, etc.) and their respective health status. Moreover, they must merge various plant information, to measure agronomic indices, to classify them and finally to extract data to enable the agriculturist or expert to make a relevant decision. We propose a real-time method to acquire, process and register multispectral images fused to 3D. The sensors system, consisting of a multispectral camera and a RGB-D sensor, can be embedded on a ground robot or other terrestrial vehicles. Experiments conducted in the vineyard field demonstrate that agronomic analyses are allowed.

@article{
 title = {LiDAR-based Structure Tracking for Agricultural Robots: Application to Autonomous Navigation in Vineyards},
 type = {article},
 year = {2021},
 keywords = {nehme2021lidarbasedstructure},
 pages = {1-16},
 volume = {103},
 id = {2e13fdd8-dd38-38e4-9643-b777e7b6354c},
 created = {2022-09-21T05:45:27.894Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:09.537Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4c7c81ce-f24b-44ae-bc2a-bf60600a3a24,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Autonomous navigation is a key defining feature that allows agricultural robots to perform automated farming tasks. Global navigation satellite system (GNSS) technology is providing autonomous navigation solutions for current commercial robotic platforms that can achieve centimeter-level accuracy when real-time kinematic (RTK) corrections are available. However, GNSS-based solutions are expensive and require a long preparation phase where the field has to be surveyed with a GNSS rover to collect waypoints for the navigation path. An alternative navigation approach can be provided by Local perception sensors, such as LiDAR scanners, by tracking geometric features in the perceived scene. This paper presents a robust LiDAR-based solution for structure tracking along vine rows. The proposed method does not require prior field surveying, and it is insensitive to crop characteristics such as row width and spacing. Moreover, the proposed algorithm identifies and builds an online regression model of the structure. This is done by applying the Hough transform with a parameterization and search method motivated by a practical interpretation of point cloud statistics. The proposed method was tested on a commercial robotic platform in two configurations of vineyards. The experiments show that the proposed algorithm achieves consistent and accurate row tracking, which was validated against a reliable RTK-GNSS ground truth.},
 bibtype = {article},
 author = {Nehme, Hassan and Aubry, Clément and Solatges, Thomas and Savatier, Xavier and Rossi, Romain and Boutteau, Rémi},
 doi = {10.1007/s10846-021-01519-7},
 journal = {Journal of Intelligent and Robotic Systems: Theory and Applications},
 number = {4}
}

Autonomous navigation is a key defining feature that allows agricultural robots to perform automated farming tasks. Global navigation satellite system (GNSS) technology is providing autonomous navigation solutions for current commercial robotic platforms that can achieve centimeter-level accuracy when real-time kinematic (RTK) corrections are available. However, GNSS-based solutions are expensive and require a long preparation phase where the field has to be surveyed with a GNSS rover to collect waypoints for the navigation path. An alternative navigation approach can be provided by Local perception sensors, such as LiDAR scanners, by tracking geometric features in the perceived scene. This paper presents a robust LiDAR-based solution for structure tracking along vine rows. The proposed method does not require prior field surveying, and it is insensitive to crop characteristics such as row width and spacing. Moreover, the proposed algorithm identifies and builds an online regression model of the structure. This is done by applying the Hough transform with a parameterization and search method motivated by a practical interpretation of point cloud statistics. The proposed method was tested on a commercial robotic platform in two configurations of vineyards. The experiments show that the proposed algorithm achieves consistent and accurate row tracking, which was validated against a reliable RTK-GNSS ground truth.

@article{
 title = {Smart applications and digital technologies in viticulture: A review},
 type = {article},
 year = {2021},
 keywords = {tardaguila2021smartapplicationdigital},
 pages = {100005},
 volume = {1},
 websites = {https://doi.org/10.1016/j.atech.2021.100005},
 publisher = {Elsevier B.V.},
 id = {ba8d4c0a-ca9b-3382-8ab8-0d08816919cf},
 created = {2022-09-21T05:45:28.076Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:01.529Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {It is important to continuously monitor the long-term impact of viticultural management practices and assess opportunities for improving the environmental footprint of vineyard operations. This is particularly relevant to the wine industry as growers face disruptive challenges caused by climate change, shortages of labour and escalating production costs. In recent years there has been considerable development and testing of non-invasive digital technologies, some of which have already demonstrated an improvement in the way that wine grapes are grown, managed and harvested to produce quality wines in a manner that is both environmentally and economically sustainable. In this paper, we describe a number of sensing technologies including spectroscopy, multispectral and hyperspectral imaging, chlorophyll fluorescence, thermography, electrical resistivity, laser imaging detection and ranging, and computer vision and the platforms where they are generally mounted or embedded for either proximal or remote monitoring. Artificial intelligence is also discussed as it is useful as a means of transforming data into different pieces of information used by the grape grower for making informed decisions. A key objective of using these technologies is to obtain and supply data and information to grape growers and wine producers as a basis for improving land and vine management through a more-informed decision-making process. The current and future application of these technologies and artificial intelligence in vineyards are discussed in relation to soil properties and topography, vegetative growth, canopy architecture, nutrient and water status, pests and diseases, crop forecasting, yield and fruit composition, vineyard sampling, targeted management and selective harvesting. The principles behind how these technologies operate are also described. While the technologies have enormous potential for growers, their adoption and use will depend on user-friendly software and devices, together with affordable costs, at the field scale.},
 bibtype = {article},
 author = {Tardaguila, Javier and Stoll, Manfred and Gutiérrez, Salvador and Proffitt, Tony and Diago, Maria P.},
 doi = {10.1016/j.atech.2021.100005},
 journal = {Smart Agricultural Technology},
 number = {July}
}

It is important to continuously monitor the long-term impact of viticultural management practices and assess opportunities for improving the environmental footprint of vineyard operations. This is particularly relevant to the wine industry as growers face disruptive challenges caused by climate change, shortages of labour and escalating production costs. In recent years there has been considerable development and testing of non-invasive digital technologies, some of which have already demonstrated an improvement in the way that wine grapes are grown, managed and harvested to produce quality wines in a manner that is both environmentally and economically sustainable. In this paper, we describe a number of sensing technologies including spectroscopy, multispectral and hyperspectral imaging, chlorophyll fluorescence, thermography, electrical resistivity, laser imaging detection and ranging, and computer vision and the platforms where they are generally mounted or embedded for either proximal or remote monitoring. Artificial intelligence is also discussed as it is useful as a means of transforming data into different pieces of information used by the grape grower for making informed decisions. A key objective of using these technologies is to obtain and supply data and information to grape growers and wine producers as a basis for improving land and vine management through a more-informed decision-making process. The current and future application of these technologies and artificial intelligence in vineyards are discussed in relation to soil properties and topography, vegetative growth, canopy architecture, nutrient and water status, pests and diseases, crop forecasting, yield and fruit composition, vineyard sampling, targeted management and selective harvesting. The principles behind how these technologies operate are also described. While the technologies have enormous potential for growers, their adoption and use will depend on user-friendly software and devices, together with affordable costs, at the field scale.

@article{
 title = {Smart Farming in Europe},
 type = {article},
 year = {2021},
 keywords = {moysiadis2021smartfarmingeurope},
 pages = {100345},
 volume = {39},
 websites = {https://doi.org/10.1016/j.cosrev.2020.100345},
 publisher = {Elsevier Inc.},
 id = {8c5c54ce-10a8-3175-be87-ad0c30f09a48},
 created = {2022-09-21T05:45:28.367Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T11:07:22.324Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Smart Farming is the new term in the agriculture sector, aiming to transform the traditional techniques to innovative solutions based on Information Communication Technologies (ICT). Concretely, technologies like Unmanned Aerial Vehicles (UAVs), Unmanned Ground Vehicles (UGVs), Image Processing, Machine Learning, Big Data, Cloud Computing, and Wireless Sensor Networks (WSNs), are expected to bring significant changes in this area. Expected benefits are the increase in production, the decrease in cost by reducing the inputs needed such as fuel, fertilizer and pesticides, the reduction in labor efforts, and finally improvement in the quality of the final products. Such innovative methods are crucial in recent days, due to the exponential increase of the global population, the importance of producing healthier products grown with as much fewer pesticides, where public opinion of European citizens is sensitized. Moreover, due to the globalization of the world economy, European countries face the low cost of production of other low-income countries. In this vein, Europe tries to evolve its agriculture domain using technology, aiming at the sustainability of its agricultural sector. Although many surveys exist, most of them tackle in a specific scientific area of Smart Farming. An overview of Smart Farming covering all the involved technologies and providing an extensive reference of good practices around Europe is essential. Our expectation from our work is to become a good reference for researchers and help them with their future work. This paper aims to provide a comprehensive reference for European research efforts in Smart Farming and is two-fold. First, we present the research efforts from researchers in Smart Farming, who apply innovative technology trends in various crops around Europe. Second, we provide and analyze the most significant projects in Europe in the area of Smart Farming.},
 bibtype = {article},
 author = {Moysiadis, Vasileios and Sarigiannidis, Panagiotis and Vitsas, Vasileios and Khelifi, Adel},
 doi = {10.1016/j.cosrev.2020.100345},
 journal = {Computer Science Review}
}

Smart Farming is the new term in the agriculture sector, aiming to transform the traditional techniques to innovative solutions based on Information Communication Technologies (ICT). Concretely, technologies like Unmanned Aerial Vehicles (UAVs), Unmanned Ground Vehicles (UGVs), Image Processing, Machine Learning, Big Data, Cloud Computing, and Wireless Sensor Networks (WSNs), are expected to bring significant changes in this area. Expected benefits are the increase in production, the decrease in cost by reducing the inputs needed such as fuel, fertilizer and pesticides, the reduction in labor efforts, and finally improvement in the quality of the final products. Such innovative methods are crucial in recent days, due to the exponential increase of the global population, the importance of producing healthier products grown with as much fewer pesticides, where public opinion of European citizens is sensitized. Moreover, due to the globalization of the world economy, European countries face the low cost of production of other low-income countries. In this vein, Europe tries to evolve its agriculture domain using technology, aiming at the sustainability of its agricultural sector. Although many surveys exist, most of them tackle in a specific scientific area of Smart Farming. An overview of Smart Farming covering all the involved technologies and providing an extensive reference of good practices around Europe is essential. Our expectation from our work is to become a good reference for researchers and help them with their future work. This paper aims to provide a comprehensive reference for European research efforts in Smart Farming and is two-fold. First, we present the research efforts from researchers in Smart Farming, who apply innovative technology trends in various crops around Europe. Second, we provide and analyze the most significant projects in Europe in the area of Smart Farming.

@article{
 title = {AI-powered mobile image acquisition of vineyard insect traps with automatic quality and adequacy assessment},
 type = {article},
 year = {2021},
 keywords = {Artificial intelligence,Image adequacy assessment,Image quality assessment,Insect traps,Machine learning,Mobile image acquisition,Pests monitoring,Viticulture},
 pages = {1-18},
 volume = {11},
 id = {376edf64-d47c-380e-a699-bd8bb4db4bc6},
 created = {2022-09-21T05:45:28.474Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-11-18T12:35:39.124Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {The increasing alarming impacts of climate change are already apparent in viticulture, with unexpected pest outbreaks as one of the most concerning consequences. The monitoring of pests is currently done by deploying chromotropic and delta traps, which attracts insects present in the production environment, and then allows human operators to identify and count them. While the monitoring of these traps is still mostly done through visual inspection by the winegrowers, smartphone image acquisition of those traps is starting to play a key role in assessing the pests’ evolu-tion, as well as enabling the remote monitoring by taxonomy specialists in better assessing the onset outbreaks. This paper presents a new methodology that embeds artificial intelligence into mobile devices to establish the use of hand-held image capture of insect traps for pest detection deployed in vineyards. Our methodology combines different computer vision approaches that improve several aspects of image capture quality and adequacy, namely: (i) image focus validation; (ii) shadows and reflections validation; (iii) trap type detection; (iv) trap segmentation; and (v) perspective correction. A total of 516 images were collected, divided into three different datasets and manually annotated, in order to support the development and validation of the different functionalities. By following this approach, we achieved an accuracy of 84% for focus detection, an accuracy of 80% and 96% for shadows/reflections detection (for delta and chromotropic traps, respectively), as well as mean Jaccard index of 97% for the trap’s segmentation.},
 bibtype = {article},
 author = {Faria, Pedro and Nogueira, Telmo and Ferreira, Ana and Carlos, Cristina and Rosado, Luís},
 doi = {10.3390/agronomy11040731},
 journal = {Agronomy},
 number = {4}
}

The increasing alarming impacts of climate change are already apparent in viticulture, with unexpected pest outbreaks as one of the most concerning consequences. The monitoring of pests is currently done by deploying chromotropic and delta traps, which attracts insects present in the production environment, and then allows human operators to identify and count them. While the monitoring of these traps is still mostly done through visual inspection by the winegrowers, smartphone image acquisition of those traps is starting to play a key role in assessing the pests’ evolu-tion, as well as enabling the remote monitoring by taxonomy specialists in better assessing the onset outbreaks. This paper presents a new methodology that embeds artificial intelligence into mobile devices to establish the use of hand-held image capture of insect traps for pest detection deployed in vineyards. Our methodology combines different computer vision approaches that improve several aspects of image capture quality and adequacy, namely: (i) image focus validation; (ii) shadows and reflections validation; (iii) trap type detection; (iv) trap segmentation; and (v) perspective correction. A total of 516 images were collected, divided into three different datasets and manually annotated, in order to support the development and validation of the different functionalities. By following this approach, we achieved an accuracy of 84% for focus detection, an accuracy of 80% and 96% for shadows/reflections detection (for delta and chromotropic traps, respectively), as well as mean Jaccard index of 97% for the trap’s segmentation.

@article{
 title = {State of the art of monitoring technologies and data processing for precision viticulture},
 type = {article},
 year = {2021},
 keywords = {ammoniaci2021stateartmonitoing},
 pages = {1-21},
 volume = {11},
 id = {5b7bf1d5-448b-3920-ac1c-de324d5b046e},
 created = {2022-09-21T05:45:28.513Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T10:02:24.511Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Precision viticulture (PV) aims to optimize vineyard management, reducing the use of resources, the environmental impact and maximizing the yield and quality of the production. New technologies as UAVs, satellites, proximal sensors and variable rate machines (VRT) are being de-veloped and used more and more frequently in recent years thanks also to informatics systems able to read, analyze and process a huge number of data in order to give the winegrowers a decision support system (DSS) for making better decisions at the right place and time. This review presents a brief state of the art of precision viticulture technologies, focusing on monitoring tools, i.e., re-mote/proximal sensing, variable rate machines, robotics, DSS and the wireless sensor network.},
 bibtype = {article},
 author = {Ammoniaci, Marco and Kartsiotis, Simon Paolo and Perria, Rita and Storchi, Paolo},
 doi = {10.3390/agriculture11030201},
 journal = {Agriculture (Switzerland)},
 number = {3}
}

Precision viticulture (PV) aims to optimize vineyard management, reducing the use of resources, the environmental impact and maximizing the yield and quality of the production. New technologies as UAVs, satellites, proximal sensors and variable rate machines (VRT) are being de-veloped and used more and more frequently in recent years thanks also to informatics systems able to read, analyze and process a huge number of data in order to give the winegrowers a decision support system (DSS) for making better decisions at the right place and time. This review presents a brief state of the art of precision viticulture technologies, focusing on monitoring tools, i.e., re-mote/proximal sensing, variable rate machines, robotics, DSS and the wireless sensor network.

@article{
 title = {Monitoring vineyard water status using sentinel-2 images: Qualitative survey on five wine estates in the south of france},
 type = {article},
 year = {2021},
 keywords = {larochepinel2021monitoringvineyardwater},
 pages = {115-127},
 volume = {55},
 id = {586f1d30-959e-39e6-bfb2-606f6f790682},
 created = {2022-09-21T05:45:28.538Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:00.913Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {The wine industry must face many challenges because of climate change. One of them is the increase in temperatures and droughts events. These changes sometimes lead to yield losses and can also impact the quality of the wine produced (e.g., increased alcohol content). The management of available water is also a sensitive issue as water requirements for vineyard irrigation are quickly increasing in the south of France. In this context, there is a need for a decision tool that can help evaluate the vine water status through the entire growth season at a large scale. To address this issue, we have previously developed a model (see Laroche-Pinel et al. 2021a) which predicts the vine Stem Water Potential (Ψstem) using Sentinel-2 (S2) images. This model was developed based on a field campaign over three years. The present study now aims to investigate the feedback of winegrowers on the outputs of our model. Therefore, it was applied on the plots of five wine estates that do not belong to the set used in the initial paper. The qualitative results show interesting spatial and temporal consistency in accordance with winegrower knowledge, irrigation data, and weather forecast. The predicted Ψstem highlights spatial variability in vine fields where a water source emerges and reflects the differences between vine fields with a drip or sparkling irrigation or without an irrigation system. The predicted Ψstem also clearly reacts to a peak in temperature. According to their feedback, three of the five winegrowers would be glad to use this service in the years to come.},
 bibtype = {article},
 author = {Laroche-Pinel, Eve and Duthoit, Sylvie and Costard, Anne D. and Rousseau, Jacques and Hourdel, Jérome and Vidal-Vigneron, Marie and Cheret, Véronique and Clenet, Harold},
 doi = {10.20870/oeno-one.2021.55.4.4752},
 journal = {Oeno One},
 number = {4}
}

The wine industry must face many challenges because of climate change. One of them is the increase in temperatures and droughts events. These changes sometimes lead to yield losses and can also impact the quality of the wine produced (e.g., increased alcohol content). The management of available water is also a sensitive issue as water requirements for vineyard irrigation are quickly increasing in the south of France. In this context, there is a need for a decision tool that can help evaluate the vine water status through the entire growth season at a large scale. To address this issue, we have previously developed a model (see Laroche-Pinel et al. 2021a) which predicts the vine Stem Water Potential (Ψstem) using Sentinel-2 (S2) images. This model was developed based on a field campaign over three years. The present study now aims to investigate the feedback of winegrowers on the outputs of our model. Therefore, it was applied on the plots of five wine estates that do not belong to the set used in the initial paper. The qualitative results show interesting spatial and temporal consistency in accordance with winegrower knowledge, irrigation data, and weather forecast. The predicted Ψstem highlights spatial variability in vine fields where a water source emerges and reflects the differences between vine fields with a drip or sparkling irrigation or without an irrigation system. The predicted Ψstem also clearly reacts to a peak in temperature. According to their feedback, three of the five winegrowers would be glad to use this service in the years to come.

@article{
 title = {Advances in unmanned aerial system remote sensing for precision viticulture},
 type = {article},
 year = {2021},
 keywords = {sassu2021advanceduav},
 pages = {1-21},
 volume = {21},
 id = {5c9af486-bf7e-3f1c-bb35-1a2df62439ec},
 created = {2022-09-21T05:45:28.635Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:09.609Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {New technologies for management, monitoring, and control of spatio-temporal crop variability in precision viticulture scenarios are numerous. Remote sensing relies on sensors able to provide useful data for the improvement of management efficiency and the optimization of inputs. unmanned aerial systems (UASs) are the newest and most versatile tools, characterized by high precision and accuracy, flexibility, and low operating costs. The work aims at providing a complete overview of the application of UASs in precision viticulture, focusing on the different application purposes, the applied equipment, the potential of technologies combined with UASs for identifying vineyards’ variability. The review discusses the potential of UASs in viticulture by distinguishing five areas of application: rows segmentation and crop features detection techniques; vineyard variability monitoring; estimation of row area and volume; disease detection; vigor and prescription maps creation. Technological innovation and low purchase costs make UASs the core tools for decision support in the customary use by winegrowers. The ability of the systems to respond to the current demands for the acquisition of digital technologies in agricultural fields makes UASs a candidate to play an increasingly important role in future scenarios of viticulture application.},
 bibtype = {article},
 author = {Sassu, Alberto and Gambella, Filippo and Ghiani, Luca and Mercenaro, Luca and Caria, Maria and Pazzona, Antonio Luigi},
 doi = {10.3390/s21030956},
 journal = {Sensors (Switzerland)},
 number = {3}
}

New technologies for management, monitoring, and control of spatio-temporal crop variability in precision viticulture scenarios are numerous. Remote sensing relies on sensors able to provide useful data for the improvement of management efficiency and the optimization of inputs. unmanned aerial systems (UASs) are the newest and most versatile tools, characterized by high precision and accuracy, flexibility, and low operating costs. The work aims at providing a complete overview of the application of UASs in precision viticulture, focusing on the different application purposes, the applied equipment, the potential of technologies combined with UASs for identifying vineyards’ variability. The review discusses the potential of UASs in viticulture by distinguishing five areas of application: rows segmentation and crop features detection techniques; vineyard variability monitoring; estimation of row area and volume; disease detection; vigor and prescription maps creation. Technological innovation and low purchase costs make UASs the core tools for decision support in the customary use by winegrowers. The ability of the systems to respond to the current demands for the acquisition of digital technologies in agricultural fields makes UASs a candidate to play an increasingly important role in future scenarios of viticulture application.

@inproceedings{
 title = {Grapevine Leaf Disease Identification Using Transfer Learning},
 type = {inproceedings},
 year = {2021},
 pages = {43-46},
 publisher = {IEEE},
 id = {692e80ab-a87a-383c-9d2c-471e2a9ae942},
 created = {2022-09-21T05:45:28.681Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T11:47:01.245Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4dd63b9a-2e78-4e16-b45d-b692d8e0c5c3,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Nagi, Reva and Tripathy, Sanjaya Shankar},
 booktitle = {2021 IEEE International Women in Engineering (WIE) Conference on Electrical and Computer Engineering (WIECON-ECE)},
 keywords = {nagi2021grapevineleafdisease}
}

@article{
 title = {Machine vision for ripeness estimation in viticulture automation},
 type = {article},
 year = {2021},
 keywords = {vrochidou2021machinevisionripeness},
 volume = {7},
 id = {f3444b21-39e9-3886-824f-151e4b14be5d},
 created = {2022-09-21T05:45:28.833Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2024-12-05T13:50:07.063Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {35f1a258-89fa-488b-8684-355f11ee4e9b,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Ripeness estimation of fruits and vegetables is a key factor for the optimization of field management and the harvesting of the desired product quality. Typical ripeness estimation involves multiple manual samplings before harvest followed by chemical analyses. Machine vision has paved the way for agricultural automation by introducing quicker, cost-effective, and non-destructive methods. This work comprehensively surveys the most recent applications of machine vision techniques for ripeness estimation. Due to the broad area of machine vision applications in agriculture, this review is limited only to the most recent techniques related to grapes. The aim of this work is to provide an overview of the state-of-the-art algorithms by covering a wide range of applications. The potential of current machine vision techniques for specific viticulture applications is also analyzed. Problems, limitations of each technique, and future trends are discussed. Moreover, the integration of machine vision algorithms in grape harvesting robots for real-time in-field maturity assessment is additionally examined.},
 bibtype = {article},
 author = {Vrochidou, Eleni and Bazinas, Christos and Manios, Michail and Papakostas, George A. and Pachidis, Theodore P. and Kaburlasos, Vassilis G.},
 doi = {10.3390/horticulturae7090282},
 journal = {Horticulturae},
 number = {9}
}

Ripeness estimation of fruits and vegetables is a key factor for the optimization of field management and the harvesting of the desired product quality. Typical ripeness estimation involves multiple manual samplings before harvest followed by chemical analyses. Machine vision has paved the way for agricultural automation by introducing quicker, cost-effective, and non-destructive methods. This work comprehensively surveys the most recent applications of machine vision techniques for ripeness estimation. Due to the broad area of machine vision applications in agriculture, this review is limited only to the most recent techniques related to grapes. The aim of this work is to provide an overview of the state-of-the-art algorithms by covering a wide range of applications. The potential of current machine vision techniques for specific viticulture applications is also analyzed. Problems, limitations of each technique, and future trends are discussed. Moreover, the integration of machine vision algorithms in grape harvesting robots for real-time in-field maturity assessment is additionally examined.

@article{
 title = {A Review on Grape Disease Detection and Classification using Image Processing},
 type = {article},
 year = {2021},
 keywords = {niharika2021reviewgrapedisease},
 pages = {411-425},
 volume = {8},
 id = {1b107b51-cc18-31c4-bfeb-2227ba68c414},
 created = {2022-09-21T06:54:51.023Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T08:16:07.079Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 bibtype = {article},
 author = {Niharika S, N V Sahana, Girija J},
 journal = {Irjet},
 number = {9}
}

@article{
 title = {Early stage detection of Downey and Powdery Mildew grape disease using atmospheric parameters through sensor nodes},
 type = {article},
 year = {2021},
 keywords = {Downey mildew,Grape diseases,Internet of things (IoT),Powdery mildew,Sensor},
 pages = {223-232},
 volume = {5},
 websites = {https://doi.org/10.1016/j.aiia.2021.10.001},
 publisher = {The Authors},
 id = {34a7afd9-3efb-30e4-962c-7f71d42f3804},
 created = {2022-09-21T06:54:51.201Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-12-12T12:51:04.481Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {Grape diseases are major factors causing severe diminution in its fruit development. Unfavorable climatic conditions are one of the principal dangers for grape disease development. Downy Mildew, Powdery Mildew, Anthracnose, Stem borer, Black Rot, Leaf Blight are widespread grape leaf vermin and diseases, which cause stern monetary losses to the grape industry. Devices ready to quantify the climate conditions in real-time for disease onset are hence crucial to perform timely diagnosis and precise detection of grape leaf diseases. This will ensure the healthy growth of grape plants, further controlling the spread of diseases. This paper discusses the requirements for building a consistent grape disease detection framework that would encourage headways in agribusiness. The primary aim of this work is to adapt an Internet of Things (IoT) based approach to predict the occurrence of Downey and Powdery Mildew grape diseases at an early stage. The sensor values received are transmitted to the Central Server with the help of the IoT device NodeMCU. At the server side, an analysis is made based on weather conditions. Further notification to the farmer is sent if weather properties are conducive for disease onset. The exclusivity of the system lies in using a rain gauge sensor along with the temperature sensor to predict the occurrence of grape diseases. This system realizes an overall accuracy of 94.4% for Downey Mildew and 96% for Powdery Mildew. Experimental results suggest the projected model can proficiently recognize Downey and Powdery Mildew grape diseases.},
 bibtype = {article},
 author = {Sanghavi, Kainjan and Sanghavi, Mahesh and Rajurkar, Archana M.},
 doi = {10.1016/j.aiia.2021.10.001},
 journal = {Artificial Intelligence in Agriculture}
}

Grape diseases are major factors causing severe diminution in its fruit development. Unfavorable climatic conditions are one of the principal dangers for grape disease development. Downy Mildew, Powdery Mildew, Anthracnose, Stem borer, Black Rot, Leaf Blight are widespread grape leaf vermin and diseases, which cause stern monetary losses to the grape industry. Devices ready to quantify the climate conditions in real-time for disease onset are hence crucial to perform timely diagnosis and precise detection of grape leaf diseases. This will ensure the healthy growth of grape plants, further controlling the spread of diseases. This paper discusses the requirements for building a consistent grape disease detection framework that would encourage headways in agribusiness. The primary aim of this work is to adapt an Internet of Things (IoT) based approach to predict the occurrence of Downey and Powdery Mildew grape diseases at an early stage. The sensor values received are transmitted to the Central Server with the help of the IoT device NodeMCU. At the server side, an analysis is made based on weather conditions. Further notification to the farmer is sent if weather properties are conducive for disease onset. The exclusivity of the system lies in using a rain gauge sensor along with the temperature sensor to predict the occurrence of grape diseases. This system realizes an overall accuracy of 94.4% for Downey Mildew and 96% for Powdery Mildew. Experimental results suggest the projected model can proficiently recognize Downey and Powdery Mildew grape diseases.

@article{
 title = {Potential Phenotyping Methodologies to Assess Inter- and Intravarietal Variability and to Select Grapevine Genotypes Tolerant to Abiotic Stress},
 type = {article},
 year = {2021},
 keywords = {carvalho2021potentialphenotypingmethodologies},
 volume = {12},
 id = {e54d4cc1-f945-3d54-b4cd-a98b687031d3},
 created = {2022-09-27T07:58:17.431Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T08:10:20.211Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Plant phenotyping is an emerging science that combines multiple methodologies and protocols to measure plant traits (e.g., growth, morphology, architecture, function, and composition) at multiple scales of organization. Manual phenotyping remains as a major bottleneck to the advance of plant and crop breeding. Such constraint fostered the development of high throughput plant phenotyping (HTPP), which is largely based on imaging approaches and automatized data retrieval and processing. Field phenotyping still poses major challenges and the progress of HTPP for field conditions can be relevant to support selection and breeding of grapevine. The aim of this review is to discuss potential and current methods to improve field phenotyping of grapevine to support characterization of inter- and intravarietal diversity. Vitis vinifera has a large genetic diversity that needs characterization, and the availability of methods to support selection of plant material (polyclonal or clonal) able to withstand abiotic stress is paramount. Besides being time consuming, complex and expensive, field experiments are also affected by heterogeneous and uncontrolled climate and soil conditions, mostly due to the large areas of the trials and to the high number of traits to be observed in a number of individuals ranging from hundreds to thousands. Therefore, adequate field experimental design and data gathering methodologies are crucial to obtain reliable data. Some of the major challenges posed to grapevine selection programs for tolerance to water and heat stress are described herein. Useful traits for selection and related field phenotyping methodologies are described and their adequacy for large scale screening is discussed.},
 bibtype = {article},
 author = {Carvalho, Luísa C. and Gonçalves, Elsa F. and Marques da Silva, Jorge and Costa, J. Miguel},
 doi = {10.3389/fpls.2021.718202},
 journal = {Frontiers in Plant Science},
 number = {October}
}

Plant phenotyping is an emerging science that combines multiple methodologies and protocols to measure plant traits (e.g., growth, morphology, architecture, function, and composition) at multiple scales of organization. Manual phenotyping remains as a major bottleneck to the advance of plant and crop breeding. Such constraint fostered the development of high throughput plant phenotyping (HTPP), which is largely based on imaging approaches and automatized data retrieval and processing. Field phenotyping still poses major challenges and the progress of HTPP for field conditions can be relevant to support selection and breeding of grapevine. The aim of this review is to discuss potential and current methods to improve field phenotyping of grapevine to support characterization of inter- and intravarietal diversity. Vitis vinifera has a large genetic diversity that needs characterization, and the availability of methods to support selection of plant material (polyclonal or clonal) able to withstand abiotic stress is paramount. Besides being time consuming, complex and expensive, field experiments are also affected by heterogeneous and uncontrolled climate and soil conditions, mostly due to the large areas of the trials and to the high number of traits to be observed in a number of individuals ranging from hundreds to thousands. Therefore, adequate field experimental design and data gathering methodologies are crucial to obtain reliable data. Some of the major challenges posed to grapevine selection programs for tolerance to water and heat stress are described herein. Useful traits for selection and related field phenotyping methodologies are described and their adequacy for large scale screening is discussed.

@inproceedings{
 title = {Embedded Machine Learning for the implementation of Autonomous Mobile Sensor Nodes (AMSNs)},
 type = {inproceedings},
 year = {2021},
 pages = {4-4},
 month = {12},
 publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
 day = {6},
 id = {bece71ec-0a80-33fe-bd45-7c19de366a03},
 created = {2022-09-29T07:47:52.731Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-12T10:14:07.034Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6aec054e-977e-4d9f-8f01-b49e735da52b,9e28b3ab-dc65-4cb8-8373-439f2758a335},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Di Nunzio, Luca},
 doi = {10.23919/eecsi53397.2021.9624250}
}

@inproceedings{
 title = {Embedded System for Eye Blink Detection Using Machine Learning Technique},
 type = {inproceedings},
 year = {2021},
 keywords = {Embedded system,Eye aspect,Eye blink detection,Face detection,Facial landmark detector,Machine learning,Raspberry pi 3},
 pages = {58-62},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {4e9e8810-e772-3658-b15b-b7a14875fe93},
 created = {2022-09-29T07:48:01.209Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-12T10:14:06.937Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 notes = {The &quot;Machine Learning&quot; part from this paper is basically just by using a library called &quot;dlib&quot; [8]-[10].<br/><br/>Did not bring anything new, or relevant.},
 folder_uuids = {9f2f513d-a81c-4e69-a1c5-34e938a643c8,6aec054e-977e-4d9f-8f01-b49e735da52b},
 private_publication = {false},
 abstract = {Nowadays, eye tracking and blink detection are increasingly popular among researchers and have the potential to become a more important component of future perceptual user interfaces. The real-time eye-tracking system has been a fundamental and challenging problem for machine learning problems. The main purpose of this paper is to propose a new method to design an embedded eye blink detection system that can be used for various applications with the lowest cost. This study presents an efficient technique to determine the level of eyes that are closed and opened. We offered a real-time blink detection method by using machine learning and computer vision libraries. The proposed method consists of four phases: (1) taking frame by employing a raspberry pi camera that slotted to the raspberry pi 3 platform, (2) utilizing haar cascade algorithm to identify faces in the captured frames, (3) find facial landmarks by utilizing facial landmark detector algorithm, (4) detect the eyes' region and calculate the eye aspect ratio. The proposed method obtained a high accuracy to indicate eye closing or opening. In this study, an aspect ratio method was used to implement a robust and low-cost embedded eye blink detection system on the raspberry pi platform. This method is exact resourceful, fast, and easy to perform eye blink detection.},
 bibtype = {inproceedings},
 author = {Ibrahim, Bishar R. and Khalifa, Farhad M. and Zeebaree, Subhi R.M. and Othman, Nashwan A. and Alkhayyat, Ahmed and Zebari, Rizgar R. and Sadeeq, Mohammed A.M.},
 doi = {10.1109/BICITS51482.2021.9509908},
 booktitle = {1st Babylon International Conference on Information Technology and Science 2021, BICITS 2021}
}

Nowadays, eye tracking and blink detection are increasingly popular among researchers and have the potential to become a more important component of future perceptual user interfaces. The real-time eye-tracking system has been a fundamental and challenging problem for machine learning problems. The main purpose of this paper is to propose a new method to design an embedded eye blink detection system that can be used for various applications with the lowest cost. This study presents an efficient technique to determine the level of eyes that are closed and opened. We offered a real-time blink detection method by using machine learning and computer vision libraries. The proposed method consists of four phases: (1) taking frame by employing a raspberry pi camera that slotted to the raspberry pi 3 platform, (2) utilizing haar cascade algorithm to identify faces in the captured frames, (3) find facial landmarks by utilizing facial landmark detector algorithm, (4) detect the eyes' region and calculate the eye aspect ratio. The proposed method obtained a high accuracy to indicate eye closing or opening. In this study, an aspect ratio method was used to implement a robust and low-cost embedded eye blink detection system on the raspberry pi platform. This method is exact resourceful, fast, and easy to perform eye blink detection.

@inproceedings{
 title = {Pedestrian Detection System with Edge Computing Integration on Embedded Vehicle},
 type = {inproceedings},
 year = {2021},
 keywords = {detection,edge computing,embedded vehicle},
 pages = {450-453},
 month = {4},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 day = {13},
 id = {f9ed34b6-0226-3d5f-a5a6-d0e7be2e2126},
 created = {2022-09-29T07:48:10.319Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-12T10:14:06.888Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6aec054e-977e-4d9f-8f01-b49e735da52b,9e28b3ab-dc65-4cb8-8373-439f2758a335},
 private_publication = {false},
 abstract = {The article proposes pedestrian detection system with edge computing with multi-network integration on embedded vehicle. When camera of lens design in machine learning, the proposal design uses AdaBoost, support vector machine (SVM) and convolutional neural network (CNN). The disadvantage is that a large number of samples are needed for training, and the amount of operation and the large number of parameters cannot be used in the embedded system for vehicles. This article proposes to reduce the amount of computation and the number of parameters required by the network by integrating different optimization operations between networks of different architectures, so as to achieve prediction by using the Renesas R-car H3 of the embedded system on vehicle. The proposed design can maintain above a certain accuracy and cost lower than camera of lens with sensors of radar and lidar.},
 bibtype = {inproceedings},
 author = {Su, Ching Lung and Lai, Wen Cheng and Li, Chun Te},
 doi = {10.1109/ICAIIC51459.2021.9415262},
 booktitle = {3rd International Conference on Artificial Intelligence in Information and Communication, ICAIIC 2021}
}

The article proposes pedestrian detection system with edge computing with multi-network integration on embedded vehicle. When camera of lens design in machine learning, the proposal design uses AdaBoost, support vector machine (SVM) and convolutional neural network (CNN). The disadvantage is that a large number of samples are needed for training, and the amount of operation and the large number of parameters cannot be used in the embedded system for vehicles. This article proposes to reduce the amount of computation and the number of parameters required by the network by integrating different optimization operations between networks of different architectures, so as to achieve prediction by using the Renesas R-car H3 of the embedded system on vehicle. The proposed design can maintain above a certain accuracy and cost lower than camera of lens with sensors of radar and lidar.

@article{
 title = {Run-Time Monitoring of Machine Learning for Robotic Perception: A Survey of Emerging Trends},
 type = {article},
 year = {2021},
 keywords = {Machine learning,performance evaluation,reliability,robot learning},
 pages = {20067-20075},
 volume = {9},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {a0a0bb3f-565c-3708-be7b-df5d67852407},
 created = {2022-09-29T07:48:23.953Z},
 file_attached = {true},
 profile_id = {c3c41a69-4b45-352f-9232-4d3281e18730},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-12T10:14:06.784Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {6aec054e-977e-4d9f-8f01-b49e735da52b,9e28b3ab-dc65-4cb8-8373-439f2758a335},
 private_publication = {false},
 abstract = {As deep learning continues to dominate all state-of-the-art computer vision tasks, it is increasingly becoming an essential building block for robotic perception. This raises important questions concerning the safety and reliability of learning-based perception systems. There is an established field that studies safety certification and convergence guarantees of complex software systems at design-time. However, the unknown future deployment environments of an autonomous system and the complexity of learning-based perception make the generalization of design-time verification to run-time problematic. In the face of this challenge, more attention is starting to focus on run-time monitoring of performance and reliability of perception systems with several trends emerging in the literature in the face of this challenge. This paper attempts to identify these trends and summarize the various approaches to the topic.},
 bibtype = {article},
 author = {Rahman, Quazi Marufur and Corke, Peter and Dayoub, Feras},
 doi = {10.1109/ACCESS.2021.3055015},
 journal = {IEEE Access}
}

As deep learning continues to dominate all state-of-the-art computer vision tasks, it is increasingly becoming an essential building block for robotic perception. This raises important questions concerning the safety and reliability of learning-based perception systems. There is an established field that studies safety certification and convergence guarantees of complex software systems at design-time. However, the unknown future deployment environments of an autonomous system and the complexity of learning-based perception make the generalization of design-time verification to run-time problematic. In the face of this challenge, more attention is starting to focus on run-time monitoring of performance and reliability of perception systems with several trends emerging in the literature in the face of this challenge. This paper attempts to identify these trends and summarize the various approaches to the topic.

@article{
 title = {Patch2Pix: Epipolar-Guided Pixel-Level Correspondences},
 type = {article},
 year = {2021},
 pages = {4667-4676},
 id = {98a932ee-257f-37a4-bfd9-90f691fe7887},
 created = {2022-10-04T09:36:42.744Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-10-04T09:36:53.832Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {02fb5526-03ff-44ad-8d5c-42bd496c3100},
 private_publication = {false},
 abstract = {The classical matching pipeline used for visual localization typically involves three steps: (i) local feature detection and description, (ii) feature matching, and (iii) outlier rejection. Recently emerged correspondence networks propose to perform those steps inside a single network but suffer from low matching resolution due to the memory bottleneck. In this work, we propose a new perspective to estimate correspondences in a detect-to-refine manner, where we first predict patch-level match proposals and then refine them. We present Patch2Pix, a novel refinement network that refines match proposals by regressing pixel-level matches from the local regions defined by those proposals and jointly rejecting outlier matches with confidence scores. Patch2Pix is weakly supervised to learn correspondences that are consistent with the epipolar geometry of an input image pair. We show that our refinement network significantly improves the performance of correspondence networks on image matching, homography estimation, and localization tasks. In addition, we show that our learned refinement generalizes to fully-supervised methods without re-training, which leads us to state-of-the-art localization performance. The code is available at https://github.com/GrumpyZhou/patch2pix.},
 bibtype = {article},
 author = {Zhou, Qunjie and Sattler, Torsten and Leal-Taixé, Laura},
 doi = {10.1109/CVPR46437.2021.00464},
 journal = {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition}
}

The classical matching pipeline used for visual localization typically involves three steps: (i) local feature detection and description, (ii) feature matching, and (iii) outlier rejection. Recently emerged correspondence networks propose to perform those steps inside a single network but suffer from low matching resolution due to the memory bottleneck. In this work, we propose a new perspective to estimate correspondences in a detect-to-refine manner, where we first predict patch-level match proposals and then refine them. We present Patch2Pix, a novel refinement network that refines match proposals by regressing pixel-level matches from the local regions defined by those proposals and jointly rejecting outlier matches with confidence scores. Patch2Pix is weakly supervised to learn correspondences that are consistent with the epipolar geometry of an input image pair. We show that our refinement network significantly improves the performance of correspondence networks on image matching, homography estimation, and localization tasks. In addition, we show that our learned refinement generalizes to fully-supervised methods without re-training, which leads us to state-of-the-art localization performance. The code is available at https://github.com/GrumpyZhou/patch2pix.

@article{
 title = {Back to the Feature: Learning Robust Camera Localization from Pixels to Pose},
 type = {article},
 year = {2021},
 pages = {3246-3256},
 id = {e5c899c8-b739-3e5d-8a46-62bf846b2396},
 created = {2022-10-04T09:36:42.752Z},
 file_attached = {true},
 profile_id = {bfbbf840-4c42-3914-a463-19024f50b30c},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-10-04T09:36:56.564Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {02fb5526-03ff-44ad-8d5c-42bd496c3100},
 private_publication = {false},
 abstract = {Camera pose estimation in known scenes is a 3D geometry task recently tackled by multiple learning algorithms. Many regress precise geometric quantities, like poses or 3D points, from an input image. This either fails to generalize to new viewpoints or ties the model parameters to a specific scene. In this paper, we go Back to the Feature: we argue that deep networks should focus on learning robust and invariant visual features, while the geometric estimation should be left to principled algorithms. We introduce PixLoc, a scene-agnostic neural network that estimates an accurate 6-DoF pose from an image and a 3D model. Our approach is based on the direct alignment of multiscale deep features, casting camera localization as metric learning. PixLoc learns strong data priors by end-to-end training from pixels to pose and exhibits exceptional generalization to new scenes by separating model parameters and scene geometry. The system can localize in large environments given coarse pose priors but also improve the accuracy of sparse feature matching by jointly refining keypoints and poses with little overhead. The code will be publicly available at github.com/cvg/pixloc.},
 bibtype = {article},
 author = {Sarlin, Paul Edouard and Unagar, Ajaykumar and Larsson, Måns and Germain, Hugo and Toft, Carl and Larsson, Viktor and Pollefeys, Marc and Lepetit, Vincent and Hammarstrand, Lars and Kahl, Fredrik and Sattler, Torsten},
 doi = {10.1109/CVPR46437.2021.00326},
 journal = {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition}
}

Camera pose estimation in known scenes is a 3D geometry task recently tackled by multiple learning algorithms. Many regress precise geometric quantities, like poses or 3D points, from an input image. This either fails to generalize to new viewpoints or ties the model parameters to a specific scene. In this paper, we go Back to the Feature: we argue that deep networks should focus on learning robust and invariant visual features, while the geometric estimation should be left to principled algorithms. We introduce PixLoc, a scene-agnostic neural network that estimates an accurate 6-DoF pose from an image and a 3D model. Our approach is based on the direct alignment of multiscale deep features, casting camera localization as metric learning. PixLoc learns strong data priors by end-to-end training from pixels to pose and exhibits exceptional generalization to new scenes by separating model parameters and scene geometry. The system can localize in large environments given coarse pose priors but also improve the accuracy of sparse feature matching by jointly refining keypoints and poses with little overhead. The code will be publicly available at github.com/cvg/pixloc.

@misc{
 title = {Vineyard yield estimation, prediction, and forecasting: A systematic literature review},
 type = {misc},
 year = {2021},
 source = {Agronomy},
 keywords = {barriguinha2021vineyardyieldestimation},
 volume = {11},
 issue = {9},
 month = {9},
 publisher = {MDPI},
 day = {1},
 id = {7c0a51c0-1ad8-392b-b4e2-1bff72def199},
 created = {2022-10-07T07:43:21.272Z},
 file_attached = {true},
 profile_id = {c99d2359-7f75-34c0-a217-662c4e9873ab},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:13.749Z},
 read = {true},
 starred = {true},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1619600c-2adf-4216-9e4c-d260d584753e,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Purpose—knowing in advance vineyard yield is a critical success factor so growers and winemakers can achieve the best balance between vegetative and reproductive growth. It is also essential for planning and regulatory purposes at the regional level. Estimation errors are mainly due to the high inter-annual and spatial variability and inadequate or poor performance sampling methods; therefore, improved applied methodologies are needed at different spatial scales. This paper aims to identify the alternatives to traditional estimation methods. Design/methodology/approach—this study consists of a systematic literature review of academic articles indexed on four databases collected based on multiple query strings conducted on title, abstract, and keywords. The articles were reviewed based on the research topic, methodology, data requirements, practical application, and scale using PRISMA as a guideline. Findings—the methodological approaches for yield estimation based on indirect methods are primarily applicable at a small scale and can provide better estimates than the traditional manual sampling. Nevertheless, most of these approaches are still in the research domain and lack practical applicability in real vineyards by the actual farmers. They mainly depend on computer vision and image processing algorithms, data-driven models based on vegetation indices and pollen data, and on relating climate, soil, vegetation, and crop management variables that can support dynamic crop simulation models. Research limitations—this work is based on academic articles published before June 2021. Therefore, scientific outputs published after this date are not included. Originality/value—this study contributes to perceiving the approaches for estimating vineyard yield and identifying research gaps for future developments, and supporting a future research agenda on this topic. To the best of the authors’ knowledge, it is the first systematic literature review fully dedicated to vineyard yield estimation, prediction, and forecasting methods.},
 bibtype = {misc},
 author = {Barriguinha, André and Neto, Miguel de Castro and Gil, Artur},
 doi = {10.3390/agronomy11091789}
}

Purpose—knowing in advance vineyard yield is a critical success factor so growers and winemakers can achieve the best balance between vegetative and reproductive growth. It is also essential for planning and regulatory purposes at the regional level. Estimation errors are mainly due to the high inter-annual and spatial variability and inadequate or poor performance sampling methods; therefore, improved applied methodologies are needed at different spatial scales. This paper aims to identify the alternatives to traditional estimation methods. Design/methodology/approach—this study consists of a systematic literature review of academic articles indexed on four databases collected based on multiple query strings conducted on title, abstract, and keywords. The articles were reviewed based on the research topic, methodology, data requirements, practical application, and scale using PRISMA as a guideline. Findings—the methodological approaches for yield estimation based on indirect methods are primarily applicable at a small scale and can provide better estimates than the traditional manual sampling. Nevertheless, most of these approaches are still in the research domain and lack practical applicability in real vineyards by the actual farmers. They mainly depend on computer vision and image processing algorithms, data-driven models based on vegetation indices and pollen data, and on relating climate, soil, vegetation, and crop management variables that can support dynamic crop simulation models. Research limitations—this work is based on academic articles published before June 2021. Therefore, scientific outputs published after this date are not included. Originality/value—this study contributes to perceiving the approaches for estimating vineyard yield and identifying research gaps for future developments, and supporting a future research agenda on this topic. To the best of the authors’ knowledge, it is the first systematic literature review fully dedicated to vineyard yield estimation, prediction, and forecasting methods.

@article{
 title = {Towards Vine Water Status Monitoring on a Large Scale Using Sentinel-2 Images},
 type = {article},
 year = {2021},
 keywords = {larochepinel2021towardsvinewater},
 pages = {1837},
 volume = {13},
 websites = {https://www.mdpi.com/2072-4292/13/9/1837/htm,https://www.mdpi.com/2072-4292/13/9/1837},
 month = {5},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {9},
 id = {ae3c742c-ff04-3a0c-af78-7119e195e639},
 created = {2022-10-14T09:58:44.558Z},
 accessed = {2022-10-14},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-02T07:03:56.250Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Wine growing needs to adapt to confront climate change. In fact, the lack of water becomes more and more important in many regions. Whereas vineyards have been located in dry areas for decades, so they need special resilient varieties and/or a sufficient water supply at key development stages in case of severe drought. With climate change and the decrease of water availability, some vineyard regions face difficulties because of unsuitable variety, wrong vine management or due to the limited water access. Decision support tools are therefore required to optimize water use or to adapt agronomic practices. This study aimed at monitoring vine water status at a large scale with Sentinel-2 images. The goal was to provide a solution that would give spatialized and temporal information throughout the season on the water status of the vines. For this purpose, thirty six plots were monitored in total over three years (2018, 2019 and 2020). Vine water status was measured with stem water potential in field measurements from pea size to ripening stage. Simultaneously Sentinel-2 images were downloaded and processed to extract band reflectance values and compute vegetation indices. In our study, we tested five supervised regression machine learning algorithms to find possible relationships between stem water potential and data acquired from Sentinel-2 images (bands reflectance values and vegetation indices). Regression model using Red, NIR, Red-Edge and SWIR bands gave promising result to predict stem water potential (R2=0.40, RMSE=0.26).},
 bibtype = {article},
 author = {Laroche-Pinel, Eve and Duthoit, Sylvie and Albughdadi, Mohanad and Costard, Anne D. and Rousseau, Jacques and Chéret, Véronique and Clenet, Harold},
 doi = {10.3390/RS13091837},
 journal = {Remote Sensing},
 number = {9}
}

Wine growing needs to adapt to confront climate change. In fact, the lack of water becomes more and more important in many regions. Whereas vineyards have been located in dry areas for decades, so they need special resilient varieties and/or a sufficient water supply at key development stages in case of severe drought. With climate change and the decrease of water availability, some vineyard regions face difficulties because of unsuitable variety, wrong vine management or due to the limited water access. Decision support tools are therefore required to optimize water use or to adapt agronomic practices. This study aimed at monitoring vine water status at a large scale with Sentinel-2 images. The goal was to provide a solution that would give spatialized and temporal information throughout the season on the water status of the vines. For this purpose, thirty six plots were monitored in total over three years (2018, 2019 and 2020). Vine water status was measured with stem water potential in field measurements from pea size to ripening stage. Simultaneously Sentinel-2 images were downloaded and processed to extract band reflectance values and compute vegetation indices. In our study, we tested five supervised regression machine learning algorithms to find possible relationships between stem water potential and data acquired from Sentinel-2 images (bands reflectance values and vegetation indices). Regression model using Red, NIR, Red-Edge and SWIR bands gave promising result to predict stem water potential (R2=0.40, RMSE=0.26).

@inproceedings{
 title = {Entropy Maximization and Meta Classification for Out-of-Distribution Detection in Semantic Segmentation},
 type = {inproceedings},
 year = {2021},
 pages = {5128-5137},
 id = {aa092321-6ee2-3e50-973f-f418bdcca211},
 created = {2022-10-27T10:48:00.988Z},
 accessed = {2022-10-27},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-11-15T06:28:33.650Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Deep neural networks (DNNs) for the semantic segmen-tation of images are usually trained to operate on a pre-defined closed set of object classes. This is in contrast to the "open world" setting where DNNs are envisioned to be deployed to. From a functional safety point of view, the ability to detect so-called "out-of-distribution" (OoD) samples, i.e., objects outside of a DNN's semantic space, is crucial for many applications such as automated driving. A natural baseline approach to OoD detection is to threshold on the pixel-wise softmax entropy. We present a two-step procedure that significantly improves that approach. Firstly, we utilize samples from the COCO dataset as OoD proxy and introduce a second training objective to maximize the softmax entropy on these samples. Starting from pretrained semantic segmentation networks we retrain a number of DNNs on different in-distribution datasets and consistently observe improved OoD detection performance when evaluating on completely disjoint OoD datasets. Secondly, we perform a transparent post-processing step to discard false positive OoD samples by so-called "meta classification." To this end, we apply linear models to a set of hand-crafted metrics derived from the DNN's softmax probabilities. In our experiments we consistently observe a clear additional gain in OoD detection performance, cutting down the number of detection errors by 52% when comparing the best baseline with our results. We achieve this improvement sacrificing only marginally in original segmentation performance. Therefore, our method contributes to safer DNNs with more reliable overall system performance.},
 bibtype = {inproceedings},
 author = {Chan, Robin and Rottmann, Matthias and Gottschalk, Hanno},
 booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)}
}

Deep neural networks (DNNs) for the semantic segmen-tation of images are usually trained to operate on a pre-defined closed set of object classes. This is in contrast to the "open world" setting where DNNs are envisioned to be deployed to. From a functional safety point of view, the ability to detect so-called "out-of-distribution" (OoD) samples, i.e., objects outside of a DNN's semantic space, is crucial for many applications such as automated driving. A natural baseline approach to OoD detection is to threshold on the pixel-wise softmax entropy. We present a two-step procedure that significantly improves that approach. Firstly, we utilize samples from the COCO dataset as OoD proxy and introduce a second training objective to maximize the softmax entropy on these samples. Starting from pretrained semantic segmentation networks we retrain a number of DNNs on different in-distribution datasets and consistently observe improved OoD detection performance when evaluating on completely disjoint OoD datasets. Secondly, we perform a transparent post-processing step to discard false positive OoD samples by so-called "meta classification." To this end, we apply linear models to a set of hand-crafted metrics derived from the DNN's softmax probabilities. In our experiments we consistently observe a clear additional gain in OoD detection performance, cutting down the number of detection errors by 52% when comparing the best baseline with our results. We achieve this improvement sacrificing only marginally in original segmentation performance. Therefore, our method contributes to safer DNNs with more reliable overall system performance.

@article{
 title = {Generalized Out-of-Distribution Detection: A Survey},
 type = {article},
 year = {2021},
 keywords = {Anomaly Detection,Index Terms-Out-of-Distribution Detection,Novelty Detection,Open Set Recognition,Outlier Detection !},
 websites = {https://arxiv.org/abs/2110.11334v2},
 month = {10},
 day = {21},
 id = {617f283d-122c-333c-a93a-4466811de36b},
 created = {2022-12-05T06:54:02.213Z},
 accessed = {2022-12-05},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-04-03T10:17:18.652Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,4b11b59b-846d-45b5-82b7-81d8bc3108b2,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Out-of-distribution (OOD) detection is critical to ensuring the reliability
and safety of machine learning systems. For instance, in autonomous driving, we
would like the driving system to issue an alert and hand over the control to
humans when it detects unusual scenes or objects that it has never seen during
training time and cannot make a safe decision. The term, OOD detection, first
emerged in 2017 and since then has received increasing attention from the
research community, leading to a plethora of methods developed, ranging from
classification-based to density-based to distance-based ones. Meanwhile,
several other problems, including anomaly detection (AD), novelty detection
(ND), open set recognition (OSR), and outlier detection (OD), are closely
related to OOD detection in terms of motivation and methodology. Despite common
goals, these topics develop in isolation, and their subtle differences in
definition and problem setting often confuse readers and practitioners. In this
survey, we first present a unified framework called generalized OOD detection,
which encompasses the five aforementioned problems, i.e., AD, ND, OSR, OOD
detection, and OD. Under our framework, these five problems can be seen as
special cases or sub-tasks, and are easier to distinguish. We then review each
of these five areas by summarizing their recent technical developments, with a
special focus on OOD detection methodologies. We conclude this survey with open
challenges and potential research directions.},
 bibtype = {article},
 author = {Yang, Jingkang and Zhou, Kaiyang and Li, Yixuan and Liu, Ziwei},
 doi = {10.48550/arxiv.2110.11334}
}

Out-of-distribution (OOD) detection is critical to ensuring the reliability and safety of machine learning systems. For instance, in autonomous driving, we would like the driving system to issue an alert and hand over the control to humans when it detects unusual scenes or objects that it has never seen during training time and cannot make a safe decision. The term, OOD detection, first emerged in 2017 and since then has received increasing attention from the research community, leading to a plethora of methods developed, ranging from classification-based to density-based to distance-based ones. Meanwhile, several other problems, including anomaly detection (AD), novelty detection (ND), open set recognition (OSR), and outlier detection (OD), are closely related to OOD detection in terms of motivation and methodology. Despite common goals, these topics develop in isolation, and their subtle differences in definition and problem setting often confuse readers and practitioners. In this survey, we first present a unified framework called generalized OOD detection, which encompasses the five aforementioned problems, i.e., AD, ND, OSR, OOD detection, and OD. Under our framework, these five problems can be seen as special cases or sub-tasks, and are easier to distinguish. We then review each of these five areas by summarizing their recent technical developments, with a special focus on OOD detection methodologies. We conclude this survey with open challenges and potential research directions.

@article{
 title = {Estimation of leaf area index in vineyards by analysing projected shadows using uav imagery},
 type = {article},
 year = {2021},
 keywords = {velez2021estimationlai},
 pages = {159-180},
 volume = {55},
 id = {484ffd16-6b29-300e-980d-c6db6efc0f8b},
 created = {2022-12-12T12:51:03.800Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:09.254Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 folder_uuids = {4787d485-df78-4119-ad84-be82dd804b00,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {A few decades ago, farmers could precisely monitor their croplands just by walking over the fields, but this task becomes more difficult as farm size increases. Precision viticulture can help better understand the vineyard and measure some key structural parameters, such as the Leaf Area Index (LAI). Remote Sensing is a typical approach to monitoring vegetation which measures the spectral information directly emitted and reflected from vegetation. This study explores a new method for estimating LAI which measures the projected shadows of plants using UAV (unmanned aerial vehicle) imagery. A flight mission over a vineyard was scheduled in the afternoon (15:30 to 16:00 solar time), which is the optimal time for the projection of vine shadows on the ground. Real LAI was measured destructively by removing all the vegetation from the area. Then, the projected shadows in the image were detected using machine learning methods (k-means and random forest) and analysed at pixel level using a customised R code. A strong linear relationship (R² = 0.76, RMSE = 0.160 m² m-2 and MAE = 0.139 m² m-2) was found between the shaded area and the LAI per vine. This is a quick and simple method, which is non-destructive and gives accurate results; moreover, flights can be scheduled during other periods of the day than solar noon, such as in the morning or afternoon, thus enabling pilots to extend their working day. Therefore, it may be a viable option for determining LAI in vineyards trained on Vertical Shoot Positioned (VSP) systems.},
 bibtype = {article},
 author = {Vélez, Sergio and Poblete-Echeverría, Carlos and Rubio, José Antonio and Vacas, Rubén and Barajas, Enrique},
 doi = {10.20870/oeno-one.2021.55.4.4639},
 journal = {Oeno One},
 number = {4}
}

A few decades ago, farmers could precisely monitor their croplands just by walking over the fields, but this task becomes more difficult as farm size increases. Precision viticulture can help better understand the vineyard and measure some key structural parameters, such as the Leaf Area Index (LAI). Remote Sensing is a typical approach to monitoring vegetation which measures the spectral information directly emitted and reflected from vegetation. This study explores a new method for estimating LAI which measures the projected shadows of plants using UAV (unmanned aerial vehicle) imagery. A flight mission over a vineyard was scheduled in the afternoon (15:30 to 16:00 solar time), which is the optimal time for the projection of vine shadows on the ground. Real LAI was measured destructively by removing all the vegetation from the area. Then, the projected shadows in the image were detected using machine learning methods (k-means and random forest) and analysed at pixel level using a customised R code. A strong linear relationship (R² = 0.76, RMSE = 0.160 m² m-2 and MAE = 0.139 m² m-2) was found between the shaded area and the LAI per vine. This is a quick and simple method, which is non-destructive and gives accurate results; moreover, flights can be scheduled during other periods of the day than solar noon, such as in the morning or afternoon, thus enabling pilots to extend their working day. Therefore, it may be a viable option for determining LAI in vineyards trained on Vertical Shoot Positioned (VSP) systems.

@article{
 title = {Impact of Leaf Occlusions on Yield Assessment by Computer Vision in Commercial Vineyards},
 type = {article},
 year = {2021},
 keywords = {iniguez2021impactleafocclusion},
 pages = {1003},
 volume = {11},
 websites = {https://www.mdpi.com/2073-4395/11/5/1003/htm,https://www.mdpi.com/2073-4395/11/5/1003},
 month = {5},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {18},
 id = {5b091b25-7175-32a3-b9d7-03d4bba2fecf},
 created = {2022-12-13T12:12:50.287Z},
 accessed = {2022-12-13},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:11.364Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Yield assessment has been identified as critical topic for grape and wine industry. Computer vision has been applied for assessing yield, but the accuracy was greatly affected by fruit occlusion affected by leaves and other plant organs. The objective of this work was the consistent, continuous evaluation of the impact of leaf occlusions in different commercial vineyard plots at different defoliation stages. RGB (red, green and blue) images from five Tempranillo (Vitis vinifera L.) vineyards were manually acquired using a digital camera under field conditions at three different levels of defoliation: no defoliation, partial defoliation and full defoliation. Computer vision was used for the automatic detection of different canopy features, and for the calibration of regression equations for the prediction of yield computed per vine segment. Leaf occlusion rate (berry occlusion affected by leaves) was computed by machine vision in no defoliated vineyards. As occlusion rate increased, R2 between bunch pixels and yield was gradually reduced, ranging from 0.77 in low occlusion, to 0.63.},
 bibtype = {article},
 author = {Íñiguez, Rubén and Palacios, Fernando and Barrio, Ignacio and Hernández, Inés and Gutiérrez, Salvador and Tardaguila, Javier},
 doi = {10.3390/AGRONOMY11051003},
 journal = {Agronomy 2021, Vol. 11, Page 1003},
 number = {5}
}

Yield assessment has been identified as critical topic for grape and wine industry. Computer vision has been applied for assessing yield, but the accuracy was greatly affected by fruit occlusion affected by leaves and other plant organs. The objective of this work was the consistent, continuous evaluation of the impact of leaf occlusions in different commercial vineyard plots at different defoliation stages. RGB (red, green and blue) images from five Tempranillo (Vitis vinifera L.) vineyards were manually acquired using a digital camera under field conditions at three different levels of defoliation: no defoliation, partial defoliation and full defoliation. Computer vision was used for the automatic detection of different canopy features, and for the calibration of regression equations for the prediction of yield computed per vine segment. Leaf occlusion rate (berry occlusion affected by leaves) was computed by machine vision in no defoliated vineyards. As occlusion rate increased, R2 between bunch pixels and yield was gradually reduced, ranging from 0.77 in low occlusion, to 0.63.

@article{
 title = {Data-Driven Artificial Intelligence Applications for Sustainable Precision Agriculture},
 type = {article},
 year = {2021},
 keywords = {agriculture,artificial intelligence,computer vision,data analysis,robotics},
 pages = {1227},
 volume = {11},
 websites = {https://www.mdpi.com/2073-4395/11/6/1227/htm,https://www.mdpi.com/2073-4395/11/6/1227},
 month = {6},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {17},
 id = {ed16463d-cff1-3228-9175-2becf8280490},
 created = {2022-12-13T12:34:19.425Z},
 accessed = {2022-12-13},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-05T13:32:08.128Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {One of the main challenges for the implementation of artificial intelligence (AI) in agriculture includes the low replicability and the corresponding difficulty in systematic data gathering, as no two fields are exactly alike. Therefore, the comparison of several pilot experiments in different fields, weather conditions and farming techniques enhances the collective knowledge. Thus, this work provides a summary of the most recent research activities in the form of research projects implemented and validated by the authors in several European countries, with the objective of presenting the already achieved results, the current investigations and the still open technical challenges. As an overall conclusion, it can be mentioned that even though in their primary stages in some cases, AI technologies improve decision support at farm level, monitoring conditions and optimizing production to allow farmers to apply the optimal number of inputs for each crop, thereby boosting yields and reducing water use and greenhouse gas emissions. Future extensions of this work will include new concepts based on autonomous and intelligent robots for plant and soil sample retrieval, and effective livestock management.},
 bibtype = {article},
 author = {Linaza, Maria Teresa and Posada, Jorge and Bund, Jürgen and Eisert, Peter and Quartulli, Marco and Döllner, Jürgen and Pagani, Alain and Olaizola, Igor G. and Barriguinha, Andre and Moysiadis, Theocharis and Lucat, Laurent},
 doi = {10.3390/AGRONOMY11061227},
 journal = {Agronomy 2021, Vol. 11, Page 1227},
 number = {6}
}

One of the main challenges for the implementation of artificial intelligence (AI) in agriculture includes the low replicability and the corresponding difficulty in systematic data gathering, as no two fields are exactly alike. Therefore, the comparison of several pilot experiments in different fields, weather conditions and farming techniques enhances the collective knowledge. Thus, this work provides a summary of the most recent research activities in the form of research projects implemented and validated by the authors in several European countries, with the objective of presenting the already achieved results, the current investigations and the still open technical challenges. As an overall conclusion, it can be mentioned that even though in their primary stages in some cases, AI technologies improve decision support at farm level, monitoring conditions and optimizing production to allow farmers to apply the optimal number of inputs for each crop, thereby boosting yields and reducing water use and greenhouse gas emissions. Future extensions of this work will include new concepts based on autonomous and intelligent robots for plant and soil sample retrieval, and effective livestock management.

@article{
 title = {Beyond the traditional NDVI index as a key factor to mainstream the use of UAV in precision viticulture},
 type = {article},
 year = {2021},
 keywords = {matese2021beyondtraditinoalndvi},
 pages = {1-13},
 volume = {11},
 websites = {https://www.nature.com/articles/s41598-021-81652-3},
 month = {2},
 publisher = {Nature Publishing Group},
 day = {1},
 id = {886bd9da-f0c9-3f67-bf85-8f603d13cb6a},
 created = {2022-12-13T12:44:35.218Z},
 accessed = {2022-12-13},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-11-01T07:12:57.785Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {In the last decade there has been an exponential growth of research activity on the identification of correlations between vegetational indices elaborated by UAV imagery and productive and vegetative parameters of the vine. However, the acquisition and analysis of spectral data require costs and skills that are often not sufficiently available. In this context, the identification of geometric indices that allow the monitoring of spatial variability with low-cost instruments, without spectral analysis know-how but based on photogrammetry techniques with high-resolution RGB cameras, becomes extremely interesting. The aim of this work was to evaluate the potential of new canopy geometry-based indices for the characterization of vegetative and productive agronomic parameters compared to traditional NDVI based on spectral response of the canopy top. Furthermore, considering grape production as a key parameter directly linked to the economic profit of farmers, this study provides a deeper analysis focused on the development of a rapid yield forecast methodology based on UAV data, evaluating both traditional linear and machine learning regressions. Among the yield assessment models, one of the best results was obtained with the canopy thickness which showed high performance with the Gaussian process regression models (R2 = 0.80), while the yield prediction average accuracy of the best ML models reached 85.95%. The final results obtained confirm the feasibility of this research as a global yield model, which provided good performance through an accurate validation step realized in different years and different vineyards.},
 bibtype = {article},
 author = {Matese, Alessandro and Di Gennaro, Salvatore Filippo},
 doi = {10.1038/s41598-021-81652-3},
 journal = {Scientific Reports 2021 11:1},
 number = {1}
}

In the last decade there has been an exponential growth of research activity on the identification of correlations between vegetational indices elaborated by UAV imagery and productive and vegetative parameters of the vine. However, the acquisition and analysis of spectral data require costs and skills that are often not sufficiently available. In this context, the identification of geometric indices that allow the monitoring of spatial variability with low-cost instruments, without spectral analysis know-how but based on photogrammetry techniques with high-resolution RGB cameras, becomes extremely interesting. The aim of this work was to evaluate the potential of new canopy geometry-based indices for the characterization of vegetative and productive agronomic parameters compared to traditional NDVI based on spectral response of the canopy top. Furthermore, considering grape production as a key parameter directly linked to the economic profit of farmers, this study provides a deeper analysis focused on the development of a rapid yield forecast methodology based on UAV data, evaluating both traditional linear and machine learning regressions. Among the yield assessment models, one of the best results was obtained with the canopy thickness which showed high performance with the Gaussian process regression models (R2 = 0.80), while the yield prediction average accuracy of the best ML models reached 85.95%. The final results obtained confirm the feasibility of this research as a global yield model, which provided good performance through an accurate validation step realized in different years and different vineyards.

@article{
 title = {Grape Cluster Detection Using UAV Photogrammetric Point Clouds as a Low-Cost Tool for Yield Forecasting in Vineyards},
 type = {article},
 year = {2021},
 keywords = {torressanchez2021grapeclusterdetection},
 pages = {3083},
 volume = {21},
 websites = {https://www.mdpi.com/1424-8220/21/9/3083/htm,https://www.mdpi.com/1424-8220/21/9/3083},
 month = {4},
 publisher = {Multidisciplinary Digital Publishing Institute},
 day = {28},
 id = {3742441f-762f-3c1c-b77f-30c1b68a1afa},
 created = {2022-12-13T13:07:52.672Z},
 accessed = {2022-12-13},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-10-31T13:21:11.601Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Yield prediction is crucial for the management of harvest and scheduling wine production operations. Traditional yield prediction methods rely on manual sampling and are time-consuming, making it difficult to handle the intrinsic spatial variability of vineyards. There have been significant advances in automatic yield estimation in vineyards from on-ground imagery, but terrestrial platforms have some limitations since they can cause soil compaction and have problems on sloping and ploughed land. The analysis of photogrammetric point clouds generated with unmanned aerial vehicles (UAV) imagery has shown its potential in the characterization of woody crops, and the point color analysis has been used for the detection of flowers in almond trees. For these reasons, the main objective of this work was to develop an unsupervised and automated workflow for detection of grape clusters in red grapevine varieties using UAV photogrammetric point clouds and color indices. As leaf occlusion is recognized as a major challenge in fruit detection, the influence of partial leaf removal in the accuracy of the workflow was assessed. UAV flights were performed over two commercial vineyards with different grape varieties in 2019 and 2020, and the photogrammetric point clouds generated from these flights were analyzed using an automatic and unsupervised algorithm developed using free software. The proposed methodology achieved R2 values higher than 0.75 between the harvest weight and the projected area of the points classified as grapes in vines when partial two-sided removal treatment, and an R2 of 0.82 was achieved in one of the datasets for vines with untouched full canopy. The accuracy achieved in grape detection opens the door to yield prediction in red grape vineyards. This would allow the creation of yield estimation maps that will ease the implementation of precision viticulture practices. To the authors’ knowledge, this is the first time that UAV photogrammetric point clouds have been used for grape clusters detection.},
 bibtype = {article},
 author = {Torres‐sánchez, Jorge and Mesas‐carrascosa, Francisco Javier and Santesteban, Luis Gonzaga and Jiménez‐brenes, Francisco Manuel and Oneka, Oihane and Villa‐llop, Ana and Loidi, Maite and López‐granados, Francisca},
 doi = {10.3390/S21093083},
 journal = {Sensors 2021, Vol. 21, Page 3083},
 number = {9}
}

Yield prediction is crucial for the management of harvest and scheduling wine production operations. Traditional yield prediction methods rely on manual sampling and are time-consuming, making it difficult to handle the intrinsic spatial variability of vineyards. There have been significant advances in automatic yield estimation in vineyards from on-ground imagery, but terrestrial platforms have some limitations since they can cause soil compaction and have problems on sloping and ploughed land. The analysis of photogrammetric point clouds generated with unmanned aerial vehicles (UAV) imagery has shown its potential in the characterization of woody crops, and the point color analysis has been used for the detection of flowers in almond trees. For these reasons, the main objective of this work was to develop an unsupervised and automated workflow for detection of grape clusters in red grapevine varieties using UAV photogrammetric point clouds and color indices. As leaf occlusion is recognized as a major challenge in fruit detection, the influence of partial leaf removal in the accuracy of the workflow was assessed. UAV flights were performed over two commercial vineyards with different grape varieties in 2019 and 2020, and the photogrammetric point clouds generated from these flights were analyzed using an automatic and unsupervised algorithm developed using free software. The proposed methodology achieved R2 values higher than 0.75 between the harvest weight and the projected area of the points classified as grapes in vines when partial two-sided removal treatment, and an R2 of 0.82 was achieved in one of the datasets for vines with untouched full canopy. The accuracy achieved in grape detection opens the door to yield prediction in red grape vineyards. This would allow the creation of yield estimation maps that will ease the implementation of precision viticulture practices. To the authors’ knowledge, this is the first time that UAV photogrammetric point clouds have been used for grape clusters detection.

@inproceedings{
 title = {Adversarial Reweighting for Partial Domain Adaptation},
 type = {inproceedings},
 year = {2021},
 pages = {14860--14872},
 websites = {https://github.com/XJTU-XGU/},
 id = {7fb78617-28e3-3f14-8b02-9be6e0b773f4},
 created = {2023-01-05T16:00:47.448Z},
 accessed = {2023-01-05},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-13T07:12:08.383Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Partial domain adaptation (PDA) has gained much attention due to its practical setting. The current PDA methods usually adapt the feature extractor by aligning the target and reweighted source domain distributions. In this paper, we experimentally find that the feature adaptation by the reweighted distribution alignment in some state-of-the-art PDA methods is not robust to the "noisy" weights of source domain data, leading to negative domain transfer on some challenging benchmarks. To tackle the challenge of negative domain transfer, we propose a novel Adversarial Reweighting (AR) approach that adversarially learns the weights of source domain data to align the source and target domain distributions, and the transferable deep recognition network is learned on the reweighted source domain data. Based on this idea, we propose a training algorithm that alternately updates the parameters of the network and optimizes the weights of source domain data. Extensive experiments show that our method achieves state-of-the-art results on the benchmarks of ImageNet-Caltech, Office-Home, VisDA-2017, and DomainNet. Ablation studies also confirm the effectiveness of our approach.},
 bibtype = {inproceedings},
 author = {Gu, Xiang and Yu, Xi and Yang, Yan and Sun, Jian and Xu, Zongben},
 booktitle = {Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual}
}

Partial domain adaptation (PDA) has gained much attention due to its practical setting. The current PDA methods usually adapt the feature extractor by aligning the target and reweighted source domain distributions. In this paper, we experimentally find that the feature adaptation by the reweighted distribution alignment in some state-of-the-art PDA methods is not robust to the "noisy" weights of source domain data, leading to negative domain transfer on some challenging benchmarks. To tackle the challenge of negative domain transfer, we propose a novel Adversarial Reweighting (AR) approach that adversarially learns the weights of source domain data to align the source and target domain distributions, and the transferable deep recognition network is learned on the reweighted source domain data. Based on this idea, we propose a training algorithm that alternately updates the parameters of the network and optimizes the weights of source domain data. Extensive experiments show that our method achieves state-of-the-art results on the benchmarks of ImageNet-Caltech, Office-Home, VisDA-2017, and DomainNet. Ablation studies also confirm the effectiveness of our approach.

@article{
 title = {Weed and crop classification with domain adaptation for precision agriculture},
 type = {article},
 year = {2021},
 keywords = {Agricultural automation,Domain adaptation,Semantic segmentation},
 month = {6},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 day = {9},
 id = {40ba381d-b87e-3208-8097-f9d18e20ad71},
 created = {2023-01-06T11:07:44.602Z},
 accessed = {2023-01-06},
 file_attached = {false},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-06T14:08:56.201Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Agricultural chemicals, which are frequently used in agricultural fields, have many different side effects on the environment, biological diversity and people. It is aimed to reduce the use of agricultural chemicals within the scope of sustainable agriculture. For this purpose, it is an important requirement for robots used in agricultural fields to precisely distinguish between crops and weeds. One of the biggest challenges in this regard is the wide variety of plant species. For this reason, the systems to be used in agricultural areas need to cope well with the changes caused by weed and crop varieties. In this study, a examination is made using various unsupervised domain adaptation approaches to tackle the challenge of species diversity in the weed and crop classification task. Thus, it is aimed to measure the generalize the classification process of crop and weed species that have not been seen before.},
 bibtype = {article},
 author = {Ekinci, Seydanur and Aptoula, Erchan},
 doi = {10.1109/SIU53274.2021.9477948},
 journal = {SIU 2021 - 29th IEEE Conference on Signal Processing and Communications Applications, Proceedings}
}

Agricultural chemicals, which are frequently used in agricultural fields, have many different side effects on the environment, biological diversity and people. It is aimed to reduce the use of agricultural chemicals within the scope of sustainable agriculture. For this purpose, it is an important requirement for robots used in agricultural fields to precisely distinguish between crops and weeds. One of the biggest challenges in this regard is the wide variety of plant species. For this reason, the systems to be used in agricultural areas need to cope well with the changes caused by weed and crop varieties. In this study, a examination is made using various unsupervised domain adaptation approaches to tackle the challenge of species diversity in the weed and crop classification task. Thus, it is aimed to measure the generalize the classification process of crop and weed species that have not been seen before.