Constructing a Large-Scale Landslide Database Across Heterogeneous Environments Using Task-Specific Model Updates. Nagendra, S., Kifer, D., Mirus, B., Pei, T., Lawson, K., Manjunatha, S. B., Li, W., Nguyen, H., Qiu, T., Tran, S., & Shen, C. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15:4349–4370, 2022. Conference Name: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensingdoi abstract bibtex Recent small-scale studies for pixel-wise labeling of potential landslide areas in remotely-sensed images using deep learning (DL) showed potential but were based on data from very small, homogeneous regions with unproven model transferability. In this paper we consider a more realistic and practical setting for large-scale heterogeneous landslide data collection and DL-based labeling. In this setting, remotely sensed images are collected sequentially in temporal batches, where each batch focuses on images from a particular ecoregion, but different batches can focus on different ecoregions with distinct landscape characteristics. For such a scenario, we study the following questions: (1) How well do DL models trained in homogeneous regions perform when they are transferred to different ecoregions? (2) Does increasing the spatial coverage in the data improve model performance in a given ecoregion? and (3) Can a landslide pixel labeling model be incrementally updated with new data, but without access to the old data and without losing performance on the old data? We address these questions by developing a mechanism for incremental training of semantic segmentation models. We call the resulting extension task-specific model updates (TSMU). A national compilation of landslide inventories by the U.S. Geological Survey (USGS) was used to develop a global database for this study. Our results indicate that the TSMU framework can be used to aid in the creation of new landslide inventories or expanding existing datasets, and also to rapidly develop hazard maps for situational awareness following a widespread landslide event.
@article{nagendra_constructing_2022,
title = {Constructing a {Large}-{Scale} {Landslide} {Database} {Across} {Heterogeneous} {Environments} {Using} {Task}-{Specific} {Model} {Updates}},
volume = {15},
issn = {2151-1535},
doi = {10.1109/JSTARS.2022.3177025},
abstract = {Recent small-scale studies for pixel-wise labeling of potential landslide areas in remotely-sensed images using deep learning (DL) showed potential but were based on data from very small, homogeneous regions with unproven model transferability. In this paper we consider a more realistic and practical setting for large-scale heterogeneous landslide data collection and DL-based labeling. In this setting, remotely sensed images are collected sequentially in temporal batches, where each batch focuses on images from a particular ecoregion, but different batches can focus on different ecoregions with distinct landscape characteristics. For such a scenario, we study the following questions: (1) How well do DL models trained in homogeneous regions perform when they are transferred to different ecoregions? (2) Does increasing the spatial coverage in the data improve model performance in a given ecoregion? and (3) Can a landslide pixel labeling model be incrementally updated with new data, but without access to the old data and without losing performance on the old data? We address these questions by developing a mechanism for incremental training of semantic segmentation models. We call the resulting extension task-specific model updates (TSMU). A national compilation of landslide inventories by the U.S. Geological Survey (USGS) was used to develop a global database for this study. Our results indicate that the TSMU framework can be used to aid in the creation of new landslide inventories or expanding existing datasets, and also to rapidly develop hazard maps for situational awareness following a widespread landslide event.},
journal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
author = {Nagendra, Savinay and Kifer, Daniel and Mirus, Benjamin and Pei, Te and Lawson, Kathryn and Manjunatha, Srikanth Banagere and Li, Weixin and Nguyen, Hien and Qiu, Tong and Tran, Sarah and Shen, Chaopeng},
year = {2022},
note = {Conference Name: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
keywords = {Terrestrial Ecoregions (CEC 1997)},
pages = {4349--4370},
}
Downloads: 0
{"_id":"phTX2p6LapcjxR5u5","bibbaseid":"nagendra-kifer-mirus-pei-lawson-manjunatha-li-nguyen-etal-constructingalargescalelandslidedatabaseacrossheterogeneousenvironmentsusingtaskspecificmodelupdates-2022","author_short":["Nagendra, S.","Kifer, D.","Mirus, B.","Pei, T.","Lawson, K.","Manjunatha, S. B.","Li, W.","Nguyen, H.","Qiu, T.","Tran, S.","Shen, C."],"bibdata":{"bibtype":"article","type":"article","title":"Constructing a Large-Scale Landslide Database Across Heterogeneous Environments Using Task-Specific Model Updates","volume":"15","issn":"2151-1535","doi":"10.1109/JSTARS.2022.3177025","abstract":"Recent small-scale studies for pixel-wise labeling of potential landslide areas in remotely-sensed images using deep learning (DL) showed potential but were based on data from very small, homogeneous regions with unproven model transferability. In this paper we consider a more realistic and practical setting for large-scale heterogeneous landslide data collection and DL-based labeling. In this setting, remotely sensed images are collected sequentially in temporal batches, where each batch focuses on images from a particular ecoregion, but different batches can focus on different ecoregions with distinct landscape characteristics. For such a scenario, we study the following questions: (1) How well do DL models trained in homogeneous regions perform when they are transferred to different ecoregions? (2) Does increasing the spatial coverage in the data improve model performance in a given ecoregion? and (3) Can a landslide pixel labeling model be incrementally updated with new data, but without access to the old data and without losing performance on the old data? We address these questions by developing a mechanism for incremental training of semantic segmentation models. We call the resulting extension task-specific model updates (TSMU). A national compilation of landslide inventories by the U.S. Geological Survey (USGS) was used to develop a global database for this study. Our results indicate that the TSMU framework can be used to aid in the creation of new landslide inventories or expanding existing datasets, and also to rapidly develop hazard maps for situational awareness following a widespread landslide event.","journal":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","author":[{"propositions":[],"lastnames":["Nagendra"],"firstnames":["Savinay"],"suffixes":[]},{"propositions":[],"lastnames":["Kifer"],"firstnames":["Daniel"],"suffixes":[]},{"propositions":[],"lastnames":["Mirus"],"firstnames":["Benjamin"],"suffixes":[]},{"propositions":[],"lastnames":["Pei"],"firstnames":["Te"],"suffixes":[]},{"propositions":[],"lastnames":["Lawson"],"firstnames":["Kathryn"],"suffixes":[]},{"propositions":[],"lastnames":["Manjunatha"],"firstnames":["Srikanth","Banagere"],"suffixes":[]},{"propositions":[],"lastnames":["Li"],"firstnames":["Weixin"],"suffixes":[]},{"propositions":[],"lastnames":["Nguyen"],"firstnames":["Hien"],"suffixes":[]},{"propositions":[],"lastnames":["Qiu"],"firstnames":["Tong"],"suffixes":[]},{"propositions":[],"lastnames":["Tran"],"firstnames":["Sarah"],"suffixes":[]},{"propositions":[],"lastnames":["Shen"],"firstnames":["Chaopeng"],"suffixes":[]}],"year":"2022","note":"Conference Name: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","keywords":"Terrestrial Ecoregions (CEC 1997)","pages":"4349–4370","bibtex":"@article{nagendra_constructing_2022,\n\ttitle = {Constructing a {Large}-{Scale} {Landslide} {Database} {Across} {Heterogeneous} {Environments} {Using} {Task}-{Specific} {Model} {Updates}},\n\tvolume = {15},\n\tissn = {2151-1535},\n\tdoi = {10.1109/JSTARS.2022.3177025},\n\tabstract = {Recent small-scale studies for pixel-wise labeling of potential landslide areas in remotely-sensed images using deep learning (DL) showed potential but were based on data from very small, homogeneous regions with unproven model transferability. In this paper we consider a more realistic and practical setting for large-scale heterogeneous landslide data collection and DL-based labeling. In this setting, remotely sensed images are collected sequentially in temporal batches, where each batch focuses on images from a particular ecoregion, but different batches can focus on different ecoregions with distinct landscape characteristics. For such a scenario, we study the following questions: (1) How well do DL models trained in homogeneous regions perform when they are transferred to different ecoregions? (2) Does increasing the spatial coverage in the data improve model performance in a given ecoregion? and (3) Can a landslide pixel labeling model be incrementally updated with new data, but without access to the old data and without losing performance on the old data? We address these questions by developing a mechanism for incremental training of semantic segmentation models. We call the resulting extension task-specific model updates (TSMU). A national compilation of landslide inventories by the U.S. Geological Survey (USGS) was used to develop a global database for this study. Our results indicate that the TSMU framework can be used to aid in the creation of new landslide inventories or expanding existing datasets, and also to rapidly develop hazard maps for situational awareness following a widespread landslide event.},\n\tjournal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},\n\tauthor = {Nagendra, Savinay and Kifer, Daniel and Mirus, Benjamin and Pei, Te and Lawson, Kathryn and Manjunatha, Srikanth Banagere and Li, Weixin and Nguyen, Hien and Qiu, Tong and Tran, Sarah and Shen, Chaopeng},\n\tyear = {2022},\n\tnote = {Conference Name: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},\n\tkeywords = {Terrestrial Ecoregions (CEC 1997)},\n\tpages = {4349--4370},\n}\n\n","author_short":["Nagendra, S.","Kifer, D.","Mirus, B.","Pei, T.","Lawson, K.","Manjunatha, S. B.","Li, W.","Nguyen, H.","Qiu, T.","Tran, S.","Shen, C."],"key":"nagendra_constructing_2022","id":"nagendra_constructing_2022","bibbaseid":"nagendra-kifer-mirus-pei-lawson-manjunatha-li-nguyen-etal-constructingalargescalelandslidedatabaseacrossheterogeneousenvironmentsusingtaskspecificmodelupdates-2022","role":"author","urls":{},"keyword":["Terrestrial Ecoregions (CEC 1997)"],"metadata":{"authorlinks":{}}},"bibtype":"article","biburl":"https://bibbase.org/zotero/NAAtlas","dataSources":["wszmZbW485BrvbZff"],"keywords":["terrestrial ecoregions (cec 1997)"],"search_terms":["constructing","large","scale","landslide","database","heterogeneous","environments","using","task","specific","model","updates","nagendra","kifer","mirus","pei","lawson","manjunatha","li","nguyen","qiu","tran","shen"],"title":"Constructing a Large-Scale Landslide Database Across Heterogeneous Environments Using Task-Specific Model Updates","year":2022}