Rapid Training for Vision Based Grading of Wood, Machine Vision and Applications. Affonso, C., Hietaniemi, R., Liu, L., Holmberg, T., & Silvén, O. Machine Vision and Applications, 2018. abstract bibtex In automated visual inspection of natural
materials a large number of correctly labeled training samples is often needed. Obtaining such a data set can be an overwhelming challenge for error prone humans. If the appearance of the material changes between batches, re-acquisition of the training data, or re-labeling earlier
ones, as well revising the features, could be required. We propose a convolutional neural network based active learning method to speed up the training of wood inspection systems, and compare approaches to select instances to be labeled from data sets. Experimental results with lumber and pencil slat data sets demonstrate capability to achieve good accuracy using only a
few tens of training samples. In contrast, experiments with a structural timber data set showed that strength grading is a complex problem with higher training material needs. In all cases selecting training samples close to category boundaries was the best strategy
@article{
title = {Rapid Training for Vision Based Grading of Wood, Machine Vision and Applications},
type = {article},
year = {2018},
pages = {14p},
id = {c8f9602d-0a4d-3d83-aa04-067eb7220489},
created = {2019-11-19T16:28:48.723Z},
file_attached = {false},
profile_id = {bddcf02d-403b-3b06-9def-6d15cc293e20},
group_id = {28b2996c-b80f-3c26-be71-695caf7040ac},
last_modified = {2019-11-19T16:30:18.316Z},
read = {false},
starred = {false},
authored = {false},
confirmed = {true},
hidden = {false},
citation_key = {cmv:2218},
source_type = {article},
folder_uuids = {8292f5ec-1c57-4113-a303-25778e695f8c},
private_publication = {false},
abstract = {In automated visual inspection of natural
materials a large number of correctly labeled training samples is often needed. Obtaining such a data set can be an overwhelming challenge for error prone humans. If the appearance of the material changes between batches, re-acquisition of the training data, or re-labeling earlier
ones, as well revising the features, could be required. We propose a convolutional neural network based active learning method to speed up the training of wood inspection systems, and compare approaches to select instances to be labeled from data sets. Experimental results with lumber and pencil slat data sets demonstrate capability to achieve good accuracy using only a
few tens of training samples. In contrast, experiments with a structural timber data set showed that strength grading is a complex problem with higher training material needs. In all cases selecting training samples close to category boundaries was the best strategy},
bibtype = {article},
author = {Affonso, C and Hietaniemi, R and Liu, L and Holmberg, T and Silvén, O},
journal = {Machine Vision and Applications}
}
Downloads: 0
{"_id":"qDqJM9W5xG6muxzbg","bibbaseid":"affonso-hietaniemi-liu-holmberg-silvn-rapidtrainingforvisionbasedgradingofwoodmachinevisionandapplications-2018","authorIDs":[],"author_short":["Affonso, C.","Hietaniemi, R.","Liu, L.","Holmberg, T.","Silvén, O."],"bibdata":{"title":"Rapid Training for Vision Based Grading of Wood, Machine Vision and Applications","type":"article","year":"2018","pages":"14p","id":"c8f9602d-0a4d-3d83-aa04-067eb7220489","created":"2019-11-19T16:28:48.723Z","file_attached":false,"profile_id":"bddcf02d-403b-3b06-9def-6d15cc293e20","group_id":"28b2996c-b80f-3c26-be71-695caf7040ac","last_modified":"2019-11-19T16:30:18.316Z","read":false,"starred":false,"authored":false,"confirmed":"true","hidden":false,"citation_key":"cmv:2218","source_type":"article","folder_uuids":"8292f5ec-1c57-4113-a303-25778e695f8c","private_publication":false,"abstract":"In automated visual inspection of natural\nmaterials a large number of correctly labeled training samples is often needed. Obtaining such a data set can be an overwhelming challenge for error prone humans. If the appearance of the material changes between batches, re-acquisition of the training data, or re-labeling earlier\nones, as well revising the features, could be required. We propose a convolutional neural network based active learning method to speed up the training of wood inspection systems, and compare approaches to select instances to be labeled from data sets. Experimental results with lumber and pencil slat data sets demonstrate capability to achieve good accuracy using only a\nfew tens of training samples. In contrast, experiments with a structural timber data set showed that strength grading is a complex problem with higher training material needs. In all cases selecting training samples close to category boundaries was the best strategy","bibtype":"article","author":"Affonso, C and Hietaniemi, R and Liu, L and Holmberg, T and Silvén, O","journal":"Machine Vision and Applications","bibtex":"@article{\n title = {Rapid Training for Vision Based Grading of Wood, Machine Vision and Applications},\n type = {article},\n year = {2018},\n pages = {14p},\n id = {c8f9602d-0a4d-3d83-aa04-067eb7220489},\n created = {2019-11-19T16:28:48.723Z},\n file_attached = {false},\n profile_id = {bddcf02d-403b-3b06-9def-6d15cc293e20},\n group_id = {28b2996c-b80f-3c26-be71-695caf7040ac},\n last_modified = {2019-11-19T16:30:18.316Z},\n read = {false},\n starred = {false},\n authored = {false},\n confirmed = {true},\n hidden = {false},\n citation_key = {cmv:2218},\n source_type = {article},\n folder_uuids = {8292f5ec-1c57-4113-a303-25778e695f8c},\n private_publication = {false},\n abstract = {In automated visual inspection of natural\nmaterials a large number of correctly labeled training samples is often needed. Obtaining such a data set can be an overwhelming challenge for error prone humans. If the appearance of the material changes between batches, re-acquisition of the training data, or re-labeling earlier\nones, as well revising the features, could be required. We propose a convolutional neural network based active learning method to speed up the training of wood inspection systems, and compare approaches to select instances to be labeled from data sets. Experimental results with lumber and pencil slat data sets demonstrate capability to achieve good accuracy using only a\nfew tens of training samples. In contrast, experiments with a structural timber data set showed that strength grading is a complex problem with higher training material needs. In all cases selecting training samples close to category boundaries was the best strategy},\n bibtype = {article},\n author = {Affonso, C and Hietaniemi, R and Liu, L and Holmberg, T and Silvén, O},\n journal = {Machine Vision and Applications}\n}","author_short":["Affonso, C.","Hietaniemi, R.","Liu, L.","Holmberg, T.","Silvén, O."],"bibbaseid":"affonso-hietaniemi-liu-holmberg-silvn-rapidtrainingforvisionbasedgradingofwoodmachinevisionandapplications-2018","role":"author","urls":{},"downloads":0},"bibtype":"article","creationDate":"2019-11-19T16:11:28.970Z","downloads":0,"keywords":[],"search_terms":["rapid","training","vision","based","grading","wood","machine","vision","applications","affonso","hietaniemi","liu","holmberg","silvén"],"title":"Rapid Training for Vision Based Grading of Wood, Machine Vision and Applications","year":2018}