TY - GEN
T1 - Two-step transfer learning for semantic plant segmentation
AU - Sakurai, Shunsuke
AU - Uchiyama, Hideaki
AU - Shimada, Atsushi
AU - Arita, Daisaku
AU - Taniguchi, Rin ichiro
N1 - Funding Information:
This work is supported by JSPS KAKENHI Grant Number JP15H01695 and JP17H01768, and grants from the Project of the NARO Bio-oriented Technology Research Advancement Institution (the special scheme project on regional developing strategy)
Publisher Copyright:
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved.
PY - 2018
Y1 - 2018
N2 - We discuss the applicability of a fully convolutional network (FCN), which provides promising performance in semantic segmentation tasks, to plant segmentation tasks. The challenge lies in training the network with a small dataset because there are not many samples in plant image datasets, as compared to object image datasets such as ImageNet and PASCAL VOC datasets. The proposed method is inspired by transfer learning, but involves a two-step adaptation. In the first step, we apply transfer learning from a source domain that contains many objects with a large amount of labeled data to a major category in the plant domain. Then, in the second step, category adaptation is performed from the major category to a minor category with a few samples within the plant domain. With leaf segmentation challenge (LSC) dataset, the experimental results confirm the effectiveness of the proposed method such that F-measure criterion was, for instance, 0.953 for the A2 dataset, which was 0.355 higher than that of direct adaptation, and 0.527 higher than that of non-adaptation.
AB - We discuss the applicability of a fully convolutional network (FCN), which provides promising performance in semantic segmentation tasks, to plant segmentation tasks. The challenge lies in training the network with a small dataset because there are not many samples in plant image datasets, as compared to object image datasets such as ImageNet and PASCAL VOC datasets. The proposed method is inspired by transfer learning, but involves a two-step adaptation. In the first step, we apply transfer learning from a source domain that contains many objects with a large amount of labeled data to a major category in the plant domain. Then, in the second step, category adaptation is performed from the major category to a minor category with a few samples within the plant domain. With leaf segmentation challenge (LSC) dataset, the experimental results confirm the effectiveness of the proposed method such that F-measure criterion was, for instance, 0.953 for the A2 dataset, which was 0.355 higher than that of direct adaptation, and 0.527 higher than that of non-adaptation.
UR - http://www.scopus.com/inward/record.url?scp=85052017436&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85052017436&partnerID=8YFLogxK
U2 - 10.5220/0006576303320339
DO - 10.5220/0006576303320339
M3 - Conference contribution
AN - SCOPUS:85052017436
T3 - ICPRAM 2018 - Proceedings of the 7th International Conference on Pattern Recognition Applications and Methods
SP - 332
EP - 339
BT - ICPRAM 2018 - Proceedings of the 7th International Conference on Pattern Recognition Applications and Methods
A2 - De Marsico, Maria
A2 - di Baja, Gabriella Sanniti
A2 - Fred, Ana
PB - SciTePress
T2 - 7th International Conference on Pattern Recognition Applications and Methods, ICPRAM 2018
Y2 - 16 January 2018 through 18 January 2018
ER -