Cargando…

Class incremental learning of remote sensing images based on class similarity distillation

When a well-trained model learns a new class, the data distribution differences between the new and old classes inevitably cause catastrophic forgetting in order to perform better in the new class. This behavior differs from human learning. In this article, we propose a class incremental object dete...

Descripción completa

Detalles Bibliográficos
Autores principales: Shen, Mingge, Chen, Dehu, Hu, Silan, Xu, Gang
Formato: Online Artículo Texto
Lenguaje:English
Publicado: PeerJ Inc. 2023
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10557500/
https://www.ncbi.nlm.nih.gov/pubmed/37810339
http://dx.doi.org/10.7717/peerj-cs.1583
_version_ 1785117103106293760
author Shen, Mingge
Chen, Dehu
Hu, Silan
Xu, Gang
author_facet Shen, Mingge
Chen, Dehu
Hu, Silan
Xu, Gang
author_sort Shen, Mingge
collection PubMed
description When a well-trained model learns a new class, the data distribution differences between the new and old classes inevitably cause catastrophic forgetting in order to perform better in the new class. This behavior differs from human learning. In this article, we propose a class incremental object detection method for remote sensing images to address the problem of catastrophic forgetting caused by distribution differences among different classes. First, we introduce a class similarity distillation (CSD) loss based on the similarity between new and old class prototypes, ensuring the model’s plasticity to learn new classes and stability to detect old classes. Second, to better extract class similarity features, we propose a global similarity distillation (GSD) loss that maximizes the mutual information between the new class feature and old class features. Additionally, we present a region proposal network (RPN)-based method that assigns positive and negative labels to prevent mislearning issues. Experiments demonstrate that our method is more accurate for class incremental learning on public DOTA and DIOR datasets and significantly improves training efficiency compared to state-of-the-art class incremental object detection methods.
format Online
Article
Text
id pubmed-10557500
institution National Center for Biotechnology Information
language English
publishDate 2023
publisher PeerJ Inc.
record_format MEDLINE/PubMed
spelling pubmed-105575002023-10-07 Class incremental learning of remote sensing images based on class similarity distillation Shen, Mingge Chen, Dehu Hu, Silan Xu, Gang PeerJ Comput Sci Artificial Intelligence When a well-trained model learns a new class, the data distribution differences between the new and old classes inevitably cause catastrophic forgetting in order to perform better in the new class. This behavior differs from human learning. In this article, we propose a class incremental object detection method for remote sensing images to address the problem of catastrophic forgetting caused by distribution differences among different classes. First, we introduce a class similarity distillation (CSD) loss based on the similarity between new and old class prototypes, ensuring the model’s plasticity to learn new classes and stability to detect old classes. Second, to better extract class similarity features, we propose a global similarity distillation (GSD) loss that maximizes the mutual information between the new class feature and old class features. Additionally, we present a region proposal network (RPN)-based method that assigns positive and negative labels to prevent mislearning issues. Experiments demonstrate that our method is more accurate for class incremental learning on public DOTA and DIOR datasets and significantly improves training efficiency compared to state-of-the-art class incremental object detection methods. PeerJ Inc. 2023-09-27 /pmc/articles/PMC10557500/ /pubmed/37810339 http://dx.doi.org/10.7717/peerj-cs.1583 Text en ©2023 Shen et al. https://creativecommons.org/licenses/by/4.0/This is an open access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/) , which permits unrestricted use, distribution, reproduction and adaptation in any medium and for any purpose provided that it is properly attributed. For attribution, the original author(s), title, publication source (PeerJ Computer Science) and either DOI or URL of the article must be cited.
spellingShingle Artificial Intelligence
Shen, Mingge
Chen, Dehu
Hu, Silan
Xu, Gang
Class incremental learning of remote sensing images based on class similarity distillation
title Class incremental learning of remote sensing images based on class similarity distillation
title_full Class incremental learning of remote sensing images based on class similarity distillation
title_fullStr Class incremental learning of remote sensing images based on class similarity distillation
title_full_unstemmed Class incremental learning of remote sensing images based on class similarity distillation
title_short Class incremental learning of remote sensing images based on class similarity distillation
title_sort class incremental learning of remote sensing images based on class similarity distillation
topic Artificial Intelligence
url https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10557500/
https://www.ncbi.nlm.nih.gov/pubmed/37810339
http://dx.doi.org/10.7717/peerj-cs.1583
work_keys_str_mv AT shenmingge classincrementallearningofremotesensingimagesbasedonclasssimilaritydistillation
AT chendehu classincrementallearningofremotesensingimagesbasedonclasssimilaritydistillation
AT husilan classincrementallearningofremotesensingimagesbasedonclasssimilaritydistillation
AT xugang classincrementallearningofremotesensingimagesbasedonclasssimilaritydistillation