EUVIMEDEUVIMED entdecken
Crossref

CFW-AT: Compressed Feature Weak-Class Adversarial Training for Enhancing Robustness in Few-Shot SAR Image Recognition

Fei Gao, Mingyang Li, Yanxian Bi, Jinping Sun, Amir Hussain, Yang Zheng

IEEE Transactions on Aerospace and Electronic Systems · 2026 · 62 · 19135-19157

DOI: 10.1109/taes.2026.3740120 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Aerospace and Electronic Systems, 2026. Verfasst von Fei Gao und 5 weiteren Autor:innen. Die Arbeit nennt 64 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In recent years, a large number of scholars have studied synthetic aperture radar image recognition under few-shot data conditions, aiming to solve the problem of insufficient sample data in practical scenarios. However, due to the emergence of adversarial attack technology, various few-shot learning methods exhibit extremely weak adversarial robustness, which seriously threatens the security of recognition models. In order to improve the adversarial robustness of few-shot recognition, we propose a compressed feature weak-class adversarial training (CFW-AT) method based on dual-weight calculation. First, the dual-weight calculation method of the original subclassifier and adversarial subclassifier is designed to balance the clean recognition performance and adversarial robustness. Second, the features and decision boundaries of weak classes are optimized through weak-class adversarial training. Finally, the compressed feature extraction module is introduced to eliminate feature redundancy and maximize feature differences between different categories, so as to learn robust category features. Experimental results on the MSTAR and ATRNet-STAR datasets show that CFW-AT achieves the best adversarial recognition accuracy, which is superior to other state-of-the-art methods, while maintaining high clean recognition accuracy. CFW-AT significantly enhances the adversarial robustness of few-shot recognition, giving few-shot recognition models higher application value.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Gao, F., Li, M., Bi, Y., Sun, J., Hussain, A., & Zheng, Y. (2026). CFW-AT: Compressed Feature Weak-Class Adversarial Training for Enhancing Robustness in Few-Shot SAR Image Recognition. IEEE Transactions on Aerospace and Electronic Systems, 62, 19135-19157. https://doi.org/10.1109/taes.2026.3740120

Vancouver

Gao F, Li M, Bi Y, Sun J, Hussain A, Zheng Y. CFW-AT: Compressed Feature Weak-Class Adversarial Training for Enhancing Robustness in Few-Shot SAR Image Recognition. IEEE Transactions on Aerospace and Electronic Systems. 2026;62:19135-19157. doi:10.1109/taes.2026.3740120

BibTeX
@article{Gao2026,
  author = {Gao, Fei and Li, Mingyang and Bi, Yanxian and Sun, Jinping and Hussain, Amir and Zheng, Yang},
  title = {CFW-AT: Compressed Feature Weak-Class Adversarial Training for Enhancing Robustness in Few-Shot SAR Image Recognition},
  journal = {IEEE Transactions on Aerospace and Electronic Systems},
  year = {2026},
  volume = {62},
  pages = {19135--19157},
  doi = {10.1109/taes.2026.3740120}
}
Kostenlos starten →