Cross-Domain Universal Adversarial Perturbations for Synthetic-to-Measured SAR Target Classification Attacks
IEEE Geoscience and Remote Sensing Letters · 2026 · 23 · 1-5
DOI: 10.1109/lgrs.2026.3672313 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Geoscience and Remote Sensing Letters, 2026. Verfasst von Yuntian Liu und 3 weiteren Autor:innen. Die Arbeit nennt 11 Literaturangaben. Gefördert von: Industrial Technology Development Program.
Abstract (Originalsprache)
Synthetic aperture radar (SAR) target recognition models are vulnerable to adversarial attacks, but most existing attack methods require access to large amounts of measured SAR data—an unrealistic assumption in practical settings. Although simulated SAR data are easier to obtain, the gap between the simulated and measured domains limits the effectiveness of adversarial perturbations trained solely on simulations. To address the challenge, the synthetic-to-measured universal adversarial perturbation (S2M-UAP) is proposed, a new framework tailored for robust SAR target deception. Unlike previous works that directly adopt synthetic data, the approach incorporates an attack-oriented redesign of CycleGAN. Specifically, a feature-guided CycleGAN is designed and modified to prioritize adversarial effectiveness over visual realism, explicitly aligning domain-invariant features that are most relevant to perturbation transfer. In addition, an SAR-oriented enhancement module is introduced to further improve the robustness of perturbations under real measurement conditions. Extensive experiments demonstrate that S2M-UAP substantially improves synthetic-to-measured (S2M) transferability and achieves superior attack performance compared to state-of-the-art universal adversarial perturbation (UAP) methods, even when only limited measured samples are available.
Förderung
- Industrial Technology Development Program
Zitieren
APA
Liu, Y., Qin, W., Zhang, Y., & Wang, F. (2026). Cross-Domain Universal Adversarial Perturbations for Synthetic-to-Measured SAR Target Classification Attacks. IEEE Geoscience and Remote Sensing Letters, 23, 1-5. https://doi.org/10.1109/lgrs.2026.3672313
Vancouver
Liu Y, Qin W, Zhang Y, Wang F. Cross-Domain Universal Adversarial Perturbations for Synthetic-to-Measured SAR Target Classification Attacks. IEEE Geoscience and Remote Sensing Letters. 2026;23:1-5. doi:10.1109/lgrs.2026.3672313
BibTeX
@article{Liu2026,
author = {Liu, Yuntian and Qin, Weibo and Zhang, Yiming and Wang, Feng},
title = {Cross-Domain Universal Adversarial Perturbations for Synthetic-to-Measured SAR Target Classification Attacks},
journal = {IEEE Geoscience and Remote Sensing Letters},
year = {2026},
volume = {23},
pages = {1--5},
doi = {10.1109/lgrs.2026.3672313}
}
EUVIMED