Scholarlink

<p>Addressing the core challenge of unknown radar illumination aspect angles in Synthetic Aperture Radar (SAR) Automatic Target Recognition, this study investigates how to construct a single physical perturbation to achieve robust adversarial attacks across the full aspect angle domain. Under the premise that the target can be accurately characterized by the Attributed Scattering Center Model, we propose an omnidirectional aspect domain attack method that incorporates SAR imaging physical mechanisms. This approach formulates the generation of physical adversarial examples as an optimization problem based on the intrinsic parameters of scattering centers, which is solved using the Particle Swarm Optimization algorithm, thereby establishing a direct mapping from physical parameters to data domain adversarial examples. Simulation experiments, encompassing non-targeted attacks, targeted attacks, and cross-model transfer attacks, supplemented by key parameter sensitivity analysis and ablation studies, demonstrate that the proposed method exhibits potential physical feasibility under simulated physical constraints and can stably maintain high attack effectiveness across the full aspect angle domain. This research provides a useful reference for the development of physical domain adversarial example attacks and defense technologies in remote sensing.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Full aspect angular domain-robust adversarial attack method for SAR ATR systems

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月19日

查看全文
  • Bo Wei,
  • Huagang Xiong,
  • Teng Huang

摘要

Addressing the core challenge of unknown radar illumination aspect angles in Synthetic Aperture Radar (SAR) Automatic Target Recognition, this study investigates how to construct a single physical perturbation to achieve robust adversarial attacks across the full aspect angle domain. Under the premise that the target can be accurately characterized by the Attributed Scattering Center Model, we propose an omnidirectional aspect domain attack method that incorporates SAR imaging physical mechanisms. This approach formulates the generation of physical adversarial examples as an optimization problem based on the intrinsic parameters of scattering centers, which is solved using the Particle Swarm Optimization algorithm, thereby establishing a direct mapping from physical parameters to data domain adversarial examples. Simulation experiments, encompassing non-targeted attacks, targeted attacks, and cross-model transfer attacks, supplemented by key parameter sensitivity analysis and ablation studies, demonstrate that the proposed method exhibits potential physical feasibility under simulated physical constraints and can stably maintain high attack effectiveness across the full aspect angle domain. This research provides a useful reference for the development of physical domain adversarial example attacks and defense technologies in remote sensing.