SAAIPAA: Optimizing aspect-angles-invariant physical adversarial attacks on SAR target recognition models (2511.03192v1)

Abstract: Synthetic aperture radar (SAR) enables versatile, all-time, all-weather remote sensing. Coupled with automatic target recognition (ATR) leveraging ML, SAR is empowering a wide range of Earth observation and surveillance applications. However, the surge of attacks based on adversarial perturbations against the ML algorithms underpinning SAR ATR is prompting the need for systematic research into adversarial perturbation mechanisms. Research in this area began in the digital (image) domain and evolved into the physical (signal) domain, resulting in physical adversarial attacks (PAAs) that strategically exploit corner reflectors as attack vectors to evade ML-based ATR. This paper proposes a novel framework called SAR Aspect-Angles-Invariant Physical Adversarial Attack (SAAIPAA) for physics-based modelling of reflector-actuated adversarial perturbations, which improves on the rigor of prior work. A unique feature of SAAIPAA is its ability to remain effective even when the attacker lacks knowledge of the SAR platform's aspect angles, by deploying at least one reflector in each azimuthal quadrant and optimizing reflector orientations. The resultant physical evasion attacks are efficiently realizable and optimal over the considered range of aspect angles between a SAR platform and a target, achieving state-of-the-art fooling rates (over 80% for DenseNet-121 and ResNet50) in the white-box setting. When aspect angles are known to the attacker, an average fooling rate of 99.2% is attainable. In black-box settings, although the attack efficacy of SAAIPAA transfers well between some models (e.g., from ResNet50 to DenseNet121), the transferability to some models (e.g., MobileNetV2) can be improved. A useful outcome of using the MSTAR dataset for the experiments in this article, a method for generating bounding boxes for densely sampled azimuthal SAR datasets is introduced.