Synthetic Aperture Radar Adversarial Perturbation Generation Algorithm Based on Scatterer Imaging and Momentum Gradient Optimization

Addressing the issue that most existing SAR adversarial example generation methods are primarily focused on the digital image domain and lack effective adversarial examples in real-world scenarios, this paper proposes a SAR adversarial example generation method based on scattering body imaging, capable of generating common scattering body perturbations in reality. Initially, scattering body imaging is conducted based on scattering body parameters to obtain the corresponding perturbations. Prior to this, heatmaps are utilized for visualization, and normalization operations are employed to analyze the image data, defining the regions where the perturbations occur. This enables the perturbations to act more precisely on the target, thereby enhancing the attack effect. Additionally, the momentum gradient descent optimization strategy is used to accelerate the optimization process, increase the generation speed, and improve timeliness. Moreover, previous research on scattering body perturbations did not consider stealthiness as an indicator. In this paper, structural similarity is adopted as a key evaluation metric, and a threshold is set to improve the generation quality of scattering body perturbations, ultimately enhancing their stealthiness. Experimental results show that when the number of scattering bodies is 1, 2, and 3, the attack success rates are 54.0%, 73.9%, and 85.8%, respectively, with computation time consumption of 29.14s, 30.84s, and 28.11s, and high stealthiness.