General Sparse Adversarial Attack Method for SAR Images based on Key Points

<p dir="ltr">With the continuous development of deep learning<br>in the field of Synthetic Aperture Radar (SAR) image processing,<br>it is found that image recognition is vulnerable to interference and<br>the accuracy is greatly reduced. Therefore, if a general adversarial<br>attack method is designed to generate adversarial examples(AEs)<br>capable of deceiving different types of convolutional neural network<br>models, it can protect target images from being correctly recognized<br>by adversaries and safeguard image privacy information. However,<br>most of the current adversarial attack methods have too large<br>attack scope and interference intensity, and do not have the ability<br>to universally attack different neural network models, resulting in<br>poor concealment, deception, and transfer deception of adversarial<br>attacks. In this paper, we propose a general Key Points Sparse<br>Attack (KPSA) method for SAR images, which achieves excellent<br>deception and transfer deception while controlling the attack range<br>and jamming strength. It is a general and efficient adversarial<br>attack method. KPSA uses the generator to generate intensity<br>interference images and imposes amplitude constraint, uses the key<br>points extraction method to generate position interference images to<br>realize sparse attack, uses the difference value(d-value) loss function<br>to achieve fast convergence of the training model, and achieves<br>transfer deception performance without depending on the type<br>of discriminator. The deception rate and transfer deception rate<br>experiments were conducted on MSTAR and ATRNet-STAR. By<br>comparing with the advanced methods in the field of adversarial<br>attacks, it was verified that the KPSA method is superior to the current advanced methods, and it was confirmed that the KPSA method has application value in various complex practical scenarios.</p>