Electromagnetic Signal Adversarial Example Generation Based on Perturbation Region Selection

With the widespread application of deep learning techniques in the field of electromagnetic signal recognition, the generation of adversarial examples for electromagnetic signals has gradually become a hot research topic in the field. Existing techniques for generating adversarial examples for electromagnetic signals suffer from issues such as high perturbation magnitude, low timeliness, and poor stealthiness, which makes it difficult to meet the practical application needs effectively. This thesis proposes an adversarial sample generation method based on region selection, called Region Selection-FGSM (RS-FGSM). This method selects the perturbation regions based on the magnitude of the gradient of the electromagnetic signal, and generates adversarial examples by applying FGSM perturbation within the selected regions. Experimental results demonstrate that the proposed method outperforms the FGSM method in terms of perturbation magnitude, timeliness, and stealthiness.