Analysis on adversarial attacks in optical satellite image recognition

With the wide application of deep learning techniques in optical satellite image recognition, its security issues are becoming more and more prominent. The adversarial sample attack deceives the model by adding tiny perturbations, leading to misclassification, which poses a threat to military reconnaissance, environmental monitoring and other critical areas. In this paper, we systematically study the generation principle of adversarial sample and its attack effect in optical satellite image recognition, analyze the advantages and disadvantages of existing defense strategies, and propose an improved defense method. The experimental results show that the proposed method can effectively improve the robustness of the model, which provides a new idea for the security protection of optical satellite image recognition system.