Method for Generating Adversarial Samples under Area Constraints in Ship Texture Generation

Ships, due to their complex structure, present challenges for adversarial example generation. In response to the issues of high implementation costs associated with the large size and complex structure of large vessels during adversarial attacks, a method for generating adversarial examples under restricted texture generation areas for ships is proposed. This method mainly utilizes the "attention" mechanism of object detection models to identify the optimal placement area for adversarial patches. Considering the visual salience of adversarial patches, a method for generating adversarial camouflage based on Voronoi diagrams is proposed. Finally, through analysis and comparative experiments, it is demonstrated that the algorithm proposed in this paper can effectively generate adversarial examples that maintain attack efficacy while ensuring that the visual impact is not overly prominent.