Infrared adversarial patch generation method and device based on reinforcement learning

The invention relates to a reinforcement learning-based infrared adversarial patch generation method and device, and the method comprises the steps: carrying out the multi-round training of a reinforcement learning decision network, so as to obtain a trained infrared adversarial patch decision network which can obtain a proper posting region and a proper gray scale in an infrared image, and posting a gray block; wherein multi-time-step gray block pasting is carried out in each round of training process, and when the gray block is pasted, the next time step execution action is obtained according to the coordinate and the gray level of the gray block of the previous time step. And finally, pasting gray blocks in a patch area according to the trained decision network of the infrared adversarial patch to obtain the infrared adversarial patch. The infrared adversarial patch generated by the method is simple in structure and texture, high in operability and high in attack effectiveness. 本发明涉及一种基于强化学习的红外对抗补丁生成方法及装置，