Infrared target detector attack method based on free speed and rollback variation

The invention belongs to the technical field of artificial intelligence security and computer vision, and discloses an infrared target detector attack method based on free speed and rollback variation, which comprises the following steps: S1, collecting a public thermal infrared data set, and carrying out transfer learning on the public thermal infrared data set to obtain an infrared target detection model; s2, an algorithm framework is built, noise is randomly sampled in Gaussian distribution to generate candidate adversarial samples, a free speed attack method and a rollback mutation search strategy are introduced, and further optimization is performed on the basis of the candidate adversarial samples; s3, implementing digital black box countermeasure attack on the target model; s4, hyper-parameters in the experiment are adjusted; s5, obtaining an infrared confrontation sample which can not be perceived by human eyes and has a relatively strong attack effect; according to the method, the attack performance