Visual single-target tracking confrontation defense method based on classification regression graph alignment

The invention discloses a visual single-target tracking confrontation defense method based on classification regression graph alignment. The method comprises the steps that a video data set is acquired and preprocessed; constructing a visual single-target tracking adversarial defense network AADN based on classification regression graph alignment; training Def1 in combination with a search branch; and training Def2 in combination with a template branch to obtain a trained AADN, obtaining a post-defense search area and a post-defense template area from an image frame in a video with unknown disturbance by using the trained AADN, and sending the post-defense search area and the post-defense template area to a target tracker for processing. According to the method, potential adversarial disturbance in a sample neighborhood is filtered through adversarial training learning, the defense capability of the target tracker to unknown adversarial disturbance is enhanced, the target tracker has better robustness and gen