Video recognition model attack method based on time sequence movement

• Main Authors: WEI ZHIPENG, WU ZUXUAN, CHEN JINGJING, JIANG YUGANG
• Format: Patent
• Language: Chinese; English
• Published: 19.11.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention belongs to the technical field of video recognition network model security, and particularly relates to a video recognition model attack method based on time sequence movement. According to the invention, the concept of a time sequence discrimination mode is introduced into a video recognition network model, and different video recognition models have different time sequence discrimination modes. The method comprises the following steps: executing a time sequence translation operation on video input data to obtain a series of video input data, and interfering a time sequence discrimination mode of a single model; obtaining gradient information of each video input data through gradient back propagation of the single model, and translating the gradient information to time sequence information which is the same as the original video input data; and finally, completing gradient information aggregation by using a Gaussian kernel, generating an adversarial sample with high mobility, and completing att