Physical confrontation attack method for unmanned aerial vehicle target detection system

• Main Authors: HE YUNLING, HANG QIN, HUANG NONGSEN, LIU BO, ZHONG LINGPENG, GUO JIAXIN, DING JIASONG, LYU XUE, ZHAO HONGPING, CHENG CHENG, ZHANG HENG
• Format: Patent
• Language: Chinese; English
• Published: 09.05.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention belongs to the field of artificial intelligence security, and particularly relates to a physical countermeasure attack method for an unmanned aerial vehicle target detection system. The method comprises the following steps: generating an adversarial patch by using a pre-trained generative adversarial network, and inputting the adversarial patch, a training sample and a label file thereof into a patch conversion module to obtain an adversarial sample; inputting the adversarial sample into a Yolov5 target detector, and calculating the total loss; designing a target optimization function, and training an adversarial patch by adopting an Adam optimizer in combination with the total loss; training for multiple times until the number of iterations reaches a preset value or a loss function is converged; according to the invention, the adversarial patch is installed on the automobile roof, so that the ground vehicle can avoid the unmanned aerial vehicle target detection system, and the safety detection