Abstract:With the rapid development
of artificial intelligence and computer vision, using intelligent object
detection technology to identify the aircraft and UAV has become a common
military defense means. Aiming at the feasibility of confusing and deceiving
the object detection model based on deep learning network from the digital
image level, and materializing the digital attack results in the future, this
paper proposes a limited area adversarial attack method for vision-based
aircraft intelligent object detection. The digital image of aircraft is used to
simulate the adversarial attack of YOLO v3 object detection network. The effectiveness
of the proposed method is verified by the experiments on the aircraft
categories in PASCAL_VOC_2007 dataset.The work of this paper provides the basis for further research on
whether the digital analog attack can be transferred to the aircraft entity.