Attack against UAV Infrared Object Detection Based on Gradient Descent Method

At present, infrared intelligent detection technology has been widely used in various fields. Due to the development of its technology, how to interfere with important targets such as intelligent recognition of vehicles has become a hot issue in the current technical field. Therefore, this study is based on the near-infrared band. We generate vehicle target confrontation samples to interfere with infrared detector devices. This paper proposes to first collect the single-band image of the protection target through the infrared UAV, and aligns it to align the position of the protection target in the single-band image to obtain the salient features of the protection target. Then, based on the loss function calculated by machine learning method, the adversarial samples of single-band images are generated by using the basic iterative method. The constraint function is used to jointly optimize the loss function, so that the attack of the adversarial samples on the single-band detection system of the infrared UAV can reach the target value, and the adversarial samples of the infrared image in this state are obtained. Then, it is arranged in multiple positions on the target vehicle to complete the interference to the infrared detection system. We first need to complete the simulation attack experiment in the digital domain, and then complete the confrontation with the infrared detector by using a special material that meets the requirements in the physical domain. The experimental results show that the attack success rate of this method has a certain effect. This paper has a certain reference for the future research of vehicle infrared reconnaissance and anti-reconnaissance.