Video SAR Moving Target Detection System Based on FPGA

Video synthetic aperture radar (Video SAR) has become a research hotspot in the SAR field due to its characteristic of continuous monitoring. Compared to traditional SAR, video SAR provides the ability to dynamically monitor ground targets. The defocusing and displacement generated during the target motion process result in the radar only capturing occlusion shadows, making shadow detection a new method for motion target recognition in video SAR. In this paper, we propose a multi-frame fusion convolutional neural network (CNN) based on the problem of low detection rate due to shadows being too close to the environmental background. We fully utilize the dynamic information of the video through methods such as video stabilization, histogram equalization, and multi-frame image fusion, effectively improving the detection rate. In addition, we reduce the volume of the neural network through network quantization and calibration methods and successfully deploy it in field-programmable gate arrays (FPGA), achieving low-power, small-volume real-time motion target detection.