A Low-Slow-Small UAV Detection Method Based on Fusion of Range-Doppler Map and Satellite Map

Qinxuan Wang, Haoxuan Xu, Shengtai Lin, Jiawei Zhang, Wei Zhang, Shiming Xiang, Meiguo Gao

Abstract: Unmanned aerial vehicle (UAV) targets are characterized by slow speed, low flying altitude, and small radar cross section, which provides them with great stealth properties. In the urban low-altitude environment, the detection of such UAV targets is facing complex interference. To address these issues, this article proposes a UAV detection method based on the fusion of range–Doppler map and satellite map. According to our survey results, this is the first time that satellite maps have been integrated into the field of radar target detection. The method realizes sample matching between radar and satellite map through the target spatial position detected by radar and simplifies the radar target detection task to a classification problem through deep learning. The training of a neural network model requires massive datasets with reliable labels. Due to the particularity of radar target detection tasks, real radar data are not yet widely available. Especially for distant targets, it is very difficult to accurately label radar data. To address the issue of precise annotation, this article introduces a labeling approach that enables the acquisition of a trustworthy frame-by-frame labeled echo dataset. This article also carries out rigorous comparative experiments with the other advanced methods. The outcomes demonstrate a significant enhancement in the performance of our detection method. The dataset used in the experiment is obtained by the multipulse radar independently designed by our laboratory.