Track Classification of UAVs Based on YOLOv5 for Civilian Surveillance Radars

Unmanned Aerial Vehicles (UAVs) pose growing threats to airspace security. This paper presents a novel approach to classify UAVs, which leverages the You Only Look Once version 5 (YOLOv5) algorithm for real-time object detection, employing a feature set extracted from radar tracks. The data are pre-processed and augmented to enhance model performance. Experimental results show the method's high accuracy of 93%, making it suitable for practical applications. Experimental results on a realistic dataset collected from a low-altitude civilian radar system show the effectiveness of the proposed method.