Low-altitude UAV Detection Method Based on One-staged Detection Framework

Abstract

Illegal flight of unmanned aerial vehicles (UAVs) poses serious threats to the public and national security. With the characteristics of small size and low flight height, UAVs are difficult for the traditional air-defense system to detect. Therefore, to deal with the illegal UAV flight, this paper proposed a state-of-the art low-altitude UAV detection method. Firstly, a large-scale UAV data set including multiple kinds of UAVs is collected and constructed. Then, based on one stage detection framework, the UAV detection (UAVDet) network is presented with the improvement of more detection scales, utilization of focal loss and specific data augmentation. Experiment results show that the proposed UAV detection method has significant improvement on UAV detection performance, and it is competent to achieve real-time and effective UAV detection.