A UAV Detection Algorithm Combined With Lightweight Network

Unmanned Aerial Vehicles (UAVs) are characterized by the small scale and fast flying speed, which brings certain challenges to the detection and supervision of UAVs. In response to this problem, this paper proposes an improved detection algorithm combined with a lightweight network for objects with fast speed. First of all, the original three feature scales that the You Only Look Once(YOLO) v3 algorithm detects are extended to five feature scales for detection, which improves the detection performance of scenes of small objects such as UAVs. Secondly, combined with the Ghost module in the lightweight network, a lightweight feature extraction network is constructed, and a series of cheap operations like linear transformations are applied to generate more feature maps with fewer parameters. To further improve the detection performance of the network, the channel attention mechanism is added to suppress the adverse information. Besides, a UAV dataset with an urban background for training and testing is made. The experimental results show that the improved method proposed in this paper effectively improves the detection accuracy of UAVs in a complex background of cities, meets the real-time requirements, and realizes the lightweight of the object detection algorithm, which provides the possibility to realize the object detection on the embedded platform.