UAV Detection in Multi-Motion Targets Environment Based on CA-YOLOv5-FE Network
Dongyang Jin, Xuzhe Wang, Feng He, Jun Ma · 2025
The number of accidents of collisions between UAVs (unmanned aerial vehicle) and other flying targets especially birds increase year by year, which makes the flying activity become a dangerous event. Thus, it is more critical to correctly distinguish UAVs from birds and other flying targets in the airspace surveillance. To solve this problem, this paper proposes a new UAV detection scheme, CA-YOLOv5-FE, which is skilled in detecting UAVs in birds interference environment. This detection scheme is constructed based on improving the classical YOLOv5. Three new improvements is added to the original YOLOv5: first, the BiFPN feature fusion structure is added to the feature extraction part to enhance the feature extraction ability of the network, second, the Coordinate Attention (CA) mechanism is added to the backbone network part, third, the Focal and Efficient-Intersection over Union (F-EIoU) loss function is added to the header part to provide robustness training. These three modules enhance the distinguished ability of the original network for similar flying targets. What's more, a multi-flyting-target dataset with both UAVs and birds are simultaneously constructed by ourselves, which are used to train and verify our detection networks. Experimental results reveal that the proposed algorithm achieves good detection results in environments where UAVs and birds coexist.