UAV Target Detection Method Based on Multi-detection Head and Attention in Urban Background

In order to improve the detection ability of UAVs in complex scenarios in UAV prevention and control tasks, this paper proposes an improved YOLOv8 target detection algorithm based on multiple-detection head and attention enhancement. Firstly, based on the original network of YOLOv8, the feature extraction layer is added, the feature information of the shallow network is combined to retain more details of the UAV, and a new detection head is added to the Head network to improve the detection accuracy of small targets. Secondly, in the deep network containing the feature information of small targets, the attention module CBAM combining the two dimensions of feature channel and feature space is added to improve the detection accuracy and efficiency. Finally, based on the USVEO-20 photoelectric detection sensor, the UAV targets detection verification experiment is carried out. Experimental results show that the improved YOLOv8 model and YOLOv8 identify the same datasets, the detection level and confidence are improved, especially for small target recognition in complex scenes, the effect is better than other models. Compared with YOLOv8, under Det-Fly common datasets, mAP@0.5 andmAP@0.5 and:0.95 of the improved YOLOv8 model are increased by 0.8% and 1.3% respectively. Under the self-made datasets, the mAP@0.5 and mAP@0.5 :0.95 of the improved YOLOv8 model increased by 4.3% and 6.1% respectively. The FPS is greater than 30 frames per second, which meets the real-time requirements of UAV prevention and control.