基于改进YOLO v5的野外实景视频水鸟检测方法OACSTPCD

This study introduces a real-time automated method YOLO v5_k-mixup for waterfowl detection,utilizing the YOLO v5 framework to achieve rapid and accurate identification of waterfowl under field video surveillance.The method incorporates a Mixup data enhancement module into the YOLO v5 network,effectively enhancing its ability to generalize and identify shielded waterfowl from each other.Additionally,to overcome difficulties caused by variations in waterfowl sizes,the paper introduces an approach based on k-means++clustering anchor frames to enhance the positioning accuracy of the detection frame.Compared with the unimproved YOLO v5 model,YOLO v5_k-mixup achieves an average increase in accuracy from 84.8%to 87.1%while maintaining high detection speeds.The enhanced model demonstrates high precision in recognizing and locating waterfowl even in complex environments with dense occlusion,showing strong robustness.