面向无人机载平台的轻量级小目标检测算法OA

To address the challenge of limited detection accuracy for small targets in aerial imagery,a lightweight detection framework oriented to unmanned aerial vehicle platforms is designed.Specific improvements are as follows:a high-resolution detection head is added to retain fine-grained feature information;the traditional loss function is replaced by an Efficient Intersection over Union(EIOU)loss function to optimize the efficiency of parameter update;and a dynamic sparse attention module is constructed to enhance feature selection capability while reducing computational complexity.Comparative tests conducted on the VisDrone2019 benchmark dataset show that the improved architecture achieves a significant improvement in mean Average Precision(mAP)metrics compared with the original YOLOv5s,with mAP@EIOU=0.5(mAP50)increased by 8.9%.Ablation study shows that each component significantly improved the detection accuracy and inference efficiency for all categories of targets.The optimized framework maintains the lightweight property of the model,and real-time processing performance of 42 Frames Per Second(FPS)is achieved,thus effectively solving the problem of small target recognition in aerial photography scenarios.