UAV Target Detection Algorithm Based on YOLOv4-tiny
Unmanned Aerial Vehicle(UAV)is of high research value as an emerging tool for information and material transmission.Limited by its hardware conditions,the target detection algorithm embedded in UAV requires lightweight models.The problems of large variation in target scale,image aberrations and target occlusion when UAV detecting targets,leading to failure in detection.To address the issue of low accuracy in drone target detection,an improved algorithm based on YOLOv4-tiny is proposed in this paper.It is based on the YOLOv4-tiny algorithm model,fusing the recursive feature pyramid to enhance the semantic expression of the features,and designing the feature conversion and feature fusion module,which can fuse the deep and shallow feature,to enhance algorithm performance and improve algorithm accuracy.After training and testing on Visdrone dataset,the mAP value reaches 0.146,which is better than other lightweight algorithms in the same class.
NETL
NSTL
万方数据
