Research on Object Detection and Recognition Method for UAV Aerial Images Based on Improved YOLOv5

In this paper, an object detection and recognition method based on improved YOLOv5 is proposed for application unmanned aerial vehicle (UAV) images. Firstly, we the traditional Gabor function to obtain convolutional kernels with better edge enhancement properties. We used eight enhance edges from directions, enhanced image has obvious features, thus providing best area subsequent deep feature extraction work. Secondly, added a coordinate attention (CA) mechanism backbone of YOLOv5. The plug-and-play lightweight CA considers information both spatial location channel features can accurately capture long-range dependencies positions. like eyes YOLOv5, making it easier network find region interest (ROI). Once again, replaced Path Aggregation Network (PANet) Bidirectional Feature Pyramid (BiFPN) at neck BiFPN performs weighting operations different input layers, which helps balance contribution each layer. addition, adds horizontally connected branches across nodes bidirectional fusion structure fuse more in-depth information. Finally, trained overall model our integrated dataset LSDUVD compared other models multiple datasets. results show that convergence effect mAP value, demonstrates unique advantages in processing tasks UAV