Small object detection algorithm for aerial photography based on improved YOLOv5

Aiming at the problems such as high missed detection rate and inaccurate positioning caused by complex background, large number and dense distribution of small targets in UAV aerial photography target detection tasks, this paper proposed an improved YOLOv5 based aerial small target detection model. Firstly, the reparameterized structure RepVGG is introduced into the backbone network, and the multi-branch structure is used to enhance the feature extraction capability of the model. Secondly, a hybrid domain Attention mechanism (ECA DSAM Model, EDM), Efficient Channel Attention mechanism (ECA) and Dilated Spatial attention Model (Dilated Spatial Attention Model) are designed. EDM). Experiments on UAV aerial photography data set VisDrone show that m AP0.5 of the original YOLOv5 model has improved by 2.7%, and the improved method has better performance in the detection of small targets in aerial photography.