Aerial image object detection method based on improved YOLOv5

The object features from the drone's perspective are poor, there is a lot of noise, the object scale changes drastically, and densely arranged objects, which brings huge challenges to object detection, and there are problems such as missed detection and false detection. Based on this, this article proposes an improved YOLOv5 UAV image object detection method. This method first introduces the RepVGG structure to deeply mine and enrich the semantic information of different features to alleviate the problems of poor object features and excessive noise in complex backgrounds; then it introduces the SKAttention attention mechanism to improve feature differences such as object occlusion, background interference, and multi-scale objects. representation ability under various circumstances; finally, the F-EIOU regression loss function is introduced to improve the regression speed and improve the recognition accuracy of noisy objects. Extensive experiments were conducted on the VisDrone2019 UAV aerial photography data set. Experimental results show that the average accuracy of the improved YOLOv5 (mAP@0.5) increased by 5.4%, and mAP@0.5:0.9 increased by 3.4%.