Research on Remote Sensing Image Target Detection Algorithm Based on YOLOv5

In recent years, Remote-sensing image target detection is a hot research problem of the researchers at home and abroad in the field of remote-sensing image and the target recognition, but because of the large amount of remote-sensing image data and complex, target is obscured in the image and the area of target is so small, the traditional remote-sensing image target detection is not enough to solve the problems in the research stage. Therefore, single-stage target detection based on algorithms such as YOLO series and SSD omits the steps of two-stage detection model, and gains the favor of researchers by sacrificing accuracy for speed. In this paper, by introducing DIOU-NMS and Alpha-IoU, based on the YOLOv5 network structure, the non-maximum inhibition in the original YOLOv5 is replaced by the non-maximum inhibition based on DIOU_Loss, and the original IoU system is replaced by the α-IoU system. The experimental results show that under the same training conditions, compared with the original YOLOv5 network, the improved YOLOv5 network for remote-sensing image target detection mAP@0.5 is increased by 4.2% to 0.775, and the accuracy is increased by 1.0% to 0.946.