增加下采样分支和替换解耦头的无人机图像小目标检测算法OA

In a complex environment,there are a large number of small targets in the images captured by high-altitude drones.When using YOLOv7-tiny for target detection,there are problems of low detection accuracy and missing detection.Therefore,a small object detection algorithm for UAV images is proposed,incorporating an added subsampling branch and a replaced decoupling head.A small target detec-tion layer is added to the Neck part of YOLOv7-tiny network structure to make full use of the image information for the low-scale feature map.The original coupling head is replaced with a decoupling head,which will help improve the accuracy of target classification and posi-tioning.SENet module is introduced to enhance the context information of the image and increase the attention to small objects,effectively improving the problem of missing detection.CIoU is replaced with Focal EIoU to optimize the loss function,making the calculation more efficient and increasing network robustness.A comparative experiment is carried out by using the public dataset of VisDrone2019 for small target detection captured by drones.The experimental results indicate that the improved algorithm has reduced the false negative rate,im-proved the precision of object detection,and achieved a recall rate of 4.2 percentage points higher than the YOLOv7-tiny algorithm.Addi-tionally,the mAP@0.5 has increased by 2.9 percentage points compared to the YOLOv7-tiny algorithm.