Improved YOLOv8 Remote Sensing Image Vehicle Target Detection

Vehicledata are essential indicators for evaluating urban economic activities, planning, development, and traffic conditions. Therefore, accurate vehicle detection in remote sensing images is crucial. Traditional object detection models often struggle with the unique characteristics of vehicles, such as varied detection sizes, complex scenes, frequent occlusions, and indistinct detection targets, which leads to incorrect positive identifications and missed detections. In this paper, an algorithm to improve the network structure of YOLOv8 is proposed to improve the above problems. Firstly, a layer was increased to perform feature fusion on the 160×160 feature map of the backbone part and output it to detect small vehicle targets. In addition, a detection head for small target detection of vehicles was added. Test results show that the way shows good performance on UCAS-AOD datasets. The results after many tests show that the proposed algorithm improves the accuracy of vehicle detection in remote sensing images.