SR-YOLOv8: Small Object Detection Method for Traffic Scenes

Small object detection is one of the key challenges faced by object detection in traffic scenes. In this paper, we propose SR-YOLOv8, a small object detection algorithm for traffic scenes. Firstly, we introduce the SPD-Conv architecture to solve the problem of feature information loss in the downsampling process of YOLOv8. Then, the ResBlock-CBAM module is designed to enhance the ability of YOLOv8 to extract important features. Finally, we construct the SR-YOLOv8 network structure for small object detection in traffic scenes. The effectiveness of the proposed algorithm is proved on the large-scale small object detection dataset SODA, the mAP50 and MAP50-95 are increased by 12.7% and 8.6% respectively, and the model complexity is much lower than other object detection models.