SEF-YOLO: An Improved YOLOv5 Algorithm for Traffic Sign Detection

In recent years, autonomous driving technology has been developing at a high speed, and detecting traffic signs is an integral part. Most algorithms are not lightweight enough with a slow reason speed, so an improved SEF-YOLO algorithm is proposed to fulfill the mentioned requirements. The lightweight network ShuffleNetV2 is introduced into the backbone network to compress the model; the Efficient Channel Attention Network (ECA Net) mechanism module is introduced to improve the feature extraction ability of the network and avoid the interference of object prediction by the complex environment; and the Focal_Loss is used to replace the original loss function, which improves the model’s ability to recognize a few categories of traffic signs. Compared with the original YOLOv5s algorithm on the TT100K traffic sign dataset, the parameters of the SEFYOLO model are reduced by 53.7%, the Giga Floating-point Operations Per Second (GFLOPs) is reduced by 61.6%, the mean average precision mean average precision (mAP) is increased by 1.6%, and the frame processing rate is increased 14 FPS. The experimental results thoroughly verify that the improved model is fast and accurate in real traffic sign detection scenarios and meets the lightweight requirement of vehicle deployment.