Enhancing Traffic Sign Detection with EEM-YOLO for Complex Scenarios

Traffic sign recognition (TSR) is a crucial application of computer vision, using image recognition to identify traffic signs in road scenes. Accurate detection is vital for autonomous driving, especially in complex scenarios with occlusion or environmental interference. Traditional models often suffer from low accuracy, missed detections, and false positives. To address these issues, we propose EEM-YOLO (Efficientformer Enhanced Multi-Detection YOLO). It uses the EfficientFormerV2 backbone, which efficiently extracts spatial features and reduces redundant computations. In the Neck module, eSE attention is integrated with MBConv to enhance feature fusion, and a new C2f_EC design improves regression prediction accuracy. Additionally, the Wisev3-SIoU loss function replaces CIoU to address missing/false detections and gradient issues. Extensive experiments on TT100K and CCTSDB2021 datasets show that EEM-YOLO achieves mAP of 83.1% and 85.9%, respectively, outperforming benchmark models by 9.4% and 7.8%. This demonstrates its superior detection performance and feature extraction capability, making it highly competitive among existing models.