Improved Traffic Sign Detection Algorithm Based on Lightweight YOLOv8n

Traffic sign detection constitutes a vital element in the environmental perception capabilities of autonomous driving systems. Aiming at the problem of reduced detection accuracy for small traffic signs and excessive model complexity in complex environments, an improved algorithm for traffic sign detection based on YOLOv8n is proposed. Specifically, the Squeeze Aggregated Excitation (SaE) module is integrated into C2f to optimize the backbone structure, enabling collaborative enhancement of local details and global semantics and significantly improving the detection of small targets. Then, a lightweight CNN-based Cross-scale Feature Fusion (CCFF) mechanism is adopted, effectively reducing the number of model parameters while enhancing robustness and adaptability to scale variations. In addition, the Weighted Intersection over Union (WIoU) loss function enhances the model's target localization accuracy by assigning adaptive weights to different regions. The experimental results show that the proposed algorithm improves average precision by 14.3% and the parameters decreased by 7.3% on the TT100K dataset. These enhancements not only streamline the model architecture but also enhance detection performance significantly.