A Traffic Sign Detection Model Based on Improved YOLOv8

To address the challenges of insufficient small object detection capability and inadequate multi-scale object processing in complex traffic environments, this study proposes a traffic sign detection model based on YOLOv8. To overcome these challenges, the SPD-Conv module is introduced to enhance the retention of small object features. Additionally, the Deformable Convolution (Denv3) module is integrated to improve feature extraction accuracy in complex backgrounds. Finally, the ASF-SDNet module is utilized to optimize cross-scale feature fusion and channel purification capabilities. Experimental results show that the improved model achieves significant performance improvement on the TT100K dataset, with a mean average precision ([email protected]) of $88.8 \%$, representing a 4.7 percentage point increase compared to the original model. Moreover, the enhanced model demonstrates improved robustness and adaptability in complex environments, while maintaining a parameter count of 13.3 M, indicating high computational efficiency. Comparative experiments with other mainstream models (e.g., Faster-RCNN, YOLOv5s) also highlight the superiority of the proposed model, particularly in terms of recall and precision.