YOLOv10 Traffic Sign Recognition Model Based on Lightweight Convolutional GSConv Improved with Multiscale Attention Mechanism MSAA

To improve the precision and efficiency of traffic sign detection, especially in intricate traffic scenarios, we present a streamlined traffic sign detection approach utilizing an enhanced version of YOLOv10. To overcome the accuracy challenges encountered by conventional YOLOv10 in certain contexts, we initially implement a lightweight convolutional module, GSConv, replacing the traditional convolutional components in the backbone and neck layers of YOLOv10N. This modification significantly diminishes computational complexity and accelerates processing speed. Furthermore, we introduce a multiscale attention mechanism, MSAA, to refine feature fusion capabilities and bolster the model’s accuracy and robustness. Experimental findings illustrate that the optimized model attains a mean average accuracy of 83.96% on the TT100K dataset. When compared to YOLOv10-N, the model exhibits an 8.9% reduction in size, a 3.12% improvement in accuracy, and an increase in frames per second to 83.53. The enhanced model retains high accuracy even under challenging lighting conditions. This algorithm not only performs exceptionally well in the traffic sign recognition task but also exhibits significant potential for implementation on embedded devices, rendering it suitable for realworld applications.