融合RepVGG的YOLOv5交通标志识别算法

准确检测交通标志已成为自动驾驶不可或缺的任务之一。基于现实场景中小而密集的交通标志，传统方式检测交通标志存在精度较低这一缺陷。针对此问题，提出一种融合RepVGG模块的改进YOLOv5的交通标志识别算法。首先将原算法的部分CBS模块替换为RepVGG模块，加强特征提取能力。并在Neck层融合CBAM注意力机制，强化检测模型的抗干扰能力。最后，在网络训练过程中，使用EIOU损失函数来弥补GIOU损失函数的不足，提高算法的检测精度与迭代速度。实验结果表明，改进后的YOLOv5算法，迭代速度更快，在CCTSDB交通标志数据集上的P、R、mAP值分别达到91.55%、85.04%、91.71%，相比YOLOv5算法能够更好的应用到实践当中。

YOLOv5 traffic sign recognition algorithm combined with RepVGG

Accurate detection of traffic signs has become one of the indispensable tasks of automatic driving. Based on the small and dense traffic signs in the real scene, the traditional way of detecting traffic signs has the defect of low accuracy. To solve this problem, an improved YOLOv5 traffic sign recognition algorithm is proposed. First, replaced some CBS modules of the original algorithm with RepVGG modules to enhance the ability of feature extraction. The attention mechanism of CBAM was integrated in the Neck layer to strengthen the anti-interference ability of the detection model. Finally, in the process of network training, EIOU loss function was used to make up for the deficiency of GIOU loss function and improve the detection accuracy and iteration speed of the algorithm. The results of the experiments show that the improved YOLOv5 algorithm has a faster iteration speed, and the P, R, mAP values on the CCTSDB traffic sign dataset reach 91.55%, 85.04%, and 91.71% respectively, which can be better applied to practice than YOLOv5 algorithm.