一种轻量型YOLOv5交通标志识别方法

针对为提高交通标志识别精度使得神经网络层数过深从而导致实时性不佳的问题，提出了一种轻量型YOLOv5交通标志识别方法。首先采用遗传学习算法和K-means聚类确定适合交通标志识别的锚框，然后引入Stem模块和ShufflenetV2的基础单元网络来替换YOLOv5的主干网络。相比于YOLOv5模型，在中国交通标志检测数据集上，轻量型YOLOv5模型在保持识别精度为95.9%的同时，参数量减少了95.4%，实际内存空间减少了93.9%，在GPU和CPU上运行的速度分别提升了79.7%和75%，极大地提高了交通标志识别的实时性，更适合无人驾驶环境感知系统的部署。

A light-weight YOLOv5 traffic sign recognition method

For the problem that the number of layers of neural network is too deep to improve the accuracy of traffic sign recognition,which leads to poor real-time performance,a light-weight YOLOv5 traffic sign recognition method is proposed.The anchor frame suitable for traffic sign recognition is firstly determined by genetic learning algorithm and K-means clustering,and then the Stem module and ShufflenetV2 base unit network are introduced to replace the YOLOv5 backbone network.Compared with the YOLOv5 model,the light-weight YOLOv5 model reduces the amount of parameters by 95.4% and the actual memory space by 93.9% on the Chinese traffic sign detection dataset while maintaining the recognition accuracy of 95.9%.Moreover,it runs 79.7% and 75% faster on GPU and CPU respectively,which greatly improves the real-time performance of traffic sign recognition and is more suitable for the deployment of unmanned environment awareness system.