In order to improve the comprehensive performance of the traffic sign system, this paper proposes a lightweight and efficient network model for the existing traffic sign recognition system, which is characterized by the complex structure of the convolutional neural network, the number of parameters and the computational volume is too large, and the model is not lightweight enough. The model chooses the lighter MobileNetV1 as its backbone network, and redefines the output layer of MobileNetV1 to effectively reduce the overall number of parameters of the network, so as to achieve the purpose of simplifying the network structure. The experimental results show that the proposed model can reduce the number of parameters by 25.5%, 37.9% and 53.8%, compared with YOLOv5n, YOLOv8n and YOLOv3 respectively, and the model complexity/computation (GFLOPs) and image processing speed (FPS) show outstanding advantages compared with several models, under the premise that the detection accuracy of the proposed model is considerable. show outstanding advantages. Finally, by building an unmanned vehicle testbed configured with STM32F103 CPU and OV2640 camera, the model pre-training weights are deployed on the hardware testbed for real testing, and the results show that the proposed lightweight network model meets the requirements of traffic sign recognition accuracy and improves the model's operation efficiency on the computationally resource-constrained devices.