<p>Traffic sign detection is important in intelligent transport systems such as autonomous and assisted driving. However, traffic sign detection suffers from a slight occlusion problem in snowy environments, which leads to long computation time of the detection algorithm and unsatisfactory detection rate. In order to solve these problems, this paper introduces the Ghost module to replace the Bottleneck module in the C3 module to obtain the C3Ghost module in the YOLOv5s model to reduce the computational redundancy and the number of parameters, and improve the inference speed. Secondly, the CA attention mechanism is introduced into the neural network to enhance the regression and localisation ability of the model by embedding the location information to extract important features, so as to improve the ability of the network to extract accurate location information. And the NWD loss function is used instead of the IoU loss function to improve the detection accuracy and stability of the model and ensure that the model can better capture and distinguish small target features. By comparing the results of the TT00k dataset with YOLOv5s, the computational loads (CLOPs) of the improved model are reduced by 22.5%, the model parameters are reduced by 21.9%, the mAP50 is improved by 1.7%, and the FPS is improved by 7.3.</p>