Abstract: In order to realize the accurate recognition of traffic signs by autonomous vehicle, a traffic sign image recognition algorithm YOLO-slim based on improved YOLOv4-Tiny is proposed. Convolution attention module was added to the original network and shallow features were introduced into the feature pyramid network to improve the utilization rate of feature information between different layers. Depthwise separable convolution was used to replace standard convolution to reduce the number of network parameters and compress the model weight file. Focus loss function was used to balance difficult samples in model training. Experimental results show that YOLO-slim's mean average precision is 94.41%, weight file is 4.49 MB, and detection speed is 8.0 ms. The improved algorithm has higher accuracy and smaller weight files, and is more suitable for deployment in vehicle-mounted computing units.