Traditional traffic sign detection algorithms have attracted growing attention from experts and scholars while improving the recognition accuracy and reducing the missed and/or false detection rate remains a great challenge. Here we proposed a small target traffic sign detection algorithm based on YOLOv8n. The algorithm uses a Conv-SPD module instead of step convolution to downsample and retain shallow feature information. Then adds a small object detection layer, which can effectively improve the model's ability to perceive small objects. Secondly, incorporating a multi-scale attention mechanism to fuse deep and shallow spatial semantic features for better capturing of pixel-level pairwise relationships. To further enhance model detection accuracy, the mode utilizes the MPDIoU loss function to compute the regression loss of the predicted box. Finally, it is verified on the data sets TT00K, GTSDB and CCTSDB. Experimental results show that the detection accuracy of this model reaches 87.3%, 93.2% and 98.4% respectively, and the parameter size is only 2.031MB, while meeting the real-time detection standards.