Research on ResNet Traffic Sign Classification Algorithm Based on Attention Mechanism

In the natural environment, there are a variety of objective conditions, such as weather conditions, light, background, occlusion, morphological changes and observation angles, etc., which will affect the recognition of traffic signs, and then affect the recognition accuracy. To address this issue, this article proposes a method of enhancing the neural network of the standard ResNet model by adding attention mechanisms. On the basis of dividing the selected dataset, the original ResNet network structure is optimized by adding attention mechanisms, and then the improved ResNet is used to train and validate the dataset. The experimental results show that after adding the ECANet attention mechanism, the average training accuracy of the model is 91.83%, the average recall is 88.54%, the average F1 is 88.71%, the average number of iterations per second is 12.516it/s, and the average batch training time is 14.7 seconds. The training and testing performance in the dataset is the best, so the recognition and classification of traffic signs can be completed with relatively high accuracy and efficiency.