Research on the Performance of ER Model Integrated with Deep Network in Traffic Signal Recognition

With the swift growth of the social economy and the extensive adoption of autonomous driving technology, greater demands are being placed on the high performance and high accuracy of traffic light recognition, which is the key to ensure traffic safety. This study proposes an ER-Net model that combines ResNet50 and EfficientNet-B0 to improve the effect of traffic sign recognition. ResNet50 's residual connection effectively extracts advanced features. EfficientNet-B0 improves network efficiency and captures detailed features through a composite scaling strategy, and then adds an attention SE module, which can dynamically adjust the weight of each channel, enhancing the capture of crucial features in complex scenes. The combination of the two makes the model significantly better than traditional machine learning models such as SVM, KNN and Random Forest in terms of accuracy and classification results. Using the Chinese Traffic Sign Database ( CTSDB ), after a series of data preprocessing, the ER-Net model has a recognition accuracy of up to 99.39 %, while the traditional machine learning is only 54.37 %. The former has obvious advantages in computational efficiency and response speed, and has wide application potential.