<p>Classification accuracy is limited owing to the problem of incomplete feature extraction caused by low resolution in traffic sign images. This study proposes an innovative method that integrates classification and super-resolution (SR) reconstruction networks in a cascading manner. First, a residual multi-scale cross network (RMSCN) is proposed in the SR reconstruction network. This network utilizes multi-scale convolutional blocks to accurately extract and efficiently reconstruct image features. Subsequently, the optimized traffic-sign images after SR reconstruction are input into the classification network, and a secondary classification strategy (SCS) is adopted to achieve a more refined training process. Quantitative and qualitative comparisons are conducted to compare the RMSCN with several classic image reconstruction methods. Finally, experimental verification shows that the images processed by the SR reconstruction network, RMSCN, and SCS implemented in the classification network significantly improve the accuracy of traffic-sign classification, achieving up to 92.3% accuracy. The results of this study can advance traffic-sign recognition and classification technologies that can ultimately improve road safety and traffic efficiency.</p>