Guan-Lin Zhang

A Defect Detection Method for Rail Ultrasonic B-Scan Images Based on a Multi-Scale Dense Attention Network

Experimental results show that MSDA-Net achieved an accuracy of 0.909, with a macro Precision of 0.867, macro Recall of 0.845, and macro F1-score of 0.848, demonstrating the effectiveness of the proposed network structure for rail defect classification.