基于Transformer与MobileNetv3融合的铁路钢轨表面伤损检测算法OA

Surface defects on railway rails,such as micro-cracks and corrosion,are among the critical factors affecting the safety of train operations.To address the efficiency and accuracy bottlenecks in detecting these minute defects on rail surfaces,this study proposes an intelligent recognition method for rail surface damage based on the collaborative optimization of MobileNetv3 and Transformer.By injecting spatially coordinate-aware features into the channel attention mechanism,an improved CBAM-Bneck module is integrated into the MobileNetv3 architecture,effectively enhancing the network's ability to discern and generalize rail defect features.A novel MobileNetV3-CBTr composite backbone network is constructed,comprising MobileNetv3 base modules,enhanced CBAM-Bneck units,and a lightweight Transformer encoding layer.This design ensures robust feature representation while significantly reducing the model's parameter count.A BiFPN-Lite is introduced to efficiently fuse multi-scale defect features without increasing computational load.The optimized YOLO detection head is then employed for precise localization and classification of rail damages.Experimental results on a self-built rail dataset demonstrate that the proposed algorithm achieves a mAP of 91.8%and a processing speed of 19.5 f/s,representing a 3.5%improvement over YOLOv5.This indicates that the method can effectively accomplish high-precision detection of railway rail surface damages.