<p>Rail track fastener detection faces edge feature degradation under complex conditions. However, existing research pays little attention to optimizing detection performance in edge feature degradation scenarios. This study proposes the YOLO-GEH algorithm, which uses an improved YOLO11 framework to robustly detect track fasteners. To solve the information loss caused by edge feature degradation, we designed the gradient-enhanced hierarchical edge control (GEH) module in the backbone network. Through a three-stage edge control mechanism of generation-preservation-fusion, it directionally strengthens and spreads edge features across layers, effectively relieving the gradient disappearance of shallow features in degradation scenarios. To boost feature discrimination under shape-deformation interference, we introduced the convolutional block attention module (CBAM). It strengthens key-area responses via adaptive weight allocation. Experimental results show that our YOLO-GEH model significantly improves detection accuracy. Specifically, the average accuracy increases by 2.1%. In the three defect detection tasks of deflection, occlusion, and shape deformation, the accuracy improves by 1.1%, 4.3%, and 8.1%, respectively. This proves our method is effective in accurately detecting rail track fasteners with edge feature degradation. The development and deployment of this model are underpinned by high-performance computing (HPC) paradigms, which are crucial for managing the computationally intensive training on large-scale datasets and enabling real-time, parallel processing for railway inspection systems.</p>