A T-YOLO and Overlapping Reconstruction-Based Method for Rail Fastener Defect Detection in Heavy-Haul Railway

Rail fasteners are vital components of railway infrastructure, playing a crucial role in the safe operation of heavy-haul railways. However, the operating environment of heavy-haul railways is complex. In coal transportation, spilled coal and ballast often obstruct fasteners, causing false positives like mistaken fracture detection and reducing accuracy. Fastener defects account for less than 0.1% of samples, with severe class imbalance hindering precision. Also, line-scan cameras’ limited field of view truncates edge fasteners, leading to missed detections and false alarms. These issues result in a high false alarm rate (FAR) and missed alarm rate (MAR) in existing algorithms, hindering their use in heavy-haul railway maintenance. To address these challenges, this article proposes a novel fastener defect detection algorithm based on a tailored YOLO (T-YOLO) and an overlapping reconstruction strategy. Specifically, we design a T-YOLO architecture by adding a P6 layer to the backbone network, introducing the DySample upsampling operator, and designing a CCNeck feature fusion network, in order to enhance the feature extraction capacity, thereby reducing the FAR during detection. Meanwhile, the quality focal loss (QFL) function is introduced to address the issue of category imbalance. Moreover, to tackle the high FAR and MAR caused by truncated fasteners, we propose an image overlapping reconstruction method combined with a post-removal strategy. Finally, the effectiveness of the proposed method is verified using datasets collected from over 60 km of heavy-haul tracks. Experimental results demonstrate that our approach significantly improves detection performance under complex real-world conditions, notably reducing the FAR from 56.4% to 9.3% and MAR from 24.6% to 0% in fastener defect detection, outperforming benchmark methods.