Image segmentation of track bolts based on improved YOLO model

Railway track bolt detection is critical for high-speed railway safety maintenance, yet existing methods often suffer from high false detection rates and poor robustness under complex illumination, heavy contamination, and varying scale conditions, limiting their practical deployment. To address these issues, this paper proposes an improved YOLO-based segmentation algorithm for railway track bolt images. Taking YOLOv11 as the baseline, three key optimizations are introduced: (1) A PCM feature extraction module is designed to replace the original C3K2 module in the backbone, enhancing feature representation while maintaining computational cost, which improves the mAP50 by 0.8% on the Cityscapes dataset; (2) An F2 feature fusion module is added to the neck network to strengthen multi-scale feature focusing, boosting the mAP50 by an additional 0.7%; (3) The L-Segment segmentation head is adopted instead of the traditional one, which further increases the mAP50 by 0.4% while effectively reducing floating-point operations. Experimental results show that the proposed model outperforms the original YOLOv11 by 1.9% in mAP50 on the Cityscapes dataset, and also achieves superior performance compared with mainstream lightweight segmentation models including YOLOv5n, YOLO8n, RTMDet-Ins-tiny, and Gold-YOLO. In practical scenarios, the proposed method significantly reduces false detection rates under complex illumination and contamination conditions, enhances the segmentation robustness against environmental interference, and maintains a competitive FPS of 116.4, making it suitable for edge deployment in railway maintenance. This work provides an effective and reliable solution for automated track bolt condition monitoring, supporting intelligent maintenance and improving the safety and efficiency of high-speed railway operations.