Image segmentation of track bolts based on improved YOLO model.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Railway Engineering

Background:

• Existing railway track bolt detection methods struggle with accuracy and robustness in complex environments.
• High false detection rates and poor performance under varying illumination and contamination hinder practical deployment.

Purpose of the Study:

• To develop an improved YOLO-based segmentation algorithm for accurate and robust railway track bolt detection.
• To enhance the safety and efficiency of high-speed railway maintenance through automated condition monitoring.

Main Methods:

• Proposed an improved YOLOv11 algorithm incorporating a PCM feature extraction module and an F2 feature fusion module.
• Integrated an L-Segment segmentation head to replace the traditional one for enhanced performance.
• Evaluated the model on the Cityscapes dataset and compared it with mainstream lightweight segmentation models.

Main Results:

• Achieved a 1.9% increase in mAP50 compared to the original YOLOv11 on the Cityscapes dataset.
• Demonstrated superior performance against YOLOv5n, YOLO8n, RTMDet-Ins-tiny, and Gold-YOLO.
• Significantly reduced false detection rates and improved robustness in complex conditions with a competitive FPS of 116.4.

Conclusions:

• The proposed algorithm offers an effective and reliable solution for automated track bolt condition monitoring.
• The enhanced model is suitable for edge deployment in railway maintenance, supporting intelligent maintenance practices.
• This work contributes to improving the safety and efficiency of high-speed railway operations.