Semi-Supervised Deep Learning in High-Speed Railway Track Detection Based on Distributed Fiber Acoustic Sensing

Area of Science:

• Railway Engineering
• Sensor Technology
• Artificial Intelligence

Background:

• Traditional high-speed railway track detection methods face challenges including high costs, safety risks, and time delays.
• Optical sensor-based approaches offer improved deployment cost and real-time performance for railway monitoring.
• Deep learning excels with large sensor datasets but requires extensive labeled data, which is difficult to acquire in operational railways.

Purpose of the Study:

• To develop an effective and cost-efficient track detection method for high-speed railways.
• To address the challenge of acquiring labeled data for deep learning models in operational railway environments.
• To leverage distributed optical fiber acoustic sensing (DAS) data for improved railway track monitoring.

Main Methods:

• Deployment of a distributed optical fiber acoustic system (DAS) on a high-speed railway track segment.
• Development of a semi-supervised deep learning model for track detection based on image recognition.
• Implementation of specialized pre-processing techniques for the acquired DAS dataset.
• Utilization of a greedy algorithm for optimizing model hyper-parameters.

Main Results:

• The proposed semi-supervised deep learning method demonstrated superior performance in track detection.
• The approach proved effective in both experimental settings and real-world railway applications.
• The system provided a viable alternative to traditional methods, enhancing efficiency and reducing costs.

Conclusions:

• Semi-supervised deep learning combined with DAS offers a promising solution for high-speed railway track detection.
• The developed method effectively mitigates the need for large amounts of labeled data.
• This approach enhances the safety, efficiency, and cost-effectiveness of railway infrastructure monitoring.