A railway surface defect detection model based on topology-enhanced feature association

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Railway safety and maintenance rely on accurate rail defect detection.
• Current methods face challenges in complex environments due to feature ambiguity and occlusion, causing missed and false detections.

Purpose of the Study:

• To propose a topology-enhanced relational modeling framework for robust rail defect detection.
• To overcome limitations of existing methods in complex environments and improve detection accuracy.

Main Methods:

• Constructed a semantic association graph for dynamic multi-order feature extraction and structured relationship learning.
• Introduced a multi-directional state-space modeling module for long-range dependency capture and enhanced spatial sensitivity.
• Designed a hypergraph-based interaction mechanism with hypergraph convolution for high-order relationship modeling and feature fusion.

Main Results:

• The proposed method consistently outperforms state-of-the-art approaches in rail defect detection.
• Achieved higher accuracy and stronger robustness, particularly under challenging environmental conditions.
• Effectively enhanced the representation of subtle defects and improved discrimination of heterogeneous defect patterns.

Conclusions:

• The topology-enhanced relational modeling framework offers a significant advancement in rail defect detection.
• The method demonstrates superior performance and robustness, crucial for ensuring railway safety and optimizing maintenance.
• This approach effectively addresses challenges posed by complex environments, feature ambiguity, and occlusion.