High Speed Railway Fastener Defect Detection by Using Improved YoLoX-Nano Model

Area of Science:

• Railway Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Rail fasteners are critical for high-speed train safety, requiring reliable condition monitoring.
• Existing rail-fastener detection models suffer from poor generalization, large size, and low efficiency.
• Accurate and efficient detection of rail fastener defects is essential for operational safety.

Purpose of the Study:

• To propose an improved YoLoX-Nano model for enhanced rail fastener defect detection.
• To address the limitations of current detection models in terms of accuracy, speed, and generalization.
• To provide a lightweight and efficient solution for real-time rail fastener inspection.

Main Methods:

• Integration of the Convolutional Attention (CA) mechanism into CSPDarknet and Path Aggregation Feature Pyramid Network (PAFPN).
• Implementation of Adaptively Spatial Feature Fusion (ASFF) to enhance feature representation.
• Utilizing the improved YoLoX-Nano architecture for defect detection, including fractured, displaced, and normal fasteners.

Main Results:

• Significant improvements in Average Precision (AP) for fractured (27.42%), displaced (15.88%), and normal (12.96%) fasteners.
• An 18.75% increase in mean Average Precision (mAP) compared to baseline models, outperforming Faster-RCNN.
• Achieved a detection speed of 54.35 frames per second (fps), a notable increase over SSD and YoLov3.

Conclusions:

• The improved YoLoX-Nano model offers accurate and rapid identification of rail fastener defects.
• The model demonstrates superior performance and efficiency, suitable for real-time detection applications.
• This approach provides a valuable reference for lightweight terminal deployment and broader image recognition tasks.