DP-YOLO: A Lightweight Real-Time Detection Algorithm for Rail Fastener Defects.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning
• Railway Engineering

Background:

• Accurate and efficient real-time detection of rail fastener defects is crucial for railway safety.
• Resource-constrained environments pose challenges for deploying complex defect detection algorithms.
• Existing methods may lack the necessary precision or efficiency for practical railway maintenance.

Purpose of the Study:

• To develop an advanced lightweight algorithm, DP-YOLO, for accurate and efficient real-time rail fastener defect detection.
• To optimize the YOLOv5s architecture for improved performance under resource constraints.
• To validate the effectiveness of DP-YOLO on a relevant dataset for railway applications.

Main Methods:

• Proposed DP-YOLO, a lightweight algorithm based on YOLOv5s, incorporating four key optimizations.
• Introduced a Depthwise Separable Convolution Stage Partial (DSP) module for parameter reduction and accuracy enhancement.
• Implemented a Position-Sensitive Channel Attention (PSCA) mechanism for dynamic feature recalibration.
• Utilized a GhostC3 structure in the Neck network to minimize computational costs.
• Adopted the Alpha-IoU loss function to improve multi-scale adaptability and model robustness.

Main Results:

• DP-YOLO achieved 87.1% detection accuracy on the Fastener-defect-detection Dataset.
• Outperformed the original YOLOv5s by 1.3% in mAP0.5 and 2.1% in mAP0.5:0.95.
• Reduced model parameters by 1.3% and computational load by 15.19% compared to YOLOv5s.

Conclusions:

• DP-YOLO demonstrates significant improvements in detection accuracy and efficiency for rail fastener defects.
• The optimized lightweight architecture is suitable for resource-constrained environments in railway maintenance.
• DP-YOLO offers practical value for high-precision, efficient defect detection systems in the railway industry.