Improved YOLOv8-Based Target Precision Detection Algorithm for Train Wheel Tread Defects.

Area of Science:

• Railway engineering
• Computer vision
• Artificial intelligence

Background:

• Train wheel integrity is critical for operational safety.
• Current image-based defect detection faces challenges with water stains and small defects.
• Timely defect identification enables conditional repair strategies.

Purpose of the Study:

• To improve the accuracy and efficiency of train wheel tread defect detection.
• To address limitations of existing methods, specifically misdetection of water stains and small defects.
• To enhance the YOLOv8 model for superior performance in identifying wheel tread anomalies.

Main Methods:

• Optimized the detection layer structure to mitigate water stain interference.
• Introduced an improved Spatial Pyramid Pooling Cross-Stage Partial (SPPCSPC) module for enhanced small target detection.
• Implemented the SIoU loss function to accelerate network convergence and improve accuracy.
• Validated the enhanced YOLOv8 model on a custom-built wheel tread defect dataset.

Main Results:

• The enhanced YOLOv8 model significantly outperformed the original network.
• Achieved high detection performance with average precision of 96.95%, accuracy of 96.30%, and recall of 95.31%.
• Demonstrated robust detection capabilities, overcoming challenges posed by water stains and small defects.

Conclusions:

• The proposed YOLOv8 enhancement offers a more reliable and efficient solution for train wheel tread defect detection.
• The optimized model contributes to improved railway safety through accurate and timely maintenance.
• This approach provides a strong foundation for advanced automated inspection systems in the rail industry.