Fine-Tuning YOLOv8 for Insulator Defect Detection in High-Speed Railway Systems

Fine-Tuning YOLOv8 for Insulator Defect Detection in High-Speed Railway Systems

Authors- Zhang Zheng, Md Kiron Ali

Abstract-Insulators are essential components in overhead catenary railway systems, providing electrical insulation and mechanical support. But lightning, physical damage, negative weather, and some other external factors that affect their performance, which in turn could interrupt the electricity supply. However, conventional inspection methods are both time consuming and labor-intensive and sensitive to environmental conditions. In order to address these challenges and improve detection performance on small defects, this work presents a deep learning-based approach to automatically detect insulator defects using the fine-tuned YOLOv8n model. The original YOLOv8n model is integrated with the custom loss function, the SGD optimizer, and some other parameters. The proposed model is trained on an unbalanced catenary defect detection dataset that contains seven categories of insulator images: missing, shelter, breakage, contamination, dirt, and the good class. Due to the class imbalance, a variety of data augmentation techniques are applied. We trained our same dataset with other existing methods, comparative results demonstrate that our model performs better than conventional methods, with achieving overall precision 95.3% and recall 93.4%. Experimental results also show excellent performance on contamination, shelter and good categories insulators, and produce promising results on more challenging defects like dirt, breakage, or cracks. In addition, the study also highlights that YOLOv8n can be used to automatically detect and classify insulator defects, which is more efficient and reliable in terms of maintenance and safety of the overhead contact lines.

DOI: /10.61463/ijset.vol.12.issue6.400