Research on Insulator Fault Diagnosis Based on Multi-Source Feature Fusion U-Net and XGBoost Classification

Infrared imaging enables the rapid detection of abnormal heating in composite insulators for defect identification. To enhance detection efficiency and reduce background interference in manual diagnostics, this paper proposes an automatic detection method based on multi-source feature fusion U-Net. The model learns from infrared image data to achieve precise insulator segmentation. The segmentation results are then used as input for an XGBoost classifier, and finally, a classification method for low-value and zero-value insulator states is developed by integrating the multi-source feature fusion U-Net with the XGBoost-based classifier.