Infrared image detection and deep learning-based fault recognition method for electrical equipment: intelligent diagnosis of abnormal heating in distribution cabinets

Abstract

Abnormal heating in wiring terminals, busbars, circuit breakers, and fuses is a key early symptom of electrical faults in distribution cabinets, but manual infrared inspection is sensitive to experience, ambient temperature, load variation, and weak hot-spot contrast. This paper proposes a deep learning-based infrared diagnosis method for abnormal heating recognition in distribution cabinets. The method builds a closed-loop framework consisting of thermal-field normalization, infrared image enhancement, key-component detection, hot-spot segmentation, and fault-level assessment. Relative temperature difference and local thermal gradient are first combined to enhance weak thermal anomalies. An improved YOLO-IR network is then constructed by adding a shallow small-object branch, multi-scale feature fusion, and channel-spatial attention to strengthen component localization and hot-spot detection. Finally, a lightweight segmentation module extracts abnormal regions, and an interpretable risk score integrates temperature difference, hotspot area ratio, detection confidence, and component risk weight. Experiments on a self-constructed infrared dataset show that the proposed method achieves an mAP@0.5 of 94.6%, a segmentation mIoU of 88.7%, and a fault recognition accuracy of 95.2%, supporting accurate and deployable intelligent inspection of distribution cabinets.