Abnormal internal heating in switchgear cabinets is a major hidden hazard that threatens the safe and stable operation of power systems. Traditional infrared monitoring methods have limitations in real-time performance and intelligence. This paper proposes a switchgear abnormal heating detection method based on Switchgear Abnormal Heat Detection Algorithm (SAHDA). Using You Only Look Once version 8 (YOLOv8) as the framework, to address challenges in infrared thermal imaging such as dense targets, scale diversity, and feature similarity, the proposed method adopts a lightweight MobileNetV2 backbone to reduce computational burden, designs a Cross-level Context Fusion Module (CCFM) model to enhance multi-scale contextual feature fusion, and integrates Dynamic detection head (Dyhead) to achieve dynamic perception and adaptive detection. Experiments are conducted on a self-constructed switchgear infrared thermal imaging dataset containing 5,000 abnormal heating images and 600 normal images. The results show that SAHDA achieves detection accuracies of 82.1% and 51.9% in mAP@0.5 and mAP@0.5–0.95, respectively, as well as an inference speed of 31 FPS, outperforming the YOLO series, DETR, the RCNN series, and two domain-specific algorithms in terms of the balance between accuracy and detection speed, demonstrating strong engineering application prospects and providing an algorithmic basis for rapid on-site monitoring of abnormal heating in switchgear cabinets.
Wang et al. (Tue,) studied this question.