Research on Real-Time Disconnector State Evaluation Method Based on Multi-Source Images

Infrared imager is widely used in substation inspections for early fault warning. However, infrared images have less texture information and large imaging differences in multiple states, which may lead to false or missed detection. This article proposed a disconnector state evaluation method based on infrared and visible images. This method combines the advantages of these two data sources and realizes individual-level temperature analysis. First, an optimized model disconnector-YOLOv4 (D-YOLOv4), with fewer model parameters and better detection performance than YOLOv4, is proposed to identify and locate the disconnectors in visible images. Experiments show that mean Average Precision (mAP)@0.5 of the optimized model is 92.86% and frames per second (FPS) is 105. Second, the detected area is mapped to the infrared image by image registration. Here, to address the problem arising from the different imaging characteristics of visible and infrared images, edge detection and affine transformation are introduced to extract the common features and complete image transformation. Third, we fit a $T$ – $g$ function to convert RGB value to temperature, and then the final state evaluation is conducted depending on the required standard. Experiments prove the feasibility and superiority of the proposed algorithm, which adapts the development demand of intelligent power substations, and the tendency of predictive maintenance.