Method for detecting the pollution degree of naturally contaminated insulator based on hyperspectral characteristics

Abstract

Abstract Insulator pollution degree detection is of great significance for preventing a flashover. Equivalent salt deposit density, leakage current, and surface pollution layer conductivity are commonly used to describe insulator pollution degree; however, all these parameters have limitations in field application and real‐time monitoring. Non‐contact detection methods, such as infrared thermal imaging and ultraviolet imaging, only image insulators in a specific band, which makes the extracted features limited. Hyperspectral technology is the new comprehensive image data technology based on imaging spectroscopy that has the advantages of multiband high resolution. Therefore, a method based on hyperspectral image and spectral characteristics is proposed to fully characterize natural pollution information and accurately detect the pollution degree of insulators. The hyperspectral spectral line characteristics, image texture, and colour characteristic data of insulators were extracted and fused and then used to establish the pollution degree detection model on the basis of integrated learning classification algorithms. The results show that the model based on fusion data has an accuracy rate of 95.0%, which is more accurate than the model based on spectral line features only. Consequently, hyperspectral technology can realize the non‐contact detection of pollution degree and provide some guidance for cleaning the contamination of external insulation.

Keywords

• flashover
• image texture
• infrared imaging
• insulator contamination
• leakage currents