Outdoor detection of the pollution degree of insulating materials based on hyperspectral model transfer

Abstract

Accurate detection of the insulator pollution degree is the premise to realize pollution flashover warning. In the laboratory, the pollution degree can be accurately evaluated by using hyperspectral imaging technology (HSI). However, laboratory model is difficult to be applied to outdoor due to the influence of natural light. Therefore, this study attempts to combine HSI and model transfer to extend the laboratory model to outdoor application. Spectra of pollution samples were collected indoor and outdoor. Multivariate scattering correction and the Savitzky–Golay filter were adopted to eliminate spectral noise. Laboratory model was established based on back propagation (BP) neural network, the direct standardization algorithm was used to model transfer and its effect was verified by principal component analysis (PCA). After the model transfer, the accuracy of outdoor detection reached 82.5% in cloudy day and 80% in sunny day, which is significative for the field non-contact detection of insulator pollution degree.