Method for detecting pollution degree of insulator based on Locality Preserving Projections Extreme Learning Machine

The Composite insulator is subjected to environmental influences, its surface can accumulate pollution, which can lead to an increase in flashover accidents. For the measurement of insulator surface pollution, the traditional detection methods have a certain degree of defects. In this paper, we use hyperspectral technology to detect the surface pollution of insulators from the perspective of high accuracy and non-contact. In this paper, firstly, artificially polluted insulator samples are produced, and then the data obtained by hyperspectral technology are subjected to black-and-white correction, standard normal variable distribution (SNV), and Savitzky-Golay convolution (SG) algorithms to reduce the interference of noise in the images and dark inrush current on the hyperspectral imaging results. Finally, we use locality preserving projections extreme learning machine (LPP-ELM) to achieve the classification of insulators with different degrees of pollution with an accuracy of 98.33%.