Study On Detection Of Conductance On Contaminated Insulatiing Surface Based On Hyperspectral Image Technology

With the continuous advancement of the national strategy of"West-East electricity transmission project"and"global energy internet",in recent years,the grid company has built many UHV power transmission and transformation projects,which have put forward new tests for the safe and reliable operation of insulators.Contamination flashover is one of the main types of accidents for insulators.At present,the traditional insulator contamination detection method has the disadvantages of complicated operation,low precision,and difficulty in field detection.Hyperspectral technology enables material discrimination by detecting optical characteristics of material surfaces,it has the characteristics of wide-area,fast and non-contact,it can be“image and spectral line considerations”,and widely used in remote sensing,geology,agriculture and other fields currently.In this paper,the spectroscopic information of the contamination is analyzed by hyperspectral imaging of the surface contamination of the insulator,and the method of contamination detection based on hyperspectral technology is proposed.Obtain different contaminated conductance values by changing the contamination level and water content of the sample,hyperspectral imaging of samples using a hyperspectral imaging test platform in the laboratory,and combined with the principle of hyperspectral technology,the sample spectral response of different contaminated conductance is analyzed.The study found that changes in the degree of contamination and water content lead to a certain regular change in the hyperspectral line.Therefore,it is possible to analyze the hyperspectral information of the composition,content,and degree of wetness of the contamination,to realize the detection of the conductivity value of the stain layer.In this paper,the acquired hyperspectral spectral lines are preprocessed by Savitzky-Golay smoothing and multi-scattering correction to achieve denoising,smoothing and reducing the effects of ambient scattered light.Segmentation of the region of interest for the image of the sample,using the successive projections algorithm,principal component analysis based on MATLAB for spectral line,and principal component analysis based on ENVI for image,feature bands are extracted to achieve data dimensionality reduction,information on 15,4,and 6 characteristic bands was extracted from the original spectra of256 bands.Aiming at the spectral line preprocessing method and feature band extraction method,the PLSR model and BPNN model are used to verify the spectral line processing method and establish a prediction model.This study found that the optimal spectral processing methods are different for different modeling methods.For the insulator contamination studied in this paper,the SPA is used to extract the feature bands,and the prediction effect in the input BPNN model is the best,R_p can reach 0.9624.The gray-level co-occurrence matrix is used to extract the texture features of the sample,and the second moment,moment of inertia,entropy and correlation parameters of the region of interest are obtained,as a feature vector fusion feature band input BPNN,the results show that the fusion image and spectral line features can effectively improve the prediction accuracy of the model.The constructed“original spectral-SG smoothing-MSC-SPA+texture feature parameters-BPNN”prediction model has an R_p of 0.9702.