Detection of polluted insulators using the information fusion of multispectral images

Noncontact evaluation of the pollution condition of insulators can be carried out by the infrared, ultraviolet, and other image detection methods, but these methods are difficult to be used effectively in practice because of their low accuracy. The objective of this paper is to improve the accuracy of image detection on the polluted insulators. A method of multispectral image is proposed by fusing the information of the visible light, infrared, and ultraviolet images. Firstly, the principles of multispectral image detection are analyzed, and then the insulators are polluted artificially. While under the working voltages of these polluted insulators, the multispectral images are obtained. After image processing and feature extraction, the relationship between the features of images and the ESDD (Equivalent Salt Deposit Density) of polluted insulators is analyzed. The influence of environmental relative humidity on the ability of the images to distinguish the pollution condition of insulators is studied. On this basis, a multispectral image information fusion method based on variable weights is proposed. Finally, the accuracy is tested. When using the multispectral images to evaluate the ESDD of insulators, the average error is about 0.0125 mg/cm2, significantly lower than that of the single and double spectral images. The test error of proposed method is also less than the other methods compared in this paper. Therefore, it can be concluded that the multispectral image information fusion method based on variable weights is effective on detecting the polluted insulators.