Discrimination of insulator contamination grades using information fusion of multi-light images

Flashover occurs more easily on contaminated insulators, which causes great economic losses and has bad effects on power system stability. A new method is introduced to discriminate the contamination grades of high voltage insulators, using information fusion of visible images and infrared images of the insulators. Firstly, three heating models of running insulators-No Dry Band, Dry Band and Dry Band Arc are established and solved by numerical analysis method to find out the differences in temperature rise under different running states. Secondly, we smear 10kV insulators in different contamination grades. Then we take their visible images by camera and infrared images by infrared thermal imager after applying the rated AC voltage for hours in artificial fog chamber in different relative humidity. After image processing, we extract 36 disk color features from the visible images and 7 disk temperature features from the infrared images, and screen out the best features using Fisher discrimination. Then, two BP (Back Propagation) neural networks are trained, one of which takes the selected visible image features as the inputs while another uses the selected infrared image features and relative humidity. Furthermore, we fuse the information of visible images and infrared images on feature level, which raises the accuracy in identifying the sample images. This method may provide effective information for insulator cleaning and flashover prevention.