Abstract
High voltage (HV) insulators operate under different environmental conditions such as wind, snow, icing, unexpected incidents and tolerate overvoltage or flashovers. An undesirable performance of HV insulators might occur due to the pollution layers formed over the insulator surface that function as conducting coatings in a humid environment or even in normal dry operational conditions. This may lead to the insulator breakdown. Thus, detection of the insulator's defect in the early stage is essential. From a supervised learning point of view, accurate classification of contaminated HV insulator surface requires: i) collecting insulator surface images under different conditions; and ii) developing classifiers of contamination given a surface image. To this end, in this study, we used Unmanned Aerial Vehicle (UAV) to collect several insulator surface images under different contaminations including water drop (water spray), cement, snow, soil, and a mixture of snow and water. We then employed convolutional neural networks (CNNs) to construct accurate classifiers of contamination. In developing our CNN-based classifier, we use a grid search model selection and compare the performance and efficiency of the constructed model with pre-trained CNN models that are fine-tuned on our dataset.