Abstract
The outdoor high voltage insulators are some of the essential parts for electrical and mechanical assistance of transmission lines. Monitoring the insulators’ health frequently on a regular basis is an indispensable routine for ensuring uninterrupted power transmission. Traditionally, the monitoring is accomplished manually by linemen, which is time-consuming and implies additional logistics costs due to the long distances and various terrains associated with transmission lines. One of the most economic solutions can be condition monitoring by analysis of insulators’ images using Unmanned Aerial Vehicles (UAVs). In this regard, this study focuses on outdoor high-voltage insulators’ health monitoring by deep learning-based classification of hazardous surface conditions. To this end, Convolutional Neural Networks (CNNs) with hyperparameter optimization as well as fine-tuning pretrained deep learning models are employed to classify insulator surface conditions into one of the following four categories: clean, or covered either with snow, ice, or dust. Applying cross-validation external to the CNN hyperparameter optimization and the fine-tuning process of pretrained models yielded remarkable accuracies of 92.07% and 97.80%, respectively.