Bagging and Voting Deep Learning Ensemble Methods for Binary Classifications of Solar Panel Cells Defects

Rain, extreme heat, and humidity are typical unpleasant weather conditions that solar panels are exposed to. This has an impact on photovoltaic panel cells and hence leads to various defects. Therefore, it takes time to manually inspect these defects, especially in large solar panel plants. In this study, we propose to implement the voting and bagging deep learning ensemble models' techniques to the images of photovoltaic panel cells, which are captured by drones with electroluminescence cameras, to diagnose faulty cells. To efficiently identify and classify defects, a benchmark of solar photovoltaic images that are compiled based on the electroluminescence phenomena is used. This dataset is combined to form a binary classification. One class is functional cells, and the other is defective cells. Then, an experiment of five deep learning vision models, using SENet, Vision Transformer (Vit), Xception, GoogleNet, and ResNet18, are applied as estimators to the bagging and voting ensemble techniques. Based on the results, both ensemble methods have the highest accuracy of 72.804% for voting and 72.222% for bagging as compared with 61.15% of the hybrid of ResNet50 and Inception V3 models.