Detection of Photovoltaic Panel Dust Using Drone Shooting Imagery and Deep Learning Techniques

Abstract

Regular detection of dust on photovoltaic (PV) modules is crucial for their maintenance because dust can hinder light harvesting and decrease power generation. The current study utilized drone images of panels and deep learning techniques to detect dust and evaluate the performance of a prediction model. After acquiring and preprocessing the drone images of the panels in Samcheok city, Gangwon state, we applied five deep learning techniques to identify and categorize the panels as dusty or clean. We evaluated the performance of these techniques using a confusion matrix. The convolutional neural network technique demonstrated the highest predictive performance with an area under the curve of 80.6%, classification accuracy of 66.0%, and recall of 94.0%. However, the model's high predictive performance for dusty panel cases was accompanied by a high rate of misclassification for the clean panel cases, indicating the need for improvement. The proposed analytical model can be used to quickly screen and evaluate the dusty panels in large-scale power plants.