Deep learning-based model for fault classification in solar modules using infrared images

The efforts to decrease air pollutants using renewable energies, especially photovoltaic energy, are developing rapidly worldwide. Photovoltaic powerhouses contain a large number of photovoltaic power generators called photovoltaic modules that must be investigated regularly. However, these modules cannot be investigated with traditional methods because they are time-consuming and life-threatening. In this article, a deep learning algorithm-based method was developed for photovoltaic powerhouse investigation. Two types of defects were studied in photovoltaic powerhouses, namely hotspot, and hot substring. These defects are more frequent in the photovoltaic powerhouse. Datasets used in this work contain thermal images of photovoltaic modules obtained from aerial and terrestrial images. The prepared network was evaluated by some statistical parameters, including F1 score, accuracy, and precision. Finally, the network was classified with a total accuracy of 0.98, and the obtained results were compared with those of other works.