The photovoltaic (PV) system industry is continuously developing around the world due to the high energy demand, even though the primary current energy source is fossil fuels, which are a limited source and other sources are very expensive. Solar cell defects are a major reason for PV system efficiency degradation, which causes disturbance or interruption of the generated electric current. In this study, a novel system for discovering solar cell defects is proposed, which is compatible with portable and low computational power devices. It is based on K-means, MobileNetV2 and linear discriminant algorithms to cluster solar cell images and develop a detection model for each constructed cluster. It can extract the distinct features between defective and nondefective solar cell clusters and overcome the confusion between different cell shapes. The proposed system was assessed using a benchmark dataset of electroluminescence images. The results demonstrate that it achieved the highest accuracy at a substantial rate compared to recent studies.
展开▼
