Classification of solar panel technology and photovoltaic cell status applying machine learning to electroluminescence images

Photovoltaic energy, being renewable and environmentally friendly, significantly contributes to reducing greenhouse gas emissions. Its popularity and swift technological advances have facilitated the widespread commercialization of solar panels across various sectors. Nonetheless, these panels may harbor cell defects that adversely affect their performance and longevity. Consequently, certain techniques are employed to assess the condition of photovoltaic panels. This study explored the electroluminescence technique, which enabled us to capture high-resolution images for defect analysis within a panel. Utilizing the "LumiSolarOutdoor"electroluminescence system, we applied this method to operational photovoltaic panels in grid-connected systems in Lima, Peru. This effort generated a comprehensive database instrumental in training the "ResNet-50"pre-trained neural network. This network efficiently classified each cell's technology and degradation status within the panels. For detailed analysis, the proposed algorithm undertook pre-processing, filtering, segmentation, feature extraction, and classification of the electroluminescence images.