Efficient Net-Based Deep Learning for Visual Fault Detection in Solar Photovoltaic Modules

Detecting faults in solar photovoltaic modules (PVM) is crucial for enhancing their longevity, power output, and overall reliability. Visual anomalies such as soiling, partial shading, cell damage, and glass breakage pose significant challenges for fault identification, particularly in harsh environmental conditions. Therefore, it is essential to maintain healthy PV systems with extended lifecycles and optimal performance through the quick and efficient detection of faults. This work introduces a comprehensive approach that encompasses dataset creation, preprocessing, and PV fault classification utilizing the EfficientNet B0 model. Processed RGB images serve as input for the model, enabling the classification of visual faults in PVM. The performance evaluation of the proposed deep neural network model includes metrics such as classification accuracy, F1 score, specificity, and recall. The results highlight the exceptional performance of the proposed model, achieving a classification accuracy of 97.24% for visual fault identification in PV modules. Moreover, the study underscores the model's robustness and efficacy through a comparative analysis with other classification techniques reported in the literature.