Enhancing solar panel maintenance with deep learning: A multi-class fault detection framework using resnet and vision transformers (VIT) with image preprocessing
With the growing use of solar power, solar panels have become a vital part of renewable energy systems. The panels are prone to several faults like bird droppings, dust, electrical and physical faults, and snow cover, which reduce their efficiency and lifespan. To mitigate these limitations, this paper suggests an enhanced fault detection system using deep learning models, namely a hybrid of ResNet and Vision Transformers (ViT), to classify multi-class faults in solar panels. The system is supplemented by advanced image preprocessing methods, such as Gaussian blur for noise elimination, Context-Aware CLAHE for dynamic contrast adjustment, and data augmentation for enhanced model generalization. The model successfully extracts local and global features, using CNNs to extract local features and ViT to extract global dependencies in the image data. The suggested approach exhibits exceptional performance, with an accuracy of 99.61%, precision of 99.08%, recall of 99.69%, F1-score of 99.37%, and ROC AUC of 98.77%. The findings demonstrate the potential of this framework in fault detection automation, minimizing operation costs, and maximizing the reliability and lifespan of solar power systems.