Classification of Defects in Photovoltaic Cells from Electroluminescence Images using Advanced Deep Learning Techniques

For a long time, keeping big solar power plants running has been a real challenge. This work talks about an exciting new way to detect problems in solar panels using special pictures called electroluminescence images (EL). Deep learning algorithms is used to help with this task! Our research introduces a fresh method that can automatically spot flaws by using a pre-trained deep learning Vision Transformer (ViT) model. The model is trained on ELPV Dataset with EL image, which is available for everyone. The model achieved an impressive accuracy of 94.53% on the test set! Our model was thoroughly tested against popular machine learning models like ResNet50, DenseNet, ResNet101, VGG-16, VGG-19, Inception ResNet, ResNet-152, MobileNet, and Xception. The results showed that proposed model did better than all of these earlier models. The model is compared with other top methods for finding problems in EL images. Proposed method has proved to be better than the existing models. It really showed it can accurately find flaws in electroluminescence images.