A machine learning-based image classification of silicon solar cells
H. Verma, S. D. V. S. S. Varma Siruvuri, P. R. Budarapu
Abstract
Silicon-based solar cells are a popular choice to generate electricity from sunlight. Micro-cracks are inherent in brittle silicon cells, which propagate during their service and hence impacts the efficiency. This study is focused on classifying micro-crack patterns in silicon-based solar cells with the help of convolutional neural network (CNN)-based models. A dataset comprising 3,651 electroluminescence images is categorised into five groups: poly-good, poly-cracked, poly-corroded, mono-good, and mono-cracked. Four pre-trained convolutional neural networks, namely: ResNet50, VGG-16, VGG-19, and DenseNet are employed to classify the images, where 80% of the data is used for training and the remaining 20% for testing. Results reveal that the VGG-19 network is able to categorise the images with 100% accuracy in distinguishing poly and mono-crystalline silicon cells, with an overall accuracy of 98.44%, thereby outperforming other models. Thus, a VGG-19-based CNN is recommended for classification of electroluminescence images of silicon solar cells. Such image classification helps for health monitoring and hence, a better maintenance of the PV modules.