Classification of damaged solar panel cells by deep learning methods

Solar energy is an alternative energy source. Solar cells, which form the basis of a solar energy system, are mainly of crystalline silicon structure. Errors may occur in solar panel cells during production or after installation. It is very difficult to detect errors in solar cells with conventional imaging methods. These errors in solar cells can be detected with electroluminescent images taken with special equipment. In this study, damage detection was performed from electroluminance (EL) images of solar panel cells using deep learning method. Electroluminance (EL) images of 6528 monocrystalline and polycrystalline different solar panel cells were used. In image classification, deep learning algorithms are used to obtain fast and successful results. Images with the size of 300x300 were divided into categories as intact, broken and cracked and trained with convolutional neural networks. In the study, CNN models Xception, Vgg16, Vgg19, Resnet50, DenseNet201 and MobileNet were used. Monocrystalline and polycrystalline solar cells were analyzed separately with CNN models. According to the accuracy and classification results of the models, successful determinations were made with convolutional neural network models.