Automated Detection of Solar Cell Defects with Deep Learning

Nowadays, renewable energies play an important role to cover the increasing power demand in accordance with environment protection. Solar energy, produced by large solar farms, is a fast growing technology offering environmental friendly power supply. However, its efficiency suffers from solar cell defects occurring during the operation life or caused by environmental incidents. These defects can be made visible using electroluminescence (EL) imaging. A manual classification of these EL images is very time and cost demanding and prone to subjective inter-examiner variations. For a fully automated defect detection, we introduce a deep learning based classification pipeline operating on the EL images. This includes image preprocessing for distortion correction, segmentation and perspective correction as well as a deep convolutional neural network for solar defect classification with special emphasis on dealing with highly imbalanced dataset. The impact of minority oversampling and data augmentation on the system accuracy is investigated. The performance of our proposed classification pipeline is demonstrated by applying it to a real world dataset.