Anomaly detection in electroluminescence images of heterojunction solar cells

•We report a deep-learning-based model for quality classification of HJT solar cells as well as anomalies detection in electroluminescent images.•The model is designed to be trained in a semi-supervised manner on imbalanced data containing small part of annotated relevant images with defects.•The performance of model was proved with on-plant collected dataset consisting of 68 748 electroluminescent images as well as ELPV dataset.