Anomaly detection and automatic labeling for solar cell quality inspection based on Generative Adversarial Network

In this manuscript, a pipeline to develop an inspection system for defect detection of solar cells is proposed. The pipeline is divided into two phases: In the first phase, a Generative Adversarial Network (GAN) employed in the medical domain for anomaly detection is adapted for inspection improving the detection rate and reducing the processing rates. This initial approach allows obtaining a model that does not require defective samples for training and can start detecting and location anomaly cells from the very beginning of a new production line. Then, in a second stage, as defective samples arise, they will be automatically labeled at pixel-level with the trained model and employed for supervised training of a second model. The experimental results show that the use of such automatically generated labels can improve the detection rates with respect to the anomaly detection model and the model trained on manual labels made by experts.