Auto-inspection system using deep-learning for glass materials

Abstract

As deep learning technology has been improved in recent years, it has started to be applied to image inspection process of high needs for making a high advanced automation in a factory. However, most of them are not completed by only the deep learning technology where they are installed for support of worker, in which the system inspect roughly first and then worker re-inspect in more details (resulting in benefit of saving worker compared full-worker system). Therefore, we did not use the deep learning technology to the inspection itself, but tried to use for classification of defects by rule-based system from conventional image data. We can improve inspection level successfully which can achieve full-automatic inspection system that does not rely on inspection by worker.