Construction of a label recognition system on water valves in the factory inspect process

Abstract

This paper presents a label recognition system aiming at being introduced to the inspection process in the water valve factory. This inspection process checks whether all the labels are pasted to the right place on the products. The numbering management that has been adopted in the current factory lines can help us with finding the missing or double pasting of the labels, but it never notifies the wrong-place pasting. Furthermore, this requires much time to prepare the same number of the parts including the labels. To solve these problems, we here introduce an image-based recognition method utilizing the deep-learning. Adopting the object recognition algorithm, not only the existence but also the pasted place of five pre-learned labels are inspected using several numbers of the frames within 1 second. The learning from about 3,000 sample data resulted in more than 95% correction rate in three of five labels though it fell nearly 80% for the others.

Keywords

• machine learning
• image recognition
• object detection
• image processing
• factory automation
• productivity improvement