Neural network visual inspection system with human collaborated learning system

A present sheet object production line employs high-speed operation. Because of its high speed, once the defects appear, a large amount of defective products may be generated in a short time. The neural network classification system has been introduced into this system to maintain the production machine. However, it is recognized that the recognition rate decreases when the number of defect classes increases. To meet with the problem, two steps neural network decision process has been introduced, here. Another problem is that the system is not able to learn properly when the size of teaching data is small. To solve the problem, the simulation system to generate the teaching data from its description has been developed. Experimental results show that the average recognition rate has been improved.