A refined flexible inspection method for identifying surface flaws using the skeleton and neural network

Abstract

This paper introduces a refined inspection technology which combines neural networks and range maps of object sub-skeleton pixel counts for identifying surface flaws. The proposed flexible inspection method is a low cost approach and is not restricted by changes of object position and orientation. Two stages are included. The first stage performs off-line neural network training and constructs the sub-skeleton range maps using only an object sample image. The second stage tests on-line flaws based on any of the associated neural network classifications, the sub-skeleton range matching, or a combination of the two. Experimental results demonstrate the feasibility of such an inspection approach and its improvement over the parent work.