Automatic defects classification and feature extraction optimization

Abstract

This paper introduces an automatic classification system that can identify defects on product surfaces in manufacturing, especially in processes like grinding and polishing. The identification process is based on grayscale images taken by a vision system. Some technologies that extract features from digital images are discussed. The support vector machine (SVM) is used in this paper as a multiclass classifier. It is shown that the overall classification rate can be close to the level that a skilled operator can obtain. The issues concerning the optimization of feature extraction are also covered in this paper.