Technological Challenges in Defect Detection for Metal Strip; Pattern Recognition Algorithms

This paper describes how to classify a data set containing features extracted from metal strips, using pattern recognition algorithms. In the first part, a short resume of pattern recognition principles and algorithms is presented, while in the second part the techniques are applied on steel samples obtained from the Anshan Steel Corporation (China). From the images made and pre-processed by the Institute of Bildsame Formgebung Aachen (D), features were extracted using ParsyVision from Parsytec GmbH Aachen (D) and a specially tailored parameter file. On these features, we used several types of classifiers, ranging from global linear classifiers to local density estimators. The influence of dimensionality (feature set size) and sample size was investigated. The main conclusion of our research was that more samples and a more precise labelling is required to obtain a reliable classification of the Anshan Steel Data.