Classification of periodical defects in inspection systems based on computer vision

Article type: Research Article

Authors: Bulnes, Francisco G.

Affiliations: Department of Computer Science, University of Oviedo, Campus de Viesques, 33204 Gijon (Asturias), Spain. E-mail: bulnes@uniovi.es

Abstract: This thesis focuses on the detection of defects generated periodically during the production of web materials. The stated goal of this thesis is to develop a technique capable of detecting such defects as quickly as possible and to assess the quality of such detections. Although the proposed technique is generic enough to be applicable to several problems, in this thesis steel strips were used for its development and evaluation. The proposed solution to detect periodical defects is a backtracking-based algorithm that checks whether certain characteristics of a set of defects meet certain conditions. To determine whether the detections are good, it is necessary to quantify them. To obtain this quantification, several metrics are proposed. Finally, the results provided by the proposed technique were assessed and compared with those obtained by a commercial tool widely used, showing a clear improvement.

Keywords: Intelligent systems, automated defect detection, pattern recognition

DOI: 10.3233/AIC-2012-0534

Journal: AI Communications, vol. 25, no. 4, pp. 385-386, 2012