Steel strip surface defect identification based on binarized statistical features

Zoheir Mentouri, A. Moussaoui, Djalil Boudjehem, et al.

Vol. 80, Iss. 4, 2018

U.P.B. Sci. Bull., Series B

550_597847

Abstract

In the steel hot rolling process, flat products that are shaped by a gradual reduction of the thickness and the increasing of the length may exhibit different surface defects, which should be identified. The solution, widely adopted, and still considered as a challenge is the automatic inspection. It is assumed, allowing an immediate detection with accurate identification of the defect that starts appearing during production. However, for a perfect labeling of the occurring defects, inspection system should be provided with reliable algorithms. In this paper, tools are combined to provide a high-efficiency solution. The suggested method is based on the recent Binarized Statistical Image Feature extractor used, to date, in biometrics. Combined with a relevant reduction-data method and the K nearest neighbors classifier, this solution showed improved recognition rates of the strip surface defects in the hot rolling process, outperforming, the reported results in previous works.

Keywords

Computer vision, statistical features; classification, strip surface defects, hot rolling process.