Surface Defect Classification of Steels Based on Ensemble of Extreme Learning Machines

Abstract

In recent years, iron and steel industry of China has developed rapidly, and steel surface defects recognition has attracted wide attention in the field of industrial inspection. Aiming at the problems of poor precision and low speed of traditional surface defect detection methods, we propose to use a fully learnable ensemble of Extreme Learning Machines (ELMs), which is ELM-IN-ELM, for defect classification. The Local Binary Pattern is adopted as the basic feature extraction method. The ELM-IN-ELM determines the final classification decision by automatically learning the output of M independent ELM sub-models. To further illustrate the superiority of the ELM-IN-ELM algorithm for classification, the Northeastern University (NEU) surface defect database is used to evaluate its classification effect. The experimental results demonstrate that this method works remarkably well for surface defects classification. Compared with other methods, the proposed method can identify the types of defects more accurately, which is of practical significance to steel surface defect detection.