Automatic Feature Extraction and Classification of Surface Defects in Continuous Casting

The problem of surface defects is a major quality concern in continuous casting. Although there exist methods in the literature to effectively detect surface defects, most are limited to specific defect types, and there is a paucity of research for classification of various defects. This paper presents a methodology for online detection and classification of surface defects in continuous casting using vision-based sensing technology. First, a two-stage algorithm is proposed to effectively extract potential surface defect regions from the noisy background. Then, the potential surface defects are classified into different categories using a newly developed classification method. The proposed methods are implemented and validated using data collected from a real world continuous casting process.