METHOD AND ALGORITHMS FOR CASCADE CLASSIFICATION OF SULFUR PRINT IMAGES OF BILLET TRANSVERSE TEMPLATES

Abstract

This paper presents studies of sulfur print images of continuous cast billet transverse templates. The authors determined a problem of low reliability of information about the quality of billets. When assessing an image visually, such assessment is subjective with a human factor to a large extent. The authors developed a control point layout chart to collect graphical information in the course of casting billets. It is proposed to introduce three classes of images broken up by a template brightness/foreground ratio. Regarding an increased complexity of algorithms, the authors proposed the cascade classification of images. The technique includes assessment of images by shape-generating parameters of a histogram, by a distance to reference normed histograms, and by applying fuzzy logic methods. Experimental performance of the cascade technique showed that a simplified technique by assessing shape-generating parameters had expressly identified $22\%$ of all images, by assessing a distance to reference normed histograms -- $70\%$ of the rest, and only by applying fuzzy logic methods all the rest images had been unambiguously identified. A $100\%$ success of the classification is achieved only when applying all cascades of the developed technique.