Cluster identification using image processing

Abstract

By closely examining hue, saturation and value (HSV) images of the solids holdup distribution in a riser, it can be seen that a “cluster” is the combination of a relatively stable core cluster of the highest solids holdups and constantly changing cluster clouds of solids holdups that are higher than the dilute phase. Based on this analysis, a threshold selection method maximizing the inter-class variance between the background and foreground classes is introduced. A systematic cluster identification process is therefore proposed that: (1) applies the threshold selection method to obtain the critical solids holdup threshold ɛsc to discriminate dense and dilute phases and (2) applies the method again in the dense phase regions to obtain the cluster solids holdup threshold ɛsct that identifies the core clusters. Using this systematic process, clusters of different shapes and sizes and a relatively clear boundary can be visualized clearly and identified accurately. Using ɛsct, the core cluster fraction is calculated by dividing the total number of pixels in the core cluster by the total number of image pixels. The variation of the core cluster fraction according to operating conditions is also discussed.