Development of online machine vision system using support vector regression (SVR) algorithm for grade prediction of iron ores

Abstract

The present study attempts to develop a machine vision system for continuous monitoring of grades of iron ores during transportation through conveyor belts. The machine vision system was developed using the support vector regression (SVR) algorithm. A radial basis function (RBF) kernel was used for the development of optimized hyperplane by transforming input space into large dimensional feature space. A set of 39-image features (27-colour and 12-texture) were extracted from each of the 88-captured images of iron ore samples. The grade values of iron ore samples corresponding to the 88-captured images were analyzed in the laboratory. The SVR model was developed using the optimized feature subset obtained using a genetic algorithm. The correlation coefficient between the actual grades and model predicted grades for testing samples was found to be 0.8244.