Development of machine vision-based system for Iron ore grade prediction using gaussian process regression (GPR)

Abstract

India is one of the major iron ore producing country and requires quality monitoring of iron ore. An attempt has made to develop a vision-based system for continuous iron ore grade prediction during transportation of ores through conveyors. A Gaussian process regression (GPR) algorithm was used to develop the model. To design the system, a pilot conveyor belt setup was fabricated to replicate the mine conveyor system and consists of image capturing system to capture images during transportation of ores. The images were processed and GPR was calibrated using the grade values of 26-iron ore samples. A set of 18 features (9-colors and 9-textures) were extracted from each of the 26-captured images for model development. The performance results revealed that the predicted grade has closely agreement with the actual grade of the ores. The correlation coefficient (R2) between the observed and predicted grades was found to be 0.9569.