Development of a machine vision system using the support vector machine regression (SVR) algorithm for the online prediction of iron ore grades

Document Type

Article

Publication Date

10-29-2018

Department

Department of Geological and Mining Engineering and Sciences

Abstract

The mineral industry needs fast and efficient mineral quality monitoring equipment, and a machine vision system could be a suitable alternative to the traditional quality monitoring system. This study attempts to develop a machine vision-based expert system using support vector machine regression (SVR) model for the online quality monitoring of iron ores (hereafter known as ore grades). The images of the ore samples were captured during the run of condition on the fabricated conveyor belt transportation system. A total of 280 image features were extracted from each of the selected captured images in order to evaluate its suitability in object identification. A sequential forward floating selection (SFFS) algorithm was developed using the support vector machine regression (SVR) as a criterion function for selecting the optimum set of image features. The optimised feature subset was used as input, and the iron ore grade value was used as an output parameter for the model development. The grade of iron ore corresponding to each captured image was analysed in the laboratory using X-Ray Fluorescence (XRF) for grade estimation. The model was trained using 70% of the dataset and tested using 30% of the sample dataset. The model performance was evaluated using a test dataset with the five indices viz. the sum of squared errors (SSE), root mean squared error (RMSE), normalised mean squared error (NMSE), R-square (R2 ) and bias. The SSE, RMSE, NMSE and bias values of the model were obtained as 537.5367, 5.9863, 0.0063, and 0.8875, respectively. The R2 value of the model was obtained as 0.9402. The results indicate that the model performs satisfactorily for the iron ore grade prediction from the image collected in a controlled laboratory environment. The performance of the proposed model was compared with other models used in the previous studies. It was observed that the proposed model performs better than the other studied models (Gaussian Process Regression and Artificial Neural Network).

Publication Title

Earth Science Informatics

Link to Full Text

Share

COinS