Determination of Ore Grade with Neural Network and Image Processing Technique

In this paper feasibility of determination of ore grade using neural network and image processing techniques is presented. In practice grade of 100 iron ore samples, from Choghart iron ore mine, were determined. Photographs of samples powder were then used for image characteristics study like Red, Green and Blue colors (RGB) and their standard deviation and Haralick's textural properties including Energy, Entropy, Contrast and Homogeneity. Initially the neural network was trained with 70 photo samples and then tested by another 30 photo samples. In training step, three procedures were followed. At the first methods, Red, Green and Blue colors (RGB) were trained to the neural network. Second method involved training of the system by addition of standard deviation to the RGB color and at third method Red, Green and Blue colors and Haralick's textural characteristics (entropy, contrast, energy and homogeneity) were considered and trained to the neural network. The results show that the trained neural network by the above three methods could determine the grade of iron ore with the accuracy of 95.2, 96.9 and 97.72 percent, respectively.