Development of Image Processing and Neuro-Fuzzy Based System for Ore Sorting in Mining Industry

In mining, gangues are mixed along with the ore, as these are closely associated. In the process of extracting minerals, through underground or open pit mining these gangues are also extracted along with the ore. It is necessary to separate the gangue minerals from the ore after the production of mining before dispatching in order to maintain the quality of the ore. Ore sorting using automatic sensor-based sorters are being used to pre-concentrate the feed ore prior to the beneficiation plant. Automatic sensor-based sorting examines the physical properties of ore and gangue minerals and separates based on the predefined sorting criteria. In the recent years, digital image processing techniques have been applied in the mineral industry in different mining operations such as online ore monitoring, particle size estimation, ore sorting, and classification. The objective of this research was to proposed a methodology for sorting gangues from the ore using image processing techniques. For this study, samples were collected from limestone and iron ore mines situated in Odisha state, India. The video was captured for ore and gangue samples collected from the different mines from the moving conveyor belt using a digital camera (Canon EOS 60D) for analyzing the color-texture features using digital image processing techniques. For analyzing the color and texture properties three different color spaces RGB, HSV and YCbCr were used. The color features were extracted from the chromatic information of the color spaces and the texture features were extracted using gray level co-occurrence matrix (GLCM), local binary pattern (LBP), local ternary pattern (LTP), Tamura, local binary pattern co-occurrence matrix (LBPCM) and local ternary pattern co-occurrence matrix (LTPCM) features from the illumination information of the color spaces. The features extracted were used for classification of gangue from mineral using density based (linear and quadratic discriminant analysis) and kernel-based classifiers (k-nearest neighbor and support vector machines) and adaptive neuro-fuzzy inference system (ANFIS) algorithm. Optimization techniques principal component analysis, linear discriminant analysis, t-test, sequential forward selection and sequential backward selection were used to reduce the features set. To reduce the computational time in calculating co-occurrence matrix, quantization was performed using 256, 128, 64 and 32 gray levels. It was observed that the proposed image analysis techniques gave good accuracy in the separation of gangues from limestone and iron ore mines. From the proposed studies, it was observed that the extracting texture features from the illuminance information of color spaces gave better classification accuracy as compared to the extraction of the texture features from the gray scale information of the image. It was observed that the performance of t-test method was better than the other methods. For classification of gangues from limestone and iron ore, it was observed that the performance of SVM with cubic kernel was found to be better as compared to the other methods