Mineral identification using color spaces and artificial neural networks

Identification of minerals and percentage of their area within a thin section of rock are important for identifying and naming rocks. Colors of minerals are the basic factors for identification. In this study, an artificial neural network is used for the classification of minerals. Optical data of thin sections is acquired from the rotating polarizing microscope stage. For the first analysis we selected a set of parameters based on red, green, blue (RGB) and the second based on hue, saturation, value (HSV) color spaces are extracted from the segmented minerals within each data set. A neural network with k-fold cross validation is trained with manually classified mineral samples based on their pixel values. The most successful artificial network to date is the three-layer feed forward network which uses minimum square error correction. The network uses 6 distinct input parameters to classify 5 different minerals, namely, quartz, muscovite, biotite, chlorite, and opaque. Testing the network with previously unseen mineral samples yielded successful results as high as 81–98%.