Iterative improved learning algorithm for petrographic image classification accuracy enhancement

Rock image classification using image processing has been practiced to assist trained geologists in decision making. However, the study of microstructures of rocks and their use in geological investigations offer challenges in the areas of Image processing and Pattern Classification due to the stochastic nature of the mineral textures that is revealed at the microscopic level. Locally relevant Igneous Rock Microstructure images were classified from Volcanic and Plutonic Rock subtypes. The imaging method used mineral grain size as the key physical feature of classification. Three algorithms, namely Radial Basis Function (RBF) Support Vector Machine classifier; Improved (RBF) Support Vector Machine classifier; and AdaBoost algorithm with Improved RBF Support Vector Machine algorithm as base classifier, were used as a base classifier in a novel ‘Iterative Improved Learning (IIL)’ approach. Implementing the IIL approach in the chosen algorithm resulted in accurately classified images that were added to the training set to enhance the ‘breadth and depth’ of the learning knowledge. The algorithm iterated through all available classifier approaches and compared the inter-classifier performance and knowledge of the misclassified images accumulated during the execution of all algorithms.