LAUSR

Morphological classifications of natural materials are used in a wide variety of applications. Visual classification of some natural materials such as minerals, characterized by their wide morphological variations, are challenging. In some instances, differences between categories are subtle or forming continuum, causing high misclassification rates. Supervised machine learning (ML) is a powerful tool to automate and improve classification. However, most image classification algorithms require a large labeled dataset in which the quality of the initial labeling is critical. Since visual labeling of hundreds of thousands of images is a tedious and biased process, a specific learning strategy is proposed to circumvent such issues. This paper describes the classification procedure for gold grain morphologies obtained from backscattered electron images.