Supervised mineral classification with semiautomatic training and validation set generation in scanning electron microscope energy dispersive spectroscopy images of thin sections

This paper addresses the problem of classifying minerals common in silicicl astic and carbonate rocks. Twelve chemical elements are mapped from thin se ctions by energy dispersive spectroscopy in a scanning electron microscope (SEM). Extensions to traditional multivariate statistical methods are appli ed to perform the classification. First, training and validation sets are g rown from one or a few seed points by a method that ensures spatial and spe ctral closeness of observations. Spectral closeness is obtained by excludin g observations that have high Mahalanobis distances to the training class m ean. Spatial closeness is obtained by requesting connectivity. Second, clas s consistency is controlled by forcing each class into 5-10 subclasses and checking the separability of these subclasses by means of canonical discrim inant analysis. Third, class separability is checked by means of the Jeffre ys-Matusita distance and the posterior probability of a class mean being cl assified as another class. Fourth, the actual classification is carried out based on Sorer supervised classifiers all assuming multinormal distributio ns: simple quadratic, a contextual quadratic, and two hierarchical quadrati c classifiers. Overall Ir,weighted misclassification rates for all quadrati c classifiers are very low for both the training (0.25-0.33%) and validatio n sets (0.65-1.13%). Finally the number of rejected observations in routine runs is checked to control the performance of the SEM image acquisition an d the classification. Although the contextual classifier performs marginall y best on the validation set, rile simple quadratic classifier is chosen in routine classifications because of the lower processing time required. The method is presently used as a routine petrographical analysis method at No rsk Hydro Research Centre. The data can be approximated by a Poisson distri bution. Accordingly the square root of the data has constant variance and a linear classifier can be used. Near orthogonal input data, enable the rise of a minimum distance classifier. Results from both linear and quadratic m inimum distance classifications are described briefly.