AN ALGORITHM FOR THE TRANSFORMATION OF XRF IMAGES INTO MINERAL-DISTRIBUTION MAPS

Abstract

A new numerical method has been developed that uses the pixel-intensities from X-ray fluorescence (XRF) images to produce mineral-distribution images or maps. The XRF images are obtained using a scanning X-ray analytical microscope (SXAM); the derived mineral-distribution maps can be applied to study the petrographic characterization of a rock or ores. As a test ca se, we applied this method to a granite consisting dominantly of quartz, biotite, plagioclase and K-feldspar. Maps of the major elements, including Al, Si, K, Ca and Fe, were obtained with the SXAM. XRF intensity is recorded for each element on a 256 � 256 pixel map. To transform these element-distribution maps into mineral-distribution maps, we employed the maximum likelihood method for a Gaussian distribution, i.e., a least-squares method. The coefficient between XRF intensity and the proportion of a desired mineral was determined from the average value in a few hundred pixels that record the XRF intensity of the desired mineral in isolation. Where one pixel recorded XRF intensity for more than one mineral, its intensity was assumed to have a lin ear relationship with the composition of the minerals included in the pixel. The proposed least-squares method is an alternative technique to methods using purely image-analysis operations, such as image enhancements, erosions, dilations and image Boolean operations. Experiments with the granite sample showed that the least-squares method gives appropriate mineral-distribution maps if the acquisition time for the XRF maps is sufficiently long, e.g., 48 hours for the granite. The sources of errors in the calculated proportions of the minerals are related to fluctuations of X-ray intensities and variations of chemical composition in each mineral. A method is also proposed to estimate these errors.