CLASSIFICATION OF SAND BY APPLYING PATTERN RECOGNITION TO XRF DATA

Abstract

Geology and archaeology, like other sciences, have come to rely more and more on the quantitative methods provided by modern technology. It was found that elemental analysis, especially of minor and trace elements, could provide diagnostic patterns, hence suggesting possibilities of classification. Sand has been the object of considerable study in recent years because the overall composition within a given territory tends to be homogeneous. In many instances, the composition, particularly of the major and trace elements, is a characteristic of the individual sources. 10 oxides are used to characterize the material, which was collected in California, Massachusetts, New Jersey, Virginia, Texas, Italy, and Israel.The XGT-7200V XRF analytical microscope was used in this study. This desktop unit utilizes a portable 50W X-ray source, two switchable monocapillaries, and the capability to work in vacuum, partial vacuum, as well as under ambient conditions.Spectra of sand were collected in the range of 1.00-40.96 keV for major and trace elements. These spectra were used to build a data set for Principal Component Analysis (PCA) and Partial Least Squares Discriminate Analysis (PLS-DA). Application of statistical methods, such as PCA and PLS-DA to sets of XRF data, has the capability to differentiate samples, which have very similar spectra features (concentration profiles) and release slight differences in the elemental composition of sand. To improve accuracy of PCA and PLS-DA, method of Data Fusion was applied to two sets of data. Examples of XRF spectra for major and trace elements of a sand are presented and results of analysis are compared and discussed. The Data shows that the location of unknown sample of sand may be predicted using PCA and / or PLS-DA.