PATTERN-RECOGNITION ANALYSIS OF LOW-RESOLUTION X-RAY-FLUORESCENCE SPECTRA

Using a high-resolution Si(Li) spectrometer to perform X-ray fluorescence (XRF) analysis of various geological, alloy, and paint samples, we have demonstrated that in situations where quantitative information is not the primary concern, a pattern-recognition approach may be used to obtain qualitative results very quickly and efficiently. Specifically, the pattern-recognition technique uses a single parameter, the normalized correlation coefficient, to identify and select samples with similar chemical compositions from their raw XRF spectra. The algorithm can be easily implemented on a personal computer; typically it takes only a few seconds to perform the pairwise comparison of 9 or 10 spectra. We report here the results of our attempt to extend the pattern-recognition technique to the analysis of low-resolution XRF spectra from a proportional counter where the spectral information is considerably poorer than that of the Si(Li) detector. Analyzing the XRF spectra obtained from a set of geological samples both in air and in vacuum we find that even with a proportional counter the pattern-recognition technique can nevertheless satisfactorily identify samples with similar chemical compositions.