Enhanced Chemical Classification of Raman Images in the Presence of Strong Fluorescence Interference

Abstract

Raman spectra and spectral images containing severe fluorescence interference are analyzed by using a variety of correlation and classification algorithms, both before and after preprocessing with the use of the Savitzky-Golay second-derivative (SGSD) method (and other related methods). Spectral correlation coefficient, principal component, and minimum Euclidean distance analyses demonstrate superior suppression of background and noise interference in Raman spectra when using SGSD preprocessing. The tested spectra include fluorescence interference that is more intense than the Raman features of interest and also contains broad background peaks that vary in shape and intensity from sample to sample. The high chemical information content of the SGSD-processed Raman spectra is demonstrated by using quantitative comparisons of correlation coefficients in a series of synthetic Raman spectra with either different or identical large backgrounds. The practical utility of SGSD in chemical image classification is illustrated by using an experimental Raman image of sugar microcrystals on substrates with large interfering background signals. The functional equivalence of SGSD and other windowed preprocessing algorithms is discussed.