Premium
Evaluating the Effects of Autofluorescence during Raman Hyperspectral Imaging
Author(s) -
Emry Julienne R.,
Olcott Marshall Alison,
Marshall Craig P.
Publication year - 2016
Publication title -
geostandards and geoanalytical research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.037
H-Index - 73
eISSN - 1751-908X
pISSN - 1639-4488
DOI - 10.1111/j.1751-908x.2015.00354.x
Subject(s) - autofluorescence , hyperspectral imaging , computer science , data processing , artificial intelligence , partial least squares regression , raman spectroscopy , pattern recognition (psychology) , computer vision , optics , machine learning , physics , fluorescence , operating system
Raman hyperspectral imaging is becoming a popular technique to analyse geological materials. Autofluorescence can affect the quality of the spectra that comprise hyperspectral data sets. Few studies have addressed potential misinterpretation of Raman images from hyperspectral data sets affected by autofluorescence. Additionally, little work has been done to develop methods for identifying the spatial distribution of spectra affected by autofluorescence. This study illustrates how autofluorescence may lead to misinterpretation of the distribution of materials based on intensity at a point images. A method is proposed utilising signal to axis analysis to create images that identify regions affected by autofluorescence. Post‐processing baseline correction is often used to address autofluorescence, and most software programs utilise a form of partial least squares regression modelling based on a subjective choice of polynomial order. This study shows that an inappropriate choice of polynomial order can introduce error, which may lead to misinterpretation of Raman images. A signal to axis analysis method is proposed to statistically compare seemingly ‘appropriate’ baseline correction trials. Although post‐processing of hyperspectral data sets and creating Raman images seem simple, data quality issues such as autofluorescence must be considered. If baseline correction is deemed necessary, it should be addressed as an experiment involving statistical comparison.