Mathematical decomposition and multivariate analysis applied to Raman imaging
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Chemometrics is a widely used method that is very useful in data analysis. By applying it to various types of data, more detailed information can be obtained that would have potentially been obscured otherwise. This dissertation uses various chemometric techniques in order to obtain information from various types of Raman data. First, hyperspectral Raman images of meteorite impact samples are utilized in considering component selection for the MCR-ALS method. This is followed by the analysis of various known samples via Spatial Heterodyne Raman Spectra, in which there was a flaw in the detector that produced unwanted fixed pattern noise in the Raman spectra, to identify the best method to remove the unwanted peaks without losing information about the samples. ☐ The novel method for quantitatively selecting the optimal number of components was conducted by determining the correlation between pre-determined reference spectra and the components for all MCR-ALS models for one to twenty components, and determining in which model all species are resolved and statistically stable. In smaller models, with lower numbers of components, the major species are immediately identified. However, it takes a larger number of components in the MCR-ALS model to identify minor species, due to the presence of weak intensity peaks that can be lost within a noisy baseline. This correlation method can be used to quickly identify species present in a sample, as well as differentiating between varying crystal orientations for a species, which can affect the Raman spectra for species. ☐ Also shown in this dissertation is an exploration of methods for successfully removing unwanted fixed pattern noise in Spatial Heterodyne Raman spectra. Various methods of multivariate analysis are considered to determine if the unwanted peaks can be removed without influencing the spectra of the samples or producing negative peaks in the baseline. Simpler methods, such as background subtraction or Savitzky-Golay filtering were unsuccessful in removing the unwanted fixed pattern noise peaks. However, more complex decomposition methods, such as PARAFAC, were successful in separating the unwanted peaks from the pure sample spectra.