Multivariate techniques for processing Raman spectral data
Abstract
The purpose of this study is to explore and develop multivariate classification, calibration and noise filtering techniques for processing Raman spectroscopic data including Raman hyper-spectral imaging data. The first three chapters are focused on preprocessing techniques aimed to reducing the effect of spectral artifacts. Chapter one demonstrates that the Raman spectral classification accuracy can be greatly improved using smooth second derivative spectral preprocessing. In chapter two and three, two new algorithms removing cosmic spikes from individual spectra or large hyper-spectral data matrices are introduced and demonstrated. Although Raman classification is the model application, these preprocessing algorithms should also be beneficial for other spectroscopic or more general multivariate calibration applications. In chapter four several widely used spectral classification algorithms are evaluated and tested with engineering plastic samples of very similar composition. The effects of different spectral preprocessing techniques on classification accuracy are also studied. Chapter five describes a new feature selection algorithm for classification applications, termed sequential forward-selection of ranked features. It is shown that this new method outperforms the univariate Fisher's method without a significant increase of the computational complexity.
Degree
Ph.D.
Advisors
Ben-Amotz, Purdue University.
Subject Area
Analytical chemistry
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.