Robust Raman Spectral Decomposition with Wavenumber Shifts Parametric Modelling

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Nonlinearity in Raman spectral mixtures caused by wavenumber shifts, has been investigated in this paper. The spectral shifts are mainly caused by the existence of multiple chemicals in the mixtures, with complex molecular interactions, which can change the spectral features of each constituent. While such non-linear behaviour may be negligible in some mixtures, it may lead to incorrect identification of chemicals in some instances. We investigate some real spectra and demonstrate the nature of such nonlinearity in Raman spectra. We then mathematically formulate such spectral behaviour and present an approach to compensate the nonlinearity artifacts. The nonlinearity has been modelled as a smooth transition in a parametric space, which can be locally modelled using first order approximation. Such a first order approximation can be translated to some augmented spectral libraries to be used with a linear generative model. A convex sparse approximation program, with nonlinearity considerations, has finally been introduced to decompose the spectral mixtures. Such decomposition has been used for chemical fingerprinting and quantification. The effect of new approach has been demonstrated with some real and synthetic spectra.
Original languageEnglish
Title of host publication2017 Sensor Signal Processing for Defence Conference (SSPD)
Subtitle of host publication6-7 Dec. 2017
PublisherInstitute of Electrical and Electronics Engineers
Number of pages5
ISBN (Electronic)978-1-5386-1663-5
DOIs
Publication statusPublished - 21 Dec 2017
EventSensor Signal Processing for Defence -
Duration: 6 Dec 2017 → …

Conference

ConferenceSensor Signal Processing for Defence
Period6/12/17 → …

Fingerprint

Dive into the research topics of 'Robust Raman Spectral Decomposition with Wavenumber Shifts Parametric Modelling'. Together they form a unique fingerprint.

Cite this