Blind source separation in diffuse reflectance NIR spectroscopy using independent component analysis

Maunu Toiviainen (Corresponding Author), F. Corona, Janne Paaso, Pekka Teppola

Research output: Contribution to journalArticleScientificpeer-review

11 Citations (Scopus)

Abstract

Near‐infrared (NIR) spectroscopy permits non‐contact analysis of solid samples in the diffuse reflectance (DR) measurement mode. However, uncontrolled physical variations between solid samples, such as changes in packing density and particle size distribution, have a complex nonlinearizing effect on the NIR spectra which complicates the extraction of chemical information from data.

Blind source separation (BSS) methods attempt to blindly factorize the measured mixture spectra into the pure analyte spectra and their concentration profiles. The physical interferences, however, make the application of BSS methods difficult on the NIR spectra of solids. The application of independent component analysis (ICA) on NIR DR spectra is discussed, and a three‐phase preprocessing procedure of the measured spectral signals designed to improve the separation capability of ICA is proposed in this work. The method involves the removal of nonlinear effects from the measured spectra using scatter correction, denoising with rank reduction and alteration of the sample statistics of the signals via differentiation with respect to the wavelength. The procedure is tested and the explanatory power of BSS is demonstrated using a laboratory data set comprising ternary mixtures of pharmaceutical powders.
Original languageEnglish
Pages (from-to)514-522
Number of pages9
JournalJournal of Chemometrics
Volume24
Issue number7-8
DOIs
Publication statusPublished - 2010
MoE publication typeA1 Journal article-refereed

Keywords

  • near-infrared spectroscopy
  • blind source separation
  • independent component analysis
  • spectral preprocessing

Fingerprint Dive into the research topics of 'Blind source separation in diffuse reflectance NIR spectroscopy using independent component analysis'. Together they form a unique fingerprint.

Cite this