Use of microscale heterogeneity in samples for spectral factorization-A strategy to build robust prediction models for nondestructive analyses

Abstract

Nondestructive spectroscopic analysis is widely used to evaluate food composition. However, distinguishing analytes of interest from other compounds remains challenging. Since most foods are heterogeneous when viewed under a microscope, we hypothesized that spectra measured at microscopic points would be "purer" than spectra acquired from a larger area. By coupling this data with nonnegative matrix factorization (NMF), the analytes of interest can be separated. This preliminary study discusses the quantification of glucose in mixtures of different sugars. Samples were made by mixing glucose with other powders in different ratios and Raman spectra were measured at 200 micro-points for each sample. NMF was applied to factorize the mixed spectra into spectra of pure compounds and their concentrations, leading to the accurate quantification of glucose, while eliminating the effects of other compounds. While this study targets simple powders, separation of analytes using microscale heterogeneity is applicable for measuring more complex foods.

Keywords: Chemometric methods; Hyperspectral imaging; Matrix factorization; Nondestructive analysis; Raman spectroscopy.