Mie-based correction for Vis/NIR scattering in emulsions and suspensions

Particulated media such as suspensions and emulsions significantly scatter the visible and near-infrared light. These light deflections increment the photon’s travel path through the sample. As a result, the absorbance increases relative to the sample’s theoretical absorption coefficient, while the latter is normalized for the path length and depends only on the sample’s composition. Accordingly, if the unnormalized absorbance is considered, an alteration in the sample’s physical properties (e.g. particle size), which reflects onto the scattering properties, might be misinterpreted as a change in the sample composition. To reduce the scattering effects present in the measured absorbance spectra, several empirical scatter correction techniques have been developed. Nevertheless, these techniques are not robust against a high variability in scattering (or physical) properties of the sample. In emulsions and suspensions, the scattering particles are often spherically shaped, which implies that the scattering properties might be described well with the Mie solutions of Maxwell’s equations. The objective of this research is to incorporate this expert knowledge into a multiple scattering correction technique to be used on measured absorbance spectra.
The Vis/NIR (500 – 1850 nm) total reflectance, total transmittance and unscattered transmittance spectra of 60 raw milk samples were measured. The Evans-Fournier approximation of the Mie solution was used to constrain the scattering properties, assuming the size distributions of the scattering particles (e.g. fat globules and casein micelles) to follow a Weibull function. The multiple scattering problem was solved with the adding-doubling numerical solution of the radiative transfer theory. Accordingly, the normalized absorption coefficient spectra were obtained for each sample and partial least squares regression (PLSR) was used to predict the fat, protein and lactose content. For comparison, the originally measured spectra were processed with several empirical scatter correction techniques in combination with PLSR to also predict the sample composition.
The approach followed in this study resulted in a clear reduction of the RMSEP for the prediction of the three milk components. Additionally, the obtained scattering coefficient spectra could be related accurately (R² = 0.95) to the milk fat content. Future research should focus on increasing the computational efficiency of this approach and on the extraction of physical information (e.g. exact particle size distribution) from the obtained scattering coefficient spectra.