A data fusion model merging information from near infrared spectroscopy and X-ray fluorescence. Searching for atomic-molecular correlations to predict and characterize the composition of coffee blends

Camila Assis; Ednilton Moreira Gama; Clésia Cristina Nascentes; Leandro Soares de Oliveira; Michel José Anzanello; Marcelo Martins Sena

doi:10.1016/j.foodchem.2020.126953

A data fusion model merging information from near infrared spectroscopy and X-ray fluorescence. Searching for atomic-molecular correlations to predict and characterize the composition of coffee blends

Food Chem. 2020 Apr 30:325:126953. doi: 10.1016/j.foodchem.2020.126953. Online ahead of print.

Authors

Camila Assis¹, Ednilton Moreira Gama¹, Clésia Cristina Nascentes¹, Leandro Soares de Oliveira², Michel José Anzanello³, Marcelo Martins Sena⁴

Affiliations

¹ Departamento de Química, Instituto de Ciências Exatas (ICEx), Universidade Federal de Minas Gerais (UFMG), 31270-901 Belo Horizonte, MG, Brazil.
² Departamento de Engenharia Mecânica, Escola de Engenharia, Universidade Federal de Minas Gerais (UFMG), 31270-901 Belo Horizonte, MG, Brazil.
³ Departamento de Engenharia Industrial, Universidade Federal do Rio Grande do Sul, 90035-190 Porto Alegre, RS, Brazil.
⁴ Departamento de Química, Instituto de Ciências Exatas (ICEx), Universidade Federal de Minas Gerais (UFMG), 31270-901 Belo Horizonte, MG, Brazil; Instituto Nacional de Ciência e Tecnologia em Bioanalítica, 13083-970 Campinas, SP, Brazil. Electronic address: marcsen@ufmg.br.

PMID: 32387940
DOI: 10.1016/j.foodchem.2020.126953

Abstract

This article aims to develop and validate a multivariate model for quantifying Robusta-Arabica coffee blends by combining near infrared spectroscopy (NIRS) and total reflection X-ray fluorescence (TXRF). For this aim, 80 coffee blends (0.0-33.0%) were formulated. NIR spectra were obtained in the wavenumber range 11100-4950 cm^-1 and 14 elements were determined by TXRF. Partial least squares models were built using data fusion at low and medium levels. In addition, selection of predictive variables based on their importance indices (SVPII) improved results. The best model reduced the number of variables from 1114 to 75 and root mean square error of prediction from 4.1% to 1.7%. SVPII selected NIR regions correlated with coffee components, and the following elements were chosen: Ti, Mn, Fe, Cu, Zn, Br, Rb, Sr. The model interpretation took advantage of the data fusion between atomic and molecular spectra in order to characterize the differences between these coffee varieties.

Keywords: Atomic spectroscopy; Coffee quality; Data fusion; PLS; Variable selection; Vibrational spectroscopy.