[Article]
 

Abstract.
In this work, the potential of mid-infrared diffuse reflectance spectroscopy with Fourier transform for discrimination of 29 commercial Brazilian coffee samples with different industrial processing, i.e., caffeine extraction and roasting degree, was evaluated. The statistical treatments applied to pretreated spectral data were principal component analysis and partial least squares – discriminant analysis (PLS-DA). The ordered predictors selection method was used for variable selection. The chemometric analyses of the mid-infra-red spectra allowed inferring on the lower carbohydrate, caffeine and chlorogenic acid concentration as well as on the higher water content in the decaffeinated coffee. The technique also allowed speculation on the higher lipid and lower water content in the dark roasted coffee compared with traditional roasted coffee. A clear discrimination of decaffeinated from medium and dark roasted coffees was observed in PC1. PLS-DA was used for the discrimination between medium and dark roasted coffees. A model with one latent variable correctly classified 100% of the external validation and prediction samples according to their roasting degree.

Keywords.

Keywords Plus.