Caracterização química dos compostos majoritários em grãos verdes de café

Abstract

Green coffee is the hulled coffee bean, before going through the roasting process, rich in chemical compounds that are indicative of quality, allowing classification into special or traditional. The objective of this work was to determine compounds present in green coffee beans and find the differentiation of green coffee beans into special or traditional ones through chemometrics. Furthermore, multispectral analysis was combined with machine learning models to discriminate special from traditional green coffee beans. Therefore, the levels of reducing, non-reducing, total sugars and phenolic compounds were quantified by spectrophotometry; caffeine, trigonelline, 5-hydroxymethylfurfural, 3-hydroxybenzoic, 4-hydroxybenzoic, chlorogenic, caffeic and nicotinic acids by HPLC-UV-Vis; acetaldehyde, acetone, methanol, ethanol and iso-amyl by HS-GC-FID. Principal component analysis (PCA) was used to differentiate green coffee beans through the levels obtained in spectrophotometric and chromatographic analysis and multispectral analysis through fluorescence. To evaluate the multispectral analysis data, four machine learning algorithms (SVM, RF, XGBoost and CatBoost) were used. Statistically, the contents of total phenolic compounds, caffeine, non-reducing sugars, total sugars, nicotinic acid, 5-hydroxymethylfurfural, acetaldehyde, ethanol and ethanol/methanol showed significant differences. The PCA made it possible classification of green coffee beans in special and traditional, in addition to understanding the attributes that influenced the differentiation between coffees. The attributes caffeine, nicotinic acid, caffeic acid, trigonelline, 3-hydrozibenzoic acid and iso-amyl alcohol best described the traditional samples and are linked to undesirable characteristics for coffee. Meanwhile, the special samples were described by the attributes non-reducing sugars, total sugars, altitude, ethanol/methanol, ethanol, acetone, 5-hydroxymethylfurfural, total phenolic compounds, chlorogenic acid, 4-hydroxybenzoic, acetaldehyde and methanol, which conferred desired characteristics for the coffee. In the multispectral analysis, the machine learning model SVM demonstrated better accuracy for the test dataset (0.96) and the autofluorescence data using the excitation/emission combination of 405/500 nm contributed to discriminate the special green coffee of the traditional class. The evaluation of the chemical composition, using chemometrics, allowed the differentiation of special and traditional samples, based on the 19 parameters and also using only the parameters of the spectrophotometric analysis. Autofluorescence-based multispectral imaging combined with SVM models could be a potential tool for future applications in the food industry for non-destructive, real-time classification of specialty and traditional green coffee