Quantificação do percentual de café arábica em misturas com café robusta usando FTIR, NIRS e modelos de dois estágios

Abstract

Coffee is one of the most acceptable and appreciated drinks around the world. The main species are Coffea arabica and Coffea canephora, which shows a very different chemical composition. The specie C. Arabica stands out for providing a drink with higher quality and aroma, if compared with C. Canephora. The last specie is considered a neutral drink, for blends and for the industry of instant coffee, because of his low cost. Blends are very used when it’s necessary to maintain a uniform flavor. Thus, the main objective of this work was to develop a reliable analytic method to indicate the amount of each type of coffee in mixture. Spectra of samples were obtained by Infrared Spectroscopy with Fourier Transform (FTIR) and Near Infrared Spectroscopy (NIRS). In the first stage, it was used Principal Component Analysis (PCA) and the Partial Least Square (PLS). The result shows that the spectra obtained by FTIR gave better results than the NIR spectra. The method for data dimensionality reduction more satisfactory was PLS. The ANN with the best performance was MLP, which, in general, had lower results of Mean Absolute Error (MAE) and higher correlation coefficient.