Classificação de café arábica por meio de redes neurais artificiais: softwares ClassCafe 1.0 e ClassTorr 1.0

Abstract

Coffee is a product of great importance for the economy of the country, and the state of Minas Gerais, is responsible for more than 50% of the national production. However, the traditional method of evaluating the quality of the beverage is performed subjectively, whit tasters trained through sensorial analysis, using their perceptions of aroma and flavor to classify a beverage in in patterns known as strictly soft, soft, barely soft, hard, rioysh, rio and rio zona. Many studies attempt to relate the chemical composition of raw and roasted grains to the quality of the beverage in order to develop objective analytical methodologies to complement the evaluation done by tasters. In view of the above, the objective of this work was to evaluate different samples of coffee from the state of Minas Gerais. These samples were previously classified by trained tasters and, later, with the aid of grading software (ClassCafe 1.0) and classification of roasted grains (ClassTor 1.0), again objectively classified. The chemical data inserted in the raw coffee software input layer were: potassium leaching, electrical conductivity, acidity, pH, soluble solids, enzymatic activity of polyphenoloxidase and total sugars. For the roasted coffee, the content of total sugars, reducing sugars and non-reducing sugars, pH, soluble solids, acidity and ethereal extract were evaluated. At the end, the classification given by the two methodologies was compared. The raw coffee software (ClassCafe 1.0) presented better efficiency in classifying the samples when compared to the software for roasted coffee (ClassTorr 1.0). Therefore, it is concluded that the softwares developed had a great capacity to help the classification of coffee, it is suggested that the chemical data found in this research be used to increase the power of these computerized systems, in order to increase the sensitivity in the distinction between different standards.