Métodos instrumentais alternativos para a predição da cor do café torrado

• Ano de defesa: 2019
• Autor(a) principal: Pires, Fabiana de Carvalho
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Programa de Pós-graduação em Ciência dos Alimentos; UFLA; brasil; Departamento de Ciência dos Alimentos
• Palavras-chave em Português: Nível de torra; Valor Agtron; Redes neurais artificiais; Level of roasting; Agtron value; Artificial neural networks; Torrefação; Roasting; Torra do café; Coffee roast; Tecnologia de Produtos de Origem Vegetal
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/38010

Coffee is one of the most important agricultural products for the Brazilian economy and your final quality is determined by factor like the roast, which is the thermal process dependent binomial time and temperature able to promote physical and chemical changes in coffee beans, among them the color. The color of roasted coffee beans is one of the qualitative criteria used to interrupt the roasting process can be used as an evaluation parameter. A very specific color descriptor used by the coffee industry to evaluate the roast level is the Agtron value, where #25 is the darkest brown and # 95 is the lightest brown. However, most of times, the roast level is subjectively monitored by the visual method and the experience of the roasting professional. Aiming to control, monitor and promote the roasted coffee, the present work had as objective propose two alternative methodology to color analyse, so much to roasted coffee beans and ground and, subsequently correlate them with the roast level, in the same scale Agtron variable, in order to minimize the variations of a subjective visual judgment. The first methodology consist in the prediction of the roast color of specialty coffees using near infrared spectroscopy (NIR) and partial least square (PLS), which in this work presented determination coefficients R^2=99,53 % for ground coffee and R^2=96,20 % for coffee beans. The second methodology consist in the prediction of the roast color of specialty coffees using digital images processing and artificial neural networks (RNA) for roasted and ground coffee R^2 training=99,97 % and R^2 valid.=99,97 % and for roasted coffee beans R^2 training=99,83 % and R^2 valid.=99,94 %, both the results were promising e suggest the possibilities to use this techniques to evaluate the roasted coffee in function of the color. For the use of RNAs was developed a software (FRR 1.0) capable to predict the Agtron value and the roast level at which the coffee samples were classified, directly to the operator, therefore, present a potential applicability in the coffee roasters.This software was registered at the National Institute of Industrial Property (INPI), on October 25, 2019, with registration number 512019002447-8. The third methodology consist in the prediction of Agtronvariable in function of the color parameters of the Commission Internationale de l'Eclairage, asL*, a*, b*, C* and H° with R^2=99,33 % to roasted coffee beans and R^2=99,88 % to roasted and ground coffee. And, in the prediction Agtron variable mass loss function with R^2=98,29 %. The results obtained through three methodologies were promising and show the possibility of using the prediction model adjustments for the Agtron variable in function of the parameters used in this study.