Análise multivariada de imagens digitais no controle de qualidade da torrefação na presença de adulterantes carbonáceos em amostras de café arábica

Abstract

This work proposes to monitor the quality of coffee in relation to the type of roasting and the undue addition of carbonaceous adulterants, which are generally made up of charred leaves and twigs. The type of roasting and carbonaceous adulterants affect the color of the coffee powder during the bean processing stage. Digital image analysis therefore becomes an accessible, economical and efficient alternative to be used as a resource for acquiring multivariate data based on the RGB color model for the development of analytical methods, aiming to determine the carbon content present in coffee samples. To this end, digital images of 101 samples made up of a mixture of coffee and charcoal in defined proportions (0-100% by mass of charcoal) were captured using a camera as an imaging device, and the distribution of image pixels was taken using RGB histograms and organized into a data matrix. The data was processed by chemometric tools, such as Principal Component Analysis (PCA) and through a multivariate calibration model based on Principal Component Regression (PCR) for quantitative predictions of the coal content in the samples. Two ways of arranging samples in separate containers were evaluated: Petri dish and plastic coffee cups. The PCA revealed the formation of two sample profiles: One with up to 40% coal and the other above 60% coal. The PCR calibration models provided satisfactory fits with R2 values ~0.97 and ~0.95 for samples in Petri dishes and cups, respectively. The samples arranged in the Petri Dish provided more satisfactory cross-validation results, with less interference from external lighting and shadows.