Coffecolor: desenvolvimento de um aplicativo para detecção do grau de torra de café baseado na cor

Coffee is one of the most important agricultural products for the Brazilian economy and its final quality is determined by several factors, one of them being roasting. Coffee roasting determines the sensory characteristics of the product, providing changes in flavor, aroma, texture, and color. During the roasting process, the color of the beans is used as a criterion to stop the process. The determination of the roasting degree is often determined subjectively by the roaster. There are internationally accepted standards for representing the roasting degree of the grains, such as the Agtron scale. However, the use of the Agtron scale involves the acquisition of considerably expensive equipment or visual comparison with printed disks, which, again, falls into the problem of subjectivity. The present work proposes an alternative and automated methodology for the color analysis of roasted coffee beans, based on the Agtron scale through a computational application for mobile devices. Through experiments and optimizations, a classification algorithm was developed using random forests that obtained an accuracy of 92.1% in predicting the color of roasted coffee beans. The results obtained are promising and represent a viable low-cost alternative to evaluate roasted coffee according to its color.