Análise de técnicas de segmentação para seleção de grãos de café torrados em imagens

Abstract

Brazil is the largest coffee producer in the world and the second-largest consumer. Since the arrival of coffee in Brazil, it has been a prominent product in the national economy. Approximately 71% of this production is roasted and ground coffee. The roasting process is of utmost importance, as it gives the roasted beans their characteristic aroma and flavor. There are several ways to determine the degree of coffee roast, but specialized equipment is expensive, and the use of Agtron/SCAA discs involves a manual and subjective process of comparing the color of the beans and selecting the closest match. There have been research efforts to create low-cost options using images to detect the degree of roast. However, in these studies, achieving good accuracy requires manually selecting the region of the image that contains the coffee beans. This manual process is laborious and hinders a more complete automation of the roast detection process. Therefore, with the aim of creating a more automated degree of roast detection process, this project analyzes segmentation techniques, evaluating the results both qualitatively and quantitatively.