Arabica coffee samples classification using a Multilayer Perceptron neural network

Abstract

The world’s coffee agribusiness is worth US$ 91 billion annually and involves half a billion people. It is in this market that the interest of the Brazilian coffee production chain is centered, which contributed with more than 30% of the world production in the last harvests. The coffee market is characterized by a range of activities, complexity, dynamism, and a growing level of consumer demand for beverage quality. This imposes a high quality control on producer, consumer and exporter countries. Currently, the definition of quality and hence the value of coffee is based on manual grading, ie a person performs the role of a trained (certified) classifier to qualify coffee samples. Thus, the current process of classification of coffee suffers from the subjectivity of the classifiers and a great difficulty of standardization of the process due to possible inconsistencies. The present work proposes the use of an MLP (Multilayer Perceptron) Neural Network for analysis of coffee beans samples by digital image analysis, in order to increase the speed and reduce the subjectivities involved in the current manual classification process, considering: shape, size and color. Among the benefits of the automation of the coffee classification process are the reduction of costs, the agility and the standardization of the classification.