Analysis of Machine Learning Techniques Applied to Coffee Bean Classification

Resumo

The coffee market is characterized by a set of activities of enormous complexity, dynamism, and a
growing level of demand from consumers regarding the quality of the drink. This imposes high quality control
on producer, consumer and exporter countries. Currently, the definition of the quality and, therefore, the value of
coffee is based on manual classification, that is, a person plays the role of a trained (certified) classifier to qualify
coffee samples. Thus, the current classification process suffers from the subjectivity of the classifiers and a great
difficulty in standardizing the process due to possible inconsistencies in the process. Given this scenario, the
present work proposes a comparison between three algorithms that classify coffee samples, considering shape
and imperfections. The algorithms are classifiers, one based on MLP (Multi-Layer Perceptron), another in
clustering by K-Means and the latter consists of a classifier based on Deep Learning and regional convolutional
networks (R-CNN). The objective of this work is to compare which of the algorithms is more effective in
classifying the grains according to the intrinsic defects present in the sample.