Revista Facultad Nacional de Agronomía, Medellín
versão impressa ISSN 0304-2847
SANDOVAL NINO, Zulma Liliana e PRIETO ORTIZ, Flavio Augusto. AN ARTIFICIAL VISION SYSTEM FOR CLASSIFICATION OF COFFEE BEANS. Rev.Fac.Nal.Agr.Medellín [online]. 2007, vol.60, n.2, pp. 4105-4127. ISSN 0304-2847.
An artificial vision system for classification of coffee beans, in eleven categories, according to its state of maturity was developed. The description of the coffee beans was done by using 208 characteristics (form, color and texture characteristics). The reduction of the set of characteristics from 208 to 9 was done by using two methods of characteristic selection. The final set of characteristics is composed by 4 texture characteristics, 3 color characteristics and 2 shape characteristics. This final set was evaluated in two classifiers: The Bayesian and a neuronal networks classifier. The classification error obtained by the Bayesian classifier was 5,43%, it required 5,5 ms for the classification process, while the error obtained by neuronal networks classifier was 7,46% and the classification time decreased to 0,8 ms.
Palavras-chave : Pattern recognition; coffee; pattern classification; digital image processing.