Desenvolvimento de máquinas de vetor suporte para a classificação de café arábica verde por espectroscopia de infravermelho médio

Abstract

Brazil is the world's largest producer and exporter of coffee being an important economic commodity in the country. The two species of greatest economic value are canephora and arabica, being the last one considered of greater economic value by generating a better quality beverage. Climate, species, cultivation method and industrialization are also critical for the final quality of the beverage. The objective of this study was to develop a methodology that is capable to discriminate different green arabica coffee genotypes and also their geographical origin by using mid-infrared spectroscopy with Fourier transform (FTIR) and support vector machines (SVM). Therefore, 74 FTIR spectra were collected from 20 different genotypes planted in the cities of Paranavaí, Cornélio Procópio, Mandaguari and Londrina. To analyze the spectra were built SVMs using radial basis as kernel function and the oneagainst-all multiclass approach. The developed SVM were evaluated by sensitivity and specificity for the test samples. For the geographic origin the samples were successfully classified with an average sensitivity of 97.5% and average specificity of 96.9%. Otherwise, for genotypic classification the performance was not satisfactory with an average sensitivity of 66.0% and a specificity of 95.6%. Furthermore, the geographical classification proved to be easier because fewer samples were selected as support vectors. The unbalance in the number of samples for genotype classification problem can be the cause of poor sensitivity of the SVM. Thus, it is suggested to search for other approaches to multiclass problems for the improvement of the proposed models.