Identificação de Viabilidade de Leveduras Com Corante Vital Utilizando Histogramas de Palavras Visuais em Imagens Coloridas

This article presents a proposal to automate the classification of viability for yeast Saccharomyces cerevisiae species, responsible for the commercial production of ethanol, using as attribute the color absorbed by vital dye methylene blue. The methodology is used widely in the Brazilians distilleries and consist in count the colorless cells that are considered feasible, separate them from colored blues considered non-viable. The number of viable cells per liter interferes with the industrial yields. Since this count is tiring and results in errors, an alternative technique is introduced as computer vision defined as the bag-of-word algorithm (histogram of visual words), and extensions that add color information’s and that can be added to the algorithm, because the Bag-of-word is used for images in grayscale. The attributes extracted with this algorithm with its extensions were used for testing and training of classifiers extracted from supervised learning algorithm. Among the used algorithm we can highlight J48, SMO, Naives Bayes and IBK that are implemented in the WEKA environment. The results were analyzed using ANOVA that showed value-p < 2e-16 indicating a statistical difference of the techniques. The Opponent Color technique showed better results, representing a potential application in real of the distilleries.