Open AccessViable Yeast Identification using Bag of Visual Words in Colored images
Author(s) -
JADHE CORREIA BALBINO DE SOUZA,
Vanessa Weber,
Ariadne Barbosa Gonçalves,
Marco Álvarez,
Marney Pascoli Cereda,
Wesley Nunes Gonçalves,
Valguima Odakura,
Hemerson Pistori
Publication year - 2020
Language(s) - English
Resource type - Conference proceedings
DOI - 10.5753/wvc.2020.13493
Subject(s) - artificial intelligence , naive bayes classifier , colored , pattern recognition (psychology) , support vector machine , computer science , computer vision , population , c4.5 algorithm , materials science , demography , sociology , composite material
In this research it is reported a system to automate the process of identification of viable yeasts whose population control is a crucial task in the ethanol production process. The identification and counting of yeasts made by human vision under a light microscope, is repetitive and susceptible to errors. We used computer vision techniques such as BoVW, Color Coherence Vectors (CCV), Color Moments (CM), Bag-of-Color (BoC) and Opponent Color (OpC) were applied for extracting characteristics that were classified by the Naive Bayes, KNN, SVM and J48 algorithms in 2614 images of yeasts separated into three classes: viable, non-viable and background. The results were analyzed using software R, which in the ANOVA test resulted in a p value equal to 2e16 indicating a significant difference between the techniques. The OPC with SVM classifier showed the highest performance using the PCC Percent Correct Classification metric, about 95% compared to other techniques.