Classificação multivariada de biodiesel B100 e B5 usando imagens digitais

• Ano de defesa: 2015
• Autor(a) principal: Costa, Gean Bezerra da
• Orientador(a): Veras Neto, Jose Germano
• Banca de defesa: Meneses, Carlos Henrique Salvino Gadelha, Lyra, Wellington da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual da Paraíba
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciências Agrárias - PPGCA
• Departamento: Pró-Reitoria de Pós-Graduação e Pesquisa - PRPGP
• País: BR
• Palavras-chave em Português: Biocombustíveis; Quimiometria; Webcam
• Palavras-chave em Inglês: Chemometrics; Biofuels
• Área do conhecimento CNPq: CIENCIAS AGRARIAS
• Link de acesso: https://repositorio.uepb.edu.br/handle/123456789/75420

Objective of this work is to present a simple, fast, inexpensive and non-destructive methodology based on the use of digital images and chemometric techniques, classification for biodiesel and biodiesel blends rating / diesel (B5) with the type of oil origin (cotton, sunflower, corn and soya). For this, images of biodiesel and mixtures were obtained from a webcam and then were resolved into histograms containing frequency distributions of the RGB color levels, HSI, grayscale, individually, gray + RGB, gray + HSI and a model using all histograms together (grayscale + RGB + HSI) were used. An exploratory analysis of the data was performed using the PCA to judge whether it is possible to identify similarities and differences between the data set sample used in the construction of classification models. The data obtained from each histogram have been partitioned into training and test sets using the Kennard-Stone algorithm (KS). Then were built SIMCA classification models (modeling Independent and flexible by Class Analogy), PLS- DA (discriminant analysis by Partial Least Squares) and LDA (Linear Discriminant Analysis) using variable selection algorithms SPA (Algorithm of Successive Projections). For the classification of biodiesel in terms of the origin of oil despite all the models are well adjusted, the SPA-LDA model stood out with as a result 100% of biodiesel samples classified correctly in its proper class. In the classification of the mixtures biodiesel / diesel B5 results of the four models tested were satisfactory, but the best results for SPA with a LDA model correct classification rate of 87.50% and 97.50% for the training sets and testing respectively. These results suggest that the proposed models are promising alternatives for classification biodiesel and mixtures biodiesel / diesel (B5) in terms of starting oil. Advantageously, the analysis is fast, does not use reagent and chemical characterization of the samples is not necessary.