Sousa, Luiz Paulo de Oliveira

Use este identificador para citar ou linkar para este item: http://repositorio.ufla.br/jspui/handle/1/37231

Título(s) alternativo(s):	Electronic tongue data analysis using principal curves
Autores:	Ferreira, Danton Diego Oliveira, Juliano Elvis Ferreira, Danton Diego Oliveira, Juliano Elvis A. Filho, Luciano Manhães de Silva, Leandro Rodrigues M.
Palavras-chave:	Língua eletrônica Reconhecimento de padrões Classificação Curvas principais Redes neurais artificiais Electronic tongue Pattern recognition Classification Principal curves Artificial neural networks
Data do documento:	15-Out-2019
Editor:	Universidade Federal de Lavras
Citação:	SOUSA, L. P. de O. Análise de dados de língua eletrônica utilizando curvas principais. 2019. 71 p. Dissertação (Mestrado em Engenharia de Sistemas e Automação)–Universidade Federal de Lavras, Lavras, 2019.
Resumo:	Electronic tongue systems are inspired by biological recognition systems, where sensory and instrumental techniques are used to determine the flavors or substances in the analyzed samples. Basically, electronic tongue are composed of sensor arrays that act as information collectors of the samples used. The data collected by the sensor arrays correspond to valuable sample information, useful in recognizing, identifying or quantifying the various constituents present, and can be obtained through sophisticated pattern recognition and data mining methods. The core objective of the is Dissertation was to analyse data from an electronic tongue system using Principal Curves, through the k-segment algorithm. Based on Principal Curves, a model capable of classifying different concentrations and identifying the substances is developed. These substances are flavor enhancers, which are used to bring about a marked flavor to the product. The enhancers analyzed were Monosodium Glutamate, Disodium Inosinate and Disodium Guanylate. These enhancers have been the subject of research aimed at replacing or decreasing the amount of sodium chloride for healthier products. The proposed classifier method using Principal Curves proved to be able to extract important information from the samples analyzed. Initially, the analyzes were performed to determine some parameters of the model, such as the number of segments used to build up the principal curves and which inputs (variables) are more relevant in these classifications. The structure of the electronic tongue system used, such as the electrode architectures and the frequency were investigated, in order to identify which are more discriminating to the model. Finally, for comparison purposes, other models were designed employing different types of artificial neural networks (Multi-Layer Perceptron, Radial Base Function and Self-Organizing Maps). As a cross-validation metric we used hold-out with twenty mutually exclusive subsets, as well as the analysis of the confusion matrices of the best results achieved. The models based on Principal Curves achieved accuracy results between 88,02% and 91,39% in classifying new events (both concentration and substance analysis) and the standard deviation range of from 2,27% to 1,28%. Perceptron Multi-Layer models achieved percentages of 74,50% to 97,03%, with standard deviation between 7,65% and 1,39%. Using the Radial Base Function, the models had 85,70% to 96,61% accuracy and the standard deviation between 2,80% and 0,83\%. With Self-Organizing Maps, the accuracy was 80,78% to 87,03% with a deviation between 2,85% and 1,04%. The models implemented using Perceptron Multi-Layer (with 2 hidden layers) and Radial Base Function neural networks had higher percentages of accuracy than the proposed method based on Principal Curves, but with the highest computational cost. The Perceptron Multi-Layer model performs 6 times operations more than the proposed method while the Radial Base Function requires about 158 times more.
URI:	http://repositorio.ufla.br/jspui/handle/1/37231
Aparece nas coleções:	Engenharia de Sistemas e automação (Dissertações)

Arquivos associados a este item:

Arquivo	Descrição	Tamanho	Formato
DISSERTAÇÃO_Electronic tongue data analysis using principal curves.pdf		2,5 MB	Adobe PDF	Visualizar/Abrir

Mostrar registro completo do item Recomendar este item Visualizar estatísticas