Aplicações de redes neurais e neuro fuzzy em engenharia biomédica e agronomia

Carregando...
Imagem de Miniatura

Data

2014-02-28

Autores

Silva, Inara Aparecida Ferrer [UNESP]

Título da Revista

ISSN da Revista

Título de Volume

Editor

Universidade Estadual Paulista (Unesp)

Resumo

The fuzzy and neuro fuzzy systems have been successfully used to solve problems in various fields such as medicine, manufacturing, control, agriculture and academic applications. In recent decades, neural networks have been used to the identification, assessment and diagnosis of diseases. In this thesis we performed a comparative study among fuzzy neural networks (ANFIS), multilayer perceptron neural networks (MLP), radial basis function network (RBF) and generalized regression (GRNN) in the area of biomedical engineering and agronomy. In biomedical engineering neural networks and neuro fuzzy were trained and validated with data set from patients (91 subjects, 81 healthy and 10 hemiplegic). The GRNN network had the lowest Root Mean Square Error (RMSE), but the MLP network was able to identify a case of hemiplegia. In the area of agriculture a comparative study to estimate the wheat (Triticum aestivum) productivity was proposed using neural networks. For this study it was used data from an experimental database of wheat cultivars evaluated during two years in the region of Selvíria - MS. The validation was performed by comparing the estimated productivity through the quadratic regression curve and the output of the ANFIS with the neural networks. The RMSE error calculated with the GRNN and RBF neural networks was lower than that obtained with the quadratic regression and the ANFIS. The results obtained in the study of hemiplegia were validated using the RMSE, the confusion matrix, the sensitivity, the specificity and the error accuracy. The results showed that the use of neural networks and fuzzy neural networks, in biomedical engineering, can be a viable for monitoring the progress of patients and discovery new information through a combination of parameters. In agriculture this methodology can bring benefits in combining several evaluation parameters of production to optimize production while minimize financial costs in new plantations
Os sistemas fuzzy e neuro fuzzy têm sido usados com sucesso para resolver problemas em diversas áreas, como medicina, indústria, controle, agronomia e aplicações acadêmicas. Nas últimas décadas, as redes neurais têm sido utilizadas para identificação, avaliação e previsão e dados na medicina e na agronomia. Nesta tese, realizou-se um novo estudo comparativo entre as redes neuro fuzzy (ANFIS), rede perceptron multicamadas (MLP), rede função de base radial (RBF) e regressão generalizada (GRNN) na área de engenharia biomédica. Na engenharia biomédica as redes neurais e neuro fuzzy foram treinadas e validadas com dados de pacientes hígidos e hemiplégicos (pacientes com sequela motora após acidente vascular cerebral no hemicorpo direito ou esquerdo do cérebro) coletados por meio de um baropodômetro eletrônico (91 indivíduos, sendo 81 hígidos e 10 hemiplégicos). A rede GRNN apresentou o menor erro RMSE (Raiz Quadrada do Erro Médio Quadrático), porém a rede MLP conseguiu identificar um caso de hemiplegia. Na área de agricultura foi proposto um novo estudo comparativo utilizando redes neurais para previsão de produção de trigo (Triticum aestivum). Para este estudo utilizou-se uma base de dados experimental de trigo avaliada no período dois anos na região de Selvíria-MS. A validação foi realizada comparando-se a produção estimada pelas redes neurais MLP, GRNN e RBF com a curva de regressão quadrática, comumente utilizada para este fim, e com a rede neuro fuzzy ANFIS. O erro RMSE calculado com as redes neurais GRNN e RBF foi menor do que o obtido com a regressão quadrática e com o ANFIS utilizando o treinamento (híbrido). Para validação dos resultados obtidos em hemiplegia utilizou-se o RMSE, a matriz de confusão, a sensitividade, a especificidade e a acurácia. Os resultados mostraram que a utilização das redes neurais e redes neuro fuzzy, na engenharia biomédica, pode ser uma alternativa viável para ...

Descrição

Palavras-chave

Lógica difusa, Hemiplegia, Redes neurais (Computação), Fuzzy logic

Como citar

SILVA, Inara Aparecida Ferrer. Aplicações de redes neurais e neuro fuzzy em engenharia biomédica e agronomia. 2014. 80 f. Tese (doutorado) - Universidade Estadual Paulista Júlio de Mesquita Filho, Faculdade de Engenharia de Ilha Solteira, 2014.