Sistema neural hÃbrido para reconhecimento de padrÃes em um nariz artificial / Hybrid neural system forpattern recognition in an artificial nose
AUTOR(ES)
Cleber Zanchettin
DATA DE PUBLICAÇÃO
2004
RESUMO
This dissertation investigates the use of Hybrid Intelligent Systems in the pattern recognition system of an artificial nose. The work involves five main parts: (1) an evaluation of the odors database by a multivariate statistics technique; (2) a validation of the Time Delay Neural Networks in the odors recognition; (3) an evaluation of the Wavelet Transform as preprocessing method of odors signals in the connectionist approaches; (4) an evaluation of hybrid intelligent systems for the odor recognition in artificial noses; and (5) a case study. Two intelligent hybrid architectures had been investigated: the neuro-fuzzy Feature-weighted Detector network, which allows the pattern classification, feature selection and rule extration of the network; and the neuro-fuzzy Evolving Fuzzy Neural Networks, that enables on-line and incremental learning, insertion, extration and aggregation of knowledge in its evolutive architecture. The signals generated by an artificial nose, composed by an array of eight conducting polymer sensors, exposed to the gases derived from the petroliferous industry were analyzed. The use of the Wavelet Transform improved the performance of the connectionist classifiers. In the experiments, the Time Delay Neural Networks obtained a mean classification error of 0.75%, while the Multi-Layer Perceptron obtained an error of 11.5%. In the investigated hybrid systems, the Feature-weighted Detector obtained a mean classification error of 20.72%. The Evolving Neural Fuzzy Networks obtained a mean classification error of 0.88% in the odors classification
ASSUNTO(S)
redes neurais artificiais reconhecimento de padrÃes narizes artificiais hybrid intelligent systems, pattern recognition, artificial noses, artificial neural networks sistemas hÃbridos inteligentes ciencia da computacao
Documentos Relacionados
- Um estudo comparativo de tÃcnicas conexionistas na implementaÃÃo de um sistena de reconhecimento de padrÃes para um nariz artificial
- Comparing arquitecture of neural networks for pattern ecognation system on artificial noses
- RAPID: uma metodologia para a recuperaÃÃo, armazenamento e reconhecimento de padrÃes em imagens digitalizadas.
- Sistema de reconhecimento de padrÃes para identificaÃÃo de porte de veÃculos atravÃs de anÃlise de perfil magnÃtico
- Um ambiente hÃbrido inteligente para previsÃo de acordes musicais em tempo real