Study Of Neural Network Model Of Olfactory System And Its Applications On Artificial Olfaction And Iris Recognition

This dissertation introduces a novel chaotic neural network - K set model and its applications in some fields. The model is setup according to the hierarchy of neural population based on electro-psychological experimental results. It describes the processing mechanism of electrical signal of olfactory systems. KIII model is a high-dimensional chaotic neural network, which is established in accordance with the entire neural architecture of mammalian olfactory system. It can not only simulate EEG waveform observed in experiments, but also perform bionic intelligence on pattern recognition. Significantly, it can provide a novel demonstration for the signal processing mechanism of neural system in the aspect of nonlinear dynamics.As presented in this dissertation, some numeric analyses of the K set model, including K0, KI, KII and KIII model, are performed. The consequence of these analyses leads to the fact that the KIII network can simulate the biological olfactory neural system well. Different from the conventional artificial neural networks, the KIII network works in its chaotic trajectory. The learning and pattern recognition process of the KIII network also relates to its memory basins, which are some kind of chaotic attractors.The KIII model is applied to artificial olfaction and iris recognition. And a prototype of "electronic nose-brain" is designed in artificial olfaction field. The performance of the KIII network is also compared with those results using conventional artificial neural networks. The results prove that the KIII network is a good pattern classifier.In the last part of this dissertation, the research on integrating new recognition method with the KIII network is discussed. An SVM (Supporting Vector Machine) classifier is successfully introduced to replace the Euclidian distance classifier in KIII network. And the SVM classifier integrated KIII network performs better then the KIII network using Euclidian distance classifier in recognition of two kinds of tea scents.