Research On A Bionic Model Of Olfactory Systems And Its Applications Based On KⅢ Model

In the aspect of olfactory nerve system research, the researchersconstructed corresponding sense of smell model based on neuralphysiological experiment. The existing models do not research the wholeolfactory neural system, so it is difficult to reflect the characteristics ofolfactory neural system on the overall. The thesis studied the olfactorynerve system model based on KIII model.Firstly, Based on nonlinear dynamic index, the stability of the KIIImodel was researched in this thesis. Qualitative respect, analyzed thestability of KIII model by numerical analysis and phase graph, statechange of the model can be intuitive saw from the result. Quantitativerespect, made a quantitative study when the KIII model was used topattern recognition by Lyapunov exponent based on wolf method. Thecalculation results show that KIII model can change from chaotic stage tostable stage quickly and presents obvious synchronization stage, which isconsistent with the analysis result drawn from phase graph. It is also shown that Lyapunov exponent is an effective method to judge thestability of KIII model.Secondly，according to the biology characteristics of olfactory neuralsystem, an olfactory neural system integrated model was proposed basedon bionics. The proposed olfactory model simulated olfactory neuralsystem from different aspects in order to make the model similar to thecharacteristics of olfactory neural system. To the structure of olfactoryneural system, the model simulated three key parts of olfactory neuralsystem, the delay and feedback among them based on the anatomicalstructure of olfactory neural system. To the neuron model, differentmodels were used to simulate different neurons in different parts ofolfactory neural system. To the dynamics of model, the model can giveresponse to stimulus. When the stimulus was durative, the output of modelshowed steady pattern. When the special stimulus was inputted, the outputof model showed the pattern of limit cycle.Finally, the thesis studied the olfactory nerve system integrationmodel when apply it to face recognition. Formulated the learning rule according to the model structure, extracted the characteristics of faceimage based on sub-image division. In the feature extraction process, threemethods (standard deviation coefficient, singular value decomposition,discrete cosine transform) were used to deal with. After learning, get theresult when the model was used to the face recognition. The experimentalresults show that the three methods could effective to extract the faceimage characteristics. In the high dimension characteristic, the singularvalue decomposition and discrete cosine transform can obtain a higherrecognition rate. The new olfactory nerve system model showed a goodperformance in face recognition.