Research On Odor And Induced Emotion Recognition Based On Olfactory EEG

The sense of smell is the oldest sensation in the history of biological evolution,and is closely related to human memory,learning and emotions.The cerebral cortex is the highest-level nerve center that can evaluate stimuli from various senses.Studying the ability of the brain to recognize different odors is of great significance in the evaluation and diagnosis of olfactory dysfunction,emotional regulation of patients with mental disorders such as depression.In recent years,the study of olfaction based on electroencephalogram(EEG)technology has gradually attracted the attention of scholars from various countries,and it has been used to identify odor types and emotions.However,most of the current researches are based on a few features such as power spectrum density(PSD)of EEG signals.They have limited ability to recognize odors and emotions,and lack the recognition of odor concentration and more elaborate emotions.In view of the existing problems,based on the olfactory EEG data set constructed by the research team,this dissertation focuses on three issues of odor type recognition,odor concentration recognition and olfactory induced emotion recognition,and has carres out related research from the two aspects of feature extraction and classification of EEG signals.Firstly,two olfactory EEG data sets are constructed.Based on the established olfactory EEG experiment platform,the EEG data and emotional subjective evaluation labels are collected from eleven volunteers sniffing thirteen types of odors,and the data set I is built for the identification of different types of odors and the corresponding emotions.The EEG data and emotional subjective evaluation labels collected from thirteen volunteers sniffing five different concentrations of odors(two odors in total)are used to build the data set II for the identification of different concentrations of odors and the corresponding emotions.The two olfactory EEG data sets lay the foundation for the subsequent study of olfactory EEG.Secondly,for the recognition of different odor types,a classification algorithm based on trapezoidal-difference features is proposed.The core of the algorithm is to construct an N-layer trapezoidal feature set with a size ratio of 1:2:1 for the top,bottom and height for each frequency band of each EEG sample,where N is the number of used EEG electrodes;then each layer's values of the trapezoidal feature set are sorted to get the corresponding electrode sequence(ES)codes;finally,a category's ES codes that are most similar to those of the testing sample are found,and the corresponding category is taken as the predicted category result.The experimental results based on the olfactory EEG data set I show that,for the identification of thirteen different types of odors,nine types of pleasant odors,and four types of unpleasant odors,the proposed average-frequency band division method of the EEG signal is superior to that of the biological-rhythm band division;the proposed classification algorithms have achieved superior classification performance compared with six classical classification methods.Thirdly,for the recognition of odors with different concentrations,a classification algorithm based on double-square features is proposed.The core of the algorithm is to construct two square feature sets with a number of N layers for each frequency band of each EEG sample,one is a sum feature set and the other is a difference feature set;then the values of each layer of each square feature sets are sorted to obtain the corresponding electrode sequence(ES)code;finally,the number of inconsistencies between the ES code of the test sample and that of each category is found,and the category corresponding to the smallest inconsistency number is used as the predicted category result.The algorithm based on double-square features does not need to use a special classifier while enriching features.The experimental results based on the olfactory EEG data set ? show that,for the identification of five different concentrations of pleasant 'rose' odors and five different concentrations of unpleasant 'rotten' odors,the proposed average-frequency band division of the EEG signal is better than that based on biological rhythm;and the proposed classification algorithms have achieved superior performance than the six classical algorithms,indicating that different concentrations of odors can be effectively distinguished by using EEG signals,and the proposed double square based algorithm is suitable for odor recognition with different concentrations.Finally,the recognition of emotions induced by different types of odors and different concentrations of odors is studied.For the former,recognition of pleasure and disgust,and recognition of different valence dimension emotions are carried out.The experimental results verify that,for different types of odor-induced emotion recognition,compared with the biological-rhythm band division method,the proposed EEG average-band division method can achieve better recognition accuracy;Compared with Naive Bayesian,k-nearest neighbor,voting-extreme learning machine and back propagation neural network algorithms,the support vector machine algorithm achieves better recognition results.Similarly,for the latter,recognition of pleasure and disgust,as well as different valence dimension emotions is carried out.The experimental results prove that,emotion recognition induced by different concentrations of odors can be realized through EEG signal processing,and it is the first time to prove that,for the recognition of different concentrations of odors,the recognition effect of the EEG signal Gamma band is better than that of other bands(Delta,Theta,Alpha and Beta);In addition,compared with Naive Bayesian,k-nearest neighbor,voting-extreme learning machine and back propagation neural network algorithms,the support vector machine algorithm still achieves better recognition results.