Research On EEG Emotion Recognition Based On Olfactory Stimulation

Emotion plays a vital role in people’s interpersonal communication and social activities,and influences all aspects of human daily lives.As a result,emotion recognition possess important research value and broad application prospects,and it also has become a significant research topic in the field of Human-Computer Interaction(HCI).From the perspective of information acquisition,the current research methods of emotion recognition can be summarized into non-physiological signals-based and physiological signals-based.Generally,non-physiological signals mainly include facial expressions and speech signals,and physiological signals mainly include Electroencephalography(EEG),Electrooculogram(EOG),and so on.While,because the obvious characteristics of EEG signals,such as hard to control subjectively,high signal-to-noise ratio and high precision,EEG-based emotion recognition,therefore,has become a research hotspot.Emotional elicitation is one of an important foundation of emotion recognition research.At present,the most frequently-used elicitation method is based on external stimuli.As a traditional stimulus material,video could stimulate the subject’s emotions by stimulating vision and auditory sense.As one of the human senses of olfaction has been confirmed that it is related to emotion.However,there are still few reports regarding research on EEG-based olfaction recognition.In fact,this research is beneficial to comprehend the emotional mechanism in human-brain.Hence,the EEG-based emotion recognition based on olfactory stimulation is discussed in this thesis.The main research works are as follows:(1)An Odor-Video elicited Physiological signal Database(OVPD)was established.A new experimental paradigm of emotional elicitation based on olfactory stimulation was designed,where the video stimulus was used as the basic stimulus,and the odor stimulus as the superimposed stimulus.More Concretely,32 video clips and 10 odors were used as stimulus materials to elicit the subjects’ emotions and collect their synchronous EEG signals.At the end of each stimulus,subjects were required to evaluate their emotional state,and the results of the evaluation were used as the basis for emotional classification.Based on this experimental paradigm,this thesis collected the synchronized EEG signals of eight subjects in video-only stimulation and odor-video stimulation,respectively,and built a public database of physiological signal based on odor-video stimuli.(2)A new method of EEG emotion recognition based on olfactory stimulation is proposed.Firstly,the emotional EEG signals were preprocessed to remove noise and artifacts.Secondly,adopting time windows of different lengths to divide single samples as the basic unit of emotion recognition.Thirdly,four kinds of emotional features of Power Spectral Density(PSD),Differential Entropy(DE),Differential Asymmetry(DASM),and Rational Asymmetry(RASM)were extracted.Finally,Support Vector Machine(SVM)and Convolutional Neural Network(CNN)were employed to learn and classify the positive,neutral,and negative emotion patterns.Under video-only stimulus and odor-video stimulus,the DE feature extracted from a single sample divided by a time window length of 4 s using SVM classifier obtain the best recognition accuracy,which were 97.77% and 98.95%,respectively.The results verified the effectiveness of the EEG emotion recognition based on olfactory stimulation proposed in this thesis.(3)The differences in subjective affective state label,brain topography and learning classification between video-only stimulus and odor-video stimulus were analyzed.Firstly,the deviations of the subjective emotional state labels after adding different types of olfactory stimuli.It was found that subjective emotional state label would produce different degree of deviations depending on the type of odor stimulus added,and negative odor stimuli caused an average of 1.68 maximum deviation.Then,the brain topographic maps of different emotional states under two stimuli were made in the five frequency bands(Delta,Theta,Alpha,Beta,Gamma).The results showed that activity of the lateral frontal lobe and the temporal lobe of the brain were increased,which was consistent with the findings of f MRI.Finally,under two stimulus modes,the emotional features of five single frequency bands and total frequency band were extracted to recognize emotion,respectively,the accuracy results of these features were compared between the two classifiers,and the experimental results were discussed.