Research On Eeg Signal-based Emotion Recognition Methods With Rhythm And Spatio-temporal Characteristics

Emotion plays a very important role in daily life.It not only exists in the communication between people,but also relates to the individual decision-making process and the perception of external things.At present,the establishment of emotional interactions between people and computers has aroused widespread interest and attention,and the emotion recognition is its core and foundation.Emotion recognition can be achieved by various methods such as objective scales,self-evaluation,and neurophysiology.Electroencephalographic(EEG),a kind of physiological signal,has been paid close attention by researchers because of its features of simplicity,cheapness,and ease of use.Emotion recognition based on EEG signals is a cross-disciplinary research topic involving psychology,neurophysiology,data mining,and machine learning,which is still far from mature.How to find biologically meaningful knowledge and laws from EEG data,and then serve the application of brain-based artificial intelligence research and brain-computer interface,is becoming a hot and difficult topic.Based on the related researches of emotional theory and brain mechanism,this paper uses EEG signals as a carrier to explore the rhythm and spatiotemporal characteristics of the brain in the process of emotional information processing,and then to study the biological characteristics inspired emotion recognition methods and techniques.The main work is as follows:(1)Research on EEG Signal-based Emotion Recognition Methods with Rhythmic CharacteristicsNeurophysiological studies have shown that there are multiple rhythms implicit in scalp EEG signals,and these rhythms are closely related to certain special emotional cognitive functions.In order to verify the binding relationship between rhythm and emotion,the differences in the recognition of different emotional states for different rhythms were studied.Firstly,the original EEG signals are decomposed and reconstructed by using wavelet transform to obtain the four rhythm EEG signals of ?,?,?,and ?.Then,seven statistical characteristics in time domain,power spectral density in frequency domain and entropy in the nonlinear domain are extracted for each EEG signal.Finally,these features were recognized by Support Vector Machine(SVM)and Radial Basis Function Neural Network(RBFNN).Experimental results verify that rhythm plays an important role in emotion recognition and lays a foundation for further exploration of emotion recognition methods based on the rhythm and spatiotemporal characteristics.(2)Research on EEG Signal-based Emotion Recognition Methods with Rhythm and Spatial CharacteristicsPrecise emotional feature quantification and feature selection have important influence on emotion recognition.The former can effectively measure and evaluate the emotion information in the EEG signal,and the latter can effectively improve the recognition efficiency of the classifier.In order to effectively measure emotional information in EEG signal and achieve accurate emotional features.Firstly,a time-frequency analysis and correlation analysis are used to construct an emotional feature pattern based on the rhythmic and spatial fusion of multi-channel EEG signals.It is used to comprehensively and objectively quantify the rhythmic synchronization phenomenon in the process of brain information processing.Further,a core emotion feature pattern mining method based on "SS-RS"(Stability & Significance-based Rhythmic Synchronization)guideline is proposed,which can be applied to the selection of the core channel patterns under different rhythm scales.Experimental results show that the mean recognition accuracy is 81.5% for valence and 70.78% for arousal.(3)Research on EEG Signal-based Emotion Recognition Methods with Rhythm and Spatiotemporal CharacteristicsIn order to verify the temporal properties of different evoked emotions under audio and video stimuli,a EEG-based emotion recognition method combining the optimal rhythm,spatial orientation and Long-Short Term Memory(LSTM)network was proposed.First,the optimal channel and rhythm are located by using the "SS-RS" guideline.Then,the wavelet transform method is used to extract different rhythms within the located EEG signal,and then the entire time domain process of the rhythmic brain wave is decomposed into multiple equal-length sub-time process.In the time course,each sub-process is approximately stationary,and then the LSTM network is further integrated for subsequent recognition and analysis.This method makes fully use of the memory properties of the LSTM model to effectively mine the multi-scale time characteristics of the scalp EEG signal during process and expression of emotion in Brain,and explores some effective time windows for emotion recognition under audio and video stimuli.Experimental results show that the highest recognition accuracy of the emotion recognition model based on rhythm and spatiotemporal characteristics exceeds 80% and the average recognition accuracy rate is more than 75%,which shows that this method is feasible and effective in the EEG-based emotion recognition.