Research On EEG-based Emotion Recognition Using Combined Feature Extraction Methods

Emotion recognition has become a new research hotspot in biomedical,biological information,artificial intelligence and data mining.The main research content of emotion recognition is based on the characteristics of different emotions,with the computer as a tool to deal with data,in order to achieve the purpose of human-computer interaction.Studies have shown that EEG,ECG,EMG and other physiological signals are closely related to human emotions.Among them,the EEG signal has the advantages of low research cost and objectivity in the process of emotion recognition,which has become a hot research topic for domestic and foreign scholars.Therefore,the EEG signal for emotion recognition has a strong theoretical and practical significance.In the process of emotion recognition,feature extraction and classification are two key steps.There are two methods for feature extraction in domestic and foreign scholars.Most scholars used a single feature extraction algorithm,such as discrete wavelet transform,power spectral density,fast Fourier transform,autoregressive model,approximate entropy and wavelet entropy.A few scholars combined two feature extraction algorithms,such as the combination of power spectral density and autoregressive model,the combination of approximate entropy and wavelet entropy,and so on.Aiming at the characteristics of nonlinear and non-stability of EEG,this paper proposes two kinds of emotion recognition classification methods based on combinatorial feature extraction.The main work is as follows:(1)A feature extraction method based on empirical mode decomposition(EMD)and sample entropy is proposed in this paper.The proposed method first uses the EMD algorithm to decompose the EEG signals in the F3 and C4 channels.After the original EEG signal is decomposed by EMD,a series of intrinsic mode functions(IMFs)are obtained.Secondly,according to the cumulative variance contribution rate,the first several IMFs are selected to calculate the sample entropy corresponding to the different segments.Finally,the sample entropy is sent to the support vector machine as a feature vector for training and testing.In the process of the experiment,the effect of different segment lengths under different time windows on the classification accuracy is fully considered.The experimental results show that the proposed feature extraction method has obvious advantages in EEG emotion recognition compared with other methods.(2)A feature extraction method based on wavelet packet decomposition(WPD)and autoregressive model is proposed in this paper.The proposed method first uses the coherence analysis method for the channel selection of EEG emotion,and selects the optimal three channels FP1,O1 and AF4.Secondly,the EEG signals of the three channels are decomposed by WPD.Then,the autoregressive model is calculated for the leaf nodes after WPD.The autoregressive coefficients are taken as eigenvalues.Finally,support vector machine is employed to train and test,and to evaluate the classification performance.The experimental results show that the proposed feature extraction method can obtain the high accuracy of emotion recognition and has a good classification effect.