基于脑电信号间Granger因果关系特征的情感识别
|
摘要
鉴于人情感变化时不同脑区与频段间脑电(EEG)信号会交互作用、形成因果相关信息流,其特征应是情感识别的关键点。因此拟将不同脑区通道、不同频段的EEG信号之间Granger因果系数差值作为特征在愉悦度、唤醒度以及控制度这3个维度上分别进行二分类的情感识别以进一步提升其效果。采用国际情感数据集DEAP,首先小波包变换将脑电信号分解成θ、α、β、γ4个频段,双密度双树复数小波变换对信号去噪,再分别计算每个频段以及每两个频段组合下各脑区之间因果关系值,将它们分别单独作为特征,用支持向量机(SVM)分类,最后将识别率高(>70%)的频段特征送入分类器,情感分类识别对象是单独被试,32名被试的平均识别率为96%,用功率谱密度(PSD)、不对称系数(AI)、能量、熵这4种特征的识别率为80%左右,因此将Granger因果关系用于基于脑电的情感分类与识别是一种有效提高情感识别率的方法。
When human's emotion changed,the electroencephalography( EEG) signals among different brain regions and frequency bands would interact and form causality information flow,and its feature should be the key point of emotion recognition. Doing binary classification in the three dimensions of valence,arousal and dominance to improve its effect further by using Granger causality coefficient differences between EEG signals of channels in different brain regions and different frequency bands as feature. We used DEAPdataset1 which was an international emotion dataset. First,EEG signals were decomposed into four frequency bands which were theta,alpha,beta and gamma band by wavelet packet transform. Then they were denoised with double density dual-tree complex wavelet transform. The causality values among different brain regions in each frequency band and each combination of two bands were calculated. Each of them was taken as individual feature and classified by support vector machine( SVM). The frequency band features with high recognition accuracy( > 70%) were sent to the classifier. The emotion classification and recognition object was single subject. The average recognition accuracy of 32 subjects was 96%. The recognition accuracy of power spectral density( PSD),asymmetric idex( AI),energy and entropy,was about 80%. Therefore,the application of Granger causality to emotion classification and recognition based on EEG is an effective way to improve the emotion recognition accuracy.
When human's emotion changed,the electroencephalography( EEG) signals among different brain regions and frequency bands would interact and form causality information flow,and its feature should be the key point of emotion recognition. Doing binary classification in the three dimensions of valence,arousal and dominance to improve its effect further by using Granger causality coefficient differences between EEG signals of channels in different brain regions and different frequency bands as feature. We used DEAPdataset1 which was an international emotion dataset. First,EEG signals were decomposed into four frequency bands which were theta,alpha,beta and gamma band by wavelet packet transform. Then they were denoised with double density dual-tree complex wavelet transform. The causality values among different brain regions in each frequency band and each combination of two bands were calculated. Each of them was taken as individual feature and classified by support vector machine( SVM). The frequency band features with high recognition accuracy( > 70%) were sent to the classifier. The emotion classification and recognition object was single subject. The average recognition accuracy of 32 subjects was 96%. The recognition accuracy of power spectral density( PSD),asymmetric idex( AI),energy and entropy,was about 80%. Therefore,the application of Granger causality to emotion classification and recognition based on EEG is an effective way to improve the emotion recognition accuracy.
引文
[1]MOHAMMADI Z,FROUNCHI J,AMIRI M.Wavelet-based emotion recognition system using EEG signal[J].Neural Computing&Applications,2017,28(8):1985-1990.
[2]THURAISINGHAM R A,TRAN Y,CRAIG A,etal.Frequency analysis of eyes open and eyes closed EEG signals using the Hilbert-Huang transform[C].Engineering in Medicine and Biology Society,IEEE,2012:2865-2868.
[3]GRANGER C W J.Investigating causal relations by econometric models and cross-spectral methods[J].Econometrica,1969,37(3):424-438.
[4]HU S,DAI G,WORRELL G A,etal.Causality analysis of neural connectivity:Critical examination of existing methods and advances of new methods[J].IEEETransactions on Neural Networks,2011,22(6):829-844.
[5]HU S,WANG H,ZHANG J,etal.Comparison analysis:Granger causality and new causality and their applications to motor imagery[J].IEEE Transactions on Neural Networks&Learning Systems,2016,27(7):1429-1444.
[6]CHEN D,FANG W,ZHEN W,etal.Eeg-based emotion recognition with brain network using independent components analysis and granger causality[C].International Conference on Computer Medical Applications,IEEE,2013:1-6.
[7]BRESSLER S L,SETH A K.Wiener-Granger causality:A well established methodology[J].Neuroimage,2011,58(2):323-329.
[8]ARNOLD A,LIU Y,ABE N.Temporal causal modeling with graphical granger methods[C].ACM SIGKDD International Conference on Knowledge Discovery and Data Mining,2007:66-75.
[9]GUPTA R,LAGHARI K U R,FALK T H.Relevance vector classifier decision fusion and EEG graph-theoretic features for automatic affective state characterization[J].Neurocomputing,2016,174:875-884.
[10]DAIMI S N,SAHA G.Classification of emotions induced by music videos and correlation with participants’rating[J].Expert Systems with Applications,2014,41(13):6057-6065.
[11]KOELSTRA S,MUHL C,SOLEYMANI M,etal.DEAP:A database for emotion analysis;using physiological signals[J].IEEE Transactions on Affective Computing,2012,3(1):18-31.
[12]SELESNICK I W.The double-density dual-tree DWT[J].IEEE Transactions on Signal Processing,2004,52(5):1304-1314.
[13]SARAWALE R K,CHOUGULE S R.Noise removal using double-density dual-tree complex DWT[C].IEEE Second International Conference on Image Information Processing,2014:219-224.
[14]MAO C L,SHEN H.Image denoising by directional complex diffusion processes based on double-density dual-tree DWT[C].International Congress on Image and Signal Processing,IEEE,2010:698-702.
[15]NAKAYAMA K,INAGAKI K.A brain computer interface based on neural network with efficient pre-processing[C].International Symposium on Intelligent Signal Processing and Communications,IEEE,2007:673-676.
[16]NAIJEN H,PALANIAPPAN R.Classification of mental tasks using fixed and adaptive autoregressive models of EEG signals[C].Annual International Conference of the IEEE Engineering in Medicine and Biology Society,2005:507-510.
[17]ROEBROECK A,FORMISANO E,GOEBEL R.Mapping directed influence over the brain using Granger causality and f MRI[J].Neuroimage,2005,25(1):230-242.
[18]CUI J,XU L,BRESSLER S L,etal.2008 Special issue:BSMART:A MATLAB/C toolbox for analysis of multichannel neural time series[J].Neural Networks,2008,21(8):1094-1104.
[2]THURAISINGHAM R A,TRAN Y,CRAIG A,etal.Frequency analysis of eyes open and eyes closed EEG signals using the Hilbert-Huang transform[C].Engineering in Medicine and Biology Society,IEEE,2012:2865-2868.
[3]GRANGER C W J.Investigating causal relations by econometric models and cross-spectral methods[J].Econometrica,1969,37(3):424-438.
[4]HU S,DAI G,WORRELL G A,etal.Causality analysis of neural connectivity:Critical examination of existing methods and advances of new methods[J].IEEETransactions on Neural Networks,2011,22(6):829-844.
[5]HU S,WANG H,ZHANG J,etal.Comparison analysis:Granger causality and new causality and their applications to motor imagery[J].IEEE Transactions on Neural Networks&Learning Systems,2016,27(7):1429-1444.
[6]CHEN D,FANG W,ZHEN W,etal.Eeg-based emotion recognition with brain network using independent components analysis and granger causality[C].International Conference on Computer Medical Applications,IEEE,2013:1-6.
[7]BRESSLER S L,SETH A K.Wiener-Granger causality:A well established methodology[J].Neuroimage,2011,58(2):323-329.
[8]ARNOLD A,LIU Y,ABE N.Temporal causal modeling with graphical granger methods[C].ACM SIGKDD International Conference on Knowledge Discovery and Data Mining,2007:66-75.
[9]GUPTA R,LAGHARI K U R,FALK T H.Relevance vector classifier decision fusion and EEG graph-theoretic features for automatic affective state characterization[J].Neurocomputing,2016,174:875-884.
[10]DAIMI S N,SAHA G.Classification of emotions induced by music videos and correlation with participants’rating[J].Expert Systems with Applications,2014,41(13):6057-6065.
[11]KOELSTRA S,MUHL C,SOLEYMANI M,etal.DEAP:A database for emotion analysis;using physiological signals[J].IEEE Transactions on Affective Computing,2012,3(1):18-31.
[12]SELESNICK I W.The double-density dual-tree DWT[J].IEEE Transactions on Signal Processing,2004,52(5):1304-1314.
[13]SARAWALE R K,CHOUGULE S R.Noise removal using double-density dual-tree complex DWT[C].IEEE Second International Conference on Image Information Processing,2014:219-224.
[14]MAO C L,SHEN H.Image denoising by directional complex diffusion processes based on double-density dual-tree DWT[C].International Congress on Image and Signal Processing,IEEE,2010:698-702.
[15]NAKAYAMA K,INAGAKI K.A brain computer interface based on neural network with efficient pre-processing[C].International Symposium on Intelligent Signal Processing and Communications,IEEE,2007:673-676.
[16]NAIJEN H,PALANIAPPAN R.Classification of mental tasks using fixed and adaptive autoregressive models of EEG signals[C].Annual International Conference of the IEEE Engineering in Medicine and Biology Society,2005:507-510.
[17]ROEBROECK A,FORMISANO E,GOEBEL R.Mapping directed influence over the brain using Granger causality and f MRI[J].Neuroimage,2005,25(1):230-242.
[18]CUI J,XU L,BRESSLER S L,etal.2008 Special issue:BSMART:A MATLAB/C toolbox for analysis of multichannel neural time series[J].Neural Networks,2008,21(8):1094-1104.