Research On Emotional Physiological Features And Feature Variabilities

Psychological illness is rising all over the world.At present,the prevention and diagnosis of clinical psychological illness mainly depends on the psychological doctor’s experience,without a more scientific and objective evaluation basis.Emotion recognition can objectively and scientifically prevent and diagnose psychological illness,which provides a reliable basis for the treatment of psychological illness.The scientific objective evaluation of each individual’s emotional state based on physiological signals has important social significance and clinical value for the prevention and diagnosis of mental illness.However,the existing studies mainly focus on the classification accuracy tests based on the different machine learning methods,without a comprehensive exploration for the inherent links between emotion states and physiological features.Aiming at the problem of the unclear specificity relationship between physiological characteristics and emotional state,this article will investigate the physiological features and feature variabilities among different emotional states.The time domain,frequency domain and nonlinear indices of heart rate variability(HRV)and pulse transit time variability(PTTV)are calculated by extracting RR series and pulse transit time series from electrocardiography(ECG)and photoplethysmography(PPG)signals to analyze the differences of the physiologic variability characteristics among different emotion states.And finally,a support vector machine(SVM)classifier based on the physiologic variability and the grid search parameters optimization is builded to classify emotions.Main works are as follows:(1)Calculating the features of HRV and PTTV.Wavelet transform(sym8 wavelet)was used for ECG and PPG pre-processing.Adaptive difference threshold method is used to locate the R peaks of ECG and thus to construct RR interval(RRI)time series.A wapb method is used to detect the onset of PPG and thus to construct pulse transit time(PTT)series.Then the HRV indices and PTTV indices,including time-domain(MEAN,SDNN,RMSSD and PNN50),frequency-domain(LFn,HFn and LF/HF)and nonlinear indices(SampEn and FuzzyMEn)are calculated.(2)Investigating the differences of HRV and PTTV indices between two opposite emotion states:happiness and sadness.Precise physiological signal acquisition experiments are performed to synchronously collect the ECG and PPG signals under the two emotion states.Then nine HRV indices and eight PTTV indices are calculated,and the differences,Pearson correlation coefficients and the HR-and MAP-related changes of of all indices between two emotion states are analyzed.The results shows that multiple HRV and PTTV indices have significant differences between happiness and sadness emotion states,and the change trends of HRV and PTTV indices is consistent.(3)Investigating the differences of HRV and PTTV indices among various emotions.A more complete physiological signal acquisition experiments that contains six emotions(neutral,happiness,sadness,fear,anger and disgust)are performed.The HRV and PTTV indices are calculated for each emotion state.The differences between neutral and other emotion states are compared.The results show that different emotion has different impact on HRV and PTTV indices,thus the HRV and PTTV indices can be used as characteristic parameters of emotion recognition.(4)Establishing emotion recognition models based on physiological variability characteristics,grid search algorithm and SVM.All variability characteristics mentioned above and statistically significantly difference variability characteristics are respectively used as feature vector.Decision-making machines of different emotion states are trained and established through GSA-SVM.The recognition effects of two groups of features as the input vector are compared.The results show that emotion recognition using physiological variability characteristics is effective.