The Affective State Recognition From Electrocardiography Signal Based On Nonlinear Characteristics

The physiological signals along with the emotional state of bio-electric signals are generated by the body’s internal organs, with a very strong objectivity. Therefore, it is a more objective and accurate way of using the physiological signals to identify emotions. ECG is one of the most important physiological signals which has been subject to a high degree of attention in emotion recognition field. The analysis method can be divided into linear and nonlinear. The common nonlinear analysis methods are fractal dimension, complexity, approximate entropy, Lyapunov exponent and Kolmogorov entropy, and so on.In this paper, the research signals are ECG and HRV (Heart Rate Variability), which are used to do the recognition of different emotions states. By the work of signal acquisition, signal denoising, waveform positioning and nonlinear features extracting have done, finally some results of emotional states recognition on the non-linear characteristics are obtained.Specific work steps are as follows:1.Build a library of ECG data samples with different emotions. Through a predetermined careful signal acquisition program, we select the freshman as participants with physical and psychological health in our college whose ages are among17-22. And use the way of watching videos to induce their different emotions, to collect their ECG of different emotions.2.Do the signal preprocessing with the experimental ECG datas. Use the method of wavelet transform to denoising and P-QRS-T waveform location for ECG datas, then get the signals for feature extraction and HRV extraction.3.Study the different informations with emotional and emotionless states, which use the nonlinear analysis methods such as scatter plot of the QT-RR intervals, the power spectrum of the1 /f distribution and Poincare cross-sectional view. And also analyze the effect of emotion recognition contrast to the statistical characteristics of ECG in emotional identification.4.Extract non-linear characteristics (Lempel-ziv complexity, ApEn approximate entropy, Lyapunov exponent and Kolmogorov entropy) of HRV with the emotions of angry, disgust, fear, grief, happy, surprise and calm, to analyze the effects of emotion recognition in seven types of emotional states with SVM classifier.Through the analysis of emotional ECG and HRV signals’nonlinear characteristics, use the support vector machine (SVM) classifier for classification, we can get some emotional state recognition results. They are as follows:1.With the non-linear characteristics of HRV, there is a more clear distinction in the two emotional states of non-excited emotional state like Calm, and excited emotional states such as angry, disgust, fear, grief, happy, surprise.2.Through the HRV’s non-linear characteristics analysis of Lempel-ziv complexity, ApEn approximate entropy, the largest Lyapunov exponent and Kolmogorov entropy, it found that the values of characteristics order the rules are as follows:the value of fear is the maximum and calm is the minimum. Grief, angry, and happy’s values are smaller than fear’s, next are disgust and surprise.3.For the HRV’s non-linear characteristics of Lempel-ziv complexity, ApEn approximate entropy, Lyapunov exponent and Kolmogorov entropy, the characteristic values of the female are larger than male’s.4.Between the emotionless state of calm and emotional states of another six emotions, whether one-to-one emotion recognition or one-to-many emotion recognition, it has achieved a better recognition results. In the recognition of six excited emotional states, the emotions of Fear, Happy, Disgust and Angry’s recognition are better in the one-to-one emotion recognition and the one-to-many emotion recognition, but the emotions of Grief and Surprise’s recognition are not good.5.From the optimal feature combination of emotion recognition, found that the Lempel-Ziv complexity and the largest Lyapunov exponent are the effective nonlinear characteristics.From the above results and analysis, it found that the non-linear characteristics are feasible in the study of emotional identification based ECG and HRV signals, and have certain regularities.