Research On Emotional Recognition Based On ECG And PPG Signals

Emotion recognition and affective computing are the key parts and the hot research topics in the field of human-computer emotional interaction.Currently,the emotion recognition is mainly based on voice intonation,facial expressions,body postures,text content,physiological signals and so on.Among them,the research on emotion recognition based on physiological signals is particularly prominent.It mainly includes the research of physiological signal collection,feature extraction and feature recognition.The thesis focuses on the experimental design of emotion-inducing paradigm,the collection and pretreatment of ECG and PPG signals,the extraction of affective physiological signal features,and the emotional recognition of physiological features in different emotional models.The main research contents are as follows:1)Emotion induced experimental paradigms design.The different video clips of 8 emotions,including happiness,sadness,surprise,pride,anger,fear,sensation,and disgust,were obtained through the rigorous questionnaires to emotion intensity as emotional evoked materials,and the Superlab software was used to design the emotional evoked experimental paradigm,then the subjects were induced to produce corresponding positive and negative emotions.2)ECG and PPG signals collect and preprocess.Using MP150 physiological signal measuring instrument,the physiological signals were collected under the different emotional states.A band-stop filter of Chebyshev type II was used to filter out 50 Hz power frequency interference.And the zero phase shift digital filter was used to remove the baseline drift of the signal;The Butterworth band-pass filter and the wavelet threshold denoising method were designed to eliminate the motion artifacts and high frequency noise in the measurement and obtain pure emotional physiological signals.3)Emotional physiological signal features extract.The methods to peak detection and wavelet transform were used to extract the linear features of ECG and PPG signals in different emotional states,including the time-domain statistical feature and wavelet time-frequency feature.The method to nonlinear dynamic analysis was applied to obtain the nonlinear characteristics of emotional physiological signals.4)ECG and PPG emotional feature recognize.Design Bayesian classifier to classify the positive and negative sentiment based on linear feature of ECG and PPG signals.And the classification accuracy rates of 79.1% and 75.9% were obtained respectively.Design deep neural network classifiers and stack self-coding deep learning algorithms,perform emotional recognition on the linear and nonlinear fusion features to ECG and PPG signals,and the highest classification recognition accuracy rate reached 97.6% and 96.3% respectively.5)Compare with the recognition results of three emotional modes,including the positive and negative emotion fusion patterns,single emotion patterns,and basic emotion patterns.The results have shown that the recognition accuracy rates is higher based on ECG feature than PPG feature under three different emotional modes,through the fusion feature vector than by the basic linear feature vector under the same classifier,and using the stack self-coding deep learning algorithm than using Bayesian method.Therefore,for three different emotion modes,we can use ECG signal,linear and nonlinear fusion features and stack self-encoding deep learning algorithm for obtaining good emotion recognition.The research results of the dissertation have important scientific and applied values in the areas of emotional robots,medical big data,psychology,situational learning,multimedia game development,and business.