Research On Emotion Recognition Based On Multiple Physiological Signals

Emotion is a complex state,which is an instinctive response of people when facing different stimuli.It plays an important role in people’s life,work and study.Positive emotions can improve people’s quality of life,while negative emotions can have a bad impact on people’s physical and mental health.Therefore,the monitoring and recognition of emotions are of great significance to the development of human beings.This thesis designs a set of experimental schemes for emotion recognition based on multiple physiological signals.Video clips are used as emotion-inducing materials to induce different emotional states of subjects,and different physiological signal sensors are used to collect physiological signals of subjects in different emotional states.The relationship between different physiological signals and different emotional states is analyzed from two perspectives of statistics and machine learning.From the perspective of statistics,regression analysis is used to explore the relationship between body temperature signals,heart rate signals and emotional arousal levels.At this stage,according to the VA two-dimensional emotional model,the emotional arousal level is divided into three levels:emotional excitement,emotional calmness and emotional depression.Taking physiological signal as independent variable and arousal level of emotion as dependent variable,correlation analysis is carried out to get the conclusion that the correlation between dependent variable and independent variable is very high.Then curve estimation is used to fit the curve between independent variable and dependent variable,and the analysis shows that the data distribution is the most consistent with the linear function,and a general predictive equation between physiological signals and emotional arousal levels is obtained.From the perspective of machine learning,five physiological signals of brain electricity,electrocardiogram,respiration,pulse wave and skin electricity are used for emotion recognition.At this stage,according to relevant research at home and abroad,emotions are divided into four basic emotional states: happy,sad,angry,and fearful.After processing and feature extraction of physiological signals,a variety of classification models are constructed and emotion recognition is performed.And select several kinds of physiological signals that can best reflect the emotional state and combine them,and find a set of physiological signals that still have a good emotional recognition effect when the number of signals is as few as possible.In the process of measurement,it is found that the accuracy of emotion recognition is related to the number of signals and the size of the data set.The more the number of signals,the larger the data set,the higher the accuracy of emotion recognition;the less the number of signals,the smaller the data set,the lower the accuracy of emotion recognition.It can be seen from the accuracy that the random forest algorithm has a better effect on emotion recognition.The highest accuracy of five kinds of signals for emotion recognition is 90%,the highest accuracy of recognition after removing the electrical skin signal is 86.5%,and the highest recognition accuracy of removing the respiratory signal again is 78.5%.After comparative analysis,it can be seen that the respiratory signal and the electrical skin signal have the least impact on emotion recognition,and the remaining three physiological signals can best reflect the emotional state,and the accuracy rate is relatively high.Since the accuracy rate begins to decline when there are only three physiological signals,if the number of signals is reduced,the effect of emotion recognition may be worse.Therefore,EEG signal,ECG signal and PPG signal are the best combination that can be found when the number of signals is as small as possible.