Heart Rate Variability-based Emotion Recognition Research

Emotions are the psychological and physical states that are produced along with consciousness and cognitive processes.They play an important role in human communication.Therefore,effective recognition for emotions is of great significance.The emotion recognition method based on physiological signals is not easy to disguise and can obtain more objective and effective results.Among them,based on EEG emotion recognition has the best effect.However,the EEG equipment is expensive,the wearing and collecting process is complicated,and it can only be collected and analyzed in a fixed place and under specific conditions.With the development of portable wearable ECG acquisition equipment,real-time emotion detection based on ECG signal is possible.However,the effect of emotion classification based on ECG signal needs to be improved.In order to improve the accuracy of emotion recognition,this article studies the feature extraction method,feature screening and classification algorithm of heart rate variability.The main research contents of this article are as follows:(1)The ECG emotion data set of user independent model is established.Aiming at the collection of user independent model data,the emotion-inducing video and emotion-inducing paradigm are designed using the film editing method.The anger,joy,sadness,peace and fear emotions were induced by experiments,and the sample size of each emotion was nearly 100 cases.The joy,sadness and fear three kinds of emotions are used in the research of this article,which provides data set support for the emotion analysis of user independent model.(2)40 kinds of heart rate variability features are extracted.Four types of extraction methods in time domain,frequency domain,nonlinearity and time-frequency domain are studied,and statistical methods are used to analyze the difference of heart rate variability features of different emotions.The results show that most of the features are statistically different between different emotional states.(3)The maximum information coefficient method is proposed for the selection of heart rate variability features.This feature selection method is combined with support vector machine,random forest and K nearest neighbor algorithm to build five classification models,which are used for the joy,anger,sadness,and pleasure of the user dependent model and the joy,sadness,and fear of the user independent model.The results show that the maximum information coefficient feature selection method proposed in this article effectively improves the accuracy of emotion recognition and reduces the feature dimension required for modeling.(4)The GA-BP-adaboost algorithm is proposed for the classification of positive and negative emotions.The recognition rate of this algorithm reaches 84.27%,which is significantly improved compared with other algorithms such as BP neural network.And we use the algorithm to design and implement a positive and negative emotion recognition system,which can effectively detect positive and negative emotions.In this article,emotion recognition research is carried out based on the heart rate variability features.The emotion data set of the user independent model is established by designing the experimental paradigm.The proposed maximum information coefficient feature selection method and GA-BP-adaboost modeling algorithm have achieved high recognition results in the Aubt emotion database and self-collected emotion data set.Finally,the software for emotion recognition is designed,which provides support and reference for real-time portable emotion analysis based on ECG signals.