Emotion Recognition Methods Based On Multi-Physiological Information Fusion

Along with the increase of life pace and pressure, people have paid more attention on mental health. Early diagnosis and treatment of mental illness is a precondition to maintain mental health. And, an accurate grasp of the human emotional state is the basis of diagnosis and treatment of mental illness. Thus, emotion recognition is the most important issue of mental health research. It is also emphasis and difficulty in the research.The human emotion can be expressed by many forms, such as languages, facial expressions, movements, and physiological signals. Physiological signals are an objective expression of physiological and psychological activities. So they are without interference by cultural background, customs and so on. And they cannot be deliberately concealed. Therefore, physiological signals become a relatively accurate criterion of emotion recognition. Meanwhile life is a multi-variable system. And the relationship between emotion and physiological parameters is complex. It cannot accurately reflect the change of emotion by a single physiological character. Therefore, multi-characters fusion based on physiological signals is an effective method for emotion recognition.In this paper, the key technology of emotion recognition based on the integrating information of physiological signals is studied. A series of work is carried out around the selection of features and design of classifier. Base on the ECG, pulse and respiratory generation mechanism and collection methods, signal pre-processing and feature extraction, a method of combining of Principal Component Analysis(PCA) and Support Vector Machine(SVM) is developed to recognize the emotions. This thesis is organized as follows:Firstly, experiments were designed to truly induce the six basis emotions. Meanwhile ECG, pulse, respiratory, and video of subjects are recorded. Effective data are extracted based on the questionnaire and video information. Effective data is useful for emotional recognition.Secondly, ECG, respiratory and pulse in corresponding emotions are obtained. The collected physiological signals are pre-processed by Butterworth filter. Then the data are stratified using wavelet transform. And the extreme points and singular points are found out from the bottom to the top. The feature points of ECG, pulse and respiratory are found out. Further, the features are extracted.Thirdly, a method of combining of Principal Component Analysis and Support Vector Machine is prospered. The extracted features are analyzed using PCA, and their sort in the principal component is obtained. Furthermore combining the SVM, emotion recognition rate as evaluation criteria, the best feature subset and the best recognition rate are found out.Fourthly, the validity of algorithm is verified by experimental data. Traditional method-no using feature select algorithm, using Binary Particle Swarm Optimization(BPSO),and using PCA-SVM in this paper are compared. And the result proved that this method can be effectively applied to emotion recognition based multi-physiological parameters. Also, the algorithm has the feature of high recognition rate and short running time. Simultaneously, women’ and men’ different variable quantity of heartbeats under six emotions are analyzed. It can provide scientific basis for gender difference in emotion recognition. And it can provide theoretical basis for emotion study.