The Research Of Affective State Recognition From Electrocardiography Signal Based On QPSO Algorithm

With thorough applications of computer in human society, Harmonious human-computer interaction environment has been paid more and more attention by the whole society. In order to achieve the real harmony and nature of human-computer interaction, computer must be able to recognize and express human emotion. The study of affection computing has made the computer with affective competence possible.Recognizing affection from physiological signal has become an important research content in the field of human-computer interaction and affection recognition for the objectivity and facticity of physiological signal. The study of affection recognition of physiological signal is achieved by analyzing the collected affective physiological signal and extracting characteristics can represent the specific affection, then constructing model of affection recognition can be used to recognize affection. Electro-cardio-signal (ECG) is regarded as an important physiological signal, and is possible to contain affective physiological response based on reliability has been proved. Meanwhile, ECG is an important object of study in medicine, and its signal processing technology has been mature. In this paper, we focus on ECG and finally construct the model of ECG affection recognition though a series of researches by acquisition of affective ECG, signal preprocessing, affection feature extraction and feature selection etc. Specific work plans are as follows:(1) The formulation of affection-evoked experimental projects and affective ECG acquisition projects. Based on the study of affection-evoked experiences in existing references and we find out that movie clips inspire human emotion more successfully than pictures and music, etc. In this paper, we edit film clips that can evoke specific emotion from a lot of movies as materials of evoking emotion. In order to avoid the cross-impact of affections, scenery pictures and light music are added for a certain minutes among every movie clip. Testees need to fill out questionnaires including their own emotional state and emotional intensity after watching every affection-evoked movie clip. We use MP150 physiological signal recorder of BIOPAC corporation collects the affective ECG of testees and use camera records the facial expressions and body positions of testees synchronously.(2) The establishment of emotional ECG sample database. ECG of many healthy freshmen without cardiac disease history are collected from affection-evoked experience based on the six affective states of anger, disgust, fear, grief, joy, and surprise. According to the analysis of data validity, affective ECG sample database is established by cutting off 80 seconds’data as a sample from affective-evoked valid ECG.(3) The emotional feature extraction. Decomposition and reconstruction of the collected affective ECG are given by wavelet transformation, and baseline drift noise, etc are removed, and the SNR of signal is improved.111-dimensional initial feature set is constituted by extracting affective features from data of different affective states after pinpointing the P-QRS-T wave.(4) The correlation dimensionality reduction of the initial feature set. Because the initial feature set contains a lot of redundant features and in order to reduce difficulty of feature selection and increase efficiency of feature selection, in this paper, based on correlation analysis theory, the relevant characteristics of the initial feature set are extracted and only one is kept to realize dimensionality reduction of the initial feature set.(5) Feature selection. Discrete binary quantum particle swarm algorithm is used in feature selection. To solve ECG emotional features choosing, an improved algorithm (IBQPSO) based on discrete binary quantum particle swarm optimization algorithm is proposed. The performance test results show that the algorithm IBQPSO has better global search properties than the original algorithm. The IBQPSO algorithm is combined with the Fisher classifier and SVM respectively to extract feature subset selection in the dimensionality reduced feature set.(6) The establishment of ECG affective recognition model. According to the results of feature selection, ECG affective recognition model is established and the performance test of it is conducted.The experimental results demonstrate that:(1) It is feasible to realize the dimensionality reduction of the initial feature set of the ECG signal by using the correlation theory, and parts of correlated features in the initial feature set are removed and the difficulty of the feature selection is reduced via the dimensionality reduction. (2) The affection recognition models of the two different ECG signals are constructed in this paper have good affection recognition capabilities and the some generalization capabilities. (3) the affection recognition model is constructed by the selected feature subset based on IBQPSO algorithm and Fisher classifier has better performance when recognizing two emotions of disgust and grief; the affection recognition model is constructed by the selected feature subset based on IBQPSO algorithm and SVM has better performance when recognizing four emotions of fear, joy, anger and surprise. In conclusion, the affection recognition model is constructed by the selected feature subset based on IBQPSO algorithm and SVM has better capability of affection recognition.