Emotion Recognition Method Research Based On Chaos Feature Extraction Of Multiple Physiological Information

Since the physiological information can objectively reflect the true emotional state,the study based on physiological signals of emotion recognition method has important practical meaning. In this paper, based on the work done by our predecessors, focusing on the problems of chaos theory apply for multiple physiological information feature extraction and C5.0 decision tree apply to emotion recognition. We first extract six kinds of chaos characteristics(Maximum Lyapunov index, Correlation dimension, Box dimension,Information entropy, Approximate entropy and Complexity) from the Physiological data(ECG, EMG, SC, RSP) offered by the University of Augsburg in Germany, then using C5.0algorithm to classify and identify four emotions under these chaos characteristic parameters.C5.0 algorithm is the latest decision tree classifier that solve the problems of machine learning in the case of a large sample, make the result of sample classification more accurate. The classification result is ideal, the recognition rate for Joy, Anger, Sadness and Pleasure all reached 100%, it can completely correct classification of these four kinds of emotions. There are two reasons for the ideal recognition, one is using chaos theory for the extraction of multiple physiological signal characteristic parameters, the second is using the latest C5.0 decision tree to classify chaos characteristic property. Using an existing laboratory equipment to acquire three kinds of physiological signals(ECG, SC and RSP)from one subject under four kinds of emotions, based on the relevant algorithms in this paper to extract chaos feature and recognize emotion, the result showed that the recognition rates of 4 kinds of emotions have decreased, there are three reasons of recognition rates decrease: First, because the experiment conditions,we did not collect EMG which has the largest contribution to recognize the emotion of Joy and Sadness, EMG. Second, the emotion evoked materials is too single cause the low arousal emotion of Pleasure recognition rate decreased. Third, the collected number of emotional samples is too small.