The Research Of Affect Recognition For GSR Based On The Combination Of Nonlinear Features

The emotion recognition based on the physiological signal has become an important research content in the field of affective computing because of the authenticity and objectivity of the physiological signal. The purpose of the research is to analyze the collected physiological signals, extract the feature which can represent the specific emotion, and build the emotion recognition models. The galvanic skin response (GSR) is regarded as a kind of important physiological signal, which has been proved that it contains the reliable emotional physiological response feature, moreover, compared with other signals, the collection methods of the GSR is not only simple, but also has less signal noise. If the feature of the GSR emotional signals can be extracted accurately, the emotion recognition is carried out more effectively, finally the emotion recognition model is established.The nonlinear feature of the GSR is conducted on the basis of the previous research work, the analysis on the GSR of human based on the emotional states of happy, surprise, grief, disgust, anger and fear is implemented by a series of work, such as the collection of the GSR physiological signals, signal preprocessing, the extraction and classification of the emotion characteristics, etc, finally the emotional recognition model of the GSR signal is established.(1) The formulation of the experimental scheme of the signal collection. The movies clips which can arouse a single passion are used as the aroused materials. The video materials containing six kinds of emotions mentioned above are regarded as the six clips of the emotion aims, the neutral videos used as the adjustment between the transition emotions is inserted between the clips. The multi-channel physiology signal recorder BIOPAC MP150made in America is selected to collect the GSR signals of the subjects.(2) The emotion sample database of GSR. A large number of experimental subjects are recruited and the GSR signals are collected based on the six kinds of emotional states, including happy, surprise, grief, disgust, anger and fear. The emotion sample database of GSR is realized eventually through the effective analysis of the data and intercepting80s data with good inducement.(3) The emotion feature extraction of the GSR. The collected signals are filtered, standardized and normalized, meanwhile the nonlinear feature of the GSR is extracted by the nonlinear method, its features including the time delay, correlation dimension, min-embedding dimension and sample entropy, etc.(4) The verification of the features. In order to determine the availability of the nonlinear feature, the IH-PSO is introduced to be the feature selection; these features are mixed with the representative statistical characteristics which are proved. If selected, it can be used as the emotion recognition.(5) Establishment of the GSR emotion recognition model. The nonlinear feature is extracted and the analysis of the experiments are performed by the methods mentioned above, the feature combinations of GSR emotion signals are obtained based on the happy, surprise, grief, disgust, anger and fear, and the two-class emotion recognition model of each emotion of GSR is established.The results show that:(1)The four nonlinear features can be used to classify and recognize the GSR signals through the verification of the IH-PSO algorithm.(2) The min-embedding dimension and the sample entropy are able to represent the change characteristics of the good mood, the time delay and the min-embedding dimension can reflect the changes of the sad mood, the greatest contribution of the recognition of the angry emotion is the min-embedding dimension, the time delay can reflect the changes of the surprised mood, the sample entropy has a great contribution to the recognition of the disgust, the combination of the sample entropy and the min-embedding dimension reflects the change of the fear.Based on an overall consideration of analysis mentioned above, the study on the classification and recognition of the related emotions of the GSR signals can be carried out by the nonlinear features, such as the min-embedding dimension, the sample entropy, the time delay, and the correlation dimension, etc; the emotion model is established eventually.