The Affective Recognition Of Regret Based On 64 Channels EEG Signal

The affective recognition for physiological signals is an important component in the research of affective computing. It mainly focuses on the physiological signals of basic emotions. In recent years, the kinds of emotions in the affective recognition researches have expanded to not only the basic emotions but also the complex social emotions, such as anxiety and disappointment. However, studies about regret are still rare in the area of affective recognition. Regret is a kind of negative social emotions which depends on personal cognitive judgment under different social conditions. In order to maintain human beings’ physical and psychological health, it is important to identify regret effectively and to determine the levels of regret.Based on current researches and methods, this paper has studied on the regret EEG signals from 25 subjects, who had participated in a gambling task in order to recognize and identify their regrets. Details are as follows:(1) Design an affective induction experiment to acquire the regret EEG signals. In order to induce the regret of subjects, this paper imports a gambling task which is designed to evoke the regret in psychology. In the laboratory, all subjects are required to finish a gambling game designed by computer programming. The results of the gambling game, including feedback categories(partial or complete) and outcome valences(reward or punishment) could be controlled by the experimenter. Thus, specific subjects would produce specific emotional experiences, such as happiness, disappointment, rejoice or regret. The EEG signals will be recorded by the 64 channels.(2) Preprocess the primitive data in order to obtain pure regret EEG signals comparatively. According to the event-related potential(ERP) researches in psychology, this paper preprocesses the EEG signals of regret and divides them from the other three emotions. After re-reference, down-sampling and filtering, the EEG data of regret is extracted, according to the key time points setting in the experiment in advance. After removing electromyography(EMG) artifact and ocular artifact, this paper combines manual operation and independent component analysis(ICA) to obtain the pure emotional signals.(3) Analyze the EEG signal with Event-related spectrum perturbation(ERSP). With ERSP, this paper analyzes the energy changes in different time intervals and frequency bands to explore the neurophysiological mechanism of regret. The results indicate that the significant differences between regret and other three emotions in the frequency band are theta(4~8Hz) and alpha(8~12Hz).(4) Use feature extraction to analyze the EEG signals for the affective recognition of regret. This paper extracts power spectrum estimation, energy and approximate entropy of EEG signals to deal with the classification recognition. This paper also considers the non-stationary, time intervals, frequency bands and nonlinear characteristics of EEG signals when it extracts the characteristics of EEG signals.(5) Classify the recognition of regret. For the three different characteristics of power spectrum estimation, energy and approximate entropy, this paper uses two-classification pattern recognition to deal with the regrets in turn with the other three emotions by Fisher classifier respectively. The two-classification pattern recognition will proceed again with 7 channels(Fz, F2, FCz, FC2, C1, Cz, C2) which perform well in three types of feature classifications. The two-classification pattern recognition rate for regret compared with the other three emotions is all about 75%. Then this paper uses SVM to process the 7 channels of the three different characteristics with two-classification pattern recognition. This time the result is over 80%, having been improved about 5% than the Fisher classifier.