The Research Of The Emotional Feature Extraction And Recognition Classification Of GSR

In recent years, the field of intellective man-machine interaction has been widely concerned. Research of the affective identification based on the physiological signal is a very important part of this field. Affective computing makes computer to have affective skills, and proposes a requirement of affective computing. Through the affective identification, it makes computer systems to own feedback ability of perception and user’s responds, and tries to apply this computer to medicine, entertainment, and technological development etc, and then improves people’s pressure and work environment. University of Augsburg in Germany and MIT Media Laboratory in America first obtain a successful research results in the affective identification based on the physiological signal. On precursors’works foundation, this paper main commits to research of the affective identification of six emotions (happy, surprise, grief, disgust, anger, fear) based on the Galvanic Skin Response.The Galvanic Skin Response contain abundant affective information, its variation is under control of sympathetic nervous system. The emotional variation would affect the variation of endocrine system, and lead to the variation of sympathetic nervous system, final the affective variation would appear in Galvanic Skin Response. In order to make a model affective identification based on Galvanic Skin Response, this lab recruits many freshmen of southwest university to watch videos with good-induced effects on motions, namely happiness, surprise, disgust, grief, anger and fear. The experimental equipment is the multi-channel physiological recorder MP150 produced by USA. Biopac, Inc. it uses this equipment to measure and record the Galvanic Skin Response, and then set up a data base. Aimed at the characters of GSR, through Butterworth filter to filtering the signal and signal denoise, to set up a affective identification model which can be used widely, this paper process data standardized treatment. And consulting the feature extraction method of University of Augsburg in Germany, this paper extract 30 dimension statistical feature which best can stand for the variation of Galvanic Skin Response, and then make this feature as the primitive character set. This text uses basic particle swarm optimization algorithm, modified immune particle swarm algorithm and simulated annealing immune particle swarm optimization algorithm, combine with Fisher classifier to process feature selection and affective classify distinguish of GSR, and obtain a better classification and recognition effect. So it proved the feasibility of building an affective distinguish model.Research shows that the follows:(1) The recognition effect of Galvanic Skin Response is the best for surprise and fear, and the better for happy, angry and grief, the worse for disgust; (2) The Particle Swarm Optimization algorithm with the immune mechanism put the diversity, immune memory and adaptivity of the characteristics antibody together. Linear diminishing weighting factor can through regulating big or small to improve the balance of the general develop skill and the partial explore skill of particles. And also can effectively solve the easy to fall into the local extremum faults of basic particle swarm algorithm; (3) It puts simulated annealing into immune the Particle Swarm Optimization algorithm and the location update and pace update of every single particle. In order to improve the diversity of particle, the fitness value of evolved grain follows Metropolis, it means accept the good result and select some worse results in proportion. It bring adaptive regulate in annealing temperature until the system cool down so the particle can exactly search the general optimal position, and then the recognition rate of affective is improved, the dimension of feature subset is reduced; (4) This paper selects the target affection from six affections and obtains the best characteristic set through analysis and compare the three particle Swarm algorithm and finds the common features, then these features can response the affective alteration of target affection.The Galvanic Skin Response contains abundant effective affection information, so this confirms the feasibility of setting up the emotion recognition model.