Research Of Affect Recognition From GSR Using Two Improved PSO Algorithms

Now galvanic skin response (GSR) monitoring system can be used to detect human physiological states, however, few researches have been done on affective recognition from GSR signal. This paper mainly studies the corresponding relationship between six kinds of affects (happiness, surprise, disgust, grief, anger and fear) and GSR features, and the course of the research is mainly divided into two stages:training and validation.To build rich GSR affective physiological database, long-term subject recruiting is needed, and carefully selected movie clips are used to stimulate the corresponding affects, namely, happiness, surprise, disgust, grief, anger and fear. then the multi-channel physiological signal acquisition instrument MP150 is used for GSR signal recording. To ensure the validity of data, the experimenters should pay close attention to the subject and make markers while the subject is watching movies. The collected GSR signals will be filtered, standardized and normalized. The selected valid data include:194 groups happiness data,159 group surprise data,82 groups disgust data,221 groups grief data,224 groups anger data and 189 groups fear data.30 statistical features that represent changes of GSR signal are extracted from the selected data, which makes preparation for feature selection.Traditional sequential backward selection (SBS) algorithm and two improved particle swarm optimization (PSO) algorithms are used for feature selection, respectively. Fisher classifier is adopted in the process of classification, and feature selection results are analyzed and compared. Throughout the process above, the goal affect is seen as the first kind and five other affects as the second. According to the case that basic PSO algorithm is easily get into the local optimum, we choose two kinds of improved PSO algorithms, one is Hybrid Particle Swarm Optimization (HPSO) adding self-adapting inertia, neighborhood search and crossover-mutation operator, the other is called IH-PSO (Immune Hybrid Particle Swarm Optimization) which imports self-adjustment concept of biological immune system and self-adapting inertia set into the HPSO algorithms.After comparing and analysing the optimal feature combination validation recognition effect of the foregoing feature selection algorithms, we attain the following conclusions: (1) SBS and PSO algorithms can recognize fear and surprise obviously, while disgust affect recognition effect is not so beautiful.(2) When we identify particular feelings, the method using intelligent optimization algorithm PSO can obtain superior validation recognition effects to that using traditional SBS algorithm.(3) Identification verification results of the two improved PSO algorithms has showed that target emotion recognition correct rate, the second type of emotion recognition correct rate, and optimal feature combination validation fitness of IH-PSO solution are all finer than the HPSO situation, which state that it can effectively solve fault easily into local minimum of the HPSO when added immunity mechanism.(4) In order to find out the optimal features which can represent changes of corresponding affects, we select 15 kinds of one-to-one individuals'identification from the six affects, after comparing each optimal feature combination in one-to-one results, we find out relatively optimal feature in the corresponding affective identification. Finally we compare the one-to-one results with IH-PSO solution to find some characteristics can most reflect the corresponding affective change.