免疫机制对皮肤电信号情感特征选择的影响

Effect of immune mechanism for emotion feature selection from GSR signal

以带有典型情感色彩的视频片段为情感诱发素材,采集皮肤电(galvanic skin response,GSR)信号构成了情感识别研究的初始数据库,并在该数据库的基础上研究了免疫机制对识别过程中的特征选择的影响。首先从GSR信号中提取了30个统计特征,并用平静状态下的相应特征值对其进行标准化;然后在混合粒子群算法(HPSO)的基础上增加免疫操作,形成免疫混合粒子群算法(immune hybrid particle swarm optimization,IH-PSO)进行特征选择,测试特征选择效果时,采用Fisher分类器进行分类;最后分别用两种算法选择出的特征组合进行了情感识别验证。验证结果显示,与HPSO相比,IH-PSO能以较少特征获得较高识别率,这说明免疫机制的应用能够使特征选择过程变得更优。

Used the video clips with emotional colors as the inducing materials,and collected GSR signals as database for emotion recognition,studied then the effect of immune operations on feature selection. Firstly,extracted 30 statistic features from GSR signals,and normalized by the values of corresponding features under emotion calm. Then through adding immune operations to HPSO,presented IH-PSO for feature selection. Adopted Fisher classifier to test the selection effect. Finally,used both of the selected feature combinations caught by the two algorithms for emotion identification and verification. The verification results show that compared with HPSO,IH-PSO can obtain higher recognition rate with fewer features. All those illus-trate that the application of the immune system can earn much better feature selection effect.