摘要
皮肤电信号作为一种重要生理信号,已证明其中包含可靠情感信息.在实验室诱发情感生理信号过程中,采用2遍情感视频播放机制,在第二遍观看视频过程中获取了记录被试主观情绪体验的"情感重评按键文件",据此可截取可靠的情感皮肤电信号.采用多种非线性分析方法,计算相应的非线性特征,如最大Lyapunov指数、关联维、近似熵、递归定量分析和多重去趋势波动分析等.基于所提取特征,采用多种分类器KNN,Fisher判别,SVM进行情感识别性能的比较研究,结果显示SVM具有更好的分类精度.之后,采用SVM分类器比较传统的统计特征与非线性特征在识别目标情感性能上的差异,结果表明非线性特征能获得更好的识别精度.研究结果显示,基于非线性特征构建情感识别模型是可行的.
Galvanic Skin Response （GSR） is a most important physiological signal, which has been proven to contain reliable affective information. When inducing the three kinds of objective emotions （happiness, sadness and fear） by affective movies fragments, the SC signals are collected by Biopac MP 150 synchro- nously. In an independent experiment the affective videos are played twice, and in the second presentation the file of subjects＇ affective experience is obtained, which can help to intercept the reliable affective GSR signal. After preprocessing the original GSR signal, several kinds of nonlinear algorithms, including the largest Lyapunov exponent, correlation dimension, approximate entropy, recursion quantitative analysis and multiple detrended fluctuation analysis, are used to extract the affective features. Based on these non- linear features, three kinds of classifier, i. e. Fisher discriminant analysis, KNN, and SVM, are used to compare the accuracy of classification. The result indicates that SVM has the highest classification accura- cy. Then we compare the classification accuracy between traditional statistical features and nonlinear fea- tures by SVM, and the result shows that the nonlinear features have better accuracy. All experiments prove that the affective model based on nonlinear features is feasible.
出处
《西南大学学报（自然科学版）》
CAS
CSCD
北大核心
2014年第6期186-194,共9页
Journal of Southwest University(Natural Science Edition)
基金
教育部科学技术研究重大项目资助(311032)
中央高校创新团队项目(XDJK2013A020)
