摘要
人们的情绪变化本质是大脑皮层上的高级神经活动。情绪认知应用是未来重要的一个趋势,现在以脑机接口为主流工具的研究,在脑电情绪主观世界和信号客观世界之间建立了桥梁。使用多种分类器来对情绪识别,选择有监督机器学习的Fisher、贝叶斯、SVM和无监督机器学习的DBN分类器进行研究。结果表明:在分类精度上,贝叶斯要优于Fisher,DBN要优于SVM,在运行时间上,贝叶斯运行时间最短。DBN有更高的分类精度和更低的标准偏差,平均最佳分类精度为84.01%,最低标准偏差为9.74%,比较适合脑电情绪识别。
The emotion belongs to higher nervous activity in the cerebral cortex of human. Now many researchers use BCI in formal analysis,simulation,and phototyping to explore predicted system behavior between the subjective world of emotion and the objective world of the signal. This paper compares various classifiers of emotion recognition,and then applies two sets of classifiers. The unsupervised classification include DBN,the supervised classification include Bayes classifier and Fisher classifier and SVM. The results have shown that the DNB method performed better than SVM in classification accruracy,and the Bayes classifier is better than Fisher classifier in running time. DBN has a higher classification accuracy and lower standard deviation,and it is more suitable for EEG emotion recognition. Moreover,the average classification accuracy is 84.01% and the minimum standard deviation is about 9.74%.
作者
马新斐
刘志宏
姜添浩
MA Xin-fei;LIU Zhi-hong;JIANG Tian-hao(College of Electronic Engineering,Chengdu University of Information Technology,Chengdu 610225,China)
出处
《成都信息工程大学学报》
2018年第4期365-369,共5页
Journal of Chengdu University of Information Technology
基金
四川省教育厅重点资助项目(2013ZZ0001)
关键词
脑电
情绪认知
分类精度
机器学习
EEG
emotion recognition
classification accuracy
machine learning
