摘要
为提高脑电信号情感识别分类准确率,结合经验模态(EMD)分解和能量熵提出一种新的脑电特征提取方法。本研究主要介绍了EMD分解的基本原理,分析了传统EMD算法中的"端点效应",采用分段幂函数插值算法改善了EMD分解的精度和性能,然后将改进后的算法应用到脑电信号特征提取,获取脑电信号的IMF分量后计算出IMF能量熵作为情感识别的特征,最后通过分类实验对比改进后的EMD算法和传统EMD算法对脑电情感特征的分类准确率。实验结果显示改进的EMD算法能使识别率提高15%左右,并且以IMF能量熵为特征的平均识别率在80%以上,实验结果表明将IMF能量熵用于脑电信号情感识别是可行的。
A new method,which combines electric motor driven( EMD) and energy entropy,was presented in order to raise the classification accuracy rate. The EMD principle was introduced in this paper and the defect of"endpoint effect"was analyzed in detail.Piecewise power function interpolation method was used to remove the"endpoint effect"to improve the performance of EMD. The improved EMD was then applied to EEG feature extraction experiment to acquire IMF components and IMF energy entropy was calculated as the emotion feature. Finally,a contrast result between traditional EMD and inproved EMD was given to show that the accuracy rate raised about 15%,and the mean classification accuracy rate of IMF energy entropy is reached 80%,which proved that it is feasible for EEG emotion recognition by using IMF energy entropy.
出处
《生物医学工程研究》
北大核心
2016年第2期71-74,80,共5页
Journal Of Biomedical Engineering Research
基金
国家自然科学基金资助项目(61071085)
上海市科委科技创新行动计划生物医药领域产学研合作项目(12DZ1940903)
关键词
经验模态分解
端点效应
分段幂函数插值
能量熵
情感识别
Empirical mode decomposition
Endpoint effect
Piecewise power function interpolation
Energy entropy
Emotion recognition