二次滑动粗粒化的快速样本熵脑电情感分析

针对传统单一尺度样本熵对脑电信号(EEG)序列特征提取不明显、多尺度熵在粗粒化过程中会遗漏重要信息导致情感分类性能下降以及样本熵算法效率不高的问题,提出了一种基于二次滑动均值粗粒化的多尺度快速样本熵脑电特征提取方法。由于不同情感的脑电信号存在差异性,先采用二次滑动均值粗粒化对脑电信号进行多尺度处理,然后利用快速样本熵算法提取不同时间尺度的样本熵值作为特征向量,结合随机森林(RF)分类模型来识别不同的情感状态。提出的方法对多模态标准情感数据库DEAP进行了研究,发现大脑额区和右脑对情感比较敏感,正性、中性和负性情感在大脑侧额区获得了88.75%的平均分类准确率。实验结果表明,该方法可以有效地提取脑电特征,并且能够保证算法的效率。

Fast sample entropy electroencephalogram emotion analysis of double sliding coarse granulation

Aiming at the problems of traditional single-scale sample entropy that couldn’t be obvious to extract electroencephalogram(EEG)sequence features,multi-scale entropy would miss important information in the coarse granulation process that decreased the performance of emotion classification,and the efficiency of sample entropy algorithm was not high,this paper proposed a multi-scale fast sample entropy EEG feature extraction method based on double sliding mean coarse granulation.Firstly,it processed the double sliding mean coarse-grained EEG signals in multiple scales because of the difference of different emotional EEG signals.Secondly,it used the fast sample entropy algorithm to extract sample entropy values of different time scales as eigenvectors.Lastly,it used the random forest(RF)classification model to identify different emotional states.This paper studied the proposed method in DEAP,a multi-mode standard emotion database,and it was found that the frontal area of the brain and the right brain were relatively sensitive to emotions,and the positive,neutral and negative emotions achieved an average classification accuracy of 88.75%in the lateral frontal area of the brain.Experimental results show that the proposed method can effectively extract EEG features and ensure the efficiency of the algorithm.