基于水平可视图多元联合模体熵的多维EEG情感脑电信号识别

Multivariate emotional EEG signal recognition based on multivariate joint motif entropy of a horizontal visibility graph

目前,许多基于深度学习和神经网络的算法被应用于脑电(electroencephalogram, EEG)信号情感识别.然而,现有研究大多采用提取单维脑电信号特征的方法.随着多传感技术的更新,更具全面性和系统性的多维信号特征提取需求出现.本文尝试将复杂网络研究应用到多维情感脑电识别中,提出一种基于水平可视图多元联合模体熵的情感识别算法,该方法可以有效避免人工选取特征对实验结果的影响,保持原始序列的非线性动力学特征.首先利用水平可视图算法将多维情感脑电信号分别转换为多路可视图网络,提取模体熵特征识别情感脑电研究中的关键频带和关键通道.在此基础上,将水平可视图网络两两联合,提取多元水平联合模体熵向量,作为输入参数对情感脑电信号进行识别.由于情感脑电序列长度会对识别效果产生影响,我们将脑电信号切割成大小不一的窗口,对比不同窗口大小对分类准确率的影响.实验结果表明,当切割窗口大小为10 s时,多元水平联合模体熵对情感脑电信号的识别效果最佳,对积极脑电/消极脑电、积极脑电/中性脑电、消极脑电/中性脑电的分类准确率分别达到95.07%, 97.73%, 90.26%,优于其他二维连接参数.同时,三分类的准确率为93.67%,本文算法无论在识别复杂度和准确率上,与已有算法相比均有较大提高.

At present,many algorithms based on deep learning and neural networks have already been applied to the emotion recognition of EEG signals.However,most existing studies extract features of only single-dimensional EEG signals.With the update of multisensing technology,multidimensional signal feature extraction needs to become more comprehensive and systematic.This paper attempts to apply a complex network to emotion EEG recognition and proposes an emotion recognition algorithm based on the multivariate joint motif entropy of a horizontal visibility graph.This method can effectively avoid the influence of artificial feature selection and maintain the nonlinear dynamic characteristics of an original sequence.The horizontal visibility algorithm is used to convert multidimensional emotion EEG signals into multiplex visibility networks,and the key frequency bands and key channels of emotional research are identified by extracting motif entropy.On this basis,the horizontal visibility network is combined in pairs to extract multivariate joint motif entropy vectors,which are input parameters for identifying emotional EEG signals.Because the length of an emotional EEG sequence can affect the recognition effect,we cut the EEG into different-sized windows and compare the recognition conditions under different window sizes.Experimental results show that when the cutting window size is 10 s,the multivariate joint motif entropy attaches the best recognition effect,and the accuracy rate of classifying positive EEG/negative EEG,positive EEG/neutral EEG,and negative EEG/neutral EEG is 95.07%,97.73%,and 90.26%,respectively,better than other connection parameters.Moreover,the accuracy rate of these three classifications is 93.67%,an improvement over the existing algorithm.