摘要： 传统运动想象脑电信号判别分析方法存在提取特征数量多、不能反映脑电信号本质特征等问题。为此，提出一种基于贝叶斯网络结构的直观判别分析方法，用于描述进行左右手运动想象时各个导联脑电信号所组成网络的结构差异。在结构学习中引入各个导联的位置信息，利用连续高斯分布对其进行描述，以充分反映脑电信号的高时间、空间分布特征，实现对左右手运动想象脑电信号的网络建模。分别在国际脑电比赛数据集及实验室采集的数据集上进行实验，结果表明，该方法能准确反映各个导联脑电信号的特征及导联之间的关联情况，与PCA+fisherscore方法相比，具有较高的识别率和稳定性。

关键词: 脑电信号, 运动想象, 贝叶斯网络, 结构学习, 导联位置, 高斯分布

Abstract: In order to solve the problems such as huge amounts of features, unable to reflect the distribution of Electroencephalography (EEG) signal in the traditional methods of EEG feature extraction, a new method based on Bayesian network for motor imagery EEG discriminant analysis is proposed, which intuitively shows the differences in the network structure made up of channels in the left and right hand motor imagery. By applying the information of channel position into structure learning, using continuous Gaussian distribution to model the nodes, and making full use of the high time spatial distribution characteristics of EEG signals, it realizes modeling the network structure of left and right hand motor imagery EEG signal. Experimental results show that the proposed method can effectively reflect the relationship between channels, and it has higher recognition rate and stability compared with PCA+fisherscore method.

Key words: Electroencephalography(EEG) signal, motor imagery, Bayesian network, structure learning, channel position, Gaussian distribution

中图分类号: 

• TP18