基于共同时频空间模式的MI-EEG分类研究

MI-EEG Classification Based on Common Time-Frequency Spatial Patterns

公共空间模式(Common Spatial Patterns,CSP)算法是目前应用于基于运动想象脑机接口(Motor Imagery-Brain Computer Interface,MI-BCI)系统中提取脑电图特征的最常用的空间滤波方法。CSP算法的有效性取决于从脑电信号中选择最优的频带和时间窗。目前已有许多算法设计用于优化CSP的频带选择,但很少有算法寻求优化时间窗。提出了一种新框架,称为共同时频空间模式(Common Time-Frequency-Spatia Patterns,CTFSP),用于在多个时间窗口中从多波段滤波的脑电数据中提取稀疏的CSP特征。具体而言,首先使用滑动时间窗方法将整个MI周期分割成多个子序列。其次,在每个时间窗内从多个频带提取稀疏CSP特征;最后,训练具有径向基函数(Radial Basis Function,RBF)内核的多个支持向量机(Support Vector Machine,SVM)分类器来识别MI任务,这些分类器的投票结果决定了BCI的最终输出。采集了12名被试的左右手和脚的运动想象实验数据,将提出的CTFSP算法应用于数据集来验证其有效性,并与其他几种最先进的方法进行了比较。实验结果表明,所提算法是提高MI-BCI系统性能的有效方法。

CSP(Common Spatial Patterns)algorithm is currently the most commonly used spatial filtering method applied to extract EEG(Electroencephalography)features in MI-BCI(Motion Imagery-Brain Computer Interface)based systems.The effectiveness of the CSP algorithm depends on the selection of the optimal frequency band and time window from the EEG signal.Currently,many algorithms are designed to optimize the frequency band selection of CSP,but few seek to optimize the time window.This paper proposes a new framework called CTFSP(Common Time-Frequency-Spatial Patterns),which is used to extract sparse CSP features from multi-band filtered EEG data in multiple time windows.Specifically,the whole MI cycle is first segmented into multiple sub-sequences using a sliding time window method.Then,sparse CSP features are extracted from multiple frequency bands in each time window.Finally,multiple SVM(Support Vector Machine)classifiers with RBF(Radial Basis Function)cores are trained to recognize the MI tasks,and the voting results of these classifiers determine the final output of the BCI.In this paper,experimental data on motor imagery of the hands and feet of 12 subjects are collected,and the proposed CTFSP algorithm is applied to the dataset to validate its effectiveness and compare it with several other state-of-the-art methods.The experimental results indicate that the algorithm is an effective method to improve the performance of MI-BCI system.