基于降维Householder变换的多任务运动想象脑电信号特征提取研究

近年来,针对传统的左右手运动想象BCI系统信息传输速率低这一现状,众多脑-机接口(BCI)研究团队开始着眼于对多任务运动想象脑电信号的研究,相比于两类模式识别,多类模式识别能够有效提高BCI系统的信息传输速率。如何准确提取出多任务脑电信号的特征,是实现多任务BCI系统的关键。采用了基于初等反射变换(又称Householder变换)的矩阵近似联合对角化算法,将CSP算法应用于多任务运动想象脑电信号的特征提取,对EEG信号采集效果较好的受试者,四任务运动想象脑电信号的分类准确率提升至80%以上,为在线BCI系统的实现奠定了坚实的基础。

Research on Feature Extraction in Multi-class Motor Imagery EEG based on dimensionality reduction Householder transform

In recent years, the traditional left and right hand motor imagery BCI system information transfer rate is low in this situation, many BCI research team began to focus on the study of Multi-class Motor Imagery EEG.Compared with two types of pattern recognition, multi-class BCI system can improve the information transmission rate. The key to multi-class BCI system is how to effectively extract features multitasking EEG signal .In this paper, the approximate joint diagonalization algorithm is applied to the CSP multitasking motor imagery EEG signals based on elementary reflection transform (also known as the Householder transformation) matrix of feature extraction .People who had better EEG signal acquisition can make four classification accuracy of motor imagery task EEG increased to more than 80%, and laid a solid foundation for the realization of the online BCI system.