基于运动想象脑机接口的AuFBCSP方法

摘要

This paper proposes the augmented Filter Bank Common Spatial Pattern, a novel feature extraction al- gorithm for the motor imagery-based brain-computer interface. Compared with the Filter Bank Common Spatial Pattern, our method takes into full consideration the specificity temporal information, in which ERD happens in motor imagery. In this way, more features are acquired, and the higher accuracy and kappa value are obtained by these features. Two kinds of feature selection algorithms are employed in the BCI competition dataset. The results show that the accuracy based method yields superior classification accuracy compared with that based on mutual information.%针对基于运动想象的脑机接口问题，提出了一种新的特征提取方法，即加强的滤波带宽共同空间模式方法。与传统的滤波带宽共同空间模式方法相比，文中提出的方法充分考虑了被试在进行运动想象时发生事件相关去同步时段的特异性，从而得到更多的特征，并利用这些特征分类，取得更高的准确率和kappa值。在特征的选择上，利用BCI的竞赛数据，将基于互信息的方法和基于准确率的方法进行对比，发现基于准确率的特征选择方法优于基于互信息的特征选择方法。