基于运动想象的脑-机接口系统是脑-机接口中的一个主要研究方向,共空间模式(CSP)算法是一种流行的运动想象数据分析特征提取方法.共空间模式的性能依赖于恰当的带通滤波,通常高度依赖于神经生理先验知识.本研究提出一种称为共迭代时空模式(ICSTP)的运动想象时空特征提取方法,该算法用与空域滤波器设计相同的广义特征值问题优化时域滤波器,并给出了算法收敛性的证明.真实脑电数据实验结果表明算法的收敛只需数个循环,且平均正确率高于人工选择时域滤波器的标准CSP方法.%The motor imagery-based brain-computer interface (BCI) system is an important research theme in BCIs. A popular feature extraction method for motor imagery data analysis is the common spatial patterns (CSP) algorithm. The performance of the CSP feature extraction is contingent on appropriate band-pass filtering, which usually highly depends on the prior neurophysiologic knowledge. In this paper we present an algorithm termed iterative common spatial-temporal patterns (ICSTP) for learning spatio-temporal features from motor imagery EEG data. The algorithm optimizes temporal filters by solving a generalized eigenvalue problem in the same way as CSP does in learning spatial filters. A proof for the convergence of the algorithm is provided.Experimental results on real EEG data demonstrate that the algorithm can converge rapidly within a few iterations. The average accuracy is higher than that of standard CSP using manually chosen temporal filters.
展开▼