眼电伪迹和噪声是导致脑电信号低信噪比的重要原因,会降低运动想象任务的分类性能。提出一种改进的基于少通道数的分块欠定盲源分离的滤波方法,通过分块的思想把非平稳的脑电信号变为近平稳的分块信号,利用二阶欠定混合矩阵盲识别方法估计混合分离矩阵,然后通过基于最小均方误差的波速形成器提取源信号,接着通过得分准则自动去除噪声信号并重构信号,最后提取共空间模式特征进行分类。想象运动的真实脑电信号实验仿真结果表明,分块欠定盲源分离方法能很好地恢复源信号并能有效地去除眼电等伪迹和噪声,共空间模式特征则提高了想象任务识别率。 EEG artifacts and noise are the important reasons that lead to low signal-to-noise ratio of EEG, which will reduce the classification performance of motion imaging tasks. This paper proposes an improved filtering method based on the number of channels with undefined blind source separation. By dividing the non-stationary EEG signal into a sub-stationary block signal, the second-order under-determined hybrid matrix is ​​blind Then, the source signal is extracted by the velocity profiler based on the least mean square error, and then the noise signal is automatically removed by the score criterion and the signal is reconstructed. Finally, the features of the co-space mode are extracted for classification. The experimental results of real EEG signals of imagination show that the method of block underdetermined blind source separation can recover the source signals well and effectively remove artifacts and noise such as electrooculogram. The common spatial pattern feature improves the recognition of task rate.