中文摘要：

脑电信号（EEG）特征提取和分类是脑机接口（BCI）系统的核心问题之一。由于BCI系统中EEG信号多通道采样和特征向量的高维性，有效的特征选择算法已经成为研究中不可分割的一部分。针对EEG特征选择问题采用一种新方法：基于封装式稀疏组lasso的EEG融合特征的同时通道和特征选择方法。实验中将该方法与现有的通道选择和特征选择方法进行比较，结果表明，该方法更适用于高维融合特征的最优特征子集选择问题，且该算法稳定、时间成本低。此外，在保证错误率相当或较低的情况下，该方法能够同时实现通道和特征选择。国际BCI竞赛Ⅳ的两类运动想象信号的测试错误率为15．28％。

英文摘要：

Feature extraction and classification of EEG signals is one of the core issues of brain-computer interface （BCI）. Be- cause of the multi-channel sampling of EEG signal and the high dimension of the feature vector in BCI system, in the past ten years, ef- fective feature selection algorithm has become an integral part of the study. In this paper, aiming at the feature selection issue of EEG signals, a novel wrapped sparse group lasso method is presented to achieve simultaneous channel and feature selection of the fused fea- tures of EEG signals. In experiment, the proposed method was compared with existing channel selection and feature selection methods, the results show that the novel method is more suitable for the optimum feature subset selection of high-dimension fused feature, and the method is stable and its time cost is low. Besides, under the condition of ensuring low or considerable error rate, the proposed algorithm can simultaneously achieve channel and feature selection. The test error rate for the two class motor imagery signals in international BCI CompetitionⅣ reaches 15.28%.