基于多域特征与随机子空间集成的脑电分类

EEG Classification Based on Multi-domain Features and Random Subspace Ensemble

针对脑机接口(Brain-Computer Interface,BCI)中脑电信号预处理、特征提取、分类识别等过程,提出一种基于多域特征的随机子空间集成方法实现运动想象脑电分类。该方法的基本思想是通过事件相关同步/事件相关去同步特性分析,提取出最佳时频段的多域特征作为特征向量,结合交叉验证自适应地选择特征随机子空间的集成规模,集成线性判别分析分类器实现脑电信号分类。实验结果表明,多域特征和随机子空间集成分类正确率可达90.71%、Kappa系数可达0.63,均优于BCI竞赛第一名成绩,从而证明了该算法在脑电分类中的有效性和实用性。

Aiming at the preprocessing feature extraction and classification recognition in BCI system, a method for EEG classification of motion imagery based on random subspaces ensemble learning of multi-domain features is proposed. Based on the analysis on the ERD/ERS characteristics of motion imagery(MI) signals, the multi-domain features of best effective time and frequency bands are extracted as the feature vectors, and the scale of the random subspace ensemble with cross-validation is adaptively chosen, and the EEG classification is realized by using linear discriminant analysis(LDA) classifiers ensemble. The test results show that the accuracy of the multi-domain features and random subspace ensemble can reach 90.71% and the Kappa coefficient can be 0.63,which are better than those of the first place in the competition, and thus prove the algorithm’s effectiveness and progressiveness.