基于CSP和SFFS-SFBS的两级双向脑电导联特征选取方法  

Two-stage and bi-direction feature selection method for EEG channel based on CSP and SFFS-SFBS

作　　者：张德明 殷国栋[1] 金贤建 庄伟超 Zhang Deming;Yin Guodong;Jin Xianjian;Zhuang Weichao(School of Mechanical Engineering, Southeast University, Nanjing 211189, China;School of Mechatronic Engineering and Automation, Shanghai University, Shanghai 200444, China)

机构地区：[1]东南大学机械工程学院,南京211189 [2]上海大学机电工程与自动化学院,上海200444

出　　处：《东南大学学报（自然科学版）》2019年第1期125-132,共8页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(U1664258;51575103);国家重点研发计划资助项目(2016YFB0100906)

摘　　要：针对多任务运动想象条件下脑电导联选取质量差、搜索时间长的问题,提出了一种基于公共空间模式(CSP)和顺序浮动双向选择算法(SFFS-SFBS)的两级导联特征选取方法.首先,结合空域滤波分析各个被试的时频特性,确定相应的特征时间和特征频率;然后由训练集的CSP滤波系数计算各个导联在特征提取过程中的权重大小,根据权重排序缩小导联搜索空间;最后,运用以训练集交叉检验正确率为评价准则的SFFS-SFBS算法在相应的搜索空间内双向选择最优的导联序列.实验结果表明,在保证较高分类正确率的前提下,与传统SFFS算法和改进SFFS算法相比,该方法选取的导联数量分别减少了51. 36%,47. 52%,对应的搜索时间缩短了90. 95%,80%.因此,基于CSP和SFFS-SFBS的两级特征选取方法可快速选择优质导联序列,有效提高脑机接口的实际使用性能.To solve the problem of poor quality and long searching time for the selection in electroencephalogram(EEG) channel under the condition of multi-class motor imagery,a two-stage channel selection method based on common spatial pattern( CSP) and sequential floating forward selectionsequential floating backward selection algorithm( SFFS-SFBS) was proposed. First,the characteristic time and the characteristic frequency were determined by analyzing the time-frequency properties of each subject with spatial filtering. Then,the weight about each channel in the feature extraction process was calculated using the CSP filtering coefficients of the training dataset,and the channel search space was reduced according to the rank of the weight. Finally,the SFFS-SFBS algorithm with the cross-validation accuracy of the training dataset as evaluation criterion was used to select the optimal channel sequence in the reduced search space. The experimental results demonstrate that the number of channels selected by the method is reduced by 51. 36% and 47. 52%,respectively,and the corresponding search time is shortened by 90. 95% and 80% compared with the traditional SFFS algorithm and the improved SFFS algorithm under the premise of a higher classification accuracy.Therefore,the proposed method can select the channel sequence more accurately and quickly,improving of the brain-computer interface in practice.

关 键 词：多任务运动想象 导联选取 公共空间模式 顺序浮动双向选择算法 

分 类 号：TN911.7[电子电信—通信与信息系统]