多类运动想象任务脑电信号的KNN分类研究

Study on multi-class motor imagery EEG classification based on KNN

针对基于多类任务的运动想象脑电信号的特点,使用共空间模式特征提取方法分别在"一对一"和"一对多"2种特征提取方法下提取了4类任务(想象左右手、双足以及舌头)运动想象脑电信号的特征。设计了基于多类任务模式的k最近邻分类器,针对多类任务分类过程中会出现不同类别的样本点数相等的情况,通过判断距离的方法改进了分类器,对2种特征提取方法下的共空间模式特征进行分类,分类结果的平均最大Kappa系数分别达到了0.55和0.59,说明了该特征提取及分类方法对该数据集的有效性。

Aiming at the features of multiclass motor imagery EEG （electroencephalogram） signals, CSP （ common spatial pattern） is used as the feature extraction method of 4 motor imagery tasks （ left ＆ right bands, feet and tongue）, and the features are extracted under two conditions of ＂One versus One＂ and ＂One versus Rest＂. Then, a kind of k nearest neighbor classifier based on multiclass task mode is designed. Aiming at the problem that different classes of samples have the same number of sample points in multiclass task classification, distance decision method is used to improve the classifier. The CSP feature is classified with the ＂One versus Rest＂ and ＂One versus One＂ classification methods, and the maximum mean Kappa coefficients are 0. 55 and 0. 59, respectively, which shows that the proposed method is suitable for the classification of the data.