基于LDA和KNN的下肢运动想象脑电信号分类研究

Classification of lower limb motor imagination signals based on LDA and KNN

基于运动想象的脑-机接口系统(motor imagery brain computer interface,MI-BCI)是一种新兴的康复治疗手段。如何提高MI-BCI的识别准确率,是目前研究中的热点和难点。针对左、右脚的运动想象脑电信号进行分类研究,分别采集了单纯运动想象和含有电刺激辅助范式中的运动想象脑电信号,使用滤波器组共空间模式(filter bank common spatial pattern,FBCSP)进行特征提取后采用线性判别分析(linear discriminant analysis,LDA)与K-近邻法(k-nearest neighbor,KNN)相结合的方法实现分类识别,并与支持向量机(support vector machine,SVM)得到的分类结果进行对比。实验结果表明,LDA+KNN算法在两种条件下得到的平均分类准确率分别为67.5%和84.62%,比SVM算法结果高出了5.29%和6.01%,说明这种改进后的算法适用于下肢的运动想象分类。

In recent years,the brain-computer interface system for motion imagination(MI-BCI)is a new method of rehabilitation.How to improve the recognition accuracy of MI-BCI is the focus and difficulty in the research.This research realized the classification of left and right foot movement imagination.Collect motor imagination EEG signals in the motor imagination paradigm and the hybrid paradigm with electrical stimulation assistance.Adopted the method of filter band common spatial mode(FBCSP)to extract signal features,and then use the linear discriminant analysis(LDA)and K-nearest neighbor(KNN)to classify lower limb motor imagination.Compared with the classification results obtained by the support vector machine(SVM).Analysis results shows that the average classification accuracy results based on the LDA+KNN algorithm in two paradigms are 67.5%and 84.62%,respectively,Increased by 5.29%and 6.01%than SVM results.It is proved that the algorithm combined with LDA and KNN is suitable for the classification of motor imagination of lower limbs.