基于LMD-CSP和随机森林的运动想象脑电信号分类

Classification of Motor Imagery EEG Signals Based on LMD-CSP and Random Forest

针对脑电信号具有非平稳性、非线性以及个体差异较大等特点而导致特征提取困难、分类准确率低的问题,提出一种基于LMD-CSP和随机森林(Random Forest,RF)的脑电信号分类方法。首先对脑电信号进行预处理,然后利用局部均值分解(Local Mean Decomposition,LMD)将预处理后的脑电信号分解为多个乘积函数(Product Function,PF)分量,并选出最具判别性的PF分量,再利用共空间模式(Common Spatial Pattern,CSP)分别对选出的PF分量进行特征提取,最后将得到的CSP特征输入随机森林分类器中进行分类识别。实验结果表明,该方法的平均分类准确率高达92.18%,远高于其他方法,证明了该方法的有效性。

Aiming at the problems of non-stationarity,non-linearity and large individual differences of EEG signals,which lead to the difficulty of feature extraction and low classification accuracy,a classification method of EEG signals based on LMD-CSP and random forests(RF)is proposed.Firstly,The EEG signals are preprocessed,then the preprocessed EEG signals are decomposed into multiple product function(PF)components by local mean decomposition(LMD),and the most discriminative PF components are selected.Then the selected PF components are extracted by common spatial pattern(CSP).Finally,the CSP features are input into the random forest classifier for classification and recognition.The experimental results show that the average classification accuracy of the method is as high as 92.18%,which is much higher than that of other methods,which proves the effectiveness of the method.