基于双树复小波变换和GBDT的运动想象脑电识别  

EEG Recognition of Motor Imagery Based on Dual-Tree Complex Wavelet Transform and GBDT

作　　者：刘飞翔 王爱民[1] LIU Fei-xiang;WANG Ai-min(School of Instrument Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]东南大学仪器科学与工程学院,江苏南京210096

出　　处：《测控技术》2019年第1期58-62,共5页Measurement & Control Technology

基　　金：江苏省科技支撑计划项目(BE2012740)

摘　　要：针对脑机接口研究中的脑电信号特征提取与分类问题,提出了一种基于双树复小波变换结合GBDT的想象左右手运动脑电识别的方法。该方法首先深入研究了双树复小波变换相比于小波包变换在脑电信号特征提取方面的优势并验证了ERD/ERS现象;实验数据采用了2003年国际脑机接口竞赛的标准数据集DataSetⅢ,然后,选取了4个典型的时间段进行实验对比,利用双树复小波变换分解与重构提取运动感知节律相关信号分量的能量均值作为特征进行GBDT分类。最后,实验取得了较好的分类准确度,验证了双树复小波变换结合GBDT的方法在脑电信号识别应用中的有效性。In order to solve the issue of feature extraction and classification of electroencephalography(EEG)in the research of brain-computer interface(BCI),an EEG recognition method of imagery left-right hand movements based on dual-tree complex wavelet transform(DTCWT)and GBDT(Gradient Boosting Decision Tree)was proposed.Firstly,the advantages of DTCWT over wavelet packet transform(WPT)were researched on EEG feature extraction of imagery left-right hand movements,and the ERD/ERS phenomenon was verified.The experimental data adopted the standard data set DataSetⅢof BCI Competition 2003.Then,four typical time intervals were selected for experimental comparison.The energy mean value of signal component related to sensory motor rhythms was extracted based on dual-tree complex wavelet decomposition and reconstruction,and as the feature of EEG signal classified by GBDT.Finally,The experimental results show that this method achieves classification accuracy and the effectiveness of DTCWT and GBDT on the EEG recognition is verified.

关 键 词：脑机接口 运动想象 双树复小波变换 GBDT 

分 类 号：R318.4[医药卫生—生物医学工程] TP391.4[医药卫生—基础医学]