计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

基于MODWT 的运动想象脑电信号识别

李东明 a,王典洪 b ,严 军 b ,王永涛 a,宋麦玲 a,余蓓蓓 b   

  1. (中国地质大学a. 信息技术教学实验中心;b. 机械与电子信息学院,武汉430074)
  • 收稿日期:2013-09-02 出版日期:2014-10-15 发布日期:2014-10-13
  • 作者简介:李东明(1982 - ),男,讲师、博士,主研方向:脑机接口技术,图像处理;王典洪,教授;严 军,副教授;王永涛、宋麦玲,讲师; 余蓓蓓,讲师。
  • 基金项目:
    中央高校基本科研业务费专项基金资助项目(CUGL120278);湖北省自然科学基金资助项目(2011335070)。

Movement Imagery Electroencephalogram Recognition Based on MOWDT

LI Dong-ming a ,WANG Dian-hong b ,YAN Jun b ,WANG Yong-tao a ,SONG Mai-ling a ,YU Bei-bei b   

  1. (a. Teaching and Experiment Centre of Information Technology;b. School of Mechanical and Electronic Information,China University of Geosciences,Wuhan 430074,China)
  • Received:2013-09-02 Online:2014-10-15 Published:2014-10-13

摘要: 对运动想象脑电信号进行分类识别,是脑机接口研究中的重要问题。为此,提出一种基于极大重叠小波变 换和AR 模型的脑电信号分类方法。将脑电信号波形进行极大重叠小波分解,抽取变换系数的统计特征,利用Burg 算法提取其3 层光滑的8 阶AR 模型系数以及3 层光滑部分的能量曲线特征,将这3 类特征进行组合后,使用 神经网络、支持向量机及线性判别进行分类和比较。与BCI2003 竞赛数据分类精度结果相比,该方法的识别率更 高。将模型移植入自行研制的嵌入式脑电信号控制电机转向系统中,该模式识别方法的平均准确度达到了91. 3% ,可用于嵌入式脑机接口的系统设计。

关键词: 脑机接口, 运动想象, 极大重叠小波变换, 能量曲线, 模式分类, 电机转向控制

Abstract: To classify and recognize the movement imagery electroencephalogram,is an important problem in Brain Computer Interface (BCI) research. This paper presents a novel method of extracting Electroencephalogram (EEG) features based on Maximum Overlap Wavelet Transform(MODWT) and Autoregressive(AR) model. The EEG signal is decomposed to three levels by MODWT and statistics of wavelet coefficients are computed. Meanwhile,in the EEG signal’s approximation part,the eighth-order AR coefficients are estimated by Burg’ s algorithm,and energe feature vector is also extracted. The combination features are used as an input vector for Neural Network(NN) classifier,Support Vector Machine ( SVM) classifier,and Linear Discriminant Analysis ( LDA) classifier. The recognition result using BCI2003 competition data set is compared with the best result of the competition,and the classification results show the higher recognition rate of the algorithm. Moreover,transplanting the trained successfully model into embedded motor steering control system based on EEG,and the average recognition accuracy of 91. 3% is obtained. The method provides a new idea for the study of embedded BCI system for practical application.

Key words: Brain Computer Interface(BCI), movement imagery, Maximum Overlap Wavelet Transform(MODWT), energy curve, pattern classification, motor steering control

中图分类号: