脑电信号共空间模式模糊融合的研究  

Study on Electroencephalogram Recognition Framework by Common Spatial Pattern and Fuzzy Fusion

作　　者：徐鲁强[1,2] 肖光灿[2] 黎茂锋[2] 

机构地区：[1]西南交通大学信息学院,成都610031 [2]西南科技大学计算机学院,绵阳621010

出　　处：《生物医学工程学杂志》2015年第6期1173-1178,共6页Journal of Biomedical Engineering

基　　金：四川省人社厅留学择优资助项目;绵阳网络融合工程实验室开放基金资助项目(12ZXWK07)

摘　　要：共空间模式(CSP)是脑电信号特征提取的主要方法,但它存在较严重的过拟合问题。本文提出对多通道脑电数据划分多个区域,分别用CSP算法提取各区域的脑电数据特征,对得到的各特征分别进行线性分类,用Choquet模糊积分融合各线性分类结果,有助于克服脑电信号处理的过拟合问题和提高脑电信号识别准确度,从而给出了脑电数据处理的一种新框架。采用2005年国际脑机接口(BCI)竞赛数据验证该处理框架,获得的识别准确率显著提高,并且在一定程度上解决了CSP的过拟合问题,显示了本框架处理脑电信息的有效性。Common spatial pattern （CSP） is a very popular method for spatial filtering to extract the features from electroencephalogram （EEG） signals, but it may cause serious over-fitting issue. In this paper, after the extraction and recognition of feature, we present a new way in which the recognition results are fused to overcome the over- fitting and improve recognition accuracy. And then a new framework for EEG recognition is proposed by using CSP to extract features from EEG signals, using linear discriminant analysis （LDA） classifiers to identify the user＇s men- tal state from such features, and using Choquet fuzzy integral to fuse classifiers results. Brain computer interface （BCI） competition 2005 data sets IVa was used to validate the framework. The results demonstrated that it effective- ly improved recognition and to some extent overcome the over-fitting problem of CSP. It showed the effectiveness of this framework for dealing with EEG.

关 键 词：脑电 共空间模式 信息融合 脑机接口 线性识别 Choquet 模糊积分 

分 类 号：R318[医药卫生—生物医学工程]