摘要
各种脑部、神经系统疾患的增多,使脑科学成为21世纪具有挑战性的研究。许多有神经一肌肉障碍疾病的患者无法向外界传递信息,他们基本的活动能力以及与外界正常交流的能力都被剥夺了,他们所要表达的信息可能被完全阻隔在体内。因此对脑科学研究的需求就越来越迫切,人们渴望能够通过对脑科学及相关领域的研究来克服人类自身的缺陷和不足。最近几年,随着电子技术、计算机技术的飞速发展,脑电信号处理的基础已经逐渐成熟起来,脑—机接口(Brain-Computer Interface,BCI)的研究开始成为一个热点问题。
在BCI的研究中,基于脑电信号的BCI系统因为简单、安全、无创而倍受关注。本文研究的就是基于想象运动期间脑电信号的BCI系统,目的是利用人的自发脑电,识别特定的意识任务,形成控制命令,从而实现人脑与计算机间的信息交换,希望为有严重行为障碍的残疾人提供帮助。这种基于自发脑电的BCI系统,系统简单,不需要外部刺激装置,训练过程短,适应范围广,有很好的应用前景。但从头皮上获得的自发脑电信号十分微弱,而且信号的信噪比很低,因此有效的脑电信号处理方法是BCI研究的一项关键技术。在总结前人工作的基础上,本文对脑电信号采集、处理进行了以下几个方面的研究。
1)系统信号采集模块设计
建立信号采集模块,采集脑电信号。在实验室已有的脑电采集设备及软件基础上,针对自己的课题研究,设计相应的大脑意识任务实验。按照实验要求,使用VC++6.0编写出相应的脑电信号采集软件。
2)脑电信号预处理
由于脑电信号本身极其微弱,非常容易受到各种干扰与噪声的影响,因此对脑电信号的去噪处理是必要的。本文分别采用了小波变换、数字滤波等方法对实验中采集的脑电信号进行处理,去除其中的心电、眼电、肌电等干扰信号,提取有效信息,取得了较好的效果。
3)脑电信号特征提取
有效地提取大脑思维活动时的特征信息是BCI研究的关键技术之一,是正确识别不同意识模式的基础。本论文将自适应自回归模型(Adaptiveautoregressive model,AAR)系数与事件相关去同步(evewt-relateddesynchronization,ERD)相结合提取脑电信号的一个特征,而将基于小波变换和shannon熵概念的小波熵(Wavelet Entropy)作为另一个特征,从而实现对想象左右手运动意识任务的脑电特征的提取。
4)意识任务分类器设计
分类器的设计是BCI系统中的另一十分重要的环节,分类器的性能将直接影响BCI系统的性能。本文采用了Fisher线性判别、人工神经网络和支持向量机(Support Vector Machine,SVM)几种分类方法。其中,支持向量机以统计学习理论为基础,它不仅要求最优分类面将两类样本无错误地分开,而且要使类间间隔最大,从而保证真实风险最小,能够较好地解决了小样本的分类问题。经过实验的分类比较,最终分类结果表明:支持向量机的分类正确率是最高的,能够很好的识别出想象左右手运动两种意识任务,得到理想的分类效果。
采用本课题采集的数据,对于所提取的脑电特征,支持向量机分类的正确识别率可达83.3%,应用同样的算法对2003年国际BCI竞赛中的左右手想象运动数据进行识别时,正确率可达87.9%。这说明应用AAR模型系数的ERD/ERS、小波熵和支持向量机能够作为现在BCI系统设计中一种可行分类算法。
The various kinds of brain illness and illness of neural system make brain science become the most challenging research in the 21~(st) century. Many neuromuscular disorders can disrupt the channels through which the brain communicates with and controls its external environment, which usually deprive patients; of their basic movement function and normal communication. The common channel responded to environment is sometimes useless to them. So the research on brain science has attracted more attention to satisfy the demands of the people who want to overcome their disadvantages by the brain science research and the related domains. In recent years, with the rapid development of electric and computer technology, the foundation of EEG processing has been built. And the research of Brain Computer Interface (BCI) also becomes a hot item.
In the research of BCI, the BCI system based on electroencephalogram gets much attention because of its characteristic of simplicity, safety and non-injury. In this paper, the BCI system based on imaginary motor is researched. The purpose of this thesis is to investigate BCI system based on spontaneous electroencephalographic signals, identify particular mental tasks, and form the control dictation. So the commutation between the brain and the computer can be come true and those persons who are seriously handicapped can get help. This kind of BCI system would be a promising one because of the simple structure, easy and non-injury EEG signals, no stimulation equipment, short training processing, etc. At the same time, the spontaneous electroencephalographic signals that are collected from scalp are very weak and the signal-to-noise of signals is low. So the valid method of electroencephalographic signal processing is a key technology in the research of BCI. In the foundation of the past work, this paper focuses on some topics described as following.
1) The design of information collected.
Information collected model is created to collected the electroencephalographic signals. A new experiment is designed to record the EEG according to the requirement of the subject on the basis of the EEG record equipment and the former software in the laboratory. And then a software is devised using VC++6.0 to meet the requirements of the experiment.
2) Pre-processing of EEG signals
EEG signals are very weak and are always influenced by eye movements, blinks and muscle, heart and power line noise. And so it is necessary to remove the artifacts and noised in EEG. In this paper, wavelet transformation and digital filter are applied to remove the artifacts. The excellent result has been achieved.
3) Facture extraction of the EEG signals
In BCI system, it is very important to find a meaningful EEG signal feature that contains the remarkable information of different mental actions. The adaptive autoregressive model combined with the event-related desynchronization is used in our research to obtain one feature. The wavelet entropy is treated as the other feature in our approach. So the feature extraction of EEG signals is completed
4) Design of the Mental Tasks Classifier
It is necessary for a BCI system to design a classifier with an excellent performance. For this purpose, different approaches have been investigated in detail. Recent advances in machine learning research have pointed out the advantaged of support vector machine(SVM) over other classification techniques. Solid theoretical foundations, good generalization capabilities and easy parameters updating are among the most appealing qualities of SVM for BCI applications. In the experiment, classifiers adopt Fisher discrimination, RBF neural net work and support vector machine (SVM). SVM classifier achieves the highest accuracy comparing the others in the classification results.
The dataset of 2003 international BCI competition are used for research purpose, and dataset recorded in our experiment are adopted for testing the algorithm. The classification accuracy using SVM on features extracted from EEG can reach 87.9% and 83.3%. The results show that ERD/ERS, wavelet entropy and SVM can be used as a classifier in BCI system.
引文
[1]G.Pfurtscheller,et al.Brain-computer interface-a new communication device for handicapped persons[J].J MicrocomputAppl,1993,16:293 - 299.
[2]Wolpaw JR,Birbaumer N,Mcfarl DJ,et al.Brain-computer interfaces for communication and control[J].Clinical Neurophycioligy,220,113(6):767-791.
[3]程明,任宇鹏,高小榕等,脑电信号控制康复机器人的关键技术.机器人技术与应用,2003,4:45-48.
[4]Birch,G.E,Mason,S.G,Borisoff,J.F.Current trends in brain-computer interface research at the Neil Squire foundation.IEEE Trans Rehab Eng,2003,11(2):123-126.
[5]程明,高上凯.基于脑电信号的脑-计算机接口[J].北京生物医学工程,2000,19(2):113-117.
[6]杨帮华,颜国正,丁国清等,脑机接口关键技术研究.北京生物医学工程,2005,24(4):308-310.
[7]Sutter E E.The brain response interface:communication through visually-induced electrical brain response.Microcomputer Appl,1992,15:31 - 45.
[8]杨立才,李佰敏,李光林.脑机接口技术综述.电子学报第33卷,第7期,1234-1241.
[9]Farwell L.A,Donchin E.Talking of the top of your head:toward a mental prothesis utilizing event-related brain potentials[J].Electroenceph Clin.Neurophysiol,1988,70:510-523.
[10]Donchine,Spencer K M,Wijesinghe R.The mental prosthesis:Assessing the speed of a P300-based brain-computer interface.IEEE Trans Rehab Eng,2000,8(2):174-179.
[11]G.Pfurtscheller,et al.Differentiation between finger toe and tongue movement in man based of 40Hz EEG[J].Electroenceph Clin Neurophysiol,1994,90:456- 460.
[12]Kalcher J,Flotzinger D,Neuper CH,et al.Graz brain-computer interface Ⅱ:towards communication between humans and computers based on online classification of three different EEG patterns[J].MedBiolEng Computer,1996,34(5):382-388.
[13]Wolpaw J R,et al.An EEG-based brain-computer interface for cursor control [J].Electroenceph Clin Neroophysiol,1991,78:252- 259.
[14]Wolpaw J R,et al.Multichannel EEG-based Brain-computer communication[J]. Electroenceph Clin Neroophysiol,1994,90:444-449.
[15]McMillan G R,et al.Direct brain-interface utilizing self-regulation of steady-state visual evoked response(ssver)[J].Proceedings of RESNA,June 9214,1995,693-695.
[16]Dewan EM.Occipital alpha rhythm eye position and lens accommodation[J].Nature,1967,214:975-977.
[17]Kirkup L,et al.EEG-based system for rapid on-off switching without prior learning[J].Med BiolEng Computer,1997,35:504-509.
[18]Keirn,Z.A,Aunon,J.I.A new mode of communication between man and his surroundings[J].IEEE Trans Biomedical.Eng,1990,37(12):1209- 1214.
[19]Cheng Ming,Gao Shangkai.An EEG-based cursor control system.Proceedings of the first joint BMES/EMBS Conference[J].Atlanta,GA,USA:IEEE BMES/EMBS,1999,669.
[20]谢勤岚,杨仲乐.基于IR-ERP的脑-机接口的结构与原理[J].武汉理工大学学报,2003,27(3):318-321.
[21]谭郁玲,侯沂等.临床脑电图与脑地形图学[M].北京:人民卫生出版社,1998:1-9.
[22]姚泰.生理学.北京:人民卫生出版社,2000,12,287-332.
[23]蔡建新,张唯真.生物医学电子学.北京:北京大学出版社,1997,12,13-22.
[24]赵仑.ERP实验教程.天津:天津社会科学院出版社.2004,9,40-45.
[25]Spreckelsen M,Bromm B.Estimation of single-evoked cerebral potentials by means of parametric modeling and kalman filtering[J].IEEE Trans Biomed Eng,1988,35(9):691-700.
[26]吴芝芳等.提取脑电信号的新方法研究-自适应卡尔曼浦波[J],信号处理,Vol 10No4,Dec 1994
[27]彭玉华.小波变换与工程应用[M].北京:科学出版社,1999
[28]朱洪俊.心电信号零相位数字滤波[J].北京生物医学工程.Vol.22,No.4,260-262.
[29]James N.Knight Signal Fraction Analysis and Artifact Removal in EEG.Masters Thesis,Department of Computer Science,Colorado State University,Fort Collins.CO 2003.
[30]Ramaswamy P.Brain Computer Interface Design Using Band.Powers Extracted During Mental Tasks.Processings of the 2th International IEEE EMBS Conference on Neural Engineering Arilington Virginia 2005:16-19.
[31]Anderson C,Stolz Z,Shamsunder S.Discriminating Mental Tasks Using EEG Represented by AR Models.Proceedings of the 1995 IEEE Engineering in Medicine and Biology Annual Conference,1995:Montreal,Canada.
[32]Anderson C,Stolz Z,Shamsunder S.Multivariate Autoregressive Models for Classification of Spontaneous Electroencephalogram During Mental Tasks,IEEE Transaction on Biomedical Engineering 1998:45(3):277-286.
[33]胡广书 数字信号处理.清华大学出版社2003:517-519.
[34]G.Pfurtscheller,Lopesda Silva F H.Event-related EEG/EMG Synchronization and desynchronization:basic principles.Clinical Neurophysiology,1999,110:1842-1857.
[35]郑治真.小波变换及其NATLAB工具的应用.地震出版社.2001.
[36]周伟等.NATLAB小波分析高级技术.西安:西安电子科技大学出版社,2006.
[37]Cososchi S,Strungaru R,Ungureanu A,Ungureanu,M.EEG Features Extraction for Motor Imagery.Proceedings of the 28th EMBS Annual International Conference.2006:1142- 1145.
[38]封洲燕.应用小波熵分析大鼠脑电信号的动态特性[J].生物物理学报,2002,18(3):325-330.
[39]G.Pfurtscheller,et al.Current trends in Graz brain-computer interface (BCI)research[J].IEEE Trans.Rehabil.Eng,2000,8(2):216-219.
[40]杨福生,吕扬生.生物医学信号的处理和识别.天津科技翻译出版公司.1997.12:291-315.
[41]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京:清华大学出版社,2000
[42]周志华,曹存根.神经网络及其应用.清华大学出版社,2004.
[43]韦保林,罗晓曙,汪秉宏等.EEG信号的径向基函数神经网络预测[J].中国生物医学工程学报.1999,第2卷,第6期,488-492.
[44]ComesC,VVapnik,Support-Vector networks Machine Learning,1995,20:273-297.