【作者】 袁道任；

【导师】 逯鹏；

【作者基本信息】 郑州大学 ， 控制理论与控制工程， 2013， 硕士

【摘要】 脑机接口(Brain Computer Interface,BCI)旨在建立脑思维活动的意向与行为之间的关联,实现直接的人机信息交互和控制,在脑认知和生物反馈训练等有着广阔的发展应用前景。EEG(Electroencephalogram)具有非线性、非平稳性且易受外界干扰等特点,是综合反映大脑组织电活动及大脑的功能状态的载体,具有较高的时间分辨率和响应实时性。脑思维具有高度复杂性,目前典型单模态BCI是基于单一类别EEG信号,只能在简单思维活动层面识别不同类别任务,识别效率低、系统通用性差。旨在通过多种途径分析思维任务的多模态BCI系统,基于多种形式EEG信号之间的互补性,通过融合和综合利用多类型EEG信号,提供多类型多任务意识识别,并提高系统通用性和识别正确率,已经成为BCI发展的趋势。如何快速有效地从多模态EEG信号中识别表征意识任务的特征量,区分不同的意识任务,产生反映大脑意识的控制命令,是基于多模态EEG信号的多模态BCI系统的核心问题,预处理、特征提取和分类识别算法是其中的关键技术。以实现左、右手和左、右脚的四分类意识指令识别的多模态BCI系统为目标；以稳态视觉诱发(Steady State Visual Evoked Potential, SSVEP)和运动想象(Motor Imagery, MI)的多模态EEG信号处理关键算法作为核心,以提高多模态BCI系统识别正确率和速度为切入点,构建基于多模态EEG信号的BCI系统框架,设计多模态BCI系统的实验范式,采集多模态EEG信号,设计预处理、特征提取和分类识别等关键算法,实现四分类的BCI控制。实验结果表明,系统具有高识别准确率和速度。主要研究结果包括：1)对比研究了典型基于EEG信号的BCI系统,针对单模态的EEG信号易受主观影响,特征信息单一,传输速度受制约等问题,提出融合运动想象MI和视觉稳态诱发SSVEP构建多模态EEG信号,为实现BCI系统的左、右手和左、右脚的最佳分类控制提供了研究基础；2)围绕多模态EEG信号的BCI系统的核心算法,以实现左、右手和左、右脚的四分类BCI系统为目标,设计了特定的实验范式,采集大样本被试的多模态EEG信号；确定了C3、C4、FC3、FC4和Cz等12个采集通道,选取了相干平均预处理算法,通过比较AR模型、WT、WPT和HHT等特征提取算法,以及Fisher和SVM分类识别算法的大样本被试基础上的实验结果,设计了AR+HHT的特征提取和SVM分类识别的组合算法；3)针对实现多模态BCI系统中的问题,基于上述核心算法,构建了识别大脑的左、右手和左、右脚的不同躯体部位的运动想象的BCI系统框架,并编程实现了上述系统框架,设计并进行了大样本对比实验,验证算法和系统框架的有效性,实验结果表明,基于多模态EEG信号的BCI系统达到了高识别准确率和速度。更多还原

【Abstract】 The Brain Computer Interface(BCI) system aims at establishing the correlation between the intention of brain thinking and the behavior, and finally achieving the direct human-computer interaction and control. BCI has a broad development and application prospect in brain cognition and biofeedback training, etc.. Electroencephalogram(EEG) signal has the characteristics of nonlinearity, nonstationarity and susceptivity to external interference. Having high time resolution and real-time response, EEG is the carrier that can comprehensively reflect the electric activity of brain tissue and the functional state of brain.Brain thinking is highly complex. At present, the typical single-mode BCI based on single category of EEG signals can only identify tasks on the level of simple thinking activity with low recognition rate and poor versatility. Aiming at analyzing the thinking tasks from multiple channels, the multi-mode BCI system based on complementarity of the various EEG signals can provide multi-type and multi-task of conscious recognitions and improve the recognition accuracy and versatility of system through appropriate fusion and comprehensive utilization of multi-modal EEG signals. The multi-modal BCI system has become the trend of BCI development. How to rapidly and effectively select the characteristic parameters that can symbolize the conscious task from the complex EEG signals, distinguish different conscious tasks and produce the corresponding control commands is the core for the multi-modal BCI system based on multi-mode EEG signals, and the algorithms for pretreatment, feature extraction and classification are its key technologies.In order to realize the identification of four kinds of conscious commands(left and right hand, left and right foot) in multi-modal BCI system, taking the key algorithms for processing the multi-modal EEG signals based on Steady State Visual Evoked Potential(SSVEP) and Motor Imagery(MI) as the core, improving the recognition accuracy and speed of multi-modal BCI system as the breakthrough point, this thesis constructed a BCI system framework based on multi-modal EEG signals, designed an experimental paradigm for multi-modal BCI system, collected multi-modal EEG signals, designed the key algorithms for pretreatment, feature extraction and classification&recognition and achieved the BCI control of four kinds of conscious commands. The experimental results demonstrated that this system has high recognition accuracy and speed. The main findings include:1) The typical BCI systems based on EEG signal were comparatively studied. The single-mode EEG signal is easily influenced by the subjective factors and has other problems like singleness of feature information and low speed of transmission. In accordance with these problems, the multi-modal EEG signal based on MI and SSVEP was put forward, providing basis for realizing the best classification control of left and right hand, left and right foot in BCI system;2) With identification of four kinds of conscious commands (left and right hand, left and right foot) as the objective, I designed a novel experimental paradigm using the core algorithms for the BCI system based on multi-modal EEG signals. The multi-modal EEG signals of large samples were collected. Twelve acquisition channels including C3, C4, FC3, FC4and Cz were determined. The coherent averaging preprocessing algorithm was selected. After comparison of the feature extraction algorithms of AR model, WT, WPT and HHT, classification identification algorithms of Fisher and SVM, a combined algorithm of AR+HHT feature extraction and SVM classification&identification was designed;3) Targeting at the problems in realizing the multi-modal BCI system, this thesis constructed a BCI system framework that can identify the MI of left and right hand, left and right foot based on the above algorithms, and this framework was realized by programming. A large sample contrast test was designed to verify the validity of the algorithms and system framework. The experimental results demonstrated that the BCI system based on multi-modal EEG signal has high recognition accuracy and speed.更多还原