【作者】 张磊；

【导师】 吴小培；

【作者基本信息】 安徽大学 ， 计算机应用技术， 2020， 博士

【摘要】 脑-机接口(Brain-Computer Interface,BCI)是一种特殊的人机交互技术,其以头皮表面检测到的脑电信号(Electroencephalogram,EEG)为信息载体,实现人脑对外部设备的直接控制。BCI技术的研究具有重要的理论意义和广阔的应用前景。基于EEG的BCI系统主要有四种类型:稳态视觉诱发电位(Steady-State Visual Evoked Potentials,SSVEPs)、运动想象(Motor Imagery,MI)、基于事件相关电位的P300和慢皮质电位。不同类型的BCI系统各有优缺点。混合BCI技术是通过结合两种或两种以上的传统BCI模式来弥补单一模式的缺点,开发出功能更强、稳定性更好的多模态BCI系统。尽管脑-机接口系统的性能方面获得了很大的突破,但就目前而言,BCI系统在可靠性和普适性方面还不够完美。这些不足主要是由于脑电信号低的信噪比,易受噪声干扰和非稳态特性引起的。本论文主要对BCI研究领域的运动想象和稳态视觉诱发两大方向展开相关研究。以脑-机交互的自主可控性和稳定性改善为主要目标,并注重训练和操作过程的“自然性”和“舒适性”等指标,从而设计与实现实用化的BCI在线系统。围绕上述目标,分别从脑电信号采集、信号处理、特征提取和分类识别等模块进行相关研究。研究内容主要包含两大部分,即理论研究和系统实现。理论研究的核心是EEG信号处理和模式识别算法的研究。在运动想象脑-机接口(Motor Imagery BCI,MI-BCI)方面,重点研究共同空间模式(Common Spatial Pattern,CSP)和独立分量分析(Independent Component Analysis,ICA)的空域滤波方法在MI-BCI中的合理应用。在SSVEP-BCI方面,重点研究典型相关算法(Canonical Correlation Analysis,CCA)在异步BCI中的应用。为了综合评测不同算法的性能指标,分别建立了MI-BCI和SSVEP-BCI数据库和算法评测平台。系统实现的难点是建立合理的脑-机协同控制策略,研究多模态BCI的融合技术,搭建在线和异步模式的BCI系统。在系统实现的过程中,需要解决一系列工程问题,例如软件系统的开发、数据的实时通信、算法的识别精度与执行效率之间平衡等情形。本文的主要贡献和创新点主要体现在以下几个方面:第一,针对传统的CSP算法稳定性不好和训练模型易受噪声干扰的缺点,本文综合现有多种类型CSP算法的优点,从时间-频率-空间联合优化角度出发,提出滑动频带滤波的CSP算法,用于自动寻找稳定的个性化特征参数。该方法可以有效降低检测阶段的计算成本,便于实现少导联和低成本的BCI系统。实验结果表明,所提算法不仅识别率好,而且执行效率高,满足在线模式MI-BCI系统开发的需求。第二,为了减少MI-BCI系统的训练时间和解决传统的ICA算法不易使用的劣势,通过对多种经典的ICA算法进行研究和对比分析,提出一种应用于MI-BCI在线系统检测的改进的信息极大化ICA算法。该方法可从较少的无标签训练数据中自动检测和输出与运动想象类别相关的独立分量,大幅缩短训练时间,有效提高计算效率;此外,所提算法在不同被试之间具有较好的模型迁移性能,鲁棒性好,所设计的BCI系统运行状态稳定。第三,针对现有的MI-BCI系统目标较少和信息传输率较低的局限性,本文融合自发脑电alpha节律可以自主控制和SSVEP目标数量多的优点,设计一种新型模式混合的BCI系统。同时提出一种滑动窗投票判断策略的算法,能够将计算效率高效的CCA算法应用于连续控制的异步SSVEP-BCI系统中。结果显示,系统的信息传输速率(Information Transmission Rate,ITR)和灵敏度都得到了明显的提高,这对开发实用化的多模态BCI在线系统具有重要的意义。第四,依照设计目标,分别设计与开发了基于空域滤波的MI-BCI同步在线系统、基于EEG和EOG混合的异步MI-BCI系统以及基于Alpha节律和SSVEP混合的异步BCI系统,各项指标满足设计要求。系统实现工作使得BCI理论研究成果可以落地,表明本论文所做的工作具有实际应用价值。文章最后,对本论文的研究工作和相关成果进行总结,并对下一步的研究方向进行展望。更多还原

【Abstract】 A Brain-Computer Interface(BCI)is a specialized type of technology that enables human-computer interaction characterized by the use of electroencephalography(EEG)signals as information carriers,which allows the human brain to exert direct control over external equipment.Research on BCI technology has important theoretical significance and broad application prospects.There are four main types of BCI systems based on EEG: steady-state visual evoked potentials(SSVEPs),motor imagery(MI),P300 based on event-related potentials(ERPs),and slow cortical potential(SCP).Different types of BCI systems have their own advantages and disadvantages.The hybrid BCI system compensates the various shortcomings of each mode by combining two or more traditional BCI modalities and can develop a more powerful and more stable multimodal BCI system.Although impressive improvements in BCI efficiency have been achieved,the current BCI systems are far from being perfect in terms of reliability and generalizability.This suboptimal performance can be mainly attributed to a low signal-to-noise ratio,the presence of artifacts in the data and the non-stationary nature of the EEG signal.This dissertation focuses on two major types of BCI: MI and SSVEP.The main goal is to improve the autonomous controllability and stability of BCI,and pay attention to the natural and comfort indicators of training and operation processes,so as to design and implement practical and online BCI systems.A series of research works were carried out around this goal from the EEG signal acquisition,signal processing,feature extraction,classification and recognition modules.The research content includes two major parts: theoretical research and system realization.The core content of theoretical research is the research of EEG signal processing and pattern recognition algorithms.In the motor imagery BCI(MI-BCI),this dissertation focuses on the spatial filtering methods of common spatial pattern(CSP)and independent component analysis(ICA).In terms of SSVEP-BCI,this dissertation focuses on the application of canonical correlation analysis(CCA)in the asynchronous BCI.To comprehensively evaluate the performance of different algorithms,we have established MI-BCI and SSVEP-BCI databases and algorithm evaluation platforms.The difficulty of system implementation is to establish a more reasonable brain-computer control strategy,study some key fusion technologies of hybrid BCI,and build online and asynchronous BCI systems.In the process of system implementation,a series of engineering problems needs to be solved,such as the development of software systems,real-time data communication,and the balance between the recognition accuracy and execution efficiency of algorithms.The contributions and innovations of this dissertation are given as follows:First,in view of the disadvantages of the traditional CSP algorithms,which are not stable and are sensitive to noise,this dissertation combined the advantages of various existing CSP algorithms from the perspective of time-frequency-space joint optimization and proposed a sliding frequency band CSP algorithm,which was used to automatically find the optimal personalized feature parameters for the specific subject.Compared with other algorithms,this method can effectively reduce the calculation cost in the detection step and facilitate the realization of a BCI system with fewer channels and lower cost.The experimental results show that the proposed algorithm has not only good classification performance but also fast execution efficiency,which meets the requirements of the online mode BCI system development.Second,to reduce the training time of the MI-BCI system and to solve the shortcoming that the traditional ICA algorithms are not easy to use,this dissertation comprehensively compared a variety of classic ICA algorithms and proposed an improved information maximization ICA algorithm by modifying the iterative strategy,ultimately applying this algorithm to online system detection.This method can automatically detect and output independent components related to motor imagery tasks from less unlabelled training data,greatly reducing the training time;in addition,the proposed algorithm has better spatial model transfer performance between different subjects,high execution efficiency,and good robustness,and the designed MI-BCI system has a stable operation state.Third,aiming at the shortcomings of the existing MI-BCI system with fewer goals and lower information transmission rate(ITR),this dissertation proposed a new multimodal BCI system based on Alpha rhythm and SSVEP.By applying the sliding window voting discrimination(SWVD)strategy and optimizing the experimental paradigm,the classical CCA algorithm was successfully applied for continuous control in an asynchronous BCI system.With the developed system,obvious improvements in the ITR and sensitivity were achieved,which will be beneficial for the development of practical BCI systems.Fourth,according to the design goals,we have designed and developed synchronous online MI-BCI systems based on CSP and ICA spatial filtering,an asynchronous MI-BCI system based on the hybrid of EEG and EOG,and an asynchronous SSVEP-BCI system with Alpha Rhythm and SSVEP.The implementation of these systems makes our BCI theoretical research results available,demonstrating that the work done in this dissertation has important application value.Finally,the work and research results of this dissertation are summarized,and the prospect of future research is put forward.更多还原