诱发式脑—机接口的设计与优化

发布时间：2018-07-11 21:45

本文选题：脑-机接口 + 脑电图　；参考：《国防科学技术大学》2015年博士论文

【摘要】：脑-机接口(brain-computer interface,BCI)是一种新的不依赖于外周神经和肌肉,与大脑直连的通信接口技术。诱发式BCI作为BCI领域的主要分支之一,其简单、高效和多目标的信息传输特点,已经引起了学者的广泛关注。但是近年来,尽管诱发式BCI技术的研究已经取得了较大进展,人机通信工效得到了显著提高,其系统的通信控制能力仍难以满足实际应用的需求。为了推动诱发式BCI向实用化的转变,本文以人机通信工效的提高为目标牵引,以基于事件相关电位(event-related potential,ERP)和稳态视觉诱发电位(steady-state visually evoked potential,SSVEP)的BCI系统为研究对象,分别在多模态刺激、多模态特征电位,以及被试对BCI操控任务的主观能动性三个方面展开了研究。主要成果如下:听触觉双模态刺激研究与非视觉ERP-BCI系统设计。为了验证多模态刺激对提高诱发式BCI性能的作用,本文提出了一种基于听觉和触觉刺激的双模态ERP-BCI方法。在该方法中,我们通过在同一方向上同时施加听觉和触觉随机刺激的方式,设计了一个方向一致双模态范式。由于该系统的操控不需要被试的视觉交互,故将其称之为非视觉BCI。为了提高系统在线性能,我们为每名被试选择了最佳EEG通道和最优的trial。实验结果表明所提出的双模态ERP-BCI相比于单模态的听觉和触觉ERP-BCI方法,在线平均ITR分别提高了45.43%和51.05%。据我们所知,在视线独立ERP-BCI研究中,我们所提出的双模态ERP-BCI系统达到了至今最高的在线ITR(10.77 bits/min)。该双模态ERP-BCI方法为非视觉BCI提供了一个新的研究方向。SSVEP-BCI的实时信息反馈与动态优化设计研究。为了提高系统性能,本文提出一种带实时信息反馈机制的动态优化SSVEP-BCI。在该方法中,我们首先基于实时目标识别结果设计了一种生物反馈机制,用于提高被试对目标刺激的注意力。其次,我们将行列(row/column,RC)范式引入到SSVEP-BCI中,采用6个频率的闪烁刺激,设计了一个36选项SSVEP-BCI拼写器。此外,我们设计了目标选择时间动态优化方法,并与固定优化方法进行了在线性能的对比。实验结果表明我们所提出的SSVEP-BCI拼写器的性能与传统方法相比有了显著的提高。具体地说,实时信息反馈通过增强被试对目标刺激的注意力,有效提高了拼写准确率。动态优化方法相比于固定优化达到了更高的PITR。基于P300与SSVEP的混合BCI研究与优化设计。在诱发式BCI研究领域,大多数研究仅集中在对单模态特征电位的分析识别上,系统性能正处于瓶颈状态。为了进一步提升BCI的性能,我们基于P300和SSVEP两种特征电位提出了三个混合BCI方法。具体内容如下:首先,我们论证了同时诱发P300与SSVEP特征电位的可行性,并将SSVEP特征引入P300拼写器,提出了一种新型的混合BCI方法。在该方法中,我们设计了一种由随机闪烁和周期闪烁构成的混合刺激机制。所述混合BCI拼写器中目标字符由三维时频信号的融合结果识别得到。实验结果表明,所述混合BCI拼写器的单trial在线拼写准确率达到了93.85%。这表明SSVEP特征的引入有效提高了P300拼写器的目标识别准确率,达到了更加准确、稳定的拼写效果。第二,为了提高拼写速度,我们采用P300和SSVEP平行输入的方式,提出一种新的快速混合BCI拼写方法。在该方法中,目标字符坐标由P300和SSVEP两个子拼写器同时输出的二维坐标确定。我们基于该方法设计了两种BCI拼写器范式:SL和RC范式。此外,我们设计了基于PITR曲线的最优trial优化选择方法。实验结果表明所提出的混合BCI方法相比于P300和SSVEP BCI,拼写速度有了明显的提高。其中,该混合BCI方法RC范式的平均PITR达到了53.06 bits/min。最后,为了达到P300和SSVEP特征信息更合理的运用,我们基于P300和SSVEP的得分融合提出了一种新的64选项混合BCI。在该方法中,我们将P300拼写器的RC范式和SSVEP-BCI的两步选择范式相结合,提出了两个混合BCI,即DRC和4-D范式。此外,我们提出了一种MPE融合方法用来实现P300和SSVEP得分级的信息融合。实验结果表明4-D混合范式相比于DRC范式性能更优越;MPE融合方法相比于其它四种常用的得分融合方法达到了更高的准确率。此外,13名被试中的12名采用4-D范式的单trial在线准确率均超过90%,而且平均准确率到达了95.18%。
[Abstract]:Brain-computer interface (BCI) is a new communication interface technology that does not depend on the peripheral nerve and muscle and directly connected with the brain. The induced BCI is one of the main branches of the BCI field. Its simple, efficient and multi target information transmission characteristics have aroused the widespread concern of the scholars. However, in recent years, despite the induced BCI The research of technology has made great progress, the efficiency of man-machine communication has been greatly improved, and the communication control ability of the system is still difficult to meet the needs of practical application. In order to promote the transformation of the induced BCI to practicality, this paper takes the improvement of the ergonomics as the target, based on the event related potential (event-related potential). ERP) and the BCI system of steady-state visually evoked potential (SSVEP), as the research object, are studied in three aspects of multimodal stimulation, multimodal characteristic potential, and subjective initiative of BCI manipulation tasks. The main results are as follows: the study of auditory tactile dual modal stimulation and non visual ERP-BCI. System design. In order to verify the effect of multimodal stimulation on improving the performance of induced BCI, a dual mode ERP-BCI method based on auditory and tactile stimulation is proposed in this paper. In this method, we set up a uniform dual mode paradigm by applying both auditory and tactile random stimuli in the same direction. The system's manipulation does not require the visual interaction of the subjects, so it is called a non visual BCI. to improve the system's online performance. We choose the best EEG channel and the optimal trial. experiment for each of the subjects to show that the proposed bimodal ERP-BCI is 45 higher than the single modal auditory and touch ERP-BCI method. .43% and 51.05%., as we know, in the line of sight independent ERP-BCI research, our proposed dual mode ERP-BCI system has reached the highest online ITR (10.77 bits/min). The dual mode ERP-BCI method provides a new research direction.SSVEP-BCI's real time feedback and dynamic optimization design for non visual BCI. In this paper, we propose a dynamic optimization SSVEP-BCI. with real-time information feedback mechanism. In this method, we first design a biofeedback mechanism based on real time target recognition results to improve the attention of the subjects to the target stimulus. Secondly, we introduce the row/ column (RC) paradigm to SSVEP-BCI and use 6 frequencies. A 36 option SSVEP-BCI spelling device is designed. In addition, we design the time dynamic optimization method of the target selection and compare the performance with the fixed optimization method. The experimental results show that the performance of the SSVEP-BCI spelling device is improved significantly compared with the traditional method. Specifically, real time. Information feedback improves the accuracy of spelling by enhancing the attention of the subjects to the target stimulation. The dynamic optimization method has achieved a higher PITR. based on the hybrid BCI research and optimization design based on P300 and SSVEP compared with the fixed optimization. In the field of induced BCI research, most of the studies focus only on the analysis and recognition of the single mode characteristic potential. In order to further improve the performance of BCI, we propose three mixed BCI methods based on two characteristic potentials of P300 and SSVEP. The specific contents are as follows: first, we demonstrate the feasibility of simultaneously inducing P300 and SSVEP characteristic potentials, and introduce the SSVEP features to the P300 spelling device, and propose a new type of mixture. In this method, we design a hybrid stimulation mechanism composed of random flicker and periodic flicker. The target character in the mixed BCI spelling device is identified by the fusion of three dimensional time-frequency signals. The experimental results show that the single trial spelling accuracy of the mixed BCI spelling device is 93.85%., which indicates that SSV The introduction of EP features effectively improves the target recognition accuracy of the P300 spelling device, achieves a more accurate and stable spelling effect. Second, in order to improve the spelling speed, we use the parallel input of P300 and SSVEP to propose a new fast mixed BCI spelling method. In this method, the target character coordinates are spelled by P300 and SSVEP two. The writer designs the two BCI spelling paradigm based on this method: SL and RC paradigm. In addition, we designed the optimal trial optimization method based on the PITR curve. The experimental results show that the proposed hybrid BCI method has a significant improvement in spelling speed compared to P300 and SSVEP BCI. The average PITR of the RC paradigm is 53.06 bits/min. at the end of the BCI method. In order to achieve more reasonable use of P300 and SSVEP feature information, we propose a new 64 option hybrid BCI. based on the score fusion of P300 and SSVEP in this method. We combine the RC paradigm of the P300 spelling device with the two step paradigm of SSVEP-BCI. Two mixed BCI, that is, DRC and 4-D paradigms. In addition, we propose a MPE fusion method to achieve the information fusion of P300 and SSVEP grading. The experimental results show that the 4-D hybrid paradigm is superior to the DRC paradigm; the MPE fusion method has achieved a higher accuracy compared with the other four common scoring methods. In addition, 13 The online accuracy rate of 12 single trial using 4-D paradigm was over 90%, and the average accuracy reached 95.18%..
【学位授予单位】：国防科学技术大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：R338;TN911.7

【相似文献】