SIA OpenIR  > 机器人学研究室
基于EEG EMG信号解码的脑机自然交互关键技术研究
Alternative TitleResearch on Key Technologies of Brain-machine Natural Interaction based on EEG/EMG Signal Decoding
褚亚奇
Department机器人学研究室
Thesis Advisor徐卫良 ; 赵新刚
Keyword人机交互 脑机接口 非完整脑电 同侧肢体多类运动想象 混合脑/肌电
Pages125页
Degree Discipline机械电子工程
Degree Name博士
2021-05-24
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract基于程式控制的人机交互严重桎梏了对智能机器人或装备的高效操控与自主适应能力,亟需开发能够主动理解人行为意图的新型交互方式。作为一类多学科交叉融合的新兴领域,脑机接口(Brain-Computer Interface, BCI)技术的研究获得极大发展和关注。其中,非侵入、自发式BCI技术凭借着简洁便携、自然灵活、可控性强等优势已成为当前研究的焦点,尤其是以运动想象(Motor Imagery, MI)为输入信号的BCI系统已成为国内外研究的重点。目前MI-BCI系统研究存在信噪比和空间分辨率低导致识别正确率低、可识别模式少、控制自由度低、控制稳定性差等关键问题和技术挑战,极大限制了其应用范围。本文针对上述问题开展基于脑/肌电信号解码的脑机自然交互关键技术研究,主要内容如下:1. 针对MI脑电信号含有大量噪声干扰以及数据丢失等非理想情形,提出一种面向非完整MI脑电的鲁棒性识别方法。提出利用数据点剔除和数据块剔除方法,直接去除受极大干扰和数据丢失影响的信号段,构造非完整的MI脑电;提出一种基于Lomb-Scargle周期图的特征提取方法,从非完整MI中提取可分性强、鲁棒性高的功率谱密度特征;为进一步对特征重构学习和抽象表征,构建了基于受限玻尔兹曼机的深度置信网络分类模型,实现对非完整MI的鲁棒性识别。通过开展一系列不同特征提取和分类方法的对比实验,验证了所提方法的有效性。2. 针对可识别MI任务类别少的问题,开展面向同侧肢体多类MI脑电的解码方法研究。根据人体肢体运动习惯和模式,提出同侧肢体肘关节屈曲/伸展、腕关节旋前/旋后和手部握拳/展拳六类MI任务并设计了实验范式;针对同侧肢体多类MI任务空间分辨率低的问题,提出基于黎曼流形的特征提取方法,利用黎曼空间度量提取黎曼切空间特征;随后,为避免高维特征导致的过拟合问题,提出一种基于偏最小二乘的有监督特征降维方法,提升切空间特征的泛化能力;最后,分析了同侧肢体六类MI任务脑电的时间域和空间域模态,并开展了多种解码方法的对比实验。实验结果验证了本文所提方法不仅能有效扩展可辨识的MI任务数量,同时能保持较高的识别正确率。3. 针对单一MI模态的BCI系统存在个体差异性大、控制稳定性差的问题,提出引入肌电模态,并开展基于肌电的连续运动估计方法研究。设计了上肢肘关节连续运动的实验范式,同步采集了肌电信号和关节角信号;提出基于非线性自回归网络模型的肘关节连续运动估计方法,利用时域肌电特征实现对肘关节角度的精确预测;使用所提方法开展了上肢肘关节角度连续估计实验,同时与其它连续估计方法进行了对比,验证所提方法的有效性,并分析了不同通道、不同时域特征对网络模型估计性能的影响。4. 从提升MI-BCI系统的控制自由度、稳定性和可靠性等角度出发,开展基于脑/肌电信号解码的脑机交互控制实验。首先,开展了基于传统左右手MI-BCI系统的脑控虚拟小车实验,验证了MI-BCI系统的有效性;然后,搭建混合脑/肌电的脑机交互控制系统,开展了脑控机械臂完成抓取物体任务的实验验证,有效扩展MI-BCI系统的控制指令数目同时提升整体系统的控制稳定性和可靠性;最后,分析并总结了本文所提混合脑/肌电信号解码的BCI系统存在的问题以及下一步的解决方向。综上所述,本文围绕提升MI-BCI系统识别正确率、增加控制指令数、提高控制稳定性等方面开展了关键技术的深入研究,将有助于提升MI-BCI系统的实用性和应用范围,为BCI系统真正走向日常生活应用提供重要技术支撑和理论基础。
Other AbstractThe human-machine interaction manner based on programming control severely restricts the efficient manipulation and autonomous adaptability of intelligent robots or devices. It is urgent to develop a novelty interaction manner that can actively understand the intentions of human behavior. As an emerging field of multidisciplinary integration, the researches of brain-computer interface (BCI) technology have gained great development and attention. Due to its advantages of simplicity, portability, natural flexibility, and strong controllability, the non-invasive and spontaneous BCI technology has become the focus of current researches. Particularly, the BCI system using motor imagery (MI) as input signal has become the emphasis of domestic and foreign researches. At present, the MI-BCI technology still exists some problems, such as lower recognition accuracy due to the lower signal-to-noise ratio (SNR) and spatial resolution, few identifiable patterns, lower control freedom, poor control stability, which greatly limit its application scope. To solve the above problems, the key technologies of brain-machine natural interaction based on electroencephalography (EEG) and electromyography (EMG) signal decoding are studied in this dissertation. The main contents are as follows: 1. For the non-ideal situations of MI EEG signal that contain a lot of noises and data loss, a kind of robustness recognition method for the incomplete MI EEG is proposed. To construct the incomplete MI EEG dataset, the methods of data point removal and data chunk removal are separately adopted to directly eliminate the signal segments that are affected by large interference and data loss. Then, a feature extraction method based on the Lomb-Scargle periodogram is proposed to extract the power spectral density with stronger separability and higher robustness from the incomplete MI EEG. To further reconstruction learning and abstract representation of those features, a deep belief network model based on restricted Boltzmann machine is constructed to realize robust classification for the incomplete MI EEG. The effectiveness of the proposed method is verified by a series of comparative experiments with different feature extraction and classifier methods. 2. For the problem of few identifiable MI task labels, the decoding method for multiclass MI tasks from the same ipsilateral limb is studied. Based on the movement habits and patterns of human limbs, six MI tasks from the same ipsilateral limb are selected, including elbow joint flexion/extension, wrist joint pronation/supination, and hand close/open. Accordingly, the MI experimental paradigm is also designed. To solve the problem of low spatial resolution of multiclass MI tasks from the same ipsilateral limb, a feature extraction method based on Riemannian manifold is proposed to extract Riemannian tangent space features by using Riemannian space metric. Then, to avoid the over-fitting problem caused by high-dimensional tangent space features, a supervised feature reduction method based on partial least squares regression is proposed to improve the generalization ability of tangent space features. Finally, the temporal and spatial patterns of six MI tasks are analyzed, and various comparative experiments with different decoding methods are also carried out. The experimental results show that the proposed decoding method can not only effectively expand the number of identifiable MI tasks, but also maintain higher recognition accuracy. 3. To solve the problem of large individual variability and poor control stability of BCI system with a single modality of MI, the modality of EMG is proposed to be introduced, and a method of continuous motion estimation based on EMG is researched. An experimental paradigm of continuous motion of the upper limb elbow is designed. Meanwhile, EMG signals and elbow joint angles are simultaneously recorded. A method of continuous motion estimation based on Nonlinear Autoregressive with Exogenous Inputs Model (NARX) is proposed to precisely predict the angles of the elbow joint using the EMG features in the time domain. The experiments of continuous angle estimation of elbow joint using the proposed method are carried out, and other continuous estimation methods are also compared to verify the effectiveness of the proposed method. Moreover, the effects of different EMG channels and features on the estimation performance of the NARX network model are analyzed and discussed. 4. From the perspective of improving the control degree of freedom, stability, and reliability of the MI-BCI system, the experiments of brain-machine interactive control based on EEG/EMG signal decoding are carried out. Firstly, an experiment of the brain-controlled virtual vehicle based on traditional left/right hand MI-BCI system is conducted to verify the effectiveness of the MI-BCI system. Then, a novelty brain-machine interactive control system with hybrid EEG/EMG is constructed. Experiments of brain-controlled robotic arm for completing the object-grasping tasks are carried out to verify the effectiveness of the proposed hybrid EEG/EMG BCI system, which can effectively expand the number of control commands of the MI-BCI system and improve the control stability and reliability of the whole system. Finally, the problems existing in the current hybrid EEG/EMG BCI system are analyzed, as well as the solution directions in the next step are summarized. To sum up, focusing on the aspects of improving the recognition accuracy, increasing the number of the control command, and improving the control stability of the MI-BCI system, the key technologies of the BCI system are researched in this dissertation, which will help to improve the practicability and application range of the MI-BCI system. The researches of this dissertation will provide important technical support and theoretical basis for the real application of the BCI system in daily life.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/29009
Collection机器人学研究室
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
褚亚奇. 基于EEG EMG信号解码的脑机自然交互关键技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2021.
Files in This Item:
File Name/Size DocType Version Access License
基于EEG EMG信号解码的脑机自然交互(8612KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[褚亚奇]'s Articles
Baidu academic
Similar articles in Baidu academic
[褚亚奇]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[褚亚奇]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.