盲信号分离在脑电信号伪迹去除中的应用
发布时间：2018-08-21 09:48
【摘要】：脑-机接口技术的核心思想在于将输入的观测脑电信号转换为输出的控制信号,从而驱动计算机设备。通过受试者头皮电极采集到的脑电(EEG,Electroencephalogram)非常微弱,并且伴随多种伪迹(Artifact)的干扰,给脑电信号的特征提取和后续分析增加了更大的难度。盲信号分离(BSS,Blind Source Separation)是在通信系统的输入和传输信道均未知的情况下提出来的,即对源信号的先验知识少知或不知,对传输信道特性也未知。本课题针对脑电信号处理中的问题,对基于BSS思想的自动去除EEG中伪迹的方法展开了研究。本文首先对脑电信号伪迹分离的研究背景和国内外研究状况做了介绍,然后学习了脑电信号的基本知识,详细阐述了脑电信号与伪迹信号的特性与分类,研究中着重考虑对脑电信号影响最严重的眼电伪迹和50Hz的工频干扰。其次介绍了盲信号分离的核心思想,其用于解决脑电信号伪迹分离问题时的数学模型、约束条件和预处理过程,还深入学习了盲信号分离的经典算法(JADE,FastICA)。在分析了传统算法局限性的基础之上,进一步寻求了脑电信号领域的一种全新的解决问题思路,本文首次尝试将Stone's BSS算法引入EEG信号处理领域,为脑电信号的伪迹去除引入了新方法。Stone's BSS突破了以往信号处理方法中要求源信号不能服从高斯分布和相互独立的局限性,只要求混合信号是时间可预测的,分别采用长、短滤波预测对混合信号作用,将BSS问题转变成一个广义特征分解问题,从而求得解混矩阵。文中还对Stone's BSS进行了改进,引入遗传算法用于对长、短滤波调谐,使之成为一种成熟稳定的算法。紧接着选取了一组具有代表性的模拟信号对改进的Stone's BSS与其他BSS方法的分离结果做了对比,理论上证明了改进算法在高斯型、亚高斯型信号分离中的良好性能。最后,结合目标信号的特征和性质,通过不同的实际数据对改进的Stone's BSS算法在EEG中的眨眼伪迹(EOG,Electrooculogram)和工频干扰的分离展开了大量实验研究,选用恰当的评价指标对结果进行分析判断,结果表明改进算法和传统算法都能够完成EEG中的伪迹分离,改进的Stone's BSS算法表现出更好地性能。本研究的工作也为Stone算法在生物信号领域的应用奠定了重要基础。结尾概括全文给出结论,并指出课题下一步可继续拓展深入研究的方向。
[Abstract]:The core idea of brain-computer interface technology is to convert the input observation EEG signal into the output control signal, so as to drive the computer equipment. The EEG electroencephalogram (EEG) collected by the scalp electrode is very weak, and accompanied by the interference of many artifacts (Artifact), it is more difficult to extract and analyze the EEG features. Blind signal separation (BSS) Blind Source Separation) is proposed when the input and transmission channels of the communication system are unknown, that is, the prior knowledge of the source signal is little or unknown, and the characteristics of the transmission channel are also unknown. In order to solve the problem of EEG signal processing, the method of automatically removing artifacts in EEG based on BSS is studied in this paper. In this paper, the background of EEG artifact separation and the research status at home and abroad are introduced, then the basic knowledge of EEG is studied, and the characteristics and classification of EEG and artifact are described in detail. Eye-electric artifacts and power frequency interference of 50Hz, which have the most serious effect on EEG, are considered in this study. Secondly, the core idea of blind signal separation is introduced, which is used to solve the problem of EEG artifact separation. The mathematical model, constraint conditions and pretreatment process are also introduced. The classical algorithm of blind signal separation (JADEN FastICA) is also studied in depth. On the basis of analyzing the limitation of the traditional algorithm, a new way of solving the problem in the field of EEG signal is further sought. In this paper, the Stone's BSS algorithm is introduced into the field of EEG signal processing for the first time. In this paper, a new method for removing artifacts of EEG signals is introduced. Stonews BSS breaks through the limitation of the previous signal processing methods that the source signals cannot be distributed from Gao Si and independent of each other, and only requires that the mixed signals be predictable in time and use long time, respectively. The BSS problem is transformed into a generalized eigenvalue decomposition problem by short filter prediction to the mixed signal, and the unmixing matrix is obtained. In this paper, Stone's BSS is improved, and genetic algorithm is introduced to tune long and short filter, which makes it a mature and stable algorithm. Then a group of representative analog signals are selected to compare the results of the improved Stone's BSS and other BSS methods. It is proved theoretically that the improved algorithm has good performance in the separation of Gao Si type and subGao Si type signals. Finally, according to the characteristics and properties of the target signal, a large number of experiments on the separation of the improved Stone's BSS algorithm in EEG and the power frequency interference are carried out through different actual data. The results show that both the improved algorithm and the traditional algorithm can separate artifacts in EEG, and the improved Stone's BSS algorithm shows better performance. The work of this study also lays an important foundation for the application of Stone algorithm in the field of biological signals. At the end of this paper, the conclusion is given, and the further research direction is pointed out.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7;R318
本文编号：2195330
[Abstract]:The core idea of brain-computer interface technology is to convert the input observation EEG signal into the output control signal, so as to drive the computer equipment. The EEG electroencephalogram (EEG) collected by the scalp electrode is very weak, and accompanied by the interference of many artifacts (Artifact), it is more difficult to extract and analyze the EEG features. Blind signal separation (BSS) Blind Source Separation) is proposed when the input and transmission channels of the communication system are unknown, that is, the prior knowledge of the source signal is little or unknown, and the characteristics of the transmission channel are also unknown. In order to solve the problem of EEG signal processing, the method of automatically removing artifacts in EEG based on BSS is studied in this paper. In this paper, the background of EEG artifact separation and the research status at home and abroad are introduced, then the basic knowledge of EEG is studied, and the characteristics and classification of EEG and artifact are described in detail. Eye-electric artifacts and power frequency interference of 50Hz, which have the most serious effect on EEG, are considered in this study. Secondly, the core idea of blind signal separation is introduced, which is used to solve the problem of EEG artifact separation. The mathematical model, constraint conditions and pretreatment process are also introduced. The classical algorithm of blind signal separation (JADEN FastICA) is also studied in depth. On the basis of analyzing the limitation of the traditional algorithm, a new way of solving the problem in the field of EEG signal is further sought. In this paper, the Stone's BSS algorithm is introduced into the field of EEG signal processing for the first time. In this paper, a new method for removing artifacts of EEG signals is introduced. Stonews BSS breaks through the limitation of the previous signal processing methods that the source signals cannot be distributed from Gao Si and independent of each other, and only requires that the mixed signals be predictable in time and use long time, respectively. The BSS problem is transformed into a generalized eigenvalue decomposition problem by short filter prediction to the mixed signal, and the unmixing matrix is obtained. In this paper, Stone's BSS is improved, and genetic algorithm is introduced to tune long and short filter, which makes it a mature and stable algorithm. Then a group of representative analog signals are selected to compare the results of the improved Stone's BSS and other BSS methods. It is proved theoretically that the improved algorithm has good performance in the separation of Gao Si type and subGao Si type signals. Finally, according to the characteristics and properties of the target signal, a large number of experiments on the separation of the improved Stone's BSS algorithm in EEG and the power frequency interference are carried out through different actual data. The results show that both the improved algorithm and the traditional algorithm can separate artifacts in EEG, and the improved Stone's BSS algorithm shows better performance. The work of this study also lays an important foundation for the application of Stone algorithm in the field of biological signals. At the end of this paper, the conclusion is given, and the further research direction is pointed out.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7;R318
【参考文献】
相关期刊论文 前9条
1 杜晓燕;李颖洁;朱贻盛;任秋实;赵仑;;脑电信号伪迹去除的研究进展[J];生物医学工程学杂志;2008年02期
2 杨帮华,颜国正,丁国清,于莲芝;脑机接口关键技术研究[J];北京生物医学工程;2005年04期
3 杨立才,李佰敏,李光林,贾磊;脑-机接口技术综述[J];电子学报;2005年07期
4 李小军,朱孝龙,张贤达;盲信号分离研究分类与展望[J];西安电子科技大学学报;2004年03期
5 杨竹青,李勇,胡德文;独立成分分析方法综述[J];自动化学报;2002年05期
6 刘琚,何振亚;盲源分离和盲反卷积[J];电子学报;2002年04期
7 张贤达,保铮;盲信号分离[J];电子学报;2001年S1期
8 戴晓晖,李敏强,寇纪淞;遗传算法的性能分析研究[J];软件学报;2001年05期
9 钱志勤,滕弘飞,孙治国;人机交互的遗传算法及其在约束布局优化中的应用[J];计算机学报;2001年05期
，本文编号：2195330
本文链接：https://www.wllwen.com/kejilunwen/wltx/2195330.html
