独立分量分析及其在脑电信号噪声分离中的应用

【作者】 李婷；

【导师】 邱天爽；

【作者基本信息】 大连理工大学 ， 信号与信息处理， 2005， 硕士

【摘要】 脑电是大脑神经元突触后电位的综合,是大脑电活动产生的电场经容积导体(由皮层、颅骨、脑膜及头皮构成)传导后在头皮上的电位分布,分为自发脑电(electroencephalograph,EEG)和诱发电位(evoked potential,EP)两种。脑电在临床诊断、军事医学、航天医学、生理学和生物学研究中都具有重要的意义,所以脑电信号的提取一直是神经科学领域的重要课题。 独立分量分析(ICA)方法是最近几年发展起来的一种新的统计方法。ICA方法是基于信号高阶统计特性的分析方法,经ICA方法分解出的各信号分量之间是相互独立的。正是因为这一特点,使ICA在信号处理领域受到了广泛的关注。大量的实验已表明,ICA方法能精确地从脑电信号中分辨出具有相对较大的瞬时独立分量的时间过程。近年来,ICA方法的发展十分迅速,国内外的众多研究人员都致力于研究新的算法,应用于脑电信号的噪声分离之中。 本文提出了一种新的基于带参考信号的ICA算法的脑电信号眨眼伪差的分离方法,可以得到纯净的脑电信号。这个方法的主要思路是:先选取一导眨眼伪差比较明显的数据,从中获得眨眼伪差的参考信号,再用ICA方法把眨眼伪差第一个提取出来,最后得到消除伪差后的EEG信号。本文详细讨论了使用带参考信号的ICA算法消除眨眼伪差的方法与步骤,并给出了应用于仿真信号和真实信号的实验结果。带参考信号的ICA算法具有fastICA算法的特点,可以一次只分离出一个独立分量,收敛速度快;不需要选择学习步长或其它参数;而且,不管是具有正峭度还是负峭度的分量它都能提取出来。它分解出的眨眼伪差以一个独立分量的形式出现,并且被第一个分离出来,不需判断就可直接重构脑电信号,减少了计算量,而且纯净脑电信号的重构也比较简单。 ICA的两类基本算法各有所长:基于非高斯性测度的算法收敛快,但不能在线学习;基于信息论的算法可以在线学习,但收敛较慢。本文把这两类算法的优点结合起来,提出了一种新的基于独立分量分析的诱发电位的快速估计方法。这种新方法是将两种已有的独立分量分析算法——改进的Infomax算法和fastICA算法——结合起来得到的。先用改进的Infomax算法获得权值的初值,再通过fasflCA算法获得最终结果。这不仅减少了fastICA算法所需数据的长度,也提高了fastICA算法的收敛稳定性。本文的计算机仿真实验也证明了这个新算法的有效性,它比其他两种算法需要的数据点数更少,收敛速度更快、更稳定。将此算法应用于EP信号的提取,可以有效地去除噪声,得到比较纯净的EP信号。更多还原

【Abstract】 Brain signals are bioelectrical signals of the human brain, which are classified into two categories: electroencephalograph (EEG) and evoked potential (EP). They have a number of clinical applications including the diagnosis of a variety of neurological disorders, physiological analysis and critical care and operating room monitoring. So, the extraction of EEG and EP is an important project in neuroscience.Independent component analysis (ICA) is a technique which extracts statistically independent components from a set of measured signals. It is an analytical method based on higher-order statistical characters of signals, and is widely used in signal processing. Recently, ICA develops very quickly, and more and more researchers are devoted into studying new algorithms and applying them to noise reduction of EEG and EP.Based on ICA algorithm with reference signals, a method removing blinking artifacts is proposed in this thesis. The main idea of it is: first select one channel of EEG with obvious blinking artifacts, and obtain reference signals from it. Then extract the blinking artifact first with ICA algorithm, and at last get pure EEG signals. The idea and steps of the ICA algorithm with reference signals are thoroughly discussed, and the results of processing real signals are also proposed in the thesis. The advantages of the new algorithm are: it has all the advantages of fastICA algorithm, and it can separate the artifact in one independent component; it also makes the reconstruction of EEG easier than other algorithms.Both kinds of ICA algorithms have advantages and disadvantages. Infomax algorithm can learn on line, but it converges slowly. FastICA algorithm converges very quickly, but it cannot learn on line. It also needs prewhitening and large size of data, and it is sensitive to the initial value of the demixing matrix. This thesis proposes a new algorithm that combines two existent algorithms, the improved infomax algorithm and the fastICA algorithm. Utilizing the initial weights obtained by the improved infomax algorithm, we can not only reduce the length of data which fastICA algorithm needs, but also enhance the convergence stability of fastICA algorithm. The effectiveness of the algorithm is verified by computer simulations. This new algorithm is applied in noise reduction of EP, and realizes fast estimation of EP.更多还原