基于MMTD的脑电信号去噪与特征提取方法研究

发布时间：2018-03-07 22:33

本文选题：脑电信号　切入点：中介真值程度度量　出处：《南京邮电大学》2017年硕士论文　论文类型：学位论文

【摘要】：脑电信号(Electroencephalogram,EEG)中含有能够客观地反映出人近期生理和心理状况的信息,通过对EEG信号进行研究和分析,可以获得大量有用的医学、生理和心理信息,对疾病的治疗和检测具有重要的意义。近年来,脑-机接口(Brain-Computer Interface,BCI)系统已经成为脑科学领域研究的热点,EEG信号与外部设备之间的传输是BCI研究的关键。因此,需要借助于EEG处理的方法提取出有效的信号达到人机交互的目的。然而,EEG是非平稳非线性的信号且易受噪声的干扰,给信号的去噪以及特征提取带来极大的困扰。尽管目前有许多EEG去噪以及特征提取的方法,但在精度或者效率方面都存在一定问题。针对上述问题,本文提出基于MMTD与小波阈值相结合的EEG信号去噪方法、基于EMD分解EEG的特征提取方法。本文完成的主要工作如下:(1)针对EEG采集时易受到噪声干扰的问题与小波阈值算法在EEG滤波中存在的不足,文中提出了基于MMTD与小波硬阈值的EEG信号去噪方法与基于MMTD与小波软阈值的EEG信号去噪方法,实验表明,文中提出的两种EEG信号去噪方法都是有效的。(2)针对EEG信号的产生机理非常复杂,可以直接进行分类识别的信号是很难获取的问题。本文在基于EMD信号分解的基础上,依据EEG中的ERS/ERD现象,提出把IMF分量能量相对偏离度作为EEG信号特征属性,并将IMF分量的相对能量偏离度、IMF分量的能量、IMF分量的相对能量组合成特征向量用于后续的EEG分类。(3)由于不同的分类器其泛化能力有一定的差异,因此分类器的选取对EEG信号分类识别率有较大影响,本文在基于上述选取的三种信号特征的基础上,提出基于SVM的EEG分类方法对EEG信号进行分类。实验表明,文中提出的特征提取方法是有效的。
[Abstract]:Electroencephalograms (EGG) contains information that can objectively reflect the recent physiological and psychological conditions of human beings. By studying and analyzing EEG signals, a large number of useful medical, physiological and psychological information can be obtained. In recent years, the Brain-Computer Interface (BCI) system has become a hot topic in the field of brain science. It is necessary to extract effective signals by means of EEG processing to achieve the purpose of human-computer interaction. Although there are many EEG denoising and feature extraction methods, there are some problems in accuracy or efficiency. In this paper, a EEG signal denoising method based on the combination of MMTD and wavelet threshold is proposed. The main work of this paper is as follows: 1) aiming at the problem of easy to be disturbed by noise in EEG acquisition and the deficiency of wavelet threshold algorithm in EEG filtering, the main work of this paper is as follows: (1) the method of feature extraction based on EMD decomposition EEG. In this paper, EEG signal denoising method based on MMTD and wavelet hard threshold and EEG signal denoising method based on MMTD and wavelet soft threshold are proposed. The two EEG signal denoising methods proposed in this paper are effective. (2) the mechanism of EEG signal generation is very complex, and it is difficult to obtain the signal that can be directly classified and recognized. This paper is based on the decomposition of EMD signal. According to the ERS/ERD phenomenon in EEG, the relative deviation of IMF component energy is proposed as the characteristic attribute of EEG signal. The relative energy deviation of the IMF component and the relative energy of the IMF component are combined into eigenvector to be used in the subsequent EEG classification. Therefore, the selection of classifier has a great influence on the classification and recognition rate of EEG signals. Based on the above selected three signal features, this paper proposes a EEG classification method based on SVM to classify EEG signals. The proposed feature extraction method is effective.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R741.044;TN911.7

【参考文献】