运动想象脑电信号的特征提取算法研究

发布时间：2018-01-09 05:20

本文关键词：运动想象脑电信号的特征提取算法研究　出处：《安徽大学》2014年硕士论文　论文类型：学位论文

【摘要】：脑-机接口(Brain-computer interface, BCI)是一种不依赖外周神经和肌肉等传统信息通道的特殊人-机交互技术。利用该技术,可实现大脑与外部设备之间的直接通信和控制。作为神经活动的信息载体,头皮脑电(EEG)信号能实时反映思维状态的变化,并且容易检测,因此在非植入式BCI系统中得到了广泛应用。然而由于大脑容积传导效应的存在,使得头皮脑电的空间分辨率较低。同时,非神经活动伪迹(如眼电、肌电、心电等)和环境噪声也大大降低了有用信息的信噪比。因此,在基于EEG的BCI系统实现研究中,如何从多道头皮脑电中获取思维相关的真实神经活动成分是非常关键的技术环节。本文围绕运动想象BCI系统的实现,对EEG信号处理和特征提取新方法开展研究,主要做了以下工作： (1)设计了运动想象BCI的实验范式,并采集了较丰富的运行想象EEG数据,为后续研究打下了良好的基础。 (2)针对任务相关EEG节律波的包络检测和运动想象分类问题,实现了四种包络检测方法：非线性能量算子(Nonlinear energy operator, NEO)、希尔伯特变换(Hilbert transform, HT)和两种滑动窗独立分量分析(Independent component analysis, ICA)算法。基于BCI2003竞赛数据,对四种包络检测算法在运动想象分类中的应用效果进行了分析和比较。研究了干扰伪迹对包络检测精度的影响,并提出了相应的改进思路。 (3)研究了结合时、频、空域的空域滤波新方法。ICA和共同空间模式(Common spatial pattern, CSP)是两种重要的空域滤波算法。两种算法都是提取空域滤波器后对预处理后的脑电信号进行滤波,得到与神经活动相关的隐含信号源。由于滤波器的设计原理的不同,最终所得隐含源的物理意义差别也很大。本文首先基于实测运动想象EEG数据,分析和比较两种空域滤波方法各自的性能特点。在此基础上,给出了一种结合ICA和CSP的EEG特征提取新方法,实验结果验证了所提方法的有效性。
[Abstract]:Brain computer interface (Brain-computer interface BCI) is a special one - a traditional information channel is not dependent on the peripheral nerves and muscles of the machine interaction technology. Using this technology, can realize direct communication and control between brain and external devices. As the information carrier of neural activity, electroencephalography (EEG) signals the changes reflect the real state of mind, and can easily be detected, so it is widely used in non-invasive BCI system. However, the brain volume conduction effect, the spatial resolution of scalp EEG is low. At the same time, non neural activity artifacts (such as EOG, EMG, ECG etc) and environmental noise greatly reduce the useful information of the signal-to-noise ratio. Therefore, in the BCI implementation of EEG system based on the research, thinking how to get real neural activity in the correlated components from multichannel scalp EEG is very important in technology.
This paper focuses on the implementation of the motion picture BCI system, and studies the new methods of EEG signal processing and feature extraction. The following work is done:
(1) the experimental paradigm of motion imaginary BCI was designed, and the more abundant EEG data were collected, which laid a good foundation for the follow-up study.
(2) according to the envelope detection and motion tasks related to EEG rhythm wave imagery classification problem, realized four kinds of envelope detection methods: nonlinear energy operator (Nonlinear energy, operator, NEO), Hilbert (Hilbert transform, HT transform) and two kinds of sliding window independent component analysis (Independent component analysis, ICA BCI2003) algorithm. The competition based on the data of four kinds of envelope detection algorithms are analyzed and compared in the effect of motor imagery classification. Research on interference artifact effect on envelope detection accuracy, and proposes the corresponding improvement ideas.
(3) the combination of frequency,.ICA, a new method of spatial filtering and spatial common spatial pattern (Common spatial, pattern, CSP) are two important spatial filtering algorithm. The two algorithms are extracted from the spatial filter to filter the EEG signal preprocessing, associated with neural activity implied signal source. Due to the design principle of the filter, is also a great physical meaning resulting implicit difference. Source based on the EEG data measured imagine movement, performance analysis and comparison of two kinds of spatial filtering methods respectively. Based on this, a new method of combining ICA and CSP EEG feature extraction is presented, experimental the results verify the effectiveness of the proposed method.

【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【参考文献】