基于运动想象的脑电信号分类算法与脑-机接口技术研究

发布时间：2018-03-27 19:34

本文选题：脑电图　切入点：脑-机接口　出处：《吉林大学》2017年硕士论文

【摘要】：脑-机接口作为一种新的人机交互途径逐渐引起人们的关注,基于运动想象BCI系统作为其中的一大分支也逐渐引起大家的重视,但人的脑电信号(Electroencephalogram,EEG)复杂且具有非线性非平稳等特点,这使其难以辨识和解析,其识别效果也依赖于数据集的不同,而表现不稳定。为了提高运动想象EEG信号的识别率,提供相对有效的特征提取及分类方案,本文基于脑电信号这样的特点,介绍和研究了两种基于二分类的运动想象脑电信号的处理、特征提取及分类方法,并开发了一套原理样机。1.利用总体经验模态分解(Ensemble empirical mode decomposition,EEMD)后得到的较具影响能力的固有模态函数(Intrinsic mode functions,IMFs),利用希尔伯特变换提取边际谱(Marginal spectrum,MS)及瞬时能谱(Instantaneous energy spectrum,IES)时频特征,同时通过加窗的方法提取非线性动力学特征近似熵特征,利用线性判别分类器(Linear Discriminant Analysis,LDA)作为分类器,实验对象为来自三个不同数据集的9个不同受试者,其取得平均分类识别率为82.74%,具有良好普适性;2.利用固有时间尺度分解(Intrinsic time-scale decomposition,ITD)后得到的固有旋转分量,选择较优分量信号,进而提取四类有效特征包括能量,AR参数模型,数学形态学特征以及模糊近似熵,采用单因素方差检验作为特征选择,利用LDA作为分类器进行分类,采用互信息(MI)及最大峭度的评价标准。实验结果证明提出基于ITD的识别系统能取得较为可观的识别率,结果优于BCI竞赛成绩及近期文献成果。在03年脑-机接口大赛数据中取得0.75的互信息,05年大赛数据中取得0.3699的平均最大峭度。且对基于ITD的特征提取方案对脑电信号处理和识别进行了解释和分析,该方法对提高脑-机接口系统的分类性能及在线BCI系统的实现,能提供良好的理论基础和应用参考;3.开发的一套基于ThinkGear脑电仪意念控制手臂的在线BCI系统。通过采集受试者前额处脑电信号,获取当前受试者视觉阻断信息,使其作为控制指令,最终可以使假肢手臂完成简单的任务动作。该系统的开发可为BCI技术的将来的发展及应用提供工作基础及方向参考。
[Abstract]:As a new way of human-computer interaction, brain-computer interface (BCI) has attracted more and more attention, and BCI system based on motion imagination has attracted more and more attention. But the EEG electroencephalogramme (EGG) is complex and nonlinear and non-stationary, which makes it difficult to identify and analyze, and its recognition effect depends on the difference of data set, and its performance is unstable. In order to improve the recognition rate of motion imagination EEG signal, In this paper, based on the characteristics of EEG, two methods of processing, feature extraction and classification of motion imagination EEG based on two-classification are introduced and studied. A set of principle prototype is developed. 1. The inherent mode function of intrinsic mode function is obtained by means of total empirical mode decomposition (EMD), and the time-frequency features of marginal spectrum and instantaneous energy spectrum energy spectrum are extracted by Hilbert transform. At the same time, the approximate entropy feature of nonlinear dynamics is extracted by adding windows, and linear discriminant classifier linear Discriminant analysis is used as the classifier. Nine different subjects from three different data sets are used as experimental objects. Its average classification recognition rate is 82.74, and it has good universality. By using the inherent rotation component obtained by decomposing intrinsic time-scale decompostion (ITD), the better component signal is selected and four effective features, including the energy AR parameter model, are extracted. Mathematical morphological features and fuzzy approximate entropy are classified by single factor variance test and LDA as classifier. The experimental results show that the recognition system based on ITD can achieve considerable recognition rate. The results were superior to those of BCI competition and recent literature. Mutual information of 0.75 was obtained in the 2003 brain-computer interface contest data, and the average maximum kurtosis was 0.3699 in the 2005 contest data. The feature extraction scheme based on ITD was applied to EEG. Signal processing and recognition are interpreted and analyzed, This method can improve the classification performance of brain-computer interface system and the realization of on-line BCI system. It can provide a good theoretical basis and application reference. 3. A set of online BCI system based on the idea control arm of ThinkGear EEG instrument is developed. By collecting the EEG signals in the forehead of the subjects, we can obtain the visual blocking information of the current subjects. The development of the system can provide the working basis and direction reference for the future development and application of BCI technology.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【相似文献】