基于运动想象脑电信号的机械臂控制系统研究

发布时间：2018-04-24 01:04

本文选题：脑机接口 + 小波去噪　；参考：《电子科技大学》2017年硕士论文

【摘要】：脑机接口是神经科学与机器学习结合的应用,能适应某些特殊环境下的控制需求,尤其是在传统的控制方式无法满足现实需要的条件下,脑机接口有着独特的优势,存在着广泛的应用前景。本文主要研究了基于脑电信号设计的脑机接口,重点在于提取脑电信号中与大脑运动想象相关的模式,并对大脑不同活动状态下的模式特征进行识别、挖掘和理解,最后与机械臂控制平台进行了结合。本文对基于脑电信号的脑机接口的基础理论框架进行了介绍,包括脑电信号的产生的机理、脑电信号的采集过程、运动想象脑电信号所表现出来的独特的ERS/ERD特征模式和脑电信号的预处理过程。针对包含大量的干扰且整体杂乱的脑电信号设计了一套滤波算法来提取特定频段的特异性脑电信号,又利用了基于小波变换的去噪算法对脑电信号进行平滑处理。对于脑电信号的特征提取,介绍了共空间模式算法,并在此基础上,将只适用于二分类任务的共空间模式算法通过“一对多”的思想推广到了脑电信号的多分类任务。同时考虑到序列脑电信号的非平稳性和随机性,设计了可自主调整参数的自适应共空间模式算法。在脑电信号的特征分类问题研究方面,针对采集脑电信号的二分类问题,分析了经典的线性判别式分析,同时考虑到不同类人群的脑电信号会存在差异的实际情况,对线性判别式算法做了改进,提出了具有自适应功能的线性判别式算法;在脑电信号的多分类问题上,针对高维且线性不可分的脑电信号分类问题,结合适用于脑电信号多分类任务的共空间模式算法,首次引入稀疏神经网络来对脑电信号进行二次特征提取并进行多分类任务,稀疏特征使得原始线性不可分的脑电信号变得线性可分,同时也有利于对脑电信号的特征进行深度挖掘。最后,本文给出了脑电控制系统的设计方案,对实验平台二连杆机械臂进行了建模分析和控制器设计,并通过机械臂平台实现了脑机接口识别的大脑运动想象任务。
[Abstract]:BCI is a combination of neuroscience and machine learning, which can meet the control needs of some special environment, especially under the condition that the traditional control mode can not meet the actual needs, the BCI has a unique advantage. There is a wide application prospect. In this paper, the brain-computer interface based on EEG is mainly studied. The emphasis is to extract the patterns related to the brain motion imagination, and to recognize, mine and understand the pattern features of the brain in different active states. Finally, it is combined with the manipulator control platform. In this paper, the basic theoretical framework of EEG based BCI is introduced, including the mechanism of EEG generation, the process of EEG acquisition, The characteristic mode of ERS/ERD and the preprocessing process of EEG signal are shown in motion imagination EEG signal. A set of filtering algorithm is designed to extract the specific EEG signal in a specific frequency band for the EEG signal which contains a lot of interference and the whole clutter. The denoising algorithm based on wavelet transform is used to smooth the EEG signal. For the feature extraction of EEG signals, the common space pattern algorithm is introduced, and on this basis, the common space pattern algorithm, which is only suitable for two classification tasks, is extended to the multi-classification tasks of EEG signals by the idea of "one-to-many". Considering the nonstationarity and randomness of sequential EEG signals, an adaptive common space mode algorithm is designed, which can adjust the parameters autonomously. In the aspect of feature classification of EEG signals, the classical linear discriminant analysis is analyzed in view of the two classification problems of EEG signals collected, and the actual situation that EEG signals of different groups of people will be different is taken into account. The linear discriminant algorithm is improved, and an adaptive linear discriminant algorithm is proposed. Combined with the common spatial pattern algorithm which is suitable for EEG multi-classification tasks, sparse neural network is introduced for the first time to extract the secondary features of EEG signals and carry out multi-classification tasks. The sparse feature makes the original linear indistinguishable EEG signal linearly separable, and it is also conducive to the in-depth mining of the EEG features. Finally, the design scheme of EEG control system is given, the modeling analysis and controller design of the two-link manipulator are carried out, and the brain-computer interface recognition task of brain motion imagination is realized through the manipulator platform.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP241;R318

【相似文献】