基于运动想象的脑机接口相关算法研究

发布时间：2018-05-04 21:09

本文选题：脑机接口 + 运动想象　；参考：《山东大学》2014年博士论文

【摘要】：探索大脑的奥秘是21世纪自然科学研究的重大挑战。人脑是人体中最复杂的组织结构,是中枢神经系统的最高级部分。脑部神经元电活动产生的脑电信号能够反映大脑不同状态的信息,因此,对脑电信号的研究是脑科学研究领域的重要组成部分。脑机接口(Brain Computer Interface, BCI)系统通过对脑电信号的分析和处理,提供用户与外界设备通信和控制的信道,是一种新的人机交互方式。BCI系统涉及计算机通信与控制、生物医学工程和康复医学等领域,已经成为交叉学科的热点课题。基于运动想象的BCI系统主要是将运动想象激发大脑运动皮层脑电节律变化的脑电信号作为输入,通过信号处理部分判断运动想象种类,然后由计算机将运动想象种类翻译成控制命令,最终可以实现人脑与外部设备的通信及控制功能。信号处理作为基于运动想象的BCI系统的核心部分,主要包括预处理、信道选择、特征提取和分类识别等主要步骤。基于运动想象的BCI系统的性能关键在于对感知运动节律变化特征的准确提取和对运动想象任务的正确分类。然而,由于脑电信号是一种微弱信号,容易受到干扰,具有低信噪比、动态性、瞬时性和非平稳性等特点,使得基于运动想象的BCI系统的发展和应用面临严峻的挑战。如何有效地提取脑电信号特征,以及匹配最佳分类器是基于运动想象的BCI系统信号处理部分研究的重点。本论文立足于脑电信号的预处理、信道选择、特征提取和分类识别等四个方面开展算法研究。从时域、频域、空域及非线性动力学领域对脑电信号进行分析和处理,提取脑电特征并匹配最佳分类器,同时设计信道选择算法以降低算法复杂度。本论文提出了几种有效的基于运动想象的BCI系统特征提取和分类识别算法。采用国际标准的BCI竞赛数据库验证算法有效性,本论文主要包括以下几项贡献和创新点： 1.在时频域对脑电信号进行分析和处理,提出了一种基于改进S变换的二类运动想象任务的特征提取和分类识别算法。该算法将改进S变换引入脑皮层电图(Electrocorticography,ECoG)信号的特征提取中,通过优化两个自适应参数来调节可变窗口尺寸大小,从而得到一个最优的频率独立窗口,可以准确定位包含感知运动节律变化的脑电信号的时频信息,通过计算经过改进S变换之后的ECoG信号的功率谱密度,最终得到脑电信号时频范围内的局部功率表示。与S变换相比,改进S变换具备更好的能量集中性；与其他时频方法相比,基于改进S变换的特征可以获得更好的时频分布表示和更高分辨率的频谱密度函数；与其他分类器相比,将改进S变换提取的特征与基于普通最小二乘回归的梯度Boosting分类器相结合,可以得到最好的分类效果；设计的信道选择算法能够大幅度降低算法复杂度,有效提升BCI系统性能。实验结果证明,所提出的算法可以得到较好的分类效果。 2.在时空域对脑电信号进行分析和处理,提出了一种基于局部二值模式和自回归模型相结合的特征提取和分类识别算法。该算法将广泛应用于图像纹理分析的局部二值模式算子和自回归模型构成的组合特征应用到一维脑电信号的分析中,并结合梯度Boosting分类器,对基于ECoG的运动想象任务进行分类。利用旋转不变的局部二值模式算子的直方图分布和Burg算法所估计的二阶自回归模型系数构成组合特征,在多个量化的角度和多分辨率基础上分析脑电信号,实现从时域和空域对脑电信号感知运动节律变化的描述,反映脑电信号在运动想象中感知运动节律变化。该算法采用国际标准BCI竞赛数据库进行算法验证,实验结果证明,与其他几种特征相比,该算法中的组合特征具备更高的分类准确率,可以更好地描述基于运动想象的脑电信号；与其他分类器相比,将组合特征与基于普通最小二乘回归的梯度Boosting分类器相结合,可以得到最好的分类效果。由于组合特征会导致特征向量维数的增加,因此,设计信道选择算法来降低输入的特征维数。实验结果证明该算法在保证分类准确率的同时有效地降低了算法复杂度。 3.以非线性动力学为基础,将分形几何理论应用到脑电信号的特征提取中,提出了一种分形特征和局部二值模式相结合的特征提取和分类识别算法。该算法引入广泛应用于灰度图像计算的毯子维覆盖技术,通过计算脑电信号中不同覆盖层的毯子维,得到相应的分形截距和缺项。该算法将分形截距、缺项和空域提取的局部二值模式算子进行组合来描述ECoG信号。该组合特征在多分辨率和多角度条件下分析脑电信号,可以度量脑电信号的复杂度,同时反映其幅度变化快慢。与其他特征相比,该组合特征可以得到更好的分类效果,能够更加完整和准确地定位脑电信号中运动想象节律变化信息。信道选择算法有效缓和组合特征导致的特征向量维数增加问题,从而降低运算量。实验结果证明,该算法可以得到理想的分类效果,同时能够获取分类准确率和算法复杂度之间更好地折中。本文的研究工作有助于进一步推动基于运动想象的BCI系统在技术理论、算法和实际应用中的研究。对于脑电信号的时域、频域、空域和非线性动力学分析在基于运动想象的BCI系统中的应用起到了积极的推进作用。
[Abstract]:The exploration of the mysteries of the brain is a major challenge in the study of Natural Science in the twenty-first Century. The human brain is the most complex structure in the human body and the most advanced part of the central nervous system. The EEG signals produced by the electrical activity of the brain neurons reflect the information of different states of the brain. Therefore, the study of brain electrical signals is an important part of the field of brain science. Brain Computer Interface (BCI) system provides channels for communication and control between users and external devices through the analysis and processing of EEG signals. It is a new human-computer interaction mode which has become a cross discipline in the fields of computer communication and control, biomedical engineering and rehabilitation medicine, and so on. Hot topics.
The BCI system based on motion imagination mainly uses the motion imagination to stimulate the EEG signals in the brain's motor cortex, which can be used as input to judge the kind of motion imagination through the signal processing section, and then translates the kind of motion imagination into control commands by the computer, and can finally realize the communication and control functions of the human brain and the external equipment. Signal processing is the core part of the BCI system based on motion imagination, mainly including preprocessing, channel selection, feature extraction and classification recognition. The performance key of BCI system based on motion imagination lies in the accurate extraction of the variation characteristics of perceptual motion rhythm and the correct classification of the task of motion imagination. However, because of the brain The electrical signal is a weak signal, easy to be disturbed, with low signal to noise ratio, dynamic, instantaneous and non-stationary, which makes the development and application of the BCI system based on motion imagination face severe challenges. How to effectively extract the features of the brain signal and match the best classifier is based on the BCI system signal of motion imagination The focus of the research.
This thesis is based on four aspects of EEG signal preprocessing, channel selection, feature extraction and classification recognition. The EEG signals are analyzed and processed in the domain of time domain, frequency domain, space domain and nonlinear dynamics, the EEG features are extracted and the best sorter is matched, and the channel selection algorithm is designed to reduce the complexity of the algorithm. In this paper, several effective algorithm for feature extraction and classification of BCI system based on motion imagination are proposed. The validity of the algorithm is verified by the international standard BCI competition database. The main contributions and innovations of this paper are as follows:
1. in the analysis and processing of the EEG in time and frequency domain, a feature extraction and classification recognition algorithm for two kinds of motion imaginary tasks based on improved S transform is proposed. This algorithm introduces the improved S transform to the feature extraction of Electrocorticography (ECoG) signal and adjusts the variable window by optimizing the two adaptive parameters. In order to obtain an optimal frequency independent window, it can accurately locate the time frequency information of the EEG signal including the perceptive motion rhythm. By calculating the power spectrum density of the ECoG signal after the improved S transform, the local power expression in the time frequency range of the EEG is obtained. Compared with the S transformation, the improved S change is improved. Better energy concentration; compared with other time-frequency methods, better time-frequency distribution and higher resolution spectrum density functions can be obtained based on the features of improved S transform. Compared with other classifiers, the features extracted from the S transform are combined with the gradient Boosting classifier based on the ordinary least two multiplied regression. The best classification effect can be obtained; the designed channel selection algorithm can greatly reduce the complexity of the algorithm and effectively improve the performance of the BCI system. The experimental results show that the proposed algorithm can get a better classification effect.
2. based on the analysis and processing of the EEG in the spatio-temporal domain, a feature extraction and classification recognition algorithm based on the combination of local two value model and autoregressive model is proposed. This algorithm applies the combination features of local two value pattern operators and autoregressive models to the one dimension EEG signal. In addition, it combines the gradient Boosting classifier to classify the motion imaginary tasks based on ECoG. The histogram distribution of the rotationally invariant local two value mode operator and the two order autoregressive model coefficients estimated by the Burg algorithm constitute the combination features, and the EEG signals are analyzed on the basis of multiple quantization angles and multi-resolution. The time domain and the spatial domain describe the changes in the motion rhythm of the EEG, reflecting the perceptive motion rhythm of the EEG in motion imagination. The algorithm is verified by the international standard BCI competition database. The experimental results prove that the combination features of the algorithm have a higher classification accuracy compared with the other features. To better describe the EEG signals based on motion imagination; compared with other classifiers, combining the combined features with the gradient Boosting classifier based on ordinary least squares regression, the best classification effect can be obtained. As the combination feature will cause the increase of the dimension of the feature vector, the channel selection algorithm is designed to reduce the input. Experimental results show that the algorithm can effectively reduce the complexity of the algorithm while ensuring the accuracy of classification.
3. on the basis of nonlinear dynamics, the fractal geometry theory is applied to the feature extraction of EEG signals. A feature extraction and classification recognition algorithm combined with fractal feature and local two value model is proposed. The algorithm introduces the blanket dimension covering technology widely used in gray image calculation, and calculates the different coverage of the brain electrical signals. The blanket dimension of the cover gets the corresponding fractal intercept and the missing item. The algorithm combines the fractal intercept, the missing term and the local two value mode operator extracted from the space to describe the ECoG signal. The combined feature can be used to analyze the EEG signal in the multi-resolution and multi angle conditions, and can measure the complexity of the EEG signal and reflect the rapid change of the amplitude. Slow. Compared with other features, the combination feature can get better classification results, and can more complete and accurately locate the motion picture of the EEG signal. Channel selection algorithm effectively mitigates the combination feature caused by the feature vector dimension increase, thus reducing the amount of operation. Experimental results show that the algorithm can be obtained. To achieve ideal classification results, we can get a better compromise between classification accuracy and algorithm complexity.
The research work in this paper will help to further promote the research of BCI system based on motion imagination in technical theory, algorithm and practical application. The time domain, frequency domain, space domain and nonlinear dynamic analysis of EEG have been actively promoted in the application of BCI system based on motion imagination.

【学位授予单位】：山东大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN911.7;TP334.8

【参考文献】