单通道诱发电位信号的快速提取算法研究

发布时间：2019-06-11 02:40

【摘要】：诱发电位(EP)信号为神经科学的理论研究与临床应用提供许多重要信息,它反映了相应的感觉通路及大脑皮层区域的神经电活动。深入地分析研究EP信号,对于研究大脑活动规律及其信息处理机制、发展神经电生理学理论与应用、临床诊断评价神经系统的功能以及临床术中监控等均具有重要意义。EP信号通常深深地湮没在自发脑电(EEG)信号之中。因此,从强EEG背景噪声中有效地提取出EP信号一直是生物医学信号处理领域研究的重要问题之一。目前在临床中广泛应用的相干平均法存在丢失EP信号细节以及因神经系统疲劳而导致较大测量误差等不足。基于此问题,对EP信号快速提取方法的研究成为近年来的一个研究热点与难点。所谓快速提取主要相对于相干平均法而言,是指对EP信号的少次提取、单次提取以及动态跟踪等。本文主要研究在单通道条件下对EP信号的快速提取算法,研究成果可以归纳如下： (1)深入研究了基于稀疏表示模型的单通道EP信号少次提取问题,提出了基于混合训练字典的稀疏表示方法与基于联合稀疏表示的方法用于单通道EP信号的少次提取。针对已有混合字典稀疏表示中使用通用过完备字典造成的对信号成分的错误划分问题,首先根据EP与EEG信号的不同特点提出了基于混合训练字典的稀疏表示方法,通过使用其他少次观测数据设计模板信号并训练分别与EP和EEG信号相适应的过完备字典,该方法有效地减少了使用混合字典稀疏表示过程中的错分问题。然后提出基于联合稀疏表示的EP信号少次提取方法,利用EP信号的准周期性,同时使用少次相邻观测信号进行联合稀疏表示,可以在较低的信噪比情况下更有效地提取EP信号。 (2)深入研究了基于时间自相关函数的单通道EP信号单次提取问题,提出了两种基于源信号时间自相关函数的波形估计方法并用于单通道EP信号的单次提取。利用时间自相关函数带来的信息,首先提出一种以源信号时间自相关函数作约束的波形估计方法,它利用源信号的时间自相关函数构造非线性方程组,并借助大规模方程组的数值解法,把从信噪比较低的观测数据中直接估计源信号这一较困难的问题,转化成分别估计迭代初始值与源信号时间自相关函数的问题。然后针对当源信号时间自相关函数估计精度较低时,该方法需要较多计算时间的不足,又提出了基于时间自相关函数最优化的波形估计方法。该方法能够在估计精度与计算速度之间较好地取得平衡,更加适用于对计算效率要求较高的应用。将以上两种方法应用于单通道EP信号的单次提取问题,取得了较好的效果。 (3)深入研究了脉冲噪声环境下基于径向基函数神经网络模型的单通道EP信号自适应估计问题,提出了三种韧性单通道EP信号自适应估计方法。当临床应用中EP信号的背景噪声呈现非高斯脉冲特性时,更加适合使用α稳定分布来描述。针对基于最小平均p范数准则的EP信号自适应估计方法无法在α值动态变化时较好工作的不足,首先提出了基于最小平均绝对偏差准则的韧性单通道EP信号自适应估计方法,它可以在α值动态变化的情况下较好地工作。但该方法使用的二值化变换完全丢失了误差信号的幅度信息,导致其无法在估计精度与收敛速度之间很好地平衡。针对该不足,又提出了基于非线性Sigmoid变换的韧性单通道EP信号自适应估计方法,可以应用于α值动态变化的情况,并可以较好地保留误差信号的幅度信息,具有较好的估计精度与收敛速度。最后提出一种基于最大相关熵准则的韧性单通道EP信号自适应估计方法,通过选取适当的核长参数,它可以较好地工作在α值动态变化的情况下。将以上算法应用于单通道EP信号的韧性自适应估计,均取得较好的估计结果。
[Abstract]:The evoked potential (EP) signal provides a number of important information for the theoretical study and clinical application of neuroscience, which reflects the corresponding sensory pathway and the neural electrical activity in the cerebral cortex area. In-depth analysis of the EP signal is of great significance in the study of the law of brain activity and its information processing mechanism, the development of the neuroelectrophysiology theory and the application, the function of the clinical diagnosis and evaluation of the nervous system, and the monitoring of clinical operation. The EP signal is typically deeply annihilated in a spontaneous brain (EEG) signal. Therefore, the effective extraction of the EP signal from the strong EEG background noise has been one of the most important problems in the field of biomedical signal processing. At present, the coherent average method widely used in the clinical application has the defects of losing the detail of the EP signal and the large measurement error due to the fatigue of the nervous system. Based on this problem, the research of the rapid extraction method of the EP signal has become a hot point and difficulty in recent years. The so-called fast extraction is mainly relative to the coherent average method, which means less extraction, single extraction and dynamic tracking of the EP signal. In this paper, the fast extraction algorithm of the EP signal under the single channel condition is studied, and the research results can be summarized as follows: Next: (1) In-depth study of a single-channel EP signal based on sparse representation model In this paper, a sparse representation method based on a mixed training dictionary and a method based on the combined sparse representation are proposed to reduce the single channel EP signal. In order to solve the problem of misclassification of the signal component caused by the general overcomplete dictionary in the sparse representation of the existing mixed dictionary, the sparse table based on the mixed training dictionary is first proposed according to the different characteristics of the EP and the EEG signal. The method comprises the following steps of: designing a template signal by using other less observation data and training an over-complete dictionary corresponding to the EP and the EEG signal, the method effectively reduces the error in the process of using the mixed dictionary sparse representation, In this paper, the secondary extraction method of the EP signal based on the joint sparse representation is proposed, and the quasi-periodicity of the EP signal is used, and the combined sparse representation of the adjacent observation signals is used at the same time, and the E signal can be extracted more effectively in the case of lower signal-to-noise ratio. P signal. (2) In-depth study of single-channel EP signal extraction problem based on time-autocorrelation function, two waveform estimation methods based on time-autocorrelation function of source signal are proposed and used for single-channel EP signal A waveform estimation method based on the time autocorrelation function of the source signal is proposed, which uses the time self-correlation function of the source signal to construct a nonlinear system of linear equations, and by means of large-scale equations The numerical solution of the source signal is directly estimated from the observation data with low signal-to-noise ratio, which is converted into the estimated iteration initial value and the source signal time self-correlation, respectively. The problem of the function is solved. Then, when the estimation accuracy of the source signal time from the correlation function is low, the method needs to be less than the calculation time, and then the wave based on the time autocorrelation function is proposed. The method can obtain the balance between the estimation precision and the calculation speed, and is more suitable for the requirement of the calculation efficiency. The application of the above two methods to single-channel EP signal extraction has been achieved. In this paper, the self-adaptive estimation of single-channel EP signal based on radial basis function neural network model is studied in this paper. Three kinds of flexible single-channel EP signals are put forward. The adaptive estimation method is more suitable for use when the background noise of the EP signal exhibits non-Gaussian pulse characteristics. In this paper, an EP signal adaptive estimation method based on the minimum average p-norm criterion is not able to work well at the dynamic change of the peak value, and a flexible single-channel EP signal based on the minimum mean absolute deviation criterion is first proposed. An adaptive estimation method, which can be used to dynamically change the value of the value but the binary transformation used by the method completely loses the amplitude information of the error signal, so that the error signal can not be in the estimation precision and the convergence speed, In this paper, the self-adaptive estimation method of the flexible single-channel EP signal based on the non-linear sigmoid transformation is proposed, which can be applied to the dynamic change of the amplitude value, and the amplitude information of the error signal can be better preserved. In this paper, an adaptive estimation method of a single-channel EP signal based on the maximum correlation entropy criterion is proposed. In the case of state change, the above algorithm is applied to the adaptive estimation of the toughness of the single-channel EP signal.
【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R338;TN911.7

【参考文献】