多通道表面肌电信号分解的研究

发布时间：2018-03-23 05:24

本文选题：表面肌电信号　切入点：运动单元　出处：《浙江大学》2014年博士论文　论文类型：学位论文

【摘要】：表面肌电信号(sEMG)通常被认为是由肌肉中多个运动单元(MU)生成的动作电位序列叠加而形成的,信号能通过放置在肌肉处皮肤上的电极进行测量。对表面肌电信号进行分解能够获得运动单元的发放以及募集的相关信息,能够为神经肌肉系统的研究和诊断提供重要的依据。本文提出了几种多通道表面肌电信号的分解方法。围绕多通道表面肌电信号的分解,本文的主要工作和研究成果有： 1.提出将卷积核补偿(CKC)方法和模糊C均值(FCM)聚类方法结合,进行多通道表面肌电信号的分解。先选出几个通道的信号,再用这几个通道的信号得出初始的MU发放序列,然后用模糊C均值聚类方法对初始的发放序列一些峰值对应的时刻进行聚类处理,最后用CKC方法得到最终的MU发放序列。仿真结果表明这种方法与原CKC方法比较,能够改进信号分解的性能。 2.根据已有的卷积核补偿(CKC)方法,提出了一种新的表面肌电信号分解方法。该方法结合自组织映射(SOM)神经网络,首先找出在某一时刻具有能量活动的发放序列,其次对这个发放序列的一些较大值对应的时刻用自组织映射神经网络进行聚类,然后利用聚类后的时刻所对应的多通道测量信号的值求出最终的一个MU发放序列。仿真信号测试得到的结果表明,所提出的这种方法是有效的。3.提出了一种能够逼近线性最小均方误差(LMMSE)估计量的多通道表面肌电信号的分解方法。首先用K均值对不同时刻对应的测量向量进行分类,然后估算出初始的MU发放序列。再用一种多步迭代算法对初始的发放序列不断进行迭代计算,得到最终的MU发放序列。用仿真信号和真实的表面肌电信号对该方法的性能进行了检验。采用仿真信号时,即使信噪比达到-10dB,所有的10个发放序列仍能够以大于90%的准确率被重建出来。对从手部第一背侧骨间肌采用64个通道电极阵列测量得到的真实表面肌电信号进行分解时,多于10个的MU能够被成功提取出来。并用“二源法”对真实信号的分解性能进行了验证,从两个独立的分组中提取出较多的相同MU个数和大于92%的相同MU的相同发放时刻的百分比,表明了该方法在表面肌电信号分解中的可靠性。 4.提出了两种基于测量信号相关的多通道表面肌电信号的分解方法。一种用Moore-Penrose伪逆构建测量信号矩阵的相关矩阵,另一种用奇异值分解(SVD)方法构建测量信号矩阵的相关矩阵。由于同一个MU不同发放时刻对应的测量向量有着一定程度的相似性,因此可以采用特定的迭代优化方法逐步增强所选出的测量向量与构建的相关矩阵之间的相关性,来达到分解信号的目的。并用仿真信号和真实的表面肌电信号对提出的方法进行了检验。在仿真信号进行的测试中,两种方法都能够以大于95%的准确率重建出多于48个的MU发放序列。在真实信号的测试中,两种方法都能够重建出多于15个的MU发放序列。并进一步用“二源法”对真实信号的分解结果进行了验证,结果表明了这两种方法分解的可靠性。本论文的研究获得美国国立卫生研究院(NIH K99DK082644,NIHROODK082644)的部分资助。
[Abstract]:Surface electromyography (sEMG) is usually considered by a number of muscle unit (MU) series of action potential generated by the superposition and the formation of the signal can be measured by electrodes placed on the skin muscles. The surface EMG signal decomposition to obtain relevant information and to raise the issue of motion unit, can provide an important basis for the study and diagnosis of neuromuscular system.
In this paper, several methods for decomposition of multichannel surface EMG signals are proposed. The main work and research results of this paper are as follows: the decomposition of multichannel surface EMG signals.
1. the convolution kernel compensation (CKC) method and fuzzy C means (FCM) clustering method with decomposition of multi-channel surface EMG signal. Select several channels, then the signal of several channels to get the initial MU release sequence, and then use the fuzzy C mean clustering method on the initial peak firing sequence the corresponding time clustering processing, by using the CKC method, the final MU distribution sequence is obtained. The simulation results show that compared with the original CKC method this method can improve the performance of signal decomposition.
2. according to the existing convolution kernel compensation (CKC) method, put forward a new kind of surface EMG signal decomposition method. This method combines self-organizing map (SOM) neural network, first identify the firing sequence with energy activities at a certain moment, then to release this sequence of some of the larger value corresponding to the time of clustering self organizing mapping neural network, and then use the multi-channel measurement signal corresponding to the clustering moments after the value obtained a MU final firing sequence. Simulation test results show that the proposed method is effective to approximate.3. proposed a linear minimum mean square error (LMMSE) estimation the channel surface EMG signal decomposition method. The amount of the first measurement vector with K mean at different times corresponding to the classification, and then estimate the initial MU release sequence. Then a multi step iterative algorithm to the initial The firing sequence continuous iterative calculation, obtain the final MU firing sequence. The performance of the method by simulation data and real surface EMG signal was tested. Using the simulation signal, even if the signal-to-noise ratio reached -10dB, 10 of all the firing sequence is still able to accurately rate greater than 90% was reconstructed. The decomposition of real surface EMG signal from the hands of the first dorsal interosseous muscle using 64 channel electrode array measurement, more than 10 MU can be successfully extracted. With the "two source" method to validate the decomposition performance of real signals, extract the same percentage more number of MU and more than 92% of the same MU the same time issued from two independent groups, shows the reliability of this method in surface EMG signal decomposition.
4. this paper puts forward two decomposition methods based on the measured signal related to the multi-channel sEMG signal. A correlation matrix with Moore-Penrose pseudo inverse matrix to construct the measurement signal, another with a singular value decomposition (SVD) method to construct the correlation matrix of measurement signal matrix. Since the measurement vector with a MU corresponding to different firing moment there is a certain degree of similarity, so it can gradually increase the correlation between the correlation matrix measurement vector selected and constructed the iterative optimization method used to achieve specific, decomposition of signal. And the simulation signal and real surface EMG signal of the proposed method is tested in the simulation test signal. In the two methods are able to more than 95% accuracy of reconstruction of more than 48 MU. The firing sequence of the true signal test, the two methods can reconstruct more than 15 MU The results of the decomposition of real signals are verified by the two source method. The results show the reliability of the decomposition of the two methods.
The research in this paper is partially funded by the National Institutes of Health (NIH K99DK082644, NIHROODK082644).

【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R741.044;TN911.7

【参考文献】