基于表面肌电信号的人体肘关节运动角度预测

发布时间：2018-01-16 04:31

本文关键词：基于表面肌电信号的人体肘关节运动角度预测　出处：《上海师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：表面肌电(surface electromyography,sEMG)信号是从人体皮肤表面检测到的生物电信号,它可以反映神经以及肌肉的活动状态。由于表面肌电信号的提取方式具有方便、准确、无创伤等优点,在康复医学、运动医学及智能机器人等领域都有广泛的研究与应用。目前通过表面肌电信号识别人体行为意图的研究大多集中在对表面肌电信号进行模式识别,从而区分肢体不同的运动模式。而在某个单一的肢体动作模式中,肢体的运动状态如运动速度、动作幅度和转动角度等方面的研究较少。单纯使用这种模式分类结果来控制假肢,无法实现假肢的关节像人体关节一样可控。随着信息科技的不断发展,通过sEMG预测肢体的连续的运动状态,成为肌电信号控制仿生假肢技术研究的关键点。因此,本文以手臂肘关节运动的角度作为切入点,旨在研究基于sEMG的肘关节运动角度的预测方法,重点探究表面肌电信号的特征提取和肘关节运动角度的网络预测方法。本文主要内容包括:(1)信号采集系统的构建。本文首先分析了sEMG的产生机理和特点,进而确定表面肌电信号和关节角度信号的测量方法。搭建表面肌电信号和关节角度的采集平台,包括表面肌电仪、表面电极、MEMS传感器,Xsens惯性追踪设备以及表面肌电信号和角度信号的采集软件。设计实验方案,实现对sEMG和肘关节角度的采集。(2)信号的处理和特征分析。首先分析了表面肌电信号和可能混入的噪声的主要频率段,设计滤波器对采集到的原始信号进行滤波,在此基础上进行特征分析,为了兼顾信号的时间和频谱特性,从时域和频域各提取了几个比较有代表性的特征值,为肘关节运动角度的预测模型提供基础。(3)肘关节运动角度预测。使用选定的sEMG特征值作为BP神经网络的输入,将处理后的肘关节角度信号作为神经网络的输出,构建BP人工神经网络预测模型,然后把遗传算法与BP神经网络结合建立改进的预测模型,具有较好的预测效果。实验结果表明,基于BP神经网络预测的肘关节运动角度,与实际角度比较接近,平均误差在8度左右。而经过遗传算法优化后的BP神经网络的预测误差有所下降,平均误差从8度下降到了5度,与目前国内的一些研究相比,预测效果有了较大的提升。这说明使用遗传算法对BP神经网络进行改进,能够减少误差,从而有效提升预测精度。本文研究的人体肘关节运动角度预测方法,对人体的肩关节、膝关节等其他关节的角度预测具有一定的参考价值,还可以与肢体的离散动作模式分类方法相结合,应用到仿生假肢控制中。
[Abstract]:Surface electromyography (EMG) signals are bioelectrical signals detected from human skin surfaces. It can reflect the active state of nerve and muscle. Because the method of extracting surface EMG signal has the advantages of convenience, accuracy, no trauma and so on, it is in rehabilitation medicine. Sports medicine and intelligent robot have been widely studied and applied. At present, most of the researches on recognition of human behavior intention by surface EMG signal are focused on pattern recognition of surface EMG signal. In a single limb movement mode, the movement state of the limb is such as the speed of movement. With the development of information technology, the joint of prosthesis can not be controlled as human joints. The prediction of limb motion by sEMG becomes the key point in the research of EMG control biomimetic prosthesis. Therefore, the angle of arm elbow motion is taken as the starting point in this paper. The purpose of this paper is to study the prediction method of elbow motion angle based on sEMG. This paper focuses on the feature extraction of surface EMG signals and the network prediction method of elbow motion angle. The main contents of this paper are as follows: 1). The construction of signal acquisition system. Firstly, the mechanism and characteristics of sEMG are analyzed. Then the measurement method of surface EMG signal and joint angle signal is determined. The acquisition platform of surface EMG signal and joint angle is built, including surface electromyograph, surface electrode MEMS sensor. Xsens inertial tracking equipment and surface EMG signal and angle signal acquisition software. The signal processing and feature analysis of sEMG and elbow angle acquisition are realized. Firstly, the main frequency bands of surface EMG signal and possible noise are analyzed. The filter is designed to filter the original signal. On this basis, the characteristic analysis is carried out. In order to take into account the time and spectrum characteristics of the signal, several representative eigenvalues are extracted from the time domain and the frequency domain respectively. This paper provides the basis for the prediction model of elbow motion angle. The selected sEMG eigenvalue is used as the input of BP neural network. Using the processed angle signal of elbow joint as the output of neural network, the prediction model of BP artificial neural network is constructed, and then the improved prediction model is established by combining genetic algorithm with BP neural network. Experimental results show that the elbow motion angle predicted based on BP neural network is close to the actual angle. The average error is about 8 degrees, but the prediction error of BP neural network optimized by genetic algorithm has decreased, the average error has been reduced from 8 degrees to 5 degrees, compared with some domestic research. The prediction effect has been greatly improved. This shows that using genetic algorithm to improve BP neural network can reduce the error, thus effectively improve the prediction accuracy. This paper studies the prediction method of human elbow motion angle. It has certain reference value for the angle prediction of other joints such as shoulder joint and knee joint and can be combined with the discrete motion pattern classification method of limbs and applied to the control of biomimetic prosthesis.
【学位授予单位】：上海师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R496;TN911.7;TP18

【参考文献】