下肢表面肌电信号的降维和映射分析

肌电信号的采集和分析是外骨骼式康复机器人关节预测控制的重要基础之一.肌电信号数据量大并且复杂,相关性较高,信号处理通用性和高效性低,分析和预测人体运动信息误差大.采用最大自主等长收缩标准化处理算法,大大提高了表面肌电信号的通用性和泛化能力,并基于主成分分析方法,对肌电信号降维处理,利用神经网络实现与下肢的映射分析.实验结果表明,通过对比分析不同的降维处理方式,主成分降维后处理的肌电信号平均相关性达0.93,利用神经网络预测人体正常行走的下肢三关节运动角度,具有良好的可重复性和较高的精度,可以实现人体下肢肌电信号和各关节的映射控制.

Dimension Reduction and Mapping Analysis of EMG signals on Lower Limbs

The acquisition and analysis of EMG signals is one of the important bases for the joint prediction and control of exoskeleton rehabilitation robot. The data of EMG signals are large and complex; the correlation is high; the signal processing universality is low; the efficiency is poor, and information prediction of human motion error is great. The maximum voluntary contraction standardization algorithm is used, which greatly improves the universality and generalization ability of surface EMG signals. Based on the principal component analysis algorithm, the EMG signals is reduced dimensionally and the neural network is used to map the lower limbs. The experimental results show that by comparing and analyzing different dimension reduction approaches, the average correlation of EMG signals in which the principal component analysis algorithm was reduced is 0.93. The neural network is used to predict the leg movement angle of lower limbs and this method has good repeatability and higher accuracy, and can realize the mapping control of lower limb EMG signals and joints.