基于GA-BP的表面肌电信号下肢动作模式识别研究  

Research on lower limb motion pattern recognition by sEMG signals based on GA-BP

作　　者：崔冰艳 张祥 邓嘉 CUI Bingyan;ZHANG Xiang;DENG Jia(College of Mechanical Engineering,North China University of Science and Technology,Tangshan 063000,China)

机构地区：[1]华北理工大学机械工程学院,河北唐山063000

出　　处：《传感器与微系统》2024年第9期63-67,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金青年科学基金资助项目(E51505124);河北省自然科学基金资助项目(E2017209252);河北省高等学校科学技术研究重点项目(ZD2020151);唐山市基础研究项目(23130201E);华北理工大学重点科研项目(ZD-G-202306-23)。

摘　　要：为了满足下肢康复机器人运动过程中对人体下肢不同动作模式的识别的需求,首先,通过8通道无线肌电传感器采集8种下肢常见动作的表面肌电(sEMG)信号,并对原始信号进行滤波、运动段提取、特征提取处理;然后,将处理后数据分别输入本文设计的BP、PCA-BP、GA-BP、PCA-GA-BP分类器进行训练与测试。4种分类器对下肢8种动作平均识别率分别为88.6%,90.5%,92.3%,95.1%,对每个动作平均识别率为85%以上。结果表明:基于GA-BP神经网络比BP神经网络具有更高的预测精度,并且降维处理可以提高动作分类的准确率。In order to meet the needs of recognizing different movement patterns of human lower limbs during exercise in lower limb rehabilitation robot,firstly,the surface electromyography(sEMG)signals of 8 common movements of the lower limbs are collected through 8-channel wireless EMG sensors,and the original signals are filtered,the motion segment is extracted,and the feature extraction is processed,and then the processed data are input into 4 classifiers of BP,PCA-BP,GA-BP,PCA-GA-BP for training and testing.Four classifiers are designed.The average recognition rates of the above four classifiers on the eight lower limb movements are 88.6%,90.5%,92.3%,and 95.1%,respectively,and the average recognition rate of each action is more than 85%.The results show that GA-BP-based neural network has higher prediction precision than BP neural network,and dimensionality reduction processing can improve the accuracy of action classification.

关 键 词：表面肌电信号 特征提取 遗传算法 反向传播神经网络 模式识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]