基于卷积神经网络的踝关节运动模式识别  

Ankle Joint Motion Pattern Recognition Based On CNN

作　　者：朱明霞 叶晔[1] ZHU Mingxia;YE Ye(College of Mechanical Engineering,Anhui University of Technology,Ma'anshan 243002,China)

机构地区：[1]安徽工业大学机械工程学院,安徽马鞍山243002

出　　处：《洛阳理工学院学报（自然科学版）》2023年第4期59-63,共5页Journal of Luoyang Institute of Science and Technology：Natural Science Edition

摘　　要：采用两个无线表面肌电信号采集器,采集人体小腿比目鱼肌和腓骨短肌的踝关节背曲、跖曲、内翻、外翻动作信号,对两通道的原始信号进行归一化,将归一化后的数据输入到卷积神经网络,对踝关节4种运动进行模式识别。实验采集6名健康被试,针对单个被试的最高分类准确率为98.19%,平均识别率为97.04%,模式识别模型和采集通道能以较高分类准确率区别踝关节的4种动作。针对6名被试的整体分类识别率为95.11%,说明个体之间的肌电信号存在共性。Two wireless surface electromyography(SEMG)signal collectors are used to acquire ankle dorsiflexion,plantar flexion,inversion,and eversion motion signals from the human calf soleus and peroneus brevis.The acquired raw signals were normalized,and the normalized data were input to a CNN for final pattern recognition of the four movements of the ankle joint.Experimentally,6 healthy subjects were collected,and the highest classification accuracy for a single subject was 98.19%and the average recognition rate was 97.04%,indicating that the recognition pattern and the collected channels could distinguish the 4 movements accurately.The overall classification recognition rate for the 6 subjects was 95.11%,indicating that there was commonality in their EMG signals.

关 键 词：表面肌电信号 踝关节运动 模式识别 卷积神经网络 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置]