文献类型：期刊文献

中文题名：基于躯干肌肉活动监测的手部行为识别

英文题名：Hand Behavior Recognition Based on Trunk Muscle Activity Monitoring

作者：王琦[1,2];王庆明[2]

机构：[1]上海电机学院设计与艺术学院,上海201306;[2]华东理工大学机械与动力工程学院,上海200237

年份：2021

卷号：21

期号：11

起止页码：4550

中文期刊名：科学技术与工程

外文期刊名：Science Technology and Engineering

收录：CSTPCD;;北大核心:【北大核心2020】；

基金：国家自然科学基金(61802247);闵行区重大产业技术攻关计划(2018MH208);科技创新行动计划(19441914900)。

语种：中文

中文关键词：表面肌电;躯干;行为识别;两步聚类;双向长短时神经网络

外文关键词：electromygraphy;trunk;behavior recognition;twostep clustering;bidirectional long short term memory network

摘要：为解决某些手部工作意图不能由手部的运动和肌肉行为来反映的问题,拟通过监测部分躯干肌肉的协作方式,识别手部行为意图。因此在限定任务的情况下,设计了伴有单手操作的弯伸腰实验。同步采集全身运动信号和一组椎旁肌肌电信号。调整、选择肌电信号的两步聚类细分程度。作为双向长短时神经网络的输入信号,肌肉组行为标注步骤的F 1平均值为91.37%。最终确认,从躯干肌肉群行为抽取的编码可作为信号源,识别手部精确控制、维持平衡等意图。
Certain work intention of hands can t be reflected by behaviors or myoelectric signals of the hands.Thus,the cooperation of some trunk muscles can be monitored to identify the intention of the hands.Therefore,within limited tasks,an experiment including flexion-extension tasks accompanied by single hand operation was designed.The motion signals of the whole body and myoelectric signals of a group of paravertebral muscle were collected simultaneously.The two-step clustering subdivision degree of myoelectric signals was adjusted and selected.As the input signal of bidirectional long short term neural network,the averaged F 1 value of the muscle behavior labeling step was 91.37%.Finally,it is confirmed that the codes extracted from the trunk muscle group can be used as a signal source to identify the intentions including precise control of the hands and maintenance of balance.