一种基于改进sEMG活动段检测的手部运动意图识别新方法  

A New Method for Hand Motion Intent Recognition Based on Improved sEMG Signal Active Segment Detection

作　　者：刘声中 许德章[1,2] Liu Shengzhong;Xu Dezhang(School of Artificial Intelligence,Anhui Polytechnic University;Wuhu Anpu Robot Industry Technology Research Institute,Wuhu,Anhui 241000)

机构地区：[1]安徽工程大学人工智能学院 [2]芜湖安普机器人产业技术研究院,安徽芜湖241000

出　　处：《嘉兴学院学报》2022年第6期105-112,共8页Journal of Jiaxing University

基　　金：国家自然科学基金(52005003);芜湖市科技局项目(2021cg03)。

摘　　要：针对单参数固定门限检测表面肌电信号(sEMG)活动段的端点反转、端点末端提前收尾导致手部运动意图分类效果不佳的问题,提出了一种基于改进sEMG活动段检测的手部运动意图识别方法.首先,通过改进两级判别自适应门限算法,检测sEMG活动段的端点;其次,将子活动段的时域特征作为分类模型的输入特征矩阵,对LSTM手部运动意图识别模型进行训练;最后,以UCI肌电数据集作为研究对象进行对比实验.结果表明,相比于单参数固定门限提取sEMG活动段的特征作为LSTM输入,手部运动意图的分类精度提高了15.6%,总体平均分类精度达到91.7%.Considering the poor classification of hand motion intent caused by endpoint reversal and early termination of the endpoint in single-parameter fixed threshold detection of the surface electromyographic(sEMG)signal active segments,a hand motion intent recognition method based on improved sEMG signal active segment detection is proposed.First,by improving the two-level adaptive threshold algorithm,the endpoints of the sEMG signal active segments are detected.Then,the time-domain features of the active sub-segments are used as the input feature matrix of the classification model to train the LSTM hand motion intent recognition model.Finally,the UCI EMG dataset is used as the research object to conduct comparative experiments.The results show that compared with the single-parameter fixed threshold detection,this new method for extracting the features of the sEMG signal active segments as the LSTM input can improve the classification accuracy of hand motion intent by 15.6%and the overall average classification accuracy by 91.7%.

关 键 词：表面肌电信号 活动段提取 LSTM神经网络 手部运动意图 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]