特征图组合的双流CNN手指关节角度连续运动预测方法研究  

Research on the continuous motion prediction method of finger joint angles using dual-stream CNN based on feature map combination

作　　者：武岩[1,2] 曹崇莉 李奇[1,2] 姬鹏辉 张航 WU Yan;CAO Chongli;LI Qi;JI Penghui;ZHANG Hang(School of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,China;Zhongshan Institute of Changchun University of Science and Technology,Zhongshan 528400,China)

机构地区：[1]长春理工大学计算机科学技术学院,长春130022 [2]长春理工大学中山研究院,广东中山528400

出　　处：《重庆理工大学学报（自然科学）》2024年第11期119-128,共10页Journal of Chongqing University of Technology：Natural Science

基　　金：吉林省科技发展计划国际科技合作项目(20200801035GH);吉林省科技发展计划国际联合研究中心建设项目(20200802004GH)。

摘　　要：针对基于表面肌电(surface electromyography,sEMG)信号手指关节角度连续运动预测时序信息提取不足、预测准确率较低的问题,提出了一种基于特征图组合(feature map combinations,FMC)的双流卷积神经网络(dual-stream convolutional neural network,DCNN)预测方法。提取sEMG信号的特征信息,采用滑动窗方式将特征信息进行特征图组合,表达特征的时间连贯性以提取sEMG信号的时序信息,通过DCNN网络在时间、空间维度对组合后的特征图提取深层特征,提高手指关节角度连续运动预测效果。在NinaPro-DB8数据集上进行实验,结果表明:在3类不同自由度(18个、5个、3个)的相关方法比较中,健康受试者的R2值分别提高了7.9%、16.8%和17.8%;截肢受试者的R2值分别提高了9.6%、14.3%和10.3%。To address the insufficient extraction of timing information and low accuracy in predicting continuous motion of finger joint angles based on surface electromyographic(sEMG)signals,we propose a two-stream convolutional neural network prediction method based on feature map combination(FMC).First,the feature information of the sEMG signal is extracted.Then,the feature information is integrated into feature maps(FMC)by employing a sliding window method to express the temporal coherence of the features and extract the temporal information of the sEMG signal.Finally,the dual stream convolutional neural network(DCNN)network is used to extract deep features from the combined feature maps in the temporal and spatial dimensions to improve the prediction of finger joint angles continuous motion.Experiments are conducted on the NinaPro-DB8 dataset,and our results show,compared with three different degrees of freedom(18,5,3),the R 2 values of healthy subjects increase by 7.9%,16.8%,and 17.8%respectively,while the R 2 values of amputees increase by 9.6%,14.3%,and 10.3%respectively.

关 键 词：SEMG 连续运动预测 特征图组合 双流卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]