采集下肢两通道的表面肌电信号和相应的关节运动信息，对原始表面肌电信号进行预处理。建立基 于径向基函数神经网络的开环估计模型，以预处理后的表面肌电信号为输入，关节运动量为输出。在此基础上，归 零神经网络作为一种特殊的递归神经网络被应用到开环模型中，形成一个混合的闭环预测模型。实验结果表明，所 提出的闭环模型能够有效地消除开环模型的预测误差，进而能够更加准确地识别出人体的主动运动意图，为后续康 复机器人的人机交互系统提供有价值的参考。

In order to accurately identify the active motion intention of human lower limbs, the surface electromyography (sEMG) signals of the two channels and the corresponding joint motion information were collected, and the raw sEMG signals were preprocessed. Then an open-loop prediction model based on radial basis function neural network was established, using the preprocessed sEMG signal as the input and the joint motion information as the output. On this basis, as a special recurrent neural network, the zeroing neural network was exploited to the open-loop model to form a hybrid closed-loop prediction model. The experimental results indicated that the proposed closed-loop model can effectively eliminate the prediction error of the open-loop model, and it can more accurately identify the active motion intention of human lower limbs, which lays a reliable foundation for the subsequent human-computer interaction system of the rehabilitation robot.

收稿日期：2022-03-11  录用日期：2023-04-21 

Received Date: 2022-03-11  Accepted Date: 2023-04-21 

基金项目：国家自然科学基金面上项目（61873304，62173048）；吉林省教育厅科学研究项目（JJKH20210745KJ） 

Foundation Item: National Natural Science Foundation of China(61873304, 62173048); Scientific Research Project of Department of Education of Jilin Province(JJKH20210745KJ) 

通讯作者：孙中波，Email：zhongbosun2012@163.com 

Corresponding Author: SUN Zhongbo, Email: zhongbosun2012@163.com 

引用格式：张鑫，李婉婷，陈岩，等 . 基于递归神经网络的人体下肢运动意图识别方法 [J].机器人外科学杂志（中英文），2024，5（2）： 121-129. 

Citation: ZHANG X, LI W T, CHEN Y, et al. A motion intention recognition method of human lower limbs based on recurrent neural network[J]. Chinese Journal of Robotic Surgery, 2024, 5(2): 121-129

[1] NIU C M, BAO Y, ZHUANG C, et al. Synergy-based FES for post-stroke rehabilitation of upper-limb motor functions [J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2019, 27(2): 256-264.

[2] WEN Y, LI M H, SI J, et al. Wearer-Prosthesis interaction for symmetrical gait: a study enabled by reinforcement learning prosthesis control [J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2020, 28(4): 904-913. 

[3] WEI D, LI Z J, WEI Q, et al. Human-in-the-loop control strategy of unilateral exoskeleton robots for gait rehabilitation [J]. IEEE Transactions on Cognitive and Developmental Systems, 2019, 13(1): 57-66. 

[4] YU X B, HE W, LI Y A, et al. Bayesian estimation of human impedance and motion intention for human-robot collaboration [J]. IEEE Transactions on Cybernetics, 2019, 51(4): 1822-1834.

[5] LIN M W, RUAN S J, TU Y W. A 3DCNN-LSTM hybrid framework for sEMG-based noises recognition in exercise [J]. IEEE Access, 2020. DOI: 10.1109/ ACCESS.2020.3021344. 

[6] Park S, Wan K C, Kim K. Training-Free Bayesian selfadaptive classification for sEMG pattern recognition including motion transition [J]. IEEE Transactions on Biomedical Engineering, 2019, 67(6): 1775-1786. 

[7] HU X H, ZENG H, SONG A G, et al. Robust continuous hand motion recognition using wearable array myoelectric sensor [J]. IEEE Sensors Journal, 2021, 21(18): 20596-20605. 

[8] XI X G, JIANG W J, HUA X, et al. Simultaneous and continuous estimation of joint angles based on surface electromyography state-space model [J]. IEEE Sensors Journal, 2021, 21(6): 8089-8099. 

[9] LI Z Y, ZHANG D H, ZHAO X G, et al. A temporally smoothed MLP regression scheme for continuous knee/ankle angles estimation by using multi-channel sEMG[J]. IEEE Access, 2020. DOI: 10.1109/ ACCESS.2020.2979008. 

[10] MA C F, GUO W Y, ZHANG H, et al. A novel and efficient feature extraction method for deep learning based continuous estimation [J]. IEEE Robotics and Automation Letters, 2021, 6(4): 7341-7348.

[11] WANG W Q, HOU Z G, CHENG L, et al. Toward patients’ motion intention recognition: dynamics modeling and identification of iLeg—an llrr under motion constraints [J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2017, 46(7): 980-992. 

[12] WANG G, LIU Y B, SHI T, et al. A novel estimation approach of sEMG-based joint movements via RBF neural network [C]. 2019 Chinese Automation Congress (CAC), 2019. 

[13] CHAI Y Y, LIU K P, LI C X, et al. A novel method based on long short term memory network and discretetime zeroing neural algorithm for upper-limb continuous estimation using sEMG signals [J]. Biomedical Signal Processing and Control, 2021. DOI: 10.1016/ j.bspc.2021.102416. 

[14] JIN L, LI S, HU B. RNN models for dynamic matrix inversion: a control-theoretical perspective [J]. IEEE Transactions on Industrial Informatics, 2017, 14(1): 189-199. 

[15] JIN L, YAN J K, DU X J, et al. RNN for solving timevariant generalized Sylvester equation with applications to robots and acoustic source localization [J]. IEEE Transactions on Industrial Informatics, 2020, 16(10): 6359-6369.