Research On Continuous Motion Estimation And Its Application In Rehabilitation Training For Knee And Ankle Joints

Stroke has become the leading cause of death and disability in China,and hemiplegia is the most common clinical manifestation of stroke,which seriously affects patients’ daily lives.Therefore,assisting stroke patients with hemiplegia to restore motor function has become a research hotspot in the field of rehabilitation medicine.Functional electrical stimulation is an important method for the rehabilitation of patients with hemiplegia.However,functional electrical stimulation uses passive training mode clinically,which has limited effect on the reconstruction of neural circuits and the promotion of motor function recovery.In order to achieve compliant,safe and active rehabilitation training,surface electromyography(s EMG),which contains rich motion control information,can be used to accurately estimate the continuous motion intention of the human body as control instructions.To this end,a comprehensive study of the continuous estimation method of knee and ankle joint angle based on s EMG is conducted in this paper.A continuous estimation model of knee and ankle joint angle based on transfer learning and deep learning is built and applied to rehabilitation training.The main research work includes:(1)A study of constrained independent component analysis(CICA)EMG crosstalk removal algorithm considering co-contraction.Design a signal acquisition scheme to acquire the s EMG under electrode shift,and analyze the effect of electrode shift on the signal acquired by sparse surface electrodes.Quantify and distinguish crosstalk and co-contraction based on the cross-correlation coefficient,and a CICA crosstalk removal algorithm that considers co-contraction is proposed.While removing crosstalk and electrode channel redundancy caused by electrode,it can keep the effective motion control information as much as possible.The feasibility and effectiveness of the algorithm are verified through simulation and comparison.(2)Continuous motion estimation of knee and ankle joint based on transfer learning.Using the feature transfer function of the kernel canonical correlation analysis,a high correlation between different subjects,that is,the source domain and the target domain,is built in the common feature space.Based on the convolutional neural network,a prediction model is constructed from the de-crosstalk s EMG in source domain to the knee and ankle joint angle information,thereby saving training data costs and enabling different domain data to obtain highly accurate estimation results.(3)Functional electrical stimulation control for knee and ankle joints of the affected side based on continuous motion estimation of the contralesional side.For rehabilitation training applications,the characteristics of s EMG signals during a unilateral side movement and the muscle response under electrical stimulation are analyzed to build an electrical stimulation muscle model.The independent adaptive sliding mode control method of each joint is used to accurately track the joint angle trajectory,thereby verifying the accuracy and effectiveness of the continuous motion estimation in the actual control application.