一种基于sEMG的脑卒中患者患侧踝关节动作识别方法研究

A method for Recognition of Lateral Ankle Joint in Stroke Patients Based on sEMG

基于sEMG(表面肌电信号)的动作识别被广泛应用于机器人辅助康复领域中.传统的运动识别方法多是利用训练后的参数模型,辨识出有限个已知动作.但由于患者患侧难以长时间保持某一动作,无法获得准确训练模型,导致在线分类、在线运动角度估计时存在较大误差.针对这一问题,提出一种自适应阈值分类模型,将归一化后的特征值极值与设定阈值比较即可方便快速地获取动作分类结果,然后估计得到患者关节运动角度.对于不同康复阶段的患者而言,模型可以根据患者肌电信号强弱自适应改变阈值的大小.通过踝关节动作识别实验方法有效性,相比传统线性判别分析、K近邻以及BP神经网络等方法,该方法精度更高、所用时间更短,更适用于患者患侧的主动控制.

Motion recognition based on sEMG(Surface EMG Signal)is widely used in the field of robotic assisted rehabilitation.Traditional motion recognition methods mostly use the trained parametric model to identify a limited number of known actions.However,because the patient’s affected side is difficult to maintain a certain movement for a long time,an accurate training model cannot be obtained,resulting in a large error in online classification and online motion angle estimation.Aiming at this problem,an adaptive threshold classification model is proposed.By comparing the normalized eigenvalue extremum with the set threshold,the action classification result can be obtained quickly and easily,and then the joint motion angle of the patient is estimated.For patients with different rehabilitation stages,the model can adaptively change the threshold according to the patient’s EMG signal strength.The effectiveness of the experimental method is recognized by the ankle joint motion.Compared with the traditional linear discriminant analysis,K-nearest neighbor and BP neural network,the method has higher precision and shorter time,and is more suitable for active control of the patient’s affected side.