基于表面肌电信号的膝关节角度预测方法

Knee Joint Angle Prediction Method Based on Surface Electromyography Signal

人体关节角度的连续预测对于提高人机协同控制至关重要.为了提高关节角度的预测精度,提出了一种基于特征的卷积神经网络-双向长短期记忆网络(Convolutional Neural Network-Bidirectional long short-term memory network,CNN-BiLSTM)模型并对下肢关节角度进行了连续预测.采集了人体在正常步态、上楼梯运动时下肢的表面肌电信号和膝关节角度,对信号进行预处理,利用主成分分法进行特征值融合,与传统的支持向量机、长短期记忆网络、卷积神经网络等算法预测效果进行对比,结果表明CNN-BiLSTM模型对关节角度的拟合效果最优,所提模型能够更有效地预测不同运动模式下的膝关节角度,在促进人机协作方面具有更好的性能.

Continuous prediction of human joint angles is essential for enhancing human-machine collaborative control.In order to improve prediction accuracy,this paper proposes a feature-based Convolutional Neural Network-Bidirectional Long Short Term Memory Network(CNN-BiLSTM)model to continuously predict the joint angles of lower limbs.The paper collects surface electromyography signals and knee joint angles of the lower limbs during normal gait and stair climbing movements,preprocesses the signals,uses principal component analysis for feature value fusion,and then compares the prediction performance with traditional algorithms such as support vector machines,long short-term memory network and convolutional neural network,and other algorithms,the results show that CNN-BiLSTM model has the best fitting effect on the joint angle,and the proposed model can more effectively predict the angles in different sports modes,and has better performance in promoting human-machine collaboration.