基于广义回归神经网络的上肢关节角度预测

针对人体上肢运动意图识别问题,基于上肢表面肌电信号,提出广义回归神经网络(general regression neural network,GRNN)预测受试者的上肢关节角度.GRNN预测模型的输入为处理后的表面肌电信号,预测的3个关节角作为输出,将GRNN预测结果和径向基函数(radial basis function,RBF)神经网络的预测结果对比,并用均方根误差对上肢关节角度的预测结果做评估,验证GRNN模型预测上肢关节角度的可行性.结果表明,GRNN模型能较好地估计人体关节角度.

Upper Limb Joint Angles Estimation based on General Regression Neural Network

Aiming at the recognition problem of human upper limb movement intention, a GRNN is proposed based on the surface eletromyography signals of upper limbs to predict the joint angles of upper limb joints in this paper. The inputs of GRNN model are the treated surface eletromyography signals, and the outputs are the predicted three joint angles. Compared with RBF network, the GRNN model shows the superior performance of prediction. The prediction of upper limb joint angles was evaluated with root mean square errors to verify the feasibility of GRNN predicting upper limb joint angles. The results show that the GRNN model is able to estimate joint angles of human.