Continuous Estimation of Lower Limb Prosthesis Based on GBDBN and BP Network

Abstract

A method for continuous estimation of sagittal joint angles based on surface electromyography(sEMG) signals and BP neural networks is proposed for the problem of continuous motion control of lower limb prostheses. We use the sEMG signals of eleven muscles related to knee and ankle joint motions and calculate the knee and ankle joint angles at the same time. First, a feature extraction method based on Gaussian-Bernoulli deep belief network (GBDBN) was used to extract the optimal feature vector from the multichannel sEMG signals, and then, the optimal features were mapped to the knee-ankle joint angles using a BP network. The root mean square error (RMSE) and Pearson correlation coefficient were used as evaluation indexes. The results show that the feature extraction method with the GBDBN proposed in this paper is superior to the conventional method, and the features extracted by the GBDBN and the BP network are a better combination for building prediction models.