SEMG-based continuous estimation of knee joint angle using deep learning with convolutional neural network

Human-machine interaction is a key component in the wearable robotics field. Because surface electromyography (sEMG) generates prior to the corresponding motion and reflects the motion intention directly, sEMG-based motion intention recognition can achieve better human-machine interaction and has been widely used in recent years. However, most of the relevant researches are concentrated on the discrete-motion classification which can not be used for smooth control of wearable robots. Thus, in this paper, an improved feature-based convolutional neural networks (CNN) model was proposed for analyzing the sEMG-based continuous estimation of knee joint angle. The normal walking experiments with six sEMG channels acquired system and optical motion capture system were carried out to analyze actual and desired knee angle. The sEMG-based continuous-motion regressions of knee joint angle obtained by the proposed model and other two existing neural network models, i.e. original data-based CNN model and back propagation neural network (BPNN) model were calculated and compared with experimental ones. The results showed that the proposed model can predict knee angle with a higher level of accuracy compared to BPNN and original data-based CNN models.