Continuous estimation of elbow joint angle by multiple features of surface electromyographic using grey features weighted support vector machine

Surface electromyographic (sEMG) collected from the skin covering muscles in a non-invasive manner reflects human's motion intention. How to estimate the continuous movement from sEMG signals is a significant issue. In this paper, a grey features weighted support vector machine (GFWSVM) is proposed. Continuous estimation of elbow joint angle from weighted feature sequences of sEMG using the GFWSVM is realized to avoid building a complicated biomechanical model describing the relationship between sEMG and elbow joint angle, and different weight values are given to different features of sEMG based on grey correlation degree theory. An experimental platform was built to record sEMG and elbow joint angle data from subjects. The average correlation coefficients (CC) value and the average root mean square (RMSD) value of experimental results by using the GFWSVM were 0.9228±0.0208 and 0.3875±0.0579 respectively. The estimation performance of GFWSVM algorithm was compared with the back-propagation (BP) artificial neural network, the radial basis function (RBF) artificial neural network and the scaled conjugate gradient (SCG) artificial neural network, and the results showed that the GFWSVM algorithm can be used to estimate the human movement intention from sEMG with the best performance.