Simulation of human lower limb skeletal muscle motion based on deep learning

The lower limb movement is the basic human movement, involving complex issues such as balance and walking. This research is conducted to realize the the real-time recognition and regulation of human lower limb skeletal muscle movement the deep learning concept. The article proposes a regularized transfinite learning machine, BP neural network and support vector machine regression algorithm to construct models for knee joint angle estimation. Three different deep learning algorithms are evaluated for the estimation of knee joint angle of lower limbs by using surface electromyography (sEMG). This paper systematically studies the human lower limb skeletal muscle motion simulation model based on deep learning. A large number of experiments and triple cross validation is performed and the outcomes are obtained using various indicators like root mean square error, correlation coefficient and training time. The root mean square error and correlation coefficient of regularized transfinite learning machine and support vector machine are less comparative to the BP neural network. Also, the training time of regularized transfinite learning machine model is the shortest, which is higher by two orders of magnitude than BP neural network model, about 1% of the neural network. The regularized transfinite learning machine algorithm provides feasible and promising outcomes establishing its application capability for real-time recognition and control of human lower limb skeletal muscle movement in complex environment. The comparative investigation is done in terms of training time and it was found that the training time of proposed RELM model is the shortest, which is higher by two orders of magnitude than BP neural network model. The outcomes obtained for the proposed RELM based method are optimal among the other counterparts and it has vast applicability for real-time recognition and regulation of lower limb motion.