A Two-Stage GA-based sEMG Feature Selection Method for User-Independent Continuous Estimation of Elbow Angles

Surface electromyography (sEMG) has great application potential in upper extremity rehabilitation exoskeleton. The accurate identification of elbow motion angle is crucial for the sEMG-controlled upper limb exoskeleton rehabilitation system. However, the existing high inter-subject variability in sEMG limits the generality of the model built through learning algorithms among different subjects. Aiming at the above problem, a feature selection method based on a two-stage genetic algorithm (GA) is proposed for the accurate user-independent estimation of continuous movements. And the information theory-based minimum redundancy maximum relevance criterion serves as the fitness function to evaluate the goodness of subsets. The effectiveness of the proposed method is verified by estimating the motion angle of the elbow joint using the collected sEMG data of 6 participants. The prediction performance is compared with that before the two-stage GA-based feature selection, and different metrics and statistical analyses are adopted to evaluate the results. The estimation angle error calculated after two-stage GA-based feature selection is controlled within 10°, which shows the feasibility of the proposed method for the accurate user-independent estimation of continuous joint movements.