Inter-subject Variability Evaluation of Continuous Elbow Angle from sEMG using BPNN

Abstract

As a non-invasive approach, surface electromyographic (sEMG) signal has great potential for application in human-robot interfaces, such as the upper-limb exoskeleton rehabilitation device. However, due to the differences in activity level of muscles, there exists high inter-subject variability. In this work, the influence of inter-subject variability for elbow continuous motion is evaluated through a shallow neural network (BPNN), and user-dependent and user-independent models are established respectively. In user-dependent model, training and testing sets are from the same subject, new set of the same person as during training is used as the input of network. The user-independent models are constructed by the same user and another additional user to determine inter-subject variability in model construction. To evaluate the degree of inter-subject variability, evaluation criteria and statistical method are adopted. Through the prediction results, and further the value of evaluation criteria and the plot of statistical method, it can be seen that the inter-subject variability on sEMG has a huge impact on the regression of elbow continuous angle, which can provide reference for the future study of building sEMG generalized modeling to estimate elbow angles.