Optimized spatial filter mask using genetic algorithm and system identification for modeling sEMG and finger force signals

摘要

This paper presents the investigation of the use of System Identification (SI) for modeling sEMG-Finger Force relation in the pursuit of improving the control of prosthetic hands. Finger force and sEMG data is generated by having the subject perform a number of random motions of the ring finger to simulate various force levels. Post-processing of the sEMG signal is performed using spatial filtering. The linear and nonlinear spatial filters are compared based on the 'kurtosis' improvements and also are based on the fit values of the models obtained using system identification. Some of the spatial filter masks are optimized using a Genetic Algorithm (GA) for the constrained and unconstrained cases. The resulting model fit value is utilized to serve as the cost function. The results are compared to the reported filter mask values in the literature. The unconstrained GA based filter mask values and in some instances the constrained GA based mask values performed better than the filter masks reported in literature.