ENHANCED sEMG SIGNALS PROCESS WITH TWO-STREAM CNN ON GESTURE CLASSIFICATION

Abstract

sEMG signals show huge potential in implementing control systems of mechatronics devices, and small muscle movements can generate enough sEMG signals to achieve desired control operations. Traditional methods on the sEMG signals process do not robustly decipher important information to distinguish subtle differences in gesture classification. This paper applied a novel deep learning method, a two-stream CNN architecture called Mario CNN, to process NinaPro DBI data on gesture classification and achieved higher accuracy than a single stream CNN.