Finger-wrist Gesture Recognition based on Surface Electromyography and Inertial Signals

Surface electromyography (sEMG) is often used as an ideal control source to drive an intelligent prosthetic hand. The study aims to distinguish 24 types of finger-wrist joint movements based on 8-channel sEMG signals and 6-channel inertial signals. Considering the dynamic characteristics of signal, we proposed a new method named segmented sliding window (SSW) for feature extraction. Compared with fixed window (FW) and sliding window (SW), the new method performed better. Then fisher score (FS) and sequential forward selection (SFS) were employed for feature selection. And a support vector machine (SVM) incorporating majority voting method was adopted as a classifier. The results from 14 subjects indicated that the combination of SSW and SFS got best performance, whose classification accuracy can achieve to 96.61% only base on sEMG signals within length of 300ms. Classification accuracy could be raised to 97.56<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> with inertial signals added, which verified inertial signal is helpful for motion recognition.