Proposal of a Hardware SVM Implementation for Fast sEMG Classification

Abstract

There are many studies in the machine learning field for classifying movements using electromyography (EMG) signals and some of them achieve high classification rates. The cost for good performance although, is the long time necessary to train the classifiers. This work proposes a multi-class Support Vector Machine (SVM) running in hardware. It is part of a bigger project which aims to train and classify movements maintaining good classification rates in reduced time. For testing the hardware solution, 12 channels of RMS extracted from surface EMG (sEMG) data available at Ninapro Database were used to classify 17 movements. Results reveal a 59.2% mean classification rate for the proposed system implemented in FPGA when running few milliseconds against 58.2% obtained by Matlab in the same scenario.