sEMG Data Expansion for Accurate Posture Classification

Abstract

The paper presents methods to expand data acquired from a multi-channel sEMG fabric sensor for the dexterous control of robotic prosthesis. It is able to improve a variety of pattern recognition performance in spite of fewer data and less computational time. A multilayer perceptron (MLP) is utilized for the classification of eight postures in order to compare several methods regarding the data expansion such as data expanded with normal distribution (N-dist), data expanded with median operations, and data expansion with median plus normal distribution. Of the methods, an accuracy achieved using the data expanded with median plus normal distribution arrives at 99.32% as the highest, followed by the expansion using median, the expansion using normal distribution.