EMG Feature Reduction Technique For Optimal Accuracies

Abstract

The recording of electrical activity which is produced by muscles is known as an Electromyogram or Electromyographic (EMG) signal. The generation of electric current during the contraction of muscles is measured by it. The insight of muscles dynamics and neural activation is provided by EMG signal and is thus significant for several different applications, such as the studies that try to identify deficiencies of neuromuscular. For researchers and practitioners, signal of EMG is very important to observe and evaluate the muscles condition and the outcome of the rehabilitation training. The signal of EMG features precision and factors vary correspondingly with signal of muscle, fatigue, and features. The hand movements classification based on signals of surface electromyography (sEMG) is a key problem in assistive devices and rehabilitation system control. The classification of movements of hand from sEMG is a method that has different applications like rehabilitation, interaction of human-machine and prosthetic control. The main issue is that by using increase number of features and channels of EMG in order to maximize the number of control commands can produce a feature vector of high dimensional. The major challenge is the process development to predict the current motion robustly and accurately based on incoming sEMG data. To overcome the problems of accuracy and computation linked with high dimension vector, feature reduction technique is applied that converts the data to low dimension vector space with a bit loss of valuable informative data. The aim of this thesis is to extract features and to reduce its dimensionality using PCA to improve classification success rate and compare the findings of classification accuracy before and after applied PCA technique. Six different classifiers were used on the EMG data before and after using feature reduction technique and a comparative study of finding is presented in this thesis study.