Modeling of Controller for Motor-Controlled Prosthetic Hand Based on Machine Learning Strategy in Classifying Two-Channel Surface EMG Signals

This research study presents a computationally improved system using a pattern recognition (PR) algorithm to classify fingers movement based on data acquired by a surface EMG (sEMG) sensor when muscle is contracting. Ten subjects were involved in this investigation where the forearm’s muscle activities were acquired using two-channel sEMG placed at the flexor digitorum superficialis and extensor digitorum muscles. The focus of this study is to integrate the sensor with servo motors to control the movement of artificial limbs or prosthetics based on which muscle is at work. The work involved signal processing on raw sEMG signals, followed by multiple time domain feature extraction (TD). sEMG signal is then segmented using an overlapping window of size 250 ms and increments of 50 ms. The feature extraction was used to build up the convolutional neural network (CNN) which is used to train the classes of fingers movement. The dataset was split into 80% for training and 20% for testing the classifier. The CNN model was able to incorporate most of the data’s variability while maintaining an average classification accuracy of 98%.