Implementation of Supervised Machine Learning on Embedded Raspberry Pi System to Recognize Hand Motion as Preliminary Study for Smart Prosthetic Hand

Abstract

EMG signals have random, non-linear, and non-stationary characteristics that require the selection of the suitable feature extraction and classifier for application to prosthetic hands based on EMG pattern recognition. This research aims to implement EMG pattern recognition on an embedded Raspberry Pi system to recognize hand motion as a preliminary study for a smart prosthetic hand. The contribution of this research is that the time domain feature extraction model and classifier machine can be implemented into the Raspberry Pi embedded system. In addition, the machine learning training and evaluation process is carried out online on the Raspberry Pi system. The online training process is carried out by integrating EMG data acquisition hardware devices, time domain features, classifiers, and motor control on embedded machine learning using Python programming. This study involved ten respondents in good health. EMG signals are collected at two lead flexor carpi radialis and extensor digitorum muscles. EMG signals are extracted using time domain features (TDF) mean absolute value (MAV), root mean square (RMS), variance (VAR) using a window length of 100 ms. Supervised machine learning decision tree (DT), support vector machine (SVM), and k-nearest neighbor (KNN) are chosen because they have a simple algorithm structure and less computation. Finally, the TDF and classifier are embedded in the Raspberry Pi 3 Model B+ microcomputer. Experimental results show that the highest accuracy is obtained in the open class, 97.03%. Furthermore, the additional datasets show a significant difference in accuracy (p-value <0.05). Based on the evaluation results obtained, the embedded system can be implemented for prosthetic hands based on EMG pattern recognition.