Automated Recognition of Hand Gestures From Multichannel EMG Sensor Data Using Time–Frequency Domain Deep Learning for IoT Applications

Abstract

The automated recognition of hand gestures from multichannel electromyogram (MEMG) sensor data is essential for human-computer interaction, prosthetic control, rehabilitation, and biometric-related applications. This letter proposes a time-frequency domain deep neural network (TFDDNN)-based approach to recognize hand gestures using MEMG recordings. The MEMG recordings are segmented into frames, and the average of all channel information (mean EMG signal) for each frame is evaluated. The continuous wavelet transform (CWT) is applied to the mean EMG signal to obtain the joint time-frequency representation (TFR). The TFR-based image of the mean EMG signal is used as the input to a deep representation learning network (DRLN) model to recognize hand gestures. Two publicly available databases are utilized to evaluate the performance of the proposed TFDDNN approach. The results show that the proposed TFDDNN has obtained overall accuracy values of 92.73% and 80.33% using MEMG signals from database 1 and database 2 for multiclass-based recognition of hand gestures. The suggested DRLN model has demonstrated higher accuracy for automated hand gesture recognition than existing deep-learning-based methods. We have deployed the proposed TFDDNN approach in a web application (WAPP) for IoT-enabled real-time recognition of hand gestures using MEMG sensor data.