International Journal of Research in Advanced Electronics Engineering
2024, Vol. 5, Issue 2, Part A
A hybrid deep learning model for real-time classification of gestures from high-density EMG signals
Author(s): Nguyen Van An, Tran Thị Bích Ngoc and Le Quang Dung
Abstract: High-density electromyography (HD-EMG) has emerged as a powerful tool for capturing detailed neuromuscular activity, enabling precise gesture recognition for various applications, including prosthetics and virtual reality interfaces. However, real-time classification of gestures from HD-EMG signals remains challenging due to high-dimensional data, signal variability across users, and the computational demands of advanced models. This study aimed to develop a hybrid deep learning model combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks for accurate and real-time classification of six predefined hand gestures: Flexion, Extension, Abduction, Adduction, Wrist Rotation, and Finger Tapping. A dataset was collected from 30 participants using a 64-channel HD-EMG grid, with signals pre-processed using a 4th-order Butterworth band-pass filter (20-450 Hz). Spatial and temporal features were extracted using CNN and LSTM layers, respectively. Model performance was validated on an embedded NVIDIA Jetson Nano device for real-time implementation. Results revealed consistently high classification accuracy across gestures, with Extension achieving the highest mean accuracy (92.74%) and Abduction the lowest (91.26%). Precision, Recall, and F1-Score metrics demonstrated balanced performance, and one-way ANOVA analysis (p=0.623) confirmed no statistically significant differences in accuracy between gestures. The findings highlight the hybrid model's robustness, low latency, and suitability for real-time applications. Practical recommendations include integrating recalibration protocols, exploring lightweight neural architectures for improved energy efficiency, and expanding datasets for better generalizability. This study bridges the gap between theoretical advancements and practical requirements, offering a scalable and adaptable solution for real-world HD-EMG gesture recognition systems. Future research should focus on multimodal data integration and adaptive learning frameworks to further enhance performance and usability.
DOI: 10.22271/27084558.2024.v5.i2a.42
Pages: 23-28 | Views: 48 | Downloads: 17
Download Full Article: Click Here
How to cite this article:
Nguyen Van An, Tran Thị Bích Ngoc, Le Quang Dung. A hybrid deep learning model for real-time classification of gestures from high-density EMG signals. Int J Res Adv Electron Eng 2024;5(2):23-28. DOI: 10.22271/27084558.2024.v5.i2a.42
