EMG-Based Hand Gesture Classification with Long Short-Term Memory Deep Recurrent Neural Networks.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Electromyogram (EMG) pattern recognition is crucial for controlling upper-limb prostheses.
• Traditional machine and deep learning models often overlook the temporal dependencies in muscle contractions.
• Convolutional neural networks primarily focus on spatial correlations, limiting their ability to capture dynamic muscle activity.

Purpose of the Study:

• To investigate the efficacy of recurrent deep learning networks, particularly Long Short-Term Memory (LSTM), for EMG signal classification.
• To leverage the capability of recurrent networks to learn long-term and non-linear dynamics inherent in time-series EMG data.
• To enhance the control strategies for upper-limb prostheses by improving EMG classification accuracy.

Main Methods:

• Utilized a Long Short-Term Memory (LSTM) neural network for multiclass classification of EMG signals.
• Classified six different grip gestures across three force levels (low, medium, high) from nine amputees.
• Extracted four distinct feature sets from raw EMG signals and fed them into the LSTM network.
• Evaluated generalization by testing three training approaches: same-force level, cross-force level, and all-force levels training.

Main Results:

• The LSTM-based neural network demonstrated reliable performance in EMG classification.
• Achieved an average classification error of approximately 9% across all nine amputees and force levels.
• Successfully captured the temporal dependencies and non-linear dynamics of EMG signals.
• Indicated the potential for robust prosthesis control through deep learning.

Conclusions:

• Recurrent deep learning networks, specifically LSTM, are highly effective for EMG pattern recognition in upper-limb prosthesis control.
• The LSTM model's ability to learn temporal dynamics leads to reliable classification across varying conditions.
• This study validates the applicability of deep learning, particularly LSTM, for advancing prosthetic limb functionality.