Performance Evaluation of Convolutional Neural Network for Hand Gesture Recognition Using EMG.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Surface electromyography (sEMG) shows promise for upper limb prosthesis control.
• Current deep learning models achieve high offline accuracy but face real-time application challenges due to system optimization delays.
• There is a need for optimized deep learning architectures with fine-tuned hyperparameters for improved real-time performance.

Purpose of the Study:

• To investigate the effect of hyperparameters on convolutional neural network (CNN) performance for decoding hand gestures from sEMG data.
• To identify optimal hyperparameter settings for robust and stable myoelectric control.
• To compare the performance of different hand gestures in gesture recognition.

Main Methods:

• Implemented a convolutional neural network (CNN) to decode hand gestures from sEMG data.
• Recorded sEMG data from 18 subjects performing various hand gestures.
• Systematically analyzed the impact of hyperparameters, including learning rate and number of epochs, on recognition accuracy.

Main Results:

• A learning rate of 0.0001 or 0.001 with 80-100 epochs significantly outperformed other settings (p < 0.05).
• Certain hand gestures, including close hand, flex hand, extend hand, and fine grip, demonstrated consistently higher performance.
• The deep learning approach showed potential for more robust performance compared to traditional machine learning algorithms.

Conclusions:

• Fine-tuning hyperparameters is crucial for optimizing deep learning models in sEMG-based prosthetic control.
• Identifying and prioritizing high-performing hand gestures can lead to more robust and stable myoelectric control systems.
• Deep learning offers a promising alternative to traditional methods for advanced prosthetic control.