Surface EMG-Based Instantaneous Hand Gesture Recognition Using Convolutional Neural Network with the Transfer...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Surface electromyography (sEMG) enables human-computer interaction, but current gesture recognition methods struggle with generalization to new users and gestures.
• This limitation hinders widespread application and increases the training burden for new users.

Purpose of the Study:

• To propose and evaluate a transfer learning (TL) strategy for instantaneous sEMG-based gesture recognition.
• The goal is to enhance the generalization performance of convolutional neural network models for new subjects and gestures, while reducing training time.

Main Methods:

• A transfer learning (TL) strategy was developed using a convolutional neural network architecture.
• The strategy was evaluated on the CapgMyo and NinaPro DB1 datasets, comparing performance against a non-transfer learning (non-TL) approach.

Main Results:

• The proposed TL strategy improved average accuracy for new subject recognition by 18.7% and new gesture recognition by 8.74% (with up to three repeated gestures).
• TL reduced the required training time by a factor of three.
• Experiments confirmed the transferability of spatial features and the effectiveness of the TL strategy.

Conclusions:

• The developed TL strategy effectively improves the generalization ability of sEMG-based gesture recognition systems.
• This approach offers a viable solution for enhancing accuracy and reducing the training burden in human-computer interaction applications.