TraHGR: Transformer for Hand Gesture Recognition via Electromyography.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Deep learning for hand gesture recognition (HGR) using surface electromyogram (sEMG) signals shows promise for advanced myoelectric prostheses.
• Classifying hand movements from sparse multichannel sEMG signals remains challenging for current deep learning models, often limited by single-model approaches that struggle with representative feature extraction.

Purpose of the Study:

• To address the limitations of existing deep learning models in HGR using sEMG signals.
• To propose a novel hybrid deep learning framework, Transformer for Hand Gesture Recognition (TraHGR), that leverages transformer architecture for improved feature extraction and classification accuracy.
• To evaluate the performance of TraHGR against state-of-the-art methods on a standard dataset.

Main Methods:

• Developed a hybrid deep learning architecture, TraHGR, integrating parallel processing paths within a transformer framework.
• The architecture features a fusion center using a linear layer to combine features from parallel modules.
• Evaluated TraHGR on the Ninapro DB2 dataset, comprising sEMG signals from 40 users performing 49 distinct gestures under real-life conditions.

Main Results:

• The TraHGR architecture achieved high recognition accuracies: 86.00% (DB2, 49 gestures), 88.72% (DB2-B, 17 gestures), 81.27% (DB2-C, 23 gestures), and 93.74% (DB2-D, 9 gestures).
• These results represent significant improvements over the state-of-the-art, with accuracy gains of 2.30%, 4.93%, 8.65%, and 4.20% for the respective datasets.
• Experiments confirmed the superior performance of the hybrid TraHGR architecture compared to its individual parallel paths.

Conclusions:

• The proposed TraHGR hybrid framework effectively addresses the challenges of HGR using sparse sEMG signals.
• The transformer-based architecture demonstrates superior capability in extracting representative features, leading to significant accuracy improvements in myoelectric prosthesis control.
• TraHGR represents a substantial advancement in the field, offering enhanced performance for advanced prosthetic limb applications.