Robust gesture recognition based on attention-deep fast convolutional neural network and surface electromyographic...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Surface electromyographic (sEMG) signals are crucial for human-machine interfaces (HMI) and myocontrol.
• High-density (HD) electrodes offer richer sEMG data compared to sparse multi-channel (SMC) electrodes.
• Electrode displacement or damage can significantly degrade gesture recognition accuracy in HD-sEMG systems.

Purpose of the Study:

• To develop a robust model for myocontrol that minimizes the impact of HD electrode shifts and damage.
• To leverage both spatial and temporal characteristics of HD-sEMG signals for improved gesture recognition.
• To enhance the stability and accuracy of HMI systems utilizing HD-sEMG.

Main Methods:

• Proposed an attention deep fast convolutional neural network (attention-DFCNN) model.
• Integrated spatial and temporal features from HD-sEMG signals.
• Evaluated the model's performance against classical and deep learning methods using a dataset with simulated electrode shifts (10 mm) and damage (6 channels).

Main Results:

• The attention-DFCNN model achieved significantly higher average accuracy (0.942 ± 0.04) compared to LSDA, CNN, TCN, LSTM, attention-BiLSTM, Transformer, and Swin-Transformer under electrode shift conditions.
• The model demonstrated superior performance without requiring pre-training.
• The proposed method effectively improved recognition rates despite electrode grid alterations.

Conclusions:

• The attention-DFCNN model offers a robust solution for myocontrol using HD-sEMG, particularly in scenarios with electrode displacement or damage.
• Combining spatial and temporal features is a key strategy for enhancing the reliability of sEMG-based HMI.
• This approach holds significant potential for improving the practical application of advanced myocontrol systems.