IFIANet: Frequency Attention Network for Time-Frequency in sEMG-Based Motion Intent Recognition.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Exoskeleton systems need precise movement intention recognition for natural human-machine interaction.
• Surface electromyography (sEMG) signals offer pre-movement neural activation data crucial for intention recognition.

Purpose of the Study:

• To develop an advanced deep learning framework, IFIANet, for improved sEMG-based movement intention recognition.
• To enhance the accuracy and robustness of sEMG signal processing for exoskeleton control.

Main Methods:

• A CNN-TCN based network was developed for efficient spatiotemporal feature learning from sEMG signals.
• An innovative Frequency-Informed Integration Attention (IFIA) module was designed to integrate global frequency information, improving feature discriminability.

Main Results:

• The IFIANet framework demonstrated stable performance across various prediction times.
• An average recognition accuracy exceeding 82% was achieved across nine participants on the MyPredict1 dataset.
• The model effectively fused local temporal-frequency features with global frequency priors.

Conclusions:

• IFIANet provides an efficient and reliable method for sEMG-based movement intention recognition.
• The proposed framework significantly advances the intelligent control of lower limb exoskeleton systems.