Multi-branch deep learning neural network prediction model for the development of angular biosensors based on sEMG.

Area of Science:

• Robotics and Human-Machine Interaction
• Biomedical Engineering
• Machine Learning and Artificial Intelligence

Background:

• Accurate human gait motion intention recognition is crucial for lower extremity exoskeleton robots to synchronize with natural user movement.
• Surface electromyography (sEMG) is commonly used for motion intention recognition, but its high-dimensional and nonlinear nature presents challenges.
• Deep learning neural networks excel with sEMG data, yet different network architectures have varying strengths for specific data types.

Purpose of the Study:

• To establish a multi-branch deep learning neural network model for accurate gait recognition and joint angle estimation.
• To quantify the performance of this model in gait recognition and joint angle prediction using sEMG data.

Main Methods:

• Collected sEMG and plantar pressure data during human walking.
• Filtered and denoised signals, then extracted time-domain and frequency-domain features.
• Developed a multi-branch deep learning model utilizing feature sensitivity differences for gait cycle and joint angle prediction.

Main Results:

• The multi-branch model achieved high accuracy in gait recognition, with an average of 92.16% (ranging from 90.11% to 95.42%).
• The model accurately estimated joint angles with an average error of 3.19 degrees.
• Successfully integrated time-domain and frequency-domain features for reliable predictions.

Conclusions:

• The developed model accurately recognizes human gait and predicts joint angles, enabling effective lower limb motion intention recognition.
• The model can be integrated with sEMG sensors to create angular biosensors for real-time joint angle prediction.
• This technology enhances the potential for seamless human-robot interaction in exoskeleton applications.