A novel attention-guided ECA-CNN architecture for sEMG-based gait classification

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neuroscience

Background:

• Gait recognition is crucial for identifying neurodegenerative disorders.
• Surface electromyographic (sEMG) signals offer a viable data source for gait analysis.
• Existing methods may benefit from advanced feature extraction and classification techniques.

Purpose of the Study:

• To develop and evaluate a novel gait classification model using sEMG signals.
• To integrate an Efficient Channel Attention (ECA) module with a Convolutional Neural Network (CNN) for enhanced gait recognition.
• To assess the model's effectiveness in detecting gait variations for potential clinical applications.

Main Methods:

• Collected sEMG signal datasets representing various gaits from different individuals.
• Employed a CNN to extract features from the 1D sEMG signals.
• Integrated an ECA module to facilitate cross-channel interaction and refine feature learning.
• Further processed features using subsequent convolutional layers for in-depth signal analysis.

Main Results:

• The proposed ECA-CNN model demonstrated robust performance in gait classification.
• Comparative experiments confirmed the efficacy of the ECA-CNN approach.
• The model achieved a high classification accuracy of 97.75%.

Conclusions:

• The ECA-CNN model shows significant promise for accurate gait recognition using sEMG signals.
• This technology can aid in the early detection of neurodegenerative dysfunction.
• The integration of ECA with CNN offers an effective strategy for analyzing complex biological signals.