Electroencephalogram and surface electromyogram fusion-based precise detection of lower limb voluntary movement using...

Area of Science:

• Neuroscience and Biomedical Engineering
• Human-Robot Interaction
• Signal Processing

Background:

• Fusion of electroencephalogram (EEG) and surface electromyogram (sEMG) is crucial for detecting human movement intention in human-robot interaction.
• Existing fusion methods have limitations due to the unclear internal relationship and response time differences between EEG and sEMG signals.

Purpose of the Study:

• To investigate a precise fusion method using a CNN-LSTM model for detecting lower limb voluntary movement by integrating EEG and sEMG signals.
• To analyze and compensate for the response time difference between EEG and sEMG signals to enhance fusion accuracy.

Main Methods:

• EEG and sEMG signal processing stages were analyzed to estimate the response time difference using symbolic transfer entropy.
• A hybrid CNN-LSTM model was developed, utilizing both data and feature fusion of EEG and sEMG signals.
• The model incorporated the estimated time difference for improved decoding of lower limb voluntary movement.

Main Results:

• The estimated time difference between EEG and sEMG signals ranged from 24-26 ms, with calculated values between 25-45 ms.
• Offline experiments showed data fusion achieved over 95% accuracy, significantly outperforming feature fusion.
• Online experiments demonstrated an average accuracy exceeding 87% for the data fusion-based CNN-LSTM model across all subjects.

Conclusions:

• The time difference between EEG and sEMG significantly influences the accuracy of lower limb voluntary movement detection.
• The proposed CNN-LSTM model effectively integrates EEG and sEMG signals, achieving high performance in detecting lower limb voluntary movement.
• This approach provides a stable and reliable foundation for lower limb exoskeleton control in human-robot interaction.