Real-Time Hand Gesture Recognition by Decoding Motor Unit Discharges Across Multiple Motor Tasks From Surface...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Surface electromyography (EMG) decomposition decodes motor neuron activity non-invasively for human-machine interfaces.
• Real-time neural decoding across multiple motor tasks remains a significant challenge, limiting widespread application.

Purpose of the Study:

• To propose a real-time hand gesture recognition method using a motion-wise decoding of motor unit (MU) discharges.
• To enhance neural decoding capabilities for human-machine interfaces by addressing multi-task and real-time limitations.

Main Methods:

• EMG signals were segmented by motion, with convolution kernel compensation applied individually.
• Local MU filters were iteratively calculated and reused for global EMG decomposition to trace MU discharges in real-time.
• A motion-wise decomposition method was applied to high-density EMG data from twelve hand gesture tasks, extracting discharge count for classification.

Main Results:

• An average of 164 ±34 MUs were identified per subject with a high pulse-to-noise ratio (32.1 ±5.6 dB).
• EMG decomposition averaged less than 5 ms within a 50 ms sliding window, enabling real-time application.
• Linear discriminant analysis achieved an average classification accuracy of 94.6 ±8.1%, outperforming traditional time-domain features and validated on a large EMG database.

Conclusions:

• The proposed motion-wise method demonstrates feasibility and superiority for MU identification and multi-task hand gesture recognition.
• This advancement extends the potential applications of neural decoding in sophisticated human-machine interfaces.