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

Chen Chen,Yang Yu,Xinjun Sheng,Jianjun Meng,Xiangyang Zhu

DOI: https://doi.org/10.1109/TBME.2023.3234642

Abstract:Objective: Surface electromyography (EMG) decomposition techniques have been developed to decode motor neuron activities non-invasively in the past decades, showing superior performance in human-machine interfaces such as gesture recognition and proportional control. However, neural decoding across multiple motor tasks and in real-time remains challenging, which limits its wide application. In this work, we proposed a real-time hand gesture recognition method by decoding motor unit (MU) discharges across multiple motor tasks ( 10) in a motion-wise way. Methods: The EMG signals were first divided into numerous segments related to motions. The convolution kernel compensation algorithm was applied for each segment individually. The local MU filters, which indicate the MU-EMG correlation for each motion, were calculated iteratively in each segment and reused for global EMG decomposition to trace the MU discharges across motor tasks in real-time. The motion-wise decomposition method was applied on the high-density EMG signals recorded during twelve hand gesture tasks from eleven non-disabled participants. The neural feature of discharge count was extracted for gesture recognition based on five common classifiers. Main results: On average, 164 ±34 MUs were identified for twelve motions from each subject, with a pulse-to-noise ratio of 32.1 ±5.6 dB. The average time cost of EMG decomposition in a sliding window of 50 ms was less than 5 ms. The average classification accuracy using a linear discriminant analysis classifier was 94.6 ±8.1%, which was significantly higher than that of a time-domain feature called root mean square. The superiority of the proposed method was also validated with a previously published EMG database comprising 65 gestures. Conclusion and significance: These results indicate the feasibility and superiority of the proposed method for MU identification and hand gesture recognition across multiple motor tasks, extending the potential applications of neural decoding in human-machine interfaces.