Detection of muscle fatigue by fusion of agonist and synergistic muscle sEMG signals

Muscle fatigue detection has a wide range of applications in the field of rehabilitation medicine. The existing methods only collect a single agonist signal for detection, whose accuracy and real-time detection are usually poor. To handle this issue, in this study collects a dataset containing both agonist and synergistic muscle surface electromyography (sEMG) signals, by analyzing the synergistic working principle of muscles. Through the fusion processing of different muscle group signals, the significance of detection of muscle state changes is improved. Moreover, the impact of signal non-stationarity and non-linearity on fatigue detection is reduced. Based on the collected dataset, a multichannel fusion recurrent attention network (MFRANet) is proposed. First, MFRANet enhances local anti-interference ability by fusing multi-channel EMG signals and reduces the impact of single channel signal noise on the overall detection performance. Second, MFRANet analyzes the signal from two dimensions, namely time domain and space domain. A gating mechanism is used to enhance the complex time correlation between channels, and an attention mechanism is employed to reconstruct the nonlinear relationship between channels, thus improving the generalization. Experiments show that the proposed signal fusion method of agonist and synergistic muscle sEMG signals significantly improves the accuracy of muscle fatigue detection, as well as reducing processing time compared to traditional machine learning methods.