Muscle Fatigue Algorithm Using Unsupervised Weighted Time-Frequency Fusion

Most existing classifications of muscle fatigue levels rely on subjective scoring or fused multi-modal information. This study proposes an objective characterization method for biceps fatigue based on root mean square (RMS) and average amplitude (AvgAmp) features of surface electromyography (sEMG) signals, combined with the root mean square error (RMSE) of their polynomial fitting residuals. Dynamic changes of sEMG features are investigated during static biceps contraction and developed an unsupervised feature contribution algorithm independent of subjective labels is developed. By analyzing windowing effects, a 200 ms window is identified as optimal for real-time processing. The experimental results indicate that the fatigue curve was derived from a weighted fusion of RMS, AvgAmp, and their RMSE, consisting with existing findings on muscle fatigue characterization.