A Novel Feature Assisting in the Prediction of sEMG muscle fatigue towards a wearable autonomous system

Surface Electromyography (sEMG) activity of the biceps muscle was recorded from ten subjects performing isometric contraction until fatigue. The signals were segmented into two parts (NonFatigue and Transition-to-Fatigue). The novel feature was extracted from two established muscle fatigue detection features (Instantaneous Median Frequency and Total Band Power of the signal). The proposed name for this feature is "ID spectro." Initial results of class separation using Davies Bouldin Index (DBI) showed encouraging results of 0.87 on average for all trials. Using the novel feature ID spectro for classification produced an average longitude classification (within subjects) of 85.72% with accuracy 0.77 and a cross classification (across subjects) with an average of 79%. Wilcoxon rank sum test quantified the gross change of both longitudinal and crossclassification, giving a result of (p<;0.0140). Comparison of the ID spectro with other sEMG fatigue features on the same datasets shows a significant improvement of 10.5% (p<;0.0027) on longitudinal classification when using the ID spectro in classifying Non-fatigue and Transition-to-Fatigue segments of the signal.