Statistical Class Separation Using sEMG Features Towards Automated Muscle Fatigue Detection and Prediction

Surface Electromyography (sEMG) activity of the biceps muscle was recorded from ten subjects. Data were recorded while subjects performed isometric contraction until fatigue. The signals were segmented into three parts (Non-Fatigue, Transition-to-Fatigue and Fatigue), assisted by a fuzzy classifier using arm angle and arm oscillation as inputs. Nine features were extracted from each of the three classes to quantify the potential performance of each feature, also aiding towards the differentiation of the three classes of muscle fatigue within the sEMG signal. Percent change was calculated between Non-Fatigue and Transition-to-Fatigue and also between Transition-to-Fatigue and Fatigue classes. Estimation of relative class overlap using Partition Index approach was used to show features that can best distinguish between the three classes and quantifying class separability. Results show that the selected dominant frequency best discriminate between the classes, giving the highest average percent change of 159.37% and 64.75%. Partition Index showed small values confirming the percent change calculations.