Fatigue Classification and Onset estimation using Surface EMG Signals during Strength Training

Muscle fatigue onset detection and estimation have applications in supporting athletes during strength training. In this paper, the problem of surface electromyography (sEMG) based binary classification of fatigue and non-fatigue of human muscle activity during strength training is addressed. Further, the learned model is used to determine the fatigue onset using sEMG signals corresponding to a training activity. A novelty of this work is the analysis of the proposed machine learning method under different measurement conditions, such as varying postures, experience levels and lifting loads. A novel sEMG dataset titled ElectroMyography Analysis of Human Activities - DataBase 6 (EMAHA-DB6) is developed for strength training activities. EMAHA-DB6 includes sEMG signals collected from the seven prominent muscle sites of the right arm of 11 subjects engaged in strength training. The participants performed five distinct training exercises using three load weights in two body postures. A set of wavelet and time domain features are extracted and the support vector machines are used to classify them. Based on the binary testing results from a given time series, the fatigue onset is estimated using the majority voting. The SVM model is compared with other classifiers in terms of testing accuracy in the mentioned measurement scenarios. The SVM has outperformed the other models with a test accuracy of 86.5% under all conditions scenario. In terms of the impact of measurement conditions, the SVM has 94% in the low load condition, 88.5% in the stand posture and nearly 88% in the case of subjects with intermediate training status. Finally, in the proposed approach, the estimation of the fatigue onset has an average relative error of 12%.