Difference analysis of musculation and estimation of sEMG-to-force in process of increasing force and decreasing force

The movement of the human limb is driven by muscle contraction. Surface electromyography (sEMG) is a weak bioelectric activity generated during muscle contraction, which reflects information regarding muscle activation. The increasing force and the decreasing force are reverse processes. Investigating the difference in musculation between the two processes and establishing an input–output model between sEMG and force can clarify the biodynamics mechanism of the human body. In the study, we try to find the truth about the difference in musculation using sEMG signal in the process of increasing and decreasing force, and create a model of the relationship between sEMG and force. A synchronous data acquisition device is used to collect force and sEMG signals, including the raw sEMG signal and its envelope signal. A new method for extracting the feature of the sEMG signal based on the spectrogram is introduced. Up to sixteen features are extracted from the sEMG signal, and their performances are evaluated. The experimental results indicate that sliding mean filtering can significantly improve feature performance. A processing means of isometric force and sEMG feature is proposed. Difference in musculation about force-increasing and force-decreasing is detailedly analyzed by statistical T-test. We come to the conclusion that the sEMG signal evoked via musculation is not exactly the same in the two processes, with a more significant difference when the muscle contraction strength is weaker, and a less significant difference when the muscle contraction strength is stronger. Finally, five regression models are used for sEMG-to-force estimation, and their performances are compared separately. The experimental results show that the DNN exhibits the best performance, achieving a RMSE of 12.782 and a R2 of 0.911.