A Study Of Feature Extraction From Semg Singal Based On Entropy

Providing an easy noninvasive access to studying the myoelectric features of the neuromuscular activation associated with muscle contraction, surface EMG signal has been widely applied in clinical medicine, rehabilitation medicine, sports medicine, neurophysiology, ergonomics, and etc.. The study of feature extraction from SEMG signal boosts the application. Starting from complexity analysis based on the concept of entropy, we did some valuable research both theoretically and experimentally on SEMG feature extraction and fatigue evaluation. The main creative work is as follows:Considering that SEMG signal is always short and has low singal-to-noise ratio, we proposed fuzzy approximate entropy—FuzzyEn. FuzzyEn goes beyond the application limitation in physiological signals of most nonlinear dynamic measures such as fractal dimension, Lyapunov exponent, and K-S entropy. It also avoids some problems facing ApEn and SampEn by importing the concept of fuzzy sets. Comparison analysis of several typical datasets shows that, FuzzyEn not only owns stronger relative consistency and less dependence on data length, which means less bias, but also achieves freer parameter selections and more robustness to noise. Experiments on the feature extraction from different motion SEMG also show that FuzzyEn can characterize different motions more efficiently.In this thesis, we put forward a method of studying the evolvement patterns (EP), as well as an index Dreg to measure the regularity change of a time series. Through a sliding observation window, the relative probability of EPs of a time series can be monitored as a function of time. Summing up the probability of all those patterns representing regularity evolvement, we get the index Dreg. The higher the Dreg is, the more regular the time series appears. By studying changes in the relative probability of EPs and Dreg of SEMG signals during both static sustained contractions and dynamic repetitive contractions, we find out that SEMG tends to be more regular with fatigue progressing. The finding confirms the hypothesis in the previous studies and provides strong foundation for assessing fatigue by complexity analysis.By investigating changes in band spectrum energy during fatigue process, we find that different frequency bands show different changes: there's a band whose relative band spectrum energy (RBE) almost remains constant with the passage of time, and we call it'changeless band'; the RBE of the band whose frequency is lower than that of the'changeless band'increases with time; whereas the RBE of the band whose frequency is higher than that of the'changeless band'decreases with time. Since SEMG spectrum energy distributes unequally among different frequency bands and shifts among the bands when the muscle status changes, band spectrum entropy (BE), which combined frequency analysis with complexity analysis, was proposed to extract features from SEMG during muscle fatigue. BE can characterize changes in the SEMG complexity as well as its frequency shift during muscle fatigue. Utilizing the result of FFT directly, BE wins its advantage of simple and online computation. Compared with traditional EMG fatigue indexes, BE shows better reliability when estimating muscle fatigue.The thesis also gives a study of the effect of vibration stimulation on fatigue process. Through the comparison of changes in SEMG during a control experiment where both vibration training and nonvibration training were conducted to the same group of atheletes at different stages, a conclusion was drawn that fatigue progressed more slowly under vibration training. The possible physiological mechanism was also discussed. The study provides positive support both theoretically and practically for the disputed topic whether vibration stimulation benefits the recuperation & strengthening of muscle.