Abstract:
The radial wrist flexor MMG signals of 10 subjects were acquired when they completed each griping and holding action. The relationship between the degree of dynamic fatigue of muscles and the characteristics of the MMG signals was analyzed by filtering, motion segmentation and feature extraction of MMG signals. For signal filtering, two methods, the wavelet packet decomposition (WP) and empirical mode decomposition (EMD) were adopted. In WP method, the original signal was decomposed into 7 layers with db4 wavelet base, and then 2-100 Hz frequency band signals were superimposed to reconstruct a filtered signal. In EMD method, the original signal was decomposed into intrinsic mode functions (IMFs), then, IMFs in layers 3-6 were superimposed to reconstruct a filtered signal. For motion segmentation, an adaptive unequal segmentation algorithm was proposed based on the variance threshold of the moving windowed signal. The starting point and ending point of each action were determined by calculating the variance value in each moving window, and thus individual action signal was segmented. For feature extraction, firstly, a moving window containing multiple motion signals was used to reconstruct the segmented signal; then mean power frequency (MPF) and median frequency (MDF) were selected as the characteristics of the windowed signal; finally, the characteristics were fitted by the exponential, the quadratic and the linear function, respectively. The results show that when selecting WP for de-noising, MPF as characteristics and exponential function as fitting method, it can reflect the trend of muscle fatigue better than that of the other combinations.