Multifractal characterization of electromyogram signals
In this thesis, we present an approach to the characterization and feature extraction of the electromyogram (EMG) signals. This approach is based upon the chaotic behaviour of the EMG signals and the existence of the corresponding strange attractors with low embedding dimensions. The multifractal dimensions of the strange attractors underlying this chaotic behaviour provide alternative features for analyzing the EMG signals. The multifractal dimensions describe how the entropy of these strange attractors changes as the hypervolume scales used for calculating the entropy vary. There are several considerations associated with the reconstruction of the strange attractors and the calculation of the multifractal dimensions from a single variable time series. We discuss how the length and the sampling rate of the time series effect the convergence of the multifractal dimensions. We also discuss the effect of high noise levels in increasing the minimum embedding dimension required for the reconstruction of the strange attractors. The EMG signals under study have been obtained from the anterior, posterior, and middle portions of the deltoid and upper trapezius during isometric contractions, using surface electrodes. The multifractal dimensions of these EMG signals are between 0.5 to 1.5. The experimental results show that the positive moment orders of the multifractal dimensions of the EMG signals can be used for discriminating among three functions of deltoid, i.e. abduction, extension, and flexion. The multifractal dimensions of the EMG of the muscle as a prime mover, are 0.3 larger on average, comparing to the muscle as synergist.