Properties of non-linearly scaled wavelets for feature extraction from time series (electromyographic signals)

Wavelet analysis (a time frequency analysis using wavelets) has become a powerful tool in signal processing. Typically the wavelet analysis uses a mother wavelet as a template and scales it linearly to obtain a set of similar wavelets. Time frequency space is subdivided by tiling the space optimally and assigning the wavelets to these tiles. However, this method is not appropriate in practical cases where the output of the wavelet transformed signal should visually be accessible and allow a meaningful interpretation. The proposed non-linear scaling of wavelets allows generating patterns that can be understood as a representation of power spectra resolved in time. The presentation will show the various features that can be extracted and how to work with them. Various examples will show how the wavelet transform can be used in practical tasks, one of them being the pattern recognition.