Training the children wavelets to recognise waveforms within non-stationary signals

Many authors have developed methods for automatic recognition and classification of signal patterns based on wavelet transforms and wavelet theory. However so far a method to find the wavelet family that best fits a particular class of signals is yet not evolved. We present a new method based on a combination of wavelet analysis and the training method used in the field of artificial neural networks. We define a training process applied to a family of wavelets and intended to optimise the pattern detection and location capabilities when applied to a particular class of signals. We also demonstrate how this method is used to detect and localise the interictal epileptic spikes within human EEG bio-signals.