METHOD OF PROCESSING VECTOR SIGNALS FOR PATTERN RECOGNITION ON THE BASIS OF WAVELET ANALYSIS

摘要

The present method includes: pre-processing a signal, and determining current level of information windows for each sample of a sample set, while allowing for the overlapping of adjacent windows; processing each sample of the sample set using a wavelet transform; selecting a supporting function and the maximum number of variables thereof, and subsequently constructing a family of models for display using the selected wavelet coefficient function for one target value; training each model from the family on a training set, with weight parameters being selected for the models and the best models being subsequently chosen according to a criterion calculated on a validation set; verifying the chosen models on a test sample by calculating an algorithm convergence evaluation criterion; selecting significant regions corresponding to windows of the given level of information, and passing to the next level of information within the selected significant regions corresponding to the best models containing wavelet coefficients ai, pre-calculated within these windows; applying, for each window of the regions determined at the first level of information, the corresponding wavelet transform with the smallest scale parameter to produce a detailed representation of the set of significant regions.