Wavelets pre-processing of Artificial Neural Networks classifiers

Abstract

Artificial neural networks are highly parallel structures inspired by the human brain. They have been used successfully in many human-like applications, such as pattern recognition. Performance of these networks can be enhanced if used properly in conjunction with equally powerful mathematical tools. In this paper, we used the discrete wavelet transform as a pre-processing tool for two well-known neural classifiers; competitive layer networks and learning vector networks. The wavelets transform was used successfully to approximate the input patterns of the two classifiers and thus reduced their input-layer requirements considerably. Such reduction facilitates cost-effective hardware implementations of artificial neural networks.