Feature extraction of a vectorcardiogram by employing a wavelet network guaranteeing a global minimum

Abstract

This paper describes a wavelet network with convex wavelets. The distinctive features of this network are (1) guarantee of a global minimum, (2) good designability (the number of connection weights is determined by the resolution given by a designer) and (3) high ability of generalization especially in local nonlinearity. It is applied to feature extraction from a vectorcardiogram and its feature visualization.