Research On Parametric Multiwavelets And Multiwavelets Neural Networks

AbstractMultiwavelets can satisfy orthogonality, symmetry, compact support and high-vanish moment at the same time, so they have better properties than those of traditional wavelet. In Section 2, based on two types of parametric forms about multiscale functions and multiwavelets, we discuss the relations between parameters and properties of multiscale functions and multiwavelets. Appropriately selecting the parameters, we can construct different multiscale functions and multiwavelets, including GHIM multiscale functions and multiwavelets.Wavelet decompositions and neural networks are powerfi.il tools for functional approximation. In Section 3, we propose a new type of network, called a multiwavelet networks, inspired by wavelet networks. The structure of networks and the algorithm of backpropagation type are proposed and some problems met in the course of study are discussed. In the end, simulating experiments show the superiority of the networks proposed by us.