Multi-mother Wavelet Neural Network Training Using Genetic Algorithm-Based Approach to Optimize and Improves the Robustness of Gradient-Descent Algorithms: 3D Mesh Deformation Application

摘要

This paper presents the implementation of genetic algorithm which aims at searching for an optimal or near optimal solution to the deformation 3D objects problem based on multi-mother wavelet neural network training. First, we introduce the problem of 3D high mesh deformation using Multi-Mother Wavelet Neural Network architecture (MMWNN). Furthermore, gradient training limits of wavelet networks are characterized by their inability to evade local optima. The idea is to integrate genetic algorithms into the wavelet network to avoid both insufficiency and local minima in the 3D mesh deformation technique. Simulation results validate the generalization ability and efficiency of the proposed network based on genetic algorithms (MMWNN-GA). Thus the significant improvement of the performances in terms of quality of 3D meshes deformation.