Estimative of SOM Learning Parameters Using Genetic Algorithms

摘要

The aim of this work is to optimize SAR image (Synthetic Aperture Radar Image) clustering using non-supervised Artificial Neural Network (ANN). The clustering optimization is achieved by means of the optimization of the neural network training parameters. In this way, Genetic Algorithms (GA) are used in order to explore the search space of the non-supervised Neural Network SOM's (Self-Organizing Maps) training parameters. The genetic functions (Object and Evaluation Functions) are defined considering the task of the SOM population (each chromosome is composed by two SOM's training parameters): the clustering of Synthetic Aperture Radar Image. The population was codified with two base (binary code).