Designing a neural network using a genetic algorithm with deterministic mutation and partial fitness

Abstract

In this paper a method of designing a neural pattern recognition system for a rotated coin recognition problem using a genetic algorithm (GA) with deterministic mutation (DM) and partial fitness (PF) is presented. In this method, chromosomes of individuals in the GA are divided into several parts and their PF functions are evaluated for GA operations. Furthermore, the DM, which is based on neural network learning, is introduced. The DM can evolve chromosomes of individuals to increase their fitness functions in a deterministic manner. In the pattern recognition system described in this paper, the Fourier transform is used as a preprocessor which produces rotation invariant features. These features are recognized by a multilayered neural network. The GA is utilized to reduce the number of signals, Fourier spectra, and input to the neural network. This approach using the GA is a type of feature selection problem. It is shown that the present method is better than conventional GAs with respect to convergence in learning, and results in the formation of a small neural network.