A method to design a neural pattern recognition system by using a genetic algorithm with partial fitness and a deterministic mutation

This paper presents a method using a genetic algorithm (GA) with a partial fitness (PF) and a deterministic mutation (DM) to design a neural pattern recognition system for a rotated coin recognition problem. In the method, chromosomes in the GA are divided into several parts. Their PFs are evaluated for GA operations. Furthermore, this paper introduces the DM based on a neural network learning. A coin recognition system in this paper includes as a preprocessor the Fourier transform, which produces rotation invariant features. Those features are recognized by a multilayered neural network. The GA is utilized to reduce the number of input signals, Fourier spectra, into the neural network. It is shown that the present method is better than conventional GAs on convergence in learning and makes a small-sized neural network.