NEURAL NETWORK TECHNOLOGIES FOR RECOGNITION CHARACTERS

Abstract

The process of neural networks modeling for pattern recognized problem of printed characters consid-ered in this paper. Learning for pattern recognition preparing for a limited set of synthetic characters. It assumes the two-layer neural network training. The convergence of three learning algorithms is stud-ied. They are packet-based adjustment of weights and biases, the gradient, the algorithm based on the computation of the Jacobian function weights. The article provides recommendations for the installa-tion of the initial parameters for a set of tools Neural Networks Toolbox software Matlab. Experimental results for different settings customer networks given that confirms these propositions.