Artificial neural network, Hebb network, Adaline, backpropagation, pattern recognition, classifiers

This paper describes an experimental study based on the application of neural networks for pattern recognition of numbers stamped (imprinted) on ingots. The same task was also solved using fuzzy logic. The ability of all tested neural networks is sufficient to learn all the test patterns, as was demonstrated during experimental works. Unfortunately, amount of training patterns provided by Company KMC Group, s.r.o. were very small and they were very different from test samples. In the article, appropriate types of binarization were discussed so as to extract sufficient information regarding classification via neural networks. There were the optimization of the training set proposed based on the training set analysis. Next, we also proposed way of optimization of parameters belonging to adaptation rules of used neural networks. All experimental results were mutually compared in conclusion.