GAによりチューニングされたニューロ識別システムとその海外紙幣への応用

抄録

We have proposed a paper currency recognition system by a neural network (NN) to develop new type of paper currency recognition machines. This system is constructed by some core techniques. One is the small scale neuro-recognition technique using masks. The other is the mask optimization technique using genetic algorithm (GA). In this mask optimizing technique, I have shown the effectiveness of the adopting GA. Up to now, it has been shown that the NN recognition and learning system are enough as a design and developing method in the market. However, in case of the mask optimizing system, it has not been shown yet. In this paper, I and my group implement this mask optimizing system using GA from EWS to personal computer. Then I show its pragmatic effectiveness and generalization using Italy, France, and Spain currencys. Furthermore, I show the relationship between the least masked area and mask optimized condition by the simulation result. Finally, I also refer running method of this mask optimizing system to aim at using as design and developing tools.