US Dollar Classifications by the LVQ Method Based on Reliability Criterion

Abstract

Automatic classification of bill money has been welldeveloped and it is important that the classifier has highaccuracy. Generally, accuracy of classification isrepresented as a recognition rate of sample data. Toevaluate the accuracy more strictly, we introduce areliability criterion. In the pattern recognition, neuralnetworks (NNs) have been adopted. Among them a competitiveNN has a simple structure and can explain the relationbetween the inputs and the outputs more easily than alayered NN based on the backpropagation method. Thus, weuse a competitive NN for the bill money classification anduse the learning vector quantization (LVQ) method fortraining the NN. After introducing a reliability criterionbased on a probability distribution for the classificationby the LVQ method, we classify a US dollar by the LVQmethod and show the effectiveness of the proposed method.