Recognition of New and Old Banknotes Based on SMOTE and SVM

Abstract

The recognition and recycling of old banknotes are important issues to be solved urgently in the financial industry. In reality the grades of diversely defaced banknotes to be recycled are quite different. Both the absolute and the relative quantities of samples of old banknote are far less than those of new banknote. For processing such imbalanced data sets, the traditional support vector machine (SVM) algorithm might produce poor classification performance for the minority class. To solve this problem, a new method, combining synthetic minority over-sampling technique (SMOTE) with SVM, is put forward to improve the prediction accuracy of old banknotes in this paper. First, a modified SMOTE algorithm is designed to generate new minority class samples. Then, the most easily defaced parts of banknote, the four corners and the center part, are selected as feature areas. Finally, the classification model is obtained using the SVM algorithm. The experimental results show that this method is of high efficiency and the performance can be improved about 20% compared with the method with the standard SVM algorithm.