The Research Of Infrared Characteristics Identification On Banknotes Discrimination Based On Random Forest

Infrared characteristic identification on banknote discrimination is an important research topic in the pattern recognition field.As the types of counterfeit currency emerge in endlessly and the simulation degree is getting higher and higher,it becomes more and more difficult to detect counterfeits.The national standard GB16999-2010 puts forward the idea of "identify counterfeit currency with real currency",which is a leap forward of the traditional idea of "identify counterfeit currency with counterfeit currency".However,at present,the software design of domestic financial machinery and tools industry usually still judges the real and counterfeit currency through the identification of a certain point,and does not really realize the "identify counterfeit currency with real currency",which greatly weakens the identification ability of financial machinery and tools.In order to solve the problem that counterfeit currency cannot be predicted,the infrared image of China's 100-yuan banknote was taken as the research object of the algorithm,and the Random Forest(RF)classification model based on semi-supervised learning was designed in this paper,that is,either the real currency or the fake currency,which solves the problem that counterfeit currency cannot be predicted.The main research contents are as follows:(1)The infrared image preprocessing method of banknote was studied.An appropriate method was selected to preprocess the infrared image of banknote,so as to improve the quality of infrared image of banknote and facilitate the extraction of infrared features of banknote.(2)The basic principles of Gabor wavelet and principal component analysis(PCA)were studied in depth,and the infrared features with discrimination ability on the infrared image of banknote were effectively extracted by combining these two algorithms.(3)A random forest algorithm based on semi-supervised learning was proposed.Combining with the idea of semi-supervised learning,the random forest classification model was constructed only by using the infrared features of real currency with different kinds of tags.(4)The banknote recognition algorithm parameters were optimized.This paper mainly studies from the dimension of feature space,the number of candidate attributes and the number of decision trees,so as to obtain relatively optimized algorithm parameters and improve the recognition rate of the algorithm.Experiments show that the random forest based on semi-supervised learning cansolve the problem that counterfeit money cannot be known in advance.However,it is worth mentioning that,due to the strong ability of the algorithm to identify the unknown types of counterfeit currency,the algorithm has the characteristics of universality,which is relatively weak to identify the known counterfeit currency.