Research On Image Feature Extraction And Dirty Degree Recognition Method Of Banknotes

With the rapid development of the economy and society,the circulation of banknotes is getting faster and faster,and the circulation is also increasing.At the same time,the ratio of banknotes defects and soiling has become larger and larger with the increase of circulation time.Nowadays,there are a large number of banknotes circulating all over the country.After entering the market,many people directly touch them,and a large amount of sweat,oil and dust will quickly accumulate on them.Therefore,banknotes have become one of the media that imperceptibly harm the human body,which makes the convenience of people’s lives and commodity transactions will be affected to varying degrees.In order to prevent these dirty banknotes from continuing to circulate in the market,when the banknotes reach a certain degree of contamination,banks often collect them and destroy them intensively.However,the banking and other financial departments are very arduous and inefficient in classifying the dirty banknotes.Effective methods should be adopted to classify,recycle and destroy the dirty banknotes.Based on this background,this thesis has carried out the research on the method of banknote image feature extraction and dirty degree recognition,and proposed three methods to solve the problem of banknote dirty degree recognition.(1)A multi-layer support vector machines(MLSVMs)recognition method was proposed for banknote dirtiness based on regional image texture features and threshold selection.Firstly,the contact image sensor(CIS)is used to collect the double-sided reflection gray-scale images of banknotes under blue,green,red,infrared and ultraviolet light,and the gray-scale images of green light transmission and infrared light transmission.Secondly,according to the pattern distribution of banknote images,the collected banknote image is divided into 8 areas with different sizes,and 22 texture feature parameters such as energy,contrast,and variance of the banknote images are extracted based on the gray-level co-occurrence matrix(GLCM)to describe the banknote dirty visual dirty characteristics.Next,22 GLCM texture feature parameters under different light sources in different regions are selected through thresholds.Finally,MLSVMs is used to identify the dirty degree of banknotes,and the simulation results show the effectiveness of the proposed method.(2)In most data dimension reduction tasks,feature selection is an essential preprocessing stage.A novel feature selection method for banknote dirtiness recognition based on mathematical function driven slime mold optimization algorithm(SMA)was proposed.Based on the Lévy flight operator and mathematical functions,the control parameters in the SMA are replaced to improve its global search ability.Then,the feature parameters extracted from banknotes under different light sources are selected by the improved SMA,and the banknotes dirtiness is identified by MLSVMs.Finally,according to the image pattern distribution of the banknotes,the collected banknote images are divided into 8 regions with different sizes,and the same method is used for feature selection and banknote dirtiness identification.In order to verify the performance of the proposed feature selection method,some swarm intelligence optimization algorithms,such as Path Finder Algorithm(PFA),Generalized Normal Distribution Optimization(GNDO),Butterfly Optimization Algorithm(BOA),and Atom Search Optimization(ASO),are adopted to carry out the compared experiments.The experimental results show that three improved strategies can effectively improve the performance of SMA and better maintain the balance between exploration and exploitation.(3)A feature selection method for banknote dirtiness recognition based on mathematical function-driven marine predator algorithm(MPA)was proposed to optimize the MLSVMs parameters.Texture features of banknote images are extracted based on the GLCM to describe the visual characteristics of the banknote dirtiness.Then,the feature parameters extracted from the banknotes under the full spectrum are selected by the improved MPA,and the cost function C value and the Gamma value in the kernel function of MLSVMs are optimized by hybrid SMA and MPA,and the dirty degree of the banknotes is identified at the same time.Finally,according to the image pattern distribution of the banknotes,the collected images of the left middle white area of the banknotes are used to select the feature parameters and identify the dirtiness of banknotes by the same method.In order to verify the performance of the proposed feature selection method,some swarm intelligence optimization algorithms,such as PFA,GNDO,BOA,and ASO,are adopted to carry out the compared experiments.The experimental results show that one improved strategy can effectively improve the performance of MPA and better maintain the balance between exploration and exploitation.