Research Of RMB Identification Technology Based On Texture Recognition And Convolutional Neural Network

In the financial services industry such as the banking industry and self-service industry which involves the identification of banknote,the banknote sorter plays a very important role.When it appears,it replaces a lot of boring and error-prone manual work,and makes the banknote sorting business greatly improved in speed and efficiency.However,there is still a large gap between China and other countries in the field of banknote sorting machine manufacturing.The banknote sorter used by the financial institutions,such as the large banks in our country are almost all imported from abroad,and the prices are above a million yuan.The identification of the banknote denomination is the core function of the banknote sorter.Therefore,we need to develop the high-efficiency banknote identification technology with independent intellectual property rights,to break the foreign monopoly of China's banknote sorter market.After fully investigating the domestic and international RMB banknote identification market conditions and extensively reading related literature,this thesis proposes two identification methods for RMB banknote images,the statistical theory recognition method based on Gray Level Co-occurrence Matrix(GLCM)and Mahalanobis distance,the recognition method based on Convolution Neural Network(CNN).Before the image identification of RMB banknote,a series of pre-processing such as smooth filtering,slant correction,and target segmentation are performed on the collected images.The filter denoising part adopts the median filtering method,the slant correction part adopts the Hough transform method,and the target segmentation part adopts the grayscale projection method.In the part of identification method which based on the gray level co-occurrence matrix(GLCM)and Mahalanobis distance theory of statistical,firstly,using the classical statistical method based on gray level co-occurrence matrix to extract the image texture feature parameters,and then use the Mahalanobis distance obtained from the statistical data of the texture feature parameters of the banknote as a classifier to achieve the purpose of identifying the RMB banknote.The algorithm can also identify the orientation of RMB banknote while identifying the denomination.The experimental data shows that the comprehensive recognition rate of RMB banknote of this algorithm is 99.4%.In the part of the identification of RMB banknote which based on the convolution neural network,a convolution neural network with eleven hidden layers is designed on the famous deep learning framework,and a number of parameters are trained and adjusted,finally train a convolution neural network that can achieve 99.8%recognition rate on the experimental data set,the recognition results meet the requirements of the actual application scenarios.This method is very innovative,which provides new research methods and ideas for the banknote sorter manufacturing field in China.