Research On Key Technologies Of Machine-readable Recognition Of Hong Kong Dollar Based On Infrared Spectrum Image

With the rapid development of international trade and Hong Kong's economy,the safe circulation and machine-readable recognition of Hong Kong Dollar have become particularly important.Due to the fact that the issuing banks are not unique,the issuing cycle is short,there are many kinds of denominations and versions of Hong Kong Dollar.The recognition features and serial numbers of different versions vary greatly,which brings many difficulties to the machine-readable recognition of Hong Kong Dollar.In addition,there are some factors affecting the image quality,such as fouling,aging and other factors in the use of Hong Kong Dollar.The existing machine-readable recognition algorithms are facing great challenges,and the false recognition rate is high.In order to solve the above problems,this paper uses infrared spectroscopy to acquire clear infrared spectrum image of Hong Kong Dollar,and recognizes the denomination,orientation,versions and serial numbers of Hong Kong Dollar,so as to promote the security of currency circulation.In this paper,we studied the key technologies of machine-readable recognition of Hong Kong Dollar based on infrared spectrum image.Firstly,the original infrared image acquired by CIS is corrected,which uses Gauss filtering,edge and vertex extraction algorithm based line segment detector and affine transformation.Secondly,it recognizes various denominations of Hong Kong Dollar,and distinguishes the orientation of Hong Kong Dollar based on the contrast characteristics of local gray level co-occurrence matrix about the pic.Finally,aiming at the recognition of Hong Kong Dollar versions and serial numbers,a convolutional neural network model based on feature angle learning is designed.By using the angle information of feature vectors,the features of the output of the neural network can be distinguished,and the ability of feature description and generalization of the network can be improved.Applying this model to the recognition of Hong Kong Dollar versions,we can distinguish different versions of Hong Kong Dollar,effectively reduce the influence of dirt and other factors on the recognition process,and classify them accurately.Furthermore,the model can be applied to the recognition of serial numbers.It successfully identifies different versions of serial numbers,and solves the problem of inconsistent size and shape of serial characters in Hong Kong Dollar.It also can solve the problem of low recognition rate caused by confusing numbers and letters.The experimental results show that the proposed algorithm and model can effectively and accurately recognize the banknote information of Hong Kong Dollar.Compared with the traditional algorithm,the feature angle learning model proposed in this paper improves the expressive ability of features,enhances the robustness of the network.This model solves the problem of version recognition affected by dirt and other factors,and the problem of character morphology on the process of serial numbers recognition,and improves the accuracy of machine-readable recognition of Hong Kong Dollar in infrared spectrum image.