Research On Recognition Method Of Multi-national Banknotes

Along with the rapid development of global economy,trades among countries have been increased,and higher requirement has been raised for the sorting of banknotes.By far,banknotes sorting technology is only for the banknotes with large circulation,so the banknotes sorting machine should be introduced to the global market for identifying banknotes of minor countries.Based on it,through the improvement of the traditional algorithm,a method for multinational banknote recognition is proposed.Finally,a simulation experiment is performed to verify that the method in this paper meets the technical requirements for the accuracy and real-timeness of banknote sorting.The specific work is as follows:(1)The correction and extraction of the banknote images are completed with ordinary least squares and perspective transformation.In the paper,the high-speed moving image on the conveyor belt is collected by industrial camera at first.Since there is much greater number of banknotes in the conveyor belt and there would be deviation in position and angle for mechanical reason,so the image of collected banknote is segmented with projection method to gain the image of single banknote.In the positioning process of banknotes,the paper is adopted with Hough transformation and ordinary least squares these two methods and there is algorithm comparison,then a method with higher operational efficiency is selected.The image size of banknotes is determined and it is extracted and corrected with the image through perspective transformation.(2)It is adopted with method based on HSV and fixed area grid division to extract the image feature of banknotes.In the paper,it transfer the image of banknote from RGB color space to HSV color space.For each country's banknotes,different denominations often have different colors,so the method is used to extract and build up chroma model with banknote image;then it is found with the extraction methods of width feature,free mask and mesh feature that the first two features are in poor stability while the latter could have an extensive presentation of the image information to confirm the image feature.(3)A paper currency recognition algorithm based on BP neural network and supporting vector machine sequence minimum optimization is proposed.In the paper,it is designed with a three-layers BP neural network to complete the banknote identification;and then the size,chroma model and feature blocks acquired from banknote image as entered data,so as to gain a network model to complete the banknote identification through training of BP neural network.Secondly,the support vector machine sequence minimum optimization algorithm is designed and improved,which avoids the complex iterative process through analytical method,and classifies samples accurately to complete the test and identification of banknotes.Finally,through the experimental simulation,the paper currency recognition rate of the improved algorithm is higher than that of neural network,and it is fast.The two banknotes identification algorithms in the paper are realized by the programming with VS2015 to get a higher correction rate for the banknote identification.The BP neural algorithm is equipped with good fault tolerance and strong promotion ability.The compatible vector machine recognition algorithm is more suitable for the change of environment with better overall effectiveness.