Research On The Serial Number Of The Paper Currency Of The Republic Of China Based On Deep Learning

Modern computer vision technology such as deep learning is of great significance to study cultural relics and financial currency.In recent years,the cultural relic detection and automatic recognition based on computer vision technology are widely concerned by cultural relic and computer researchers,which has become a research hotspot.However,there is a lack of effective automatic recognition scheme for paper currency of the Republic of China as the current research on coin recognition focuses on the mainstream currency of modern circulation and ancient coins.The serial number of paper currency can be used to retrieve the batch,quantity,related cultural relics,historical and cultural information associated with paper currency,which is the only number of paper currency of the Republic of China.It is an urgent problem to be solved on how to automatically identify the serial number and detect the relevant patterns through the analysis of paper currency.However,the recognition algorithm of serial number of traditional paper currency is not suitable for the paper currency of the Republic of China,and it is still in its infancy to study the paper currency of the Republic of China through the use of computer vision.There are mainly difficulties in identifying the serial number of paper currency of the Republic of China: For example,there are various kinds of paper currencies,and there may be differences in the number,style and position of serial numbers of different paper currencies;It is difficult to deal with them in a unified way because the background of paper currency is complex,some paper currencies have high similarity in background color,pattern and serial number;There are damages,stains,folds and deformities during the circulation and collection of paper currency,and the image quality of paper currency is low;There are few paper currencies in stock,it is difficult to collect and the image sample of paper currency is small.In view of the above problems,it thoroughly discusses and studies the serial number identification technology of paper currency of the Republic of China in this paper.First of all,it solves the problems of small sample,low quality and multiple kinds of serial number location by proposing the serial number detection algorithm for paper currency of the Republic of China based on deep learning target detection.Secondly,it solves the problem that it is difficult to segment the serial number of paper currency of the Republic of China due to the adhesion and incomplete part of the projection between characters by improving the traditional segmentation algorithm of character projection.Finally,it proposes single-character processing techniques and convolutional neural networks to solve the problems of traditional recognition algorithms including low accuracy,weak generalization ability,high training cost of deep learning network and high requirement of reasoning software and hardware resources.According to the experimental results,the recognition model in this paper can effectively detect the serial number and non-character pattern of paper currency of the Republic of Chin.The accuracy rate is 100% in the detection stage;The accuracy rate is 98.5% in the segmentation stage;The accuracy is higher than 99.9% in the single character recognition stage;Finally,the overall recognition accuracy is 97.5%.At the same time,it also obtains better detection results in the application of related techniques to the detection of non-text patterns.