Research On Identification Method Of Needle-printed Medical Invoice Based On Deep Learning Model

In the era of rapid development of information technology,image recognition technology plays an important role in many fields.With the development of digital image processing technology and deep learning technology,the accuracy of image recognition has been improved.Medical invoice as indispensable part of people's life,the current health care center of medical instrument,the traditional medical invoice information library work is done by manual,need medical institutions invest a lot of time and money,not only improve the operation cost of medical institutions,and manually information efficiency is low,failure rate is relatively high,to improve the overall service quality of hospital brings negative effect,is also a weak link of medical industry informatization development.Therefore,it is of great significance for medical institutions to be able to accurately and efficiently identify medical invoice images.Efficient and accurate text image recognition method plays a very important role in the warehousing of invoice information.Compared with the popular laser printed text,the traditional needle printed text has breakpoints and low recognition rate.Based on this,this paper proposes a deep learning as the core pin type printing medical invoice identification method,its basic principle is: the digital image processing technology and the combination of deep learning technology,to complete the pin type of medical invoice printed text recognition,among them,the digital image processing technology is mainly used for invoice image pre-processing,specific include: image tilt correction,image enhancement processing,gray processing,denoising threshold segmentation,breakpoints,and so on.Secondly,in order to meet the conditions of OCR character recognition,the characters in the pre-processed image are also subjected to single character cutting processing.Deep learning technology mainly learns deeper features from character images through neural networks,in order to better understand character images,so as to achieve the purpose of character recognition.Choose the appropriate network structure and combine the deep learning framework to get the final network model.Secondly,there are a large number of medical terms in the medical invoices to be recognized.In order to further improve the accuracy of needle-printed medical invoice recognition,this paper adds Semantic detection and correction functions to obtain a higher recognition rate.Finally,combined with all the previous modules,the overall framework of online identification of medical invoice is given,which verifies the feasibility of the method in this paper.In order to verify the effectiveness of the proposed method,a comparison experiment is designed for the performance of the recognition method and the network parameters.In the experiment,the self-built single-character data set and more than 7000 medical terms are used as the data sets of the convolutional neural network and the recurrent neural network to train the network.In response to the shortcomings of the traditional Alexnet network,the improved F-Alexnet network can reduce the computer resource occupation and improve the recognition rate,and the network model is used to recognize the pin printed character image.The experimental results show that compared with the single convolutional neural network recognition method,the improved F-Alexnet network and semantic detection method can achieve a higher recognition accuracy,and the recognition rate is increased by 7.5%.