Design And Implementation Of Invoice Automatic Identification System Based On Deep Learning

Invoicing refers to receipt and payment of receipts and receipts for purchase and sales of goods,provision or acceptance of services,and other business activities.In most of the domestic enterprises and government units,the processing of invoices has basically taken the form of traditional manual processing,and the number of invoices is often relatively large,making the burden on staff too large,but also causing inefficiencies and other issues.For this kind of situation,based on deep learning and image processing algorithms,this paper designs and implements an automatic invoice identification system that can accurately identify digital information.Firstly,pre-processing and tilt correction are performed on the captured invoice image,and then the invoice information area is located to obtain a region of interest.Then denoise the information area and use the projection algorithm to complete a single character cut.Separately collect the cut single characters in different areas and make a character data set.The essence of deep learning is to train sample data by constructing neural networks of multiple hidden layers,and the essence of training samples is to let the network autonomously learn the characteristics of the samples.Analyze the characteristics of invoice printing numbers and use the Alex Net network model of convolutional neural networks to perform digital training and recognition.The main research contents are as follows:1.Image acquisition,preprocessing and calibrationUse the SDK provided by the scanner for secondary development and capture invoice images in real time;due to the inclination of the invoice during the collection process or the uneven illumination,it is necessary to perform a series of preprocessing on the invoice image,and then do tilt correction.2.Information area positioning and character cuttingThis article needs to locate five information areas,namely: invoice number,amount,tax amount,purchaser taxpayer identification number,and sales party taxpayer identification number.VAT invoices are nationally uniform,have the same and fixed layout structure,and can use the characteristics of the invoice structure to obtaininformation areas.First,according to prior knowledge,the approximate range of each information area is obtained,and a coarse positioning area is obtained;then,according to the characteristics of each region's information,a template matching method is used to find the precise region of the information.Next,the digital string area is denoised to improve the accuracy of subsequent character cutting and recognition.Finally,a grey projection character cutting algorithm based on prior knowledge is proposed to achieve single character cutting.3.Digital recognition based on deep learningFirst of all,build a deep learning caffe framework,using character cutting to create a single character character data set.Then adjust the parameters of the convolutional neural network Alex Net network model and train the model.Finally,the trained model was used for digital recognition.The experimental results show that the system can quickly and accurately identify the invoice information,greatly improving the efficiency of the invoice processing work,and reducing the workload of the staff.