Research On Recognition Algorithm Of Invoice Image Information Based On Deep Learning

In recent years,the usage of invoices in China has increased sharply,among which special VAT invoices and general VAT invoices have been used greatly,accounting for 90% of the total amount of various invoices.The reimbursement process of invoice is very cumbersome,requiring manual input of key information in invoice,which wastes a lot of manpower and time.Therefore,it is particularly important to design an algorithm for intelligent identification of invoice information.At present,traditional algorithms can identify information in scanned invoice images,but there are challenges to identify information in invoice images with different zoom ratio,blur degree,brightness,tilt angle,size and background interference taken by mobile phones.The invoice image information recognition algorithm based on deep learning proposed in this paper combines computer vision technology and text detection and recognition algorithm to recognize the invoice image information taken by mobile phones,and the information accuracy is up to 99.10%.Firstly,an HTA algorithm based on Hough transform is proposed to make intelligent skew correction of invoice images in natural scenes.Then,the YOLOV3 deep learning algorithm and Open CV tailoring technology were innovatively combined to locate,segment and extract the tilted invoice image,effectively removing the redundant information on the invoice image,and classifying the information.Then,CDNets algorithm is used to detect and recognize the information extracted from the invoice area image.Finally,a format text processing method based on OCR is proposed to format the identified results.Then,the uppercase Chinese amount characters in the total area of price and tax in the invoice image are recognized.In this paper,an algorithm model of invoice image text recognition based on deep learning is proposed.The model continuously uses the convolution kernel as the convolutional layer,and then connects the5×5convolutional layer with the batch normalization layer to form a new convolutional neural network architecture.The recognition accuracy is as high as 99.79%.Finally,the deep learning-based invoice image text recognition algorithm model was compressed,and the pruning network was proposed for the invoice image text recognition algorithm.The model was pruned layer by layer,and the weight of the compressed lightweight model was 2537 KB,and the recognition accuracy was 99.38%.