Design And Implementation Of Invoice Recognition System Based On Deep Learning

With the rapid development of the age and the economy,for some companies or organizations,more and more invoices need to be sorted.Manual entering and manual retrieving are not only a waste of time,but also prone to errors.The era of artificial intelligence has arrived.More and more work can be solved by artificial intelligence.A solution for automatic recognition of invoices is needed.By using OCR technology,the information on the invoices is collected in batches and outputs are structured.Compared with the traditional manual data entry,it will greatly reduce the workload of financial staff and improve their efficiency.In order to realize the "shoot-and-get" invoice recognition in daily life,the following work was mainly carried out: ResNet was used to complete the image orientation recognition in the way of image classification.PSENet was used to complete the target text detection.Image annotation and data verification were done too.ResNet was used to complete the detection of specific fields in the way of image classification.A corresponding font template library was created based on the original invoice image,the binary image was used as the target image,and some image processing methods were used to simulate and generate the binary text images.DenseNet,BiLSTM,and CTC were used to design text content recognition models and related training and testing processes were performed.PyQt5 and Flask were used to build the user interface,and the integration of each function of the system was done.Recognition of invoice images taken directly in daily life has been realized.The detection accuracy of specific fields in the invoice image has reached over 95%,the individual recognition rate of each field has reached about 95%,and the overall recognition rate has reached over 90%.