基于深度学习的票据识别系统设计

Design of a Deep Learning-Based Invoice Recognition System

信息时代,票据给人们的生活带来了极大的便利。但是,输入票据的过程较耗时,大量的票据会给财务人员带来一定的工作压力。幸运的是,大数据分析和深度学习技术的发展,使得通过一个基于深度学习的票据内容识别技术方案简化这一过程成为可能。这个方案涉及3个阶段。第一阶段,使用ResNet分类拍摄的票据图像。第二阶段,对特定的票据进行文本定位和识别。在文本检测阶段,DBNet被用来检测和定位票据中的细粒度文本。第三阶段,通过文本序列识别(Convolutional Recurrent Neural Network,CRNN)实现文本识别任务。基于深度学习的票据识别系统比人工识别的识别速度更快,准确率更高。

In the information age,invoices bring convenience to our lives.However,the large number of bills can be stressful for finance staff,and the process of entering bills can be extremely time-consuming for tax officers.Fortunately,advances in big data analytics and deep learning technologies have made it possible to simplify this process by designing a complete deep learning-based technical solution for bill content recognition.This scheme involves several stages,starting with the classification of captured ticket images using ResNet.Next,text localization and recognition are performed for a specific ticket.In the text detection phase,DBNet is used to detect and locate fine-grained text in the invoices.Finally,the text recognition task is implemented by Convolutional Recurrent Neural Network(CRNN).The deep learning-based ticket recognition system significantly speeds up the recognition speed while maintaining a high accuracy rate compared to manual recognition in the former.