基于深度学习的多重文档结构识别方法研究  

Research on multiple document structure recognition method based on deep learning

作　　者：徐一鸣 潘伟民[1] XU Yiming;PAN Weimin(School of Computer Science&Technology,Xinjiang Normal University,Urumqi 830001,China)

机构地区：[1]新疆师范大学计算机科学技术学院,新疆乌鲁木齐830001

出　　处：《电子设计工程》2021年第21期53-56,共4页Electronic Design Engineering

摘　　要：针对多重文档识别时出现的识别效率慢与识别精准度低的问题,文中提出了一种基于深度学习的多重文档结构识别方法。利用自编码器构建多层文档学习网络,调整训练的参数,使用卷积神经网络内的多层神经网络对多重文档进行特征提取、特征学习与次抽样。估算现实输出和对应期望输出的差值,提高文档结构识别精度。利用多元函数对文档结构解析,完成多重文档结构识别。实验证明,该方法比传统方法能够更快速地识别多重文档结构,并且识别精准度较高。Aiming at the problems of slow recognition efficiency and low accuracy in multi document recognition,a multi document structure recognition method based on deep learning is proposed.The self encoder is used to construct a multi⁃layer document learning network,and the training parameters are adjusted.The multi⁃layer neural network within the convolution neural network is used to extract features,learn features and sub sample multiple documents.The difference between the actual output and the corresponding expected output is estimated to improve the accuracy of document structure recognition.Using multivariate function to analyze the document structure and complete the identification of multiple document structure.Experiments show that this method can recognize multiple document structures more quickly than traditional methods,and the recognition accuracy is high.

关 键 词：深度学习 多重文档 结构识别 特征提取 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]