Method for generating single-stage table detection network

Abstract

The invention provides a method for generating a single-stage table detection network, which comprises the following steps of: preprocessing an original data set to generate a first type of picture data, namely converting a label format of a picture, performing data enhancement on the picture with the converted label format, and generating a second type of picture data; meanwhile, mapping processing is carried out on the labels with the target frame positions changed in the enhancement process; the structure of a detection network is set, the first type of picture data sequentially passes through the detection network to generate second type of picture data, and the structure of the detection network comprises an input layer, a convolution module, a self-attention module and an output layer; and calculating a loss value of the second type of picture data in an output layer through a target function, and monitoring and optimizing a detection network according to the loss value. According to the method, the problem of accurate regression of the table is solved, the coexistence of the complexity and the precision of the detection network is realized, the problem of complicated post-processing is solved, and the economic benefit and the social benefit are relatively good.

Description

technical field [0001] The invention relates to the field of document layout analysis of artificial intelligence in the computer industry, in particular to a method for generating a single-stage form detection network. Background technique [0002] In recent years, AI has developed rapidly in the fields of CV (Computer Vision, Computer Vision) and NLP (Natural Language Processing, Natural Language Processing). Many breakthroughs have been made in both basic theoretical research and vertical field applications, and have successfully blossomed in the industry. . [0003] At the same time, due to the continuous generation, processing and transformation of documents from all walks of life and various industries, only the financial field involves various contracts such as asset management agreements, various reports such as corporate financial reports, and various bills such as running bills. Huge documents, therefore, AI research and attempts on document layout analysis are als...

AI Technical Summary

Problems solved by technology

[0014] In view of the shortcomings of the prior art described above, the purpose of the present invention is to provide a method for generating a single-stage form detection network, which is used to solve the problems of high complexity, slow speed, and low intensive reading in the existing form detection process in the prior art. question

Images