Rapid and automatic table extraction method based on deep neural network

Abstract

Design data of an overhead transmission line generally appear in the forms of a picture table, a PDF table and the like, data formats and representation forms provided by different design units are different, table data cannot be directly read by a computer, manual input is needed, the workload is large, and the speed is low. In order to solve the problem, the invention discloses a fast and automatic table extraction method based on a deep neural network, and the method comprises the steps: S1, carrying out the correction preprocessing of an image table; S2, carrying out image table frame line extraction to obtain a cut cell picture Q (i); S3, carrying out cell character positioning; S4, recognizing characters of the cells; and S5, performing sequential merging on all the cell recognition results to realize automatic and rapid table extraction. Compared with an existing table extraction method based on manual visual interpretation or single character segmentation, the method is higher in speed, higher in automation degree, more accurate and complete in recognition, higher in generalization ability and higher in efficiency, can still keep a good extraction effect on multi-source heterogeneous and complex image table data, and acceptance item judgment and acceptance work management are facilitated.

Description

technical field [0001] The invention relates to the technical field of neural network applications, in particular to a method for quickly and automatically extracting tables based on a deep neural network. Background technique [0002] Overhead transmission line design materials exist in the form of picture tables, PDF tables, etc. The types of materials are complicated. The design handover materials provided by different design units are different in content, format, and form of expression. It is difficult for the design handover data to be directly read by computers. Effective analysis requires manual input, but the manual extraction method has a large workload, long cycle, slow speed, low efficiency, high cost, and error-prone, especially for large-scale tabular data is extremely limited. At the same time, due to the lack of standardized management of design transfer data, it is difficult to quickly locate and browse design data, and it is not convenient for the judgment ...

AI Technical Summary

Problems solved by technology

In response to this problem, some people proposed to use word segmentation to extract image tables. This method needs to split a long paragraph of text into words, which is slow and inefficient. It is affected by inconsistent text spacing, table tilt, image brightness, image Influenced by contrast, etc., it is also easy to mis-extract

Images