Unstructured text analysis and table database construction of hydraulic concrete materials

Abstract

In the process of historical construction of water conservancy projects, limited by the level of text informatization, a large number of non-editable documents of hydraulic concrete materials have been accumulated in the form of paper texts and scanned images, making it difficult to directly and effectively utilize material data, greatly increasing the difficulty of applying material knowledge. A document parsing method was proposed based on machine vision and deep learning, which accurately and efficiently converts the text information and table data of hydraulic concrete materials into editable form. Furthermore, based on the interpreted table information, a database of hydraulic concrete material tables was constructed, achieving efficient querying and unified management of concrete material data. Taking the actual engineering hydraulic concrete material document as an example to verify the feasibility of new method, the result show that the accuracy of each subtask of the document parsing method is over 90%, which is helpful for the automated reuse of non-editable resources of concrete materials and improves the data service capability in the field of water conservancy engineering.

Keywords

hydraulic concrete materials / layout structure division / text detection and recognition / table database