An Instrument Spreadsheet Layout Classification Method Based on Table Structure and Cell Content

Most users in the instrumentation industry use spreadsheets to indicate their quotation requirements, and the use of an automated instrumentation spreadsheet information extraction system can effectively reduce the labor costs of instrumentation companies. Since the layout format of spreadsheets provided by users is not uniform and standardized, the primary goal of building an automated instrumentation spreadsheet information extraction system is to identify different spreadsheet layout formats. To identify various spreadsheet layout formats, this paper presents an instrument spreadsheet layout classification dataset, an instrument attribute name database, and an instrument attribute value database. Additionally, an instrument spreadsheet layout classification algorithm (ResNet-Spreadsheet) based on ResNet is proposed. The algorithm retains the original residual structure, by manually modeling the structure and cell content information of the instrument spreadsheet to make it abstract as a three-dimensional instrument spreadsheet feature tensor for input into a convolutional neural network, while using a smaller convolution to replace the convolution of the original model so that the network pays more attention to the subtle features of the instrument spreadsheet. The practical applications demonstrate that the algorithm proposed in this paper achieves good results in common classification evaluation indices. It can accurately and efficiently identify instrument spreadsheets with different layouts.