Handwritten Table Recognition Method Based on Multi-head Attention Mechanism and Knowledge Graph

With the development of smart grid, the use of digital to reduce the burden of the basic level is the efficiency of the development of power grid. However, most of the data of operation and maintenance are stored in paper reports, which cannot be extracted quickly. Among them, it is difficult to identify the handwritten form data and cannot accurately extract the knowledge logic. This paper presents a handwritten table identification method based on multi-head attention mechanism and knowledge graph, aiming to improve the efficiency and accuracy of automated processing of tabular data. In the table generation stage, we use the extracted semantic and structural information to generate accurate and consistent table results. Through experimental evaluation, we verify the table based on long attention mechanism and knowledge graph recognition and generation method identification result logic is very strong, can match the identification results of the new template, its accuracy is above 95%, the method can effectively deal with different types and complex structure of the table data, and provides strong support for subsequent table data processing task.