Document key information extraction method based on zero sample learning

Abstract

The invention discloses a document key information extraction method based on zero sample learning, and belongs to the field of computer vision. The method comprises the following steps: randomly selecting a element sample; respectively carrying out feature coding on the image corresponding to each text block in the document image, the text block content and the coordinate value of the text block box; fusing the features after feature coding to obtain a plurality of first fusion coding features; performing feature coding and fusion on the plurality of text blocks in the meta sample to obtain a plurality of second fusion coding features; and similarity comparison: selecting the entity category of the text block corresponding to the second fusion coding feature with the highest similarity as the entity category of the text block in the document image. According to the technical scheme of the invention, the method greatly reduces the energy required for collecting and labeling data, greatly reduces the size of the model, and greatly reduces the time required for training and prediction, so that the method can be better applied to an actual production scene.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a method for extracting document key information based on zero-sample learning. Background technique [0002] The information extraction process is the process of automatically extracting and converting unstructured information embedded in documents into structured data. A traditional key information extraction method is based on template matching, which has obvious limitations. First, depending on the complexity of the document layout, it takes a lot of energy and time to build a high-precision template. In addition, manpower needs to be invested in the continuous maintenance of all customized templates in the later stage to ensure that the functions of existing templates will not be affected when new templates are added. Second, no matter how finely the template is designed, it is difficult to fully consider the problems that may arise in real situations. For example, the docu...

AI Technical Summary

Problems solved by technology

Second, no matter how finely the template is designed, it is difficult to fully consider the problems that may arise in reality

Due to the huge amount of data, and the accuracy of subsequent training and model prediction has great requirements on the high precision of sample labeling, this process is time-consuming and laborious.

Third, training a large number of samples requires a lot of resources and time, but in reality, there may be insufficient resources or high time requirements, so that it is impossible to obtain a higher-precision model

Fourth, in order to solve complex tasks and obtain higher accuracy, the complexity of the model will become higher and higher, the capacity of parameters will increase accordingly, and the training time and prediction time will also increase

This makes it difficult to directly apply it to actual production scenarios even if a higher-precision model is trained

Images