Semantic Entity Recognition Model Identification Method Based on Multiple Modes

In recent years, the multi-module method in the field of computer vision has been continuously optimized, and its application covers various services such as gray level map conversion and key information extraction, and general card bill recognition, general text table recognition and information extraction of some financial characteristics bill vouchers are also easier to be recorded. Therefore, image coloring and content recognition have become an urgent problem to be solved. In the past, images can be compared with text, page layout and other information when extracting structured information by taking photos. However, due to the unsatisfactory generalization effect of traditional technical solutions of actual methods, many key information is lost during recognition, affecting the subsequent progress. Therefore, in response to this problem, a key key-value pair was designed to identify various personal information sheets and invoices with tabular nature, the key information was identified through PaddleOCR, the training data was enriched by data mining and data synthesis methods, and the fine-tuning was carried out based on the training data, and the multi-modal information was fused by UniColor methods. Finally, the PDB task is used to make the model learn the matching relationship between Box directions. The effectiveness of the proposed method is verified on the PP-OCRv2 ultra-lightweight detection pre-training model, and the efficiency and accuracy of key information identification are improved.