New Media Text Resource Extraction System Based on Rep OCR Model

The continuous development of new media has raised people's requirements for the accuracy of text extraction models. In view of the problem that the current text extraction model cannot handle the text extraction of irregular text boxes, a group feature pyramid space (Group Feature Pyramid Network, GFPN) is designed based on the feature pyramid space. In combination with deep learning technology, a deep learning model suitable for information extraction of irregular text boxes is designed. The results show that the information extraction accuracy of the designed model can reach 93.8, the recall rate can reach 87.7, and the floating point calculation can reach 20.46. The model designed in the study effectively improves the application ability of the text extraction model in the field of new media and improves people's ability to collect information on new media resources.