Research on scan identification technology of detection reports
Since there are many factors affecting the identification quality of detection report,such as differ-ent report formats,poor scanning quality,and irregular handwriting,a report identification solution is pro-posed.Firstly,the deep convolutional neural network VGG16 is used to correct the report direction,and then,the image text generation method Table-Master is introduced to analyze the report structure.Nextly,the natural image text detection model CTPN is introduced for report text position recognition,and the se-quence text recognition model Master is used for report text recognition.Finally,a multi-feature fusion clas-sification model MFFC,which integrates text information and location information,is used to extract key in-formation from the recognition results.The experiment results show that the evaluation indicators of this scheme are better than other recognition model schemes,which can effectively extract the text information from the report,realize the identification and extraction of the structured information in the scanned copy of the test report,and improve the digital input efficiency of the scanned copy.
identification of detection reportdeep learninginformation identificationinformation extrac-tionpaddlepaddle
NETL
NSTL
万方数据
