Paper version personal credit investigation content identification method and application

Abstract

The invention belongs to the field of artificial intelligence recognition, and particularly relates to a method and application capable of recognizing credit investigation content, which can accurately and quickly recognize the credit investigation content. Through a series of technical means including the steps of training and application of a deep learning model, preprocessing and optimization of credit investigation pictures, content filtering, feature enhancement and the like, a paper-edition credit investigation material can be converted into an electronic edition, the content of the electronic edition is accurately recognized and audited, and the credit investigation efficiency is improved. Therefore, the accuracy and efficiency of credit investigation content identification are greatly improved, and a more convenient and accurate method is provided for personal credit evaluation.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence recognition, and specifically designs a method and application that can accurately identify credit information content. Background Art

[0002] Personal credit content identification is based on the records of personal credit information issued by the Credit Reference Center of the People's Bank of China, including personal credit reports and corporate credit reports, which are used to query the social credit of individuals or enterprises.

[0003] Personal credit reports contain personal information and credit records that can help banks and other financial institutions assess an individual's credit risk and debt repayment ability.

[0004] Corporate credit mainly includes basic corporate information, credit transaction information and other information. Through this information, you can understand the business registration and changes of the company, check the company's credit, pledge guarantee, trade financing and other information, understand the judicial judgment and execution, and the punishment or commendation in taxation, environmental protection, quality supervision, food and drug supervision, human resources, public utilities, customs and other aspects. In addition, the credit report can also understand the company's contact information and business status.

[0005] Traditional OCR recognition methods cannot support recognition in non-single scenarios well, while OCR recognition based on deep learning has great advantages.

[0006] Deep learning methods can use any recognition object that can be converted into an image as input, including PDFs, scanned documents, etc. By detecting and recognizing the lines of text in the image, the obtained information can be processed. Preprocessing methods such as image direction judgment, tilt correction, watermark detection and filtering can better identify text accurately in complex scenarios. Summary of the invention

[0007] In view of this, the present invention will solve the problems of unstable and inaccurate recognition of personal credit content and single scenario in traditional OCR, thereby helping enterprises to quickly collect accurate credit information of users.

[0008] The present invention provides a method and application for identifying credit report content, which includes: converting paper credit reports to electronic versions, preprocessing electronic images, training deep technology models, identifying electronic content, reorganizing electronic content, filtering electronic content, enhancing electronic features, enhancing electronic content, cutting electronic content, identifying content after cutting, proofreading content, and intelligently reviewing content to complete personal credit report content.

[0009] Optionally, in any embodiment of the present invention, the paper version of the credit report is converted into an electronic version. If an electronic version exists, this step is skipped.

[0010] Optionally, in any embodiment of the present invention, the electronic image preprocessing module is further used to train a deep technology model.

[0011] Optionally, in any embodiment of the present invention, the deep technology model training module is further used to improve content recognition quality.

[0012] Optionally, in any embodiment of the present invention, the electronic content recognition module is further used to use the training model to perform text recognition on the new paper version of the credit report.

[0013] Optionally, in any embodiment of the present invention, the electronic content reorganization module is further used to reorganize and merge to obtain a clear electronic version of the credit investigation content that connects all modules of the credit investigation in series.

[0014] Optionally, in any embodiment of the present invention, the electronic version content filtering module is further used to perform watermark recognition and filtering according to the recognized content, format and characteristics.

[0015] Optionally, in any embodiment of the present invention, the electronic version feature enhancement module is further used to optimize the integrity and color of table lines according to the identified content, format and features.

[0016] Optionally, in any embodiment of the present invention, the electronic version content enhancement module is further used to perform contrast and color optimization processing according to the identified content, format and features.

[0017] Optionally, in any embodiment of the present invention, the electronic version content cutting module is further used to number and cut the block-level elements of the electronic version with titles and tables as the main forms.

[0018] Optionally, in any embodiment of the present invention, the content identification module is further used to synchronously use deep learning technology to identify the content of these small modules.

[0019] Optionally, in any embodiment of the present invention, the content proofreading module is further configured to bring the identified contents into content proofreading respectively.

[0020] Optionally, in any embodiment of the present invention, the content intelligent review module is further used to perform a final content intelligent review on the overall credit investigation content. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of a scanned copy of a paper version of a credit report in an embodiment of the present invention.

[0022] Figure 2 It is a schematic diagram of the reorganized electronic version content in an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of a collection of small pictures of credit reporting modules after the reorganized complete electronic credit reporting is split in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] First, convert the paper version of the credit report into an electronic version. You can use a scanner, camera, mobile phone or other equipment to obtain the electronic version of the credit report, such as Figure 1 shown.

[0025] With the electronic credit report, we can start to recognize the text on the electronic photos, but in order to improve the recognition accuracy, the pictures need to be pre-processed.

[0026] Preprocessing steps: denoising, binarization, contrast stretching, etc., help improve OCR results.

[0027] The recognition problems that may exist in the paper version of the credit report mainly include irregular layout size, inconsistent column spacing and page margins, uneven font thickness, etc., so we do not use simple OCR recognition technology to directly perform text recognition.

[0028] For the image set obtained in step

[0026] , we use deep learning technology to train a convolutional neural network (CNN) model for text recognition. Specifically, some open source deep learning frameworks, such as TensorFlow and PyTorch, can be used to build a CNN model.

[0029] During the training process, the text information of the paper version of the credit report needs to be used as a label and input into the model for training. During training, the data set needs to be divided into a training set and a test set. The training set is used for model training, and the test set is used to evaluate the model, and the model parameters are adjusted to optimize performance.

[0030] After the model training is completed, the model can be used to perform text recognition on the new paper version of the credit report to obtain preliminary electronic credit report content.

[0031] The steps are as follows.

[0032] Data preprocessing: First, the image needs to be preprocessed, including resizing, pixel standardization, color space conversion, etc., to improve the accuracy of the model in processing and understanding the image data.

[0033] Feature extraction: Use deep neural networks to encode and extract features from input images. The “depth” here refers to the multiple layers contained in the network, each of which can extract features at a different level of abstraction.

[0034] Build a model: Build a neural network model that takes images as input and outputs categories or labels.

[0035] Training model: Use a large amount of labeled image data to train the model so that it can make correct predictions on new images.

[0036] The initial electronic credit report content is inaccurate, and we only use it as the basis for analyzing the document structure.

[0037] According to the recognition structure, it is sorted with page number as the main parameter, and then the electronic version of the credit report is reassembled according to the defined format data to obtain a complete electronic version of the credit report picture that connects all the credit report modules, such as Figure 2 shown.

[0038] All the previous work is aimed at making this complete electronic credit report combination more accurate.

[0039] There are still many problems in it that will affect the final text recognition results, such as discontinuous and unclear table lines, different text colors, connections between text and table lines, and watermarks.

[0040] Therefore, we also need to train a second model based on the above deep technical operation steps to identify this preliminary and complete electronic credit picture.

[0041] Watermark recognition and filtering are performed based on the identified content, format and features.

[0042] Optimize table line integrity and color based on identified content, format, and features.

[0043] Content contrast is enhanced based on the identified content, format and features, including contrast, color, etc.

[0044] Following the above steps, we have obtained the final version of the complete electronic credit report picture.

[0045] The content of the credit report may vary depending on the history of each person or company, so the image may be very large and there may be conflicts on the content page, which will affect the final text recognition results.

[0046] In order to improve the accuracy and speed of recognition, the block-level elements of this electronic version with titles and tables as the main form are numbered and cut again to obtain multiple small modules, such as Figure 3 shown.

[0047] Then, deep learning technology is used synchronously on these small modules to identify the content.

[0048] The identified contents are brought into the content proofreading module respectively, and compared with the information from big data to further improve the accuracy of the identified data.

[0049] Continue to conduct content review on the overall credit report content through the big data training model. If there are any problems, manual review will be carried out. If there are no problems, it will go directly to the storage module.

[0050] Finally, we get a complete credit report and a report on data recognition accuracy based on data proofreading and data review.

Claims

1. A method for identifying credit information content, characterized in that: The following steps are involved: 1.1 Convert paper-based credit reports into electronic versions; 1.2 Preprocess the electronic images; 1.3 Use deep technology models for training; 1.4 Identify the electronic content; 1.5 Reorganize the electronic version; 1.6 Filter the electronic content; 1.7 Enhance the features of the electronic version; 1.8 Enhance the content of the electronic version; 1.9 Cut the electronic version of the content; 1.10 Identify the cut content; 1.11 Proofread the content; 1.12 Conduct intelligent review of content.

2. The method according to claim 1 or 2, characterized in that: The electronic image preprocessing module is used to train the deep technology model.

3. The method according to any one of claims 1 to 5, characterized in that: The deep technology model training module is used to improve the quality of content recognition.

4. The method according to claim 1, characterized in that: The electronic content recognition module is used to perform text recognition on the new paper version of the credit report using the training model.

5. The method according to claim 1, characterized in that The electronic content reorganization module is used to reorganize and merge to obtain a clear electronic version of the credit investigation content that connects all modules of the credit investigation in series.

6. The method according to claim 1, characterized in that The electronic version content filtering module is used to perform watermark recognition and filtering according to the recognized content, format and characteristics.

7. The method according to claim 1, characterized in that The electronic version feature enhancement module is used to optimize the integrity and color of the table lines according to the identified content, format and features.

8. The method according to claim 1, characterized in that: The electronic version content enhancement module is used to optimize contrast and color according to the identified content, format and features.

9. The method according to claim 1, characterized in that: The electronic version content cutting module is used to number and cut the block-level elements of the electronic version with titles and tables as the main forms.

10. The method according to claim 1, characterized in that The content recognition module is used to synchronously use deep learning technology to recognize the content of these small modules.

Images