A system and method for improving the ability to identify content of physical real estate certificates

Abstract

The present invention relates to the field of new generation information technology, and discloses a method and system for improving the content recognition capability of physical real estate certificates. The method comprises S1: marking the position of real estate images to obtain a position data set; S2: constructing a training set, a verification set and a test set; S3: constructing a table position detection model, and sequentially building a position feature extraction network, a position feature fusion network, a position detection network and a non-maximum suppression network; S4: training, verifying and evaluating the detection capability of the detection model; S5: using the detection model to perform table position detection and text content recognition to obtain table position and text; S6: performing variable matching according to the position, content and the form template of real estate certification materials in various regions to obtain the recognition result of the real estate content. The system comprises a real estate image input module, a model training module, a text recognition module, a variable matching module and a real estate form content output module. The present invention can realize automatic and accurate recognition of the content of real estate certification materials.

Description

Technical Field

[0001] The present invention relates to the field of new generation information technology, and in particular to a system and method for improving the content recognition capability of physical real estate certificates. Background Art

[0002] At present, when the financial industry grants credit or uses credit for customers, it mainly approves the credit limit based on the customer information it has, such as personal information, corporate information, tax information, real estate information, etc. Among these customer information, personal information, corporate information and tax information can all be obtained uniformly through connecting to the corresponding platform, but the customer's real estate information cannot currently be obtained directly from a certain platform, but is mainly obtained through physical materials submitted by the customer offline, and by identifying the real estate materials in the submitted materials, the risk control system obtains the corresponding real estate information, which is then used for credit limit approval operations.

[0003] However, although the current operation mode of identifying real estate content by using real estate information obtained offline can achieve the purpose of real estate content identification to a certain extent, there are still the following implementation difficulties in the operation process:

[0004] (1) Due to the different formats of real estate certification materials in different regions, the current real estate content recognition model has difficulty in identifying the content of certification materials in different formats, resulting in frequent failure to identify the content of the materials;

[0005] (2) Most real estate certification materials are in PDF format, and most of the tables are vertical. It is difficult to correctly identify the content using traditional PDF content recognition methods, and garbled characters may even appear;

[0006] (3) When the content of the property certificate exceeds the range of the table or there is a line break in the table, it is difficult to accurately identify the correct content;

[0007] (4) Since the table of real estate data is very complex, it is difficult to correctly identify real estate information due to the limitations of the recognition capabilities of traditional OCR, machine learning and other methods.

[0008] The above difficulties have all become the key obstacles to credit approval and credit use approval. Therefore, it is necessary to design a method and system that can accurately identify the content of real estate certification materials to help financial institutions accurately obtain customers' real estate information. Summary of the invention

[0009] The present invention aims to provide a method for improving the content recognition capability of a physical real estate certificate, so as to realize automatic and accurate recognition of the content of real estate certification materials.

[0010] To achieve the above object, the present invention adopts the following technical solution: a method for improving the content recognition capability of a physical real estate certificate, comprising the following steps:

[0011] S1: Collect and generate real estate images, mark the positions of real estate tables in the images, and obtain the real estate table position dataset required for deep neural network training;

[0012] S2: Randomly select data in the dataset and construct training set, validation set and test set respectively;

[0013] S3: Construct a table location detection model in the real estate certification material. When constructing the location detection model, first build a real estate table location feature extraction network; then build a table location feature fusion network; then build a table location detection network; and then build a non-maximum suppression network;

[0014] S4: Use the training set to train the table location detection model. After each training, use the validation set to verify the table location model and continuously optimize the weight of the detection model. Take the model weight with the best verification effect, and use the test set to evaluate the detection ability of the model to obtain the detection mode with the best performance.

[0015] S5: Using the detection model to detect the position of the real estate table to be detected in the detection mode with the best performance, and sending the detected table position to the text recognition model for text content recognition to obtain the position and text content of the table in the real estate table;

[0016] S6: According to the position and content of the form and the template of the real estate certification material form in each region, variable matching is performed to finally obtain the recognition result of the real estate content.

[0017] The principle of this scheme is:

[0018] Through step S1, the table positions from the real estate image are collected and annotated to form a data set in a unified format, which helps the model learn different table layout features, thereby improving the adaptability to certification materials in different formats, thereby helping to solve the problem of inconsistent table formats in real estate images.

[0019] In step S3, by building a special real estate table location detection model, especially using network structures such as feature extraction, feature fusion, location detection and non-maximum suppression, the table location can be more accurately located, and complex tables and vertical tables in PDF format can be effectively processed. This method can better cope with complex scenarios than traditional OCR and machine learning models, and reduce garbled characters and recognition errors.

[0020] In steps S4 and S5, by training the model, the model performs table position detection on the real estate table to be detected in the detection mode with the best performance, thereby improving the accuracy of the table position detection model, thereby ensuring that even if the content exceeds the table range or wraps in the table, these changes can be accurately captured. In conjunction with the subsequent text recognition model, the content in these special cases can be further analyzed and accurately extracted.

[0021] The template matching and variable recognition mechanism introduced in step S6 allows the system to automatically match and parse the recognized form content based on the known templates of real estate certification materials in various regions. This method not only improves the accuracy of recognition, but also enables the system to automatically adapt to the form formats and variable requirements of different regions.

[0022] The advantages of this solution are:

[0023] 1. It can improve the recognition accuracy and efficiency of the forms in the real estate certification materials:

[0024] Through the combination of deep neural networks and template matching, the recognition accuracy and efficiency of real estate certification materials have been greatly improved, reducing the need for manual review.

[0025] 2. Enhanced the versatility and adaptability of accurately identifying real estate forms in different formats:

[0026] The present invention can process real estate certification materials in different formats and in different regions, thus improving the versatility and adaptability of the system.

[0027] 3. High degree of automation:

[0028] From table position detection to content recognition, and then to template matching and variable recognition, the entire process is highly automated, which not only ensures the accuracy of recognition, but also reduces human intervention and improves work efficiency.

[0029] 4. Can reduce the error rate:

[0030] By optimizing the processing of complex tables and special format contents, the recognition error rate is significantly reduced and the data quality is improved.

[0031] 5. Can help improve customer experience:

[0032] Since this method can realize the rapid and accurate recognition of the forms in the real estate materials, it can reduce the tedious process of customers submitting materials, speed up the credit granting and credit use approval process, and help improve customer satisfaction.

[0033] In summary, this method effectively overcomes the challenges faced by the financial industry in the process of real estate information identification, and provides financial institutions with a more automatic, efficient and accurate real estate information identification solution.

[0034] Preferably, as an improvement, in step S1, first, image data of real estate certification materials and a large amount of table image data automatically generated by a python program are obtained from multiple channels to obtain the original real estate data image; secondly, when marking the position of the table, the LabelImg tool is used to mark the size of the table in the original real estate data image and the position of the table in the image to obtain a data set for deep neural network training.

[0035] Using the above solution, the format, content, layout, etc. of real estate certification materials obtained from different channels are different. Acquiring image data of real estate certification materials from multiple channels can ensure the diversity of the data set, which is helpful to train a more robust and generalized deep neural network model. In addition, a large amount of tabular image data is automatically generated using the python program. This method can quickly expand the scale of the data set and provide sufficient samples for model training, which helps to improve the recognition accuracy and stability of the model.

[0036] At the same time, LabelImg allows users to accurately annotate target areas in images, such as tables, in a graphical interface and generate corresponding annotation files, which can be directly used for deep neural network training without the need for additional data preprocessing, and can greatly reduce the complexity and time cost of manual intervention, which greatly improves training efficiency and shortens the model development cycle. At the same time, since the data set contains images of real estate certification materials from different channels and in different formats, as well as a large number of automatically generated table images, the trained model will have stronger generalization capabilities, enabling the model to better adapt to unknown or new formats of real estate certification materials and improve recognition accuracy in practical applications.

[0037] Preferably, as an improvement, in step S2, the property table location data set is randomly divided into a training set, a validation set and a test set in a ratio of 8:1:1.

[0038] Using the above scheme, random partitioning can ensure that the data between the training set, validation set, and test set are independent of each other, avoiding model overfitting or evaluation bias caused by data leakage. This independence is the basis for model training and evaluation, and is also an important prerequisite for ensuring the generalization ability of the model. The 8:1:1 partition ratio can better balance the needs of model training, verification, and evaluation. By evaluating the model performance on an independent test set, the pros and cons of the model can be evaluated more objectively. This evaluation result not only helps to select the optimal model, but also improves the credibility and reliability of the model in practical applications.

[0039] Preferably, as an improvement, in step S3, the real estate table position feature extraction network includes a convolutional learning module, a residual learning module, and a spatial pyramid pooling module; the table position feature fusion network includes an upsampling module and a multi-scale feature fusion module; the table position detection network includes a large-scale detection layer for detecting large-size real estate tables, a medium-scale detection layer for detecting medium-size real estate tables, and a small-scale detection layer for detecting small-size real estate tables; the non-maximum suppression network is used to screen out the best real estate table detection result from multiple detection results.

[0040] By adopting the above scheme, the detection accuracy of the location of the real estate table can be significantly improved through the collaborative work of the above modules. The feature extraction, feature fusion and multi-scale detection mechanism work together to enable the model to more accurately identify real estate tables of different sizes and formats. The introduction of technologies such as residual learning and multi-scale feature fusion enhances the robustness and generalization ability of the model, which enables the model to better adapt to images of real estate certification materials from different sources and qualities. The application of the non-maximum suppression algorithm effectively reduces redundant detection results and improves detection efficiency. The combination of the above operation methods has the advantages of improving detection accuracy, enhancing robustness and improving efficiency, and can achieve automatic, accurate and efficient processing in the process of real estate information recognition.

[0041] Preferably, as an improvement, in step S4, first, the parameters of the real estate table detection model are set, and the parameters include the number of training times, the type of table to be detected, the learning rate and the input real estate image size; secondly, the model is trained using the training set of real estate table detection, and the weights of the detection model are continuously optimized by back propagation and gradient descent during training, so that the model reaches the optimal solution; finally, the detection capability of the model is evaluated on the real estate table detection test set to obtain the detection mode with the best performance.

[0042] Through the above scheme, users can flexibly set key parameters such as the number of training times, the type of table to be detected, the learning rate, and the size of the input real estate image according to specific task requirements and data characteristics. This high degree of configurability enables the model to better adapt to different application scenarios and data sets. Using optimization algorithms such as backpropagation and gradient descent, the weights of the detection model can be continuously adjusted to minimize the loss function, so that the model gradually approaches the optimal solution. This training method not only improves the accuracy of the model, but also ensures the stability and efficiency of the training process. By fully training on the training set, the model can learn the general characteristics and rules of the real estate table, so that it has a strong generalization ability, so that the model can accurately detect the position of the table when facing new and unseen real estate images. Evaluating the detection ability of the model on the real estate table detection test set can objectively reflect the actual performance of the model. Through performance evaluation, the detection mode with the best performance can be obtained, which provides strong support for subsequent model deployment and application. 。

[0043] Preferably, as an improvement, in step S5: first, crop the image of each table according to the table position; second, send the cropped table image to the table position detection model to detect the table position, and send the detected table position to the text recognition model for text recognition.

[0044] In the above scheme, by cropping out the image of each table, the amount of data to be processed later can be significantly reduced. The text recognition model only needs to process the image area containing the table, rather than the entire property image, which greatly improves the processing speed and efficiency. The cropped table image usually has a higher resolution and a clearer table structure, which helps the text recognition model to more accurately recognize the text in the table. At the same time, the interference of background noise and other non-table content is avoided, further improving the accuracy of recognition.

[0045] Preferably, as an improvement, in step S6, first, the real estate table template of each region is queried; secondly, variable matching is performed according to the template, the table position and text content obtained in step S5, that is, key->value pairing; finally, the matching result is output to obtain the identified real estate content.

[0046] Through the above solution, customized processing for different tables can be achieved by querying templates for specific regions. This flexibility enables the system to adapt to different application scenarios and needs. Using templates for matching can ensure that the extracted real estate information follows a unified structure and format, which helps reduce data deviations caused by inconsistent formats or parsing errors and improves data accuracy and consistency. When you need to process new regions or update existing templates, you only need to add or modify the corresponding templates. This modular design makes the system easy to maintain and expand, and can adapt to changing business needs. Matching is performed using key->value, i.e. key-value pairs. This presentation method makes the data clearer and easier to read, which facilitates subsequent data analysis and processing.

[0047] In addition, the present invention also provides a system for improving the content recognition capability of a physical real estate certificate, comprising:

[0048] A real estate image input module, used to input real estate images;

[0049] The real estate table location detection module is used to build a table location detection model in real estate certification materials, and use the model to extract, fuse and detect table location features to obtain the table location;

[0050] The real estate table location detection model training module is used to train the table location detection model and obtain the detection mode with the best performance of the detection model;

[0051] The real estate table text recognition module is used to recognize the table text content according to the table position obtained by the table position detection model to obtain the table text content;

[0052] The variable matching module is used to match the obtained table position and text content with the form templates of real estate certification materials in various regions to obtain the real estate content recognition results;

[0053] The real estate table content output module is used to output the obtained real estate content.

[0054] In this system, by integrating modules such as image input, table position detection, text recognition and variable matching, the system realizes fully automated processing from raw real estate images to structured real estate information, improving the processing efficiency and accuracy of real estate images. The real estate table position detection model uses deep learning technology, which can automatically learn and extract table position features, so as to accurately locate the table area. This technology can handle complex and changeable real estate certification material formats and adapt to the differences in table layouts in different regions. Through the real estate table position detection model training module, the system can continuously optimize the performance of the detection model to ensure that it can maintain a high recognition rate and stability in different scenarios. This continuous optimization mechanism enables the system to adapt to the changing needs of real estate information recognition in the financial industry. The system uses the preset real estate certification material form templates for each region for variable matching, and matches the recognized table text content with the variables in the template, so as to parse out specific real estate information. This method not only improves the accuracy of information extraction, but also enables the system to flexibly respond to regulatory requirements in different regions.

[0055] Through the application of automation, intelligence and deep learning technologies, the system effectively overcomes the challenges faced by the financial industry in the process of real estate information identification, and provides financial institutions with efficient, accurate and flexible real estate information processing solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The figure is a schematic diagram of the operation flow of an embodiment of the present invention.

[0057] Figure 2 To collect and obtain image data of a certain property certification material.

[0058] Figure 3 In order to use the method and system of the present invention Figure 2 The recognition results of the real estate certificate image data are shown. DETAILED DESCRIPTION

[0059] The following is further described in detail by specific implementations, but the implementations of the present invention are not limited thereto. Unless otherwise specified, the technical means used in the following implementations are conventional means well known to those skilled in the art.

[0060] like Figure 1 to Figure 3 The present invention provides a real estate content recognition method based on target detection technology, and the specific steps are as follows:

[0061] Step 1: Build a real estate content recognition dataset.

[0062] First, collect and generate real estate images, collect and obtain real estate certification material image data from multiple channels, and a large amount of table image data automatically generated by Python program, and obtain the following Figure 2 The original data shown; secondly, the LabelImg tool is used to mark the position of the table in the original data, that is, the size of the table and the position in the image are marked, and a real estate table position detection dataset for deep neural network training is obtained. The label format of the dataset is YOLO, that is, (x, y, w, h, c), where (x, y) is the center point coordinate of the cell, (w, h) is the width and height of the cell, and c is the category of the cell. In this embodiment, there is only one category of cell.

[0063] Step 2: Construct training set, validation set, and test set for real estate content recognition.

[0064] According to the ratio of 8:1:1, the data in the data set is randomly screened, and the data set for real estate content recognition is constructed into a training set, a validation set, and a test set. Among them, the training set is used to train the real estate table detection model, and the validation set is used to dynamically adjust the parameters during model training. That is, every time the training set is trained, it is verified once with the validation set in order to adjust the parameters of subsequent training. The test set is used to evaluate the final training results of the model.

[0065] Step 3: Build a table location detection model in real estate certification materials.

[0066] First, a table position detection model in the real estate certification materials is constructed. When constructing the position detection model, a real estate table position feature extraction network is first built; the real estate table position feature extraction network includes a convolutional learning module, a residual learning module, and a spatial pyramid pooling module; the real estate table position feature extraction network in this embodiment is composed of 5 convolutional layers. Each time a convolution operation is performed, the size of the image will be reduced by half, and the network will also perform pooling operations and residual learning on the output of each convolutional layer to improve the receptive field and feature extraction capability of each layer of feature map.

[0067] Secondly, an upsampling module and a multi-scale feature fusion network for real estate table location detection are constructed, which can fuse the feature information of real estate tables at multiple scales to enhance the robustness and robustness of the model. At the same time, the features of the lower layer can be upsampled and then fused with the feature map of the upper layer to complete the fusion process of feature maps at multiple scales. In this embodiment, feature fusion is performed only on the feature maps of the 3rd, 4th, and 5th layers, because they will eventually be used as the input of the detection network.

[0068] Next, a real estate table location detection network is constructed. The table location detection network includes a large-scale detection layer for detecting large-sized real estate tables, a medium-scale detection layer for detecting medium-sized real estate tables, and a small-scale detection layer for detecting small-sized real estate tables. The table location detection network can predict the large, medium, and small scale table locations in the above-mentioned 3rd, 4th, and 5th layer feature maps. When the network is trained, the value of the loss function will be continuously reduced. The lower the loss, the better the model training. However, the network will detect a large number of results for the same table.

[0069] Finally, a non-maximum suppression network is constructed, through which the best detection results are screened out from thousands of detection results to obtain the final real estate table detection results.

[0070] Step 4: Train the table location detection model in the property certificate materials.

[0071] First, set the training parameters of the real estate certification material form location detection model. The training parameters in this embodiment are set as follows: the number of training times is 200, the type of form to be detected is 1, the learning rate is 0.01, and the input image size is 1280x1280. Secondly, the model is trained using the training set of real estate form detection. During training, the weights of the detection model are continuously optimized by back propagation and gradient descent to make the model reach the optimal solution. Each training is verified on the verification set so that the program can dynamically adjust some parameters of the model during training. Finally, after 200 trainings and parameter adjustments, a weight file for model training is obtained, referred to as the model, and the detection capability of the model is evaluated on the test set to obtain the detection mode with the best performance.

[0072] When evaluating the detection capability of the model, the evaluation indicators used are as follows:

[0073]

[0074] Wherein, formula (1) represents the accuracy, formula (2) represents the recall, formula (3) represents the average accuracy, and formula (4) represents the average precision; TP represents a positive example, FP represents a false positive example, and FN represents a false negative example; r i+1 -r i is an interval between recall rates, Pinterp(r i+1 ) represents r i To r i+1 A maximum interpolation value between ; K represents the number of categories.

[0075] The evaluation results of the present invention on the test set are shown in Table 1.

[0076] Table 1 Experimental results of the real estate table location detection model on the test set

[0077] Precision Recall <![CDATA[AP 0.5 ]]> <![CDATA[mAP 0.5-0.95 ]]> 0.953 0.911 0.957 0.752

[0078] The experimental results show that the evaluation results using this method have high accuracy.

[0079] Step 5: First, crop the image of each table according to the table position; second, send the cropped table image to the table position detection model for table position detection. During the detection, the detection model performs table position detection on the real estate table to be detected in the detection mode with the best performance, and sends the detected table position to the text recognition model for text recognition;

[0080] Step 6: Query the real estate form template for each region; according to the real estate certification material template for each region, perform variable matching on the results of the form position detection model and text recognition, that is, key->value pairing; finally, output the matching results, as shown below Figure 3 The identified property contents are shown.

[0081] In addition, this embodiment also provides a system for improving the content recognition capability of physical real estate certificates, including:

[0082] A real estate image input module, used to input real estate images;

[0083] The real estate table location detection module is used to build a table location detection model in real estate certification materials, and use the model to extract, fuse and detect table location features to obtain the table location;

[0084] The real estate table location detection model training module is used to train the table location detection model and obtain the detection mode with the best performance of the detection model;

[0085] The real estate table text recognition module is used to recognize the table text content according to the table position obtained by the table position detection model to obtain the table text content;

[0086] The variable matching module is used to match the obtained table position and text content with the form templates of real estate certification materials in various regions to obtain the real estate content recognition results;

[0087] The real estate table content output module is used to output the obtained real estate content.

[0088] The method of this embodiment can realize the rapid and accurate recognition of the forms in the real estate materials, thereby reducing the cumbersome process of customer submission of materials, speeding up the credit approval process, and helping to improve customer satisfaction. The system effectively overcomes the challenges faced by the financial industry in the process of real estate information identification through the application of automation, intelligence and deep learning technology, and provides financial institutions with efficient, accurate and flexible real estate information processing solutions.

[0089] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for improving the content recognition capability of a physical real estate certificate, characterized in that: The following steps are involved: S1: Collect and generate real estate images, mark the positions of real estate tables in the images, and obtain the real estate table position dataset required for deep neural network training; S2: Randomly select data in the dataset and construct training set, validation set and test set respectively; S3: Construct a table location detection model in the real estate certification material. When constructing the location detection model, first build a real estate table location feature extraction network; then build a table location feature fusion network; Then build a table position detection network; then build a non-maximum suppression network; S4: Use the training set to train the table location detection model. After each training, use the validation set to verify the table location model and continuously optimize the weight of the detection model. Take the model weight with the best verification effect, and use the test set to evaluate the detection ability of the model to obtain the detection mode with the best performance. S5: Using the detection model to detect the position of the real estate table to be detected in the detection mode with the best performance, and sending the detected table position to the text recognition model for text content recognition to obtain the position and text content of the table in the real estate table; S6: According to the position and content of the form and the template of the real estate certification material form in each region, variable matching is performed to finally obtain the recognition result of the real estate content.

2. A method for improving the content recognition capability of a physical real estate certificate according to claim 1, characterized in that ,In step S1, first, the image data of real estate certification materials and ,a large number of table image data automatically generated by the python ,program are obtained from multiple channels to obtain the original real estate data ,image; secondly, when marking the position of the table, the LabelImg tool is used to ,annotate the size of the table in the original real estate data image and the ,position of the table in the image to obtain the data set for deep ,neural network training.

3. A method for improving the content recognition capability of a physical real estate certificate according to claim 1, characterized in that: In step S2, the property table location dataset is randomly divided into a training set, a validation set, and a test set in a ratio of 8:1:

1.

4. A method for improving the content recognition capability of a physical real estate certificate according to claim 1, characterized in that: In step S3, the real estate table position feature extraction network includes a convolutional learning module, a residual learning module, and a spatial pyramid pooling module; the table position feature fusion network includes an upsampling module and a multi-scale feature fusion module; the table position detection network includes a large-scale detection layer for detecting large-size real estate tables, a medium-scale detection layer for detecting medium-size real estate tables, and a small-scale detection layer for detecting small-size real estate tables; the non-maximum suppression network is used to screen out the best real estate table detection result from multiple detection results.

5. A method for improving the content recognition capability of a physical real estate certificate according to claim 1, characterized in that: In step S4, first, the parameters of the real estate table detection model are set, including the number of training times, the type of table to be detected, the learning rate, and the size of the input real estate image; secondly, the model is trained using the training set of real estate table detection, and the weights of the detection model are continuously optimized by back propagation and gradient descent during training, so that the model reaches the optimal solution; finally, the detection capability of the model is evaluated on the real estate table detection test set to obtain the detection mode with the best performance.

6. A method for improving the content recognition capability of a physical real estate certificate according to claim 1, characterized in that: In step S5: first, the image of each table is cropped according to the table position; second, the cropped table image is sent to the table position detection model to detect the table position, and the detected table position is sent to the text recognition model for text recognition.

7. A method for improving the content recognition capability of a physical real estate certificate according to claim 1, characterized in that: In step S6, first, the real estate table template of each region is queried; secondly, variable matching is performed according to the template, the table position and text content obtained in step S5, that is, key->value pairing; finally, the matching result is output to obtain the identified real estate content.

8. A system for implementing the method according to any one of claims 1 to 7, characterized in that: include: A real estate image input module, used to input real estate images; The real estate table location detection module is used to build a table location detection model in real estate certification materials, and use the model to extract, fuse and detect table location features to obtain the table location; The real estate table location detection model training module is used to train the table location detection model and obtain the detection mode with the best performance of the detection model; The real estate table text recognition module is used to recognize the table text content according to the table position obtained by the table position detection model to obtain the table text content; The variable matching module is used to match the obtained table position and text content with the form templates of real estate certification materials in various regions to obtain the real estate content recognition results; The real estate table content output module is used to output the obtained real estate content.

Images