A construction land approval form automatic filling method, device, medium and product

Abstract

The application discloses a construction land approval form automatic filling method, equipment, medium and product, relates to the field of construction land, and the method comprises the following steps: obtaining construction land approval materials, and constructing a construction land approval material dataset; based on the construction land approval material dataset, an optical character recognition model is constructed; based on a text file dataset, a land natural language processing model is constructed; the trained optical character recognition model is used for text recognition on the construction land approval materials to be processed, so that the text files to be processed are obtained; the trained land natural language processing model is used for information extraction on the text files to be processed, so that the approval information to be processed is obtained; and the approval information to be processed is mapped to the approval form based on a preset rule configuration library, and the approval form is filled. The application can improve the filling efficiency and accuracy of the approval form.

Description

Technical Field

[0001] This application relates to the field of construction land, and in particular to a method, device, medium and product for automatically filling in construction land application forms. Background Technology

[0002] Currently, the approval process for construction land mainly relies on manual operation. Relevant staff manually collect various approval materials such as project approval documents, survey and demarcation reports, land space planning documents, land ownership certificates, and geological hazard assessment reports; manually read and screen the above materials one by one, manually extract key information such as project name, land area, boundary coordinates, and land type; manually enter the extracted information into various appendices; manually check the consistency of the entered data and check whether the materials comply with the approval rules; finally, manually format and organize the materials to generate the approval submission file and submit it to the approval system.

[0003] Currently, existing technologies have many problems. First, manual data entry is extremely inefficient. There are many types of application materials (no fewer than 10 categories) and complex fields (over 50 fields per parcel). Manually filling out a single parcel takes 4-6 hours, which is time-consuming and labor-intensive, and cannot meet the needs of large-scale applications. Second, the human error rate is high. Repeated copying and data entry can easily lead to errors in key data such as land area and boundary coordinates, causing application materials to be returned for correction and affecting the approval process. Third, data verification is difficult. Inconsistencies can easily occur in the same data (such as land area) between multiple documents. Manual verification is labor-intensive, prone to omissions, and difficult to guarantee data accuracy. Furthermore, multi-source file processing is difficult. It cannot achieve unified structured extraction of multiple file formats such as PDFs, scanned documents, and Excel spreadsheets, requiring manual processing separately, further increasing the workload. Finally, professional adaptability is poor. General form-filling software cannot recognize professional terms and aliases in the application field, resulting in low extraction accuracy and making it unsuitable for direct application in construction land application scenarios.

[0004] Therefore, based on the above problems, there is an urgent need to provide a method for automatically filling in the application form for construction land use, which can improve the efficiency and accuracy of filling in the application form. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, medium, and product for automatically filling in construction land application forms, which can improve the efficiency and accuracy of filling in application forms.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for automatically filling in a construction land application form, including: Obtain the application materials for construction land use approval and construct a dataset of such materials. Based on a dataset of land use approval documents, an optical character recognition (OCR) model is constructed. The OCR model includes a Swin converter, a feature pyramid network, and a prediction head unit. The Swin converter is used to extract features from the land use approval documents to obtain spatial and semantic features. The feature pyramid network is used to fuse the spatial and semantic features to obtain fused features. The prediction head unit is used to determine the text file of the land use approval documents based on the fused features. A land natural language processing model is constructed based on a text file dataset. The model includes an input layer, an encoding layer, a fusion layer, and an output layer. The input layer converts text files into text vectors. The encoding layer encodes the text vectors to obtain text features. The fusion layer uses an attention mechanism to weightedly fuse text features with knowledge vectors to obtain land-enhanced semantic vectors. The knowledge vectors are vectorized from land resource databases, land standard and normative corpora, and historical land records. The output layer performs sequence labeling on the land-enhanced semantic vectors to obtain approval information. The text file dataset is constructed based on construction land approval text materials. The trained optical character recognition model is used to perform text recognition on the construction land approval materials to be processed, and the text file to be processed is obtained. The trained land natural language processing model is used to extract information from the text file to be processed, and the approval information to be processed is obtained. Based on a pre-defined rule configuration library, the approval information to be processed is mapped to the approval form, and the approval form is populated.

[0007] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of an automatic filling method for construction land approval forms.

[0008] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of an automatic filling method for a construction land approval form.

[0009] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of an automatic filling method for construction land approval forms.

[0010] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, device, medium, and product for automatically filling in construction land approval forms. It involves acquiring construction land approval materials and constructing a dataset of these materials, then further building an Optical Character Recognition (OCR) model and a land natural language processing (NLP) model. The trained OCR model is then used to extract text from the construction land approval materials, and the trained NLP model is used to extract information from the text file. This application combines the OCR model with the NLP model, reducing manual data entry and verification, and improving the accuracy of data extraction. Based on a pre-defined rule configuration library, the application information is mapped to the approval form, and the form is filled in, significantly improving data entry efficiency and reducing the data error rate. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating an automatic filling method for a construction land approval form in one embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] In one exemplary embodiment, such as Figure 1 As shown, a method for automatically filling in a construction land approval form is provided, including the following S1 to S6. Wherein: S1: Obtain the application materials for construction land and construct a dataset of construction land application materials.

[0016] This system retrieves user-uploaded land use approval documents in batches. These documents are available in formats including PDF, scanned files, Word, and Excel. The documents include stamped documents, handwritten notes, uncommon place names, and various anti-counterfeiting textured paper. A land use approval document dataset is then constructed based on these documents.

[0017] S2: Construct an optical character recognition model based on the dataset of construction land approval materials.

[0018] The optical character recognition model includes a Swin converter, a Feature Pyramid Network (FPN), and a prediction head unit. The Swin converter is used to extract features from the land use approval materials to obtain spatial and semantic features. The Feature Pyramid Network is used to fuse the spatial and semantic features to obtain fused features. The prediction head unit is used to determine the text file of the land use approval materials based on the fused features.

[0019] Specifically, the optical character recognition model in this application adopts an end-to-end multi-task deep learning architecture, using a backbone-neck-multi-head structure. The backbone is a Swin Transformer used to extract spatial and semantic features from the application materials; the neck is a feature pyramid network used to fuse shallow high-resolution spatial features (spatial features) with deep semantic features to obtain fused features, which are used to locate the target (microtext, seal boundary); the prediction head unit includes multiple prediction heads, namely a connection-sequential classification decoding head (or attention mechanism decoding head), a lightweight segmentation head, and a classification discriminant head. The connection-sequential classification decoding head (or attention mechanism decoding head) is used to perform text recognition on the fused features to obtain a text file; the lightweight segmentation head (U-Net structure) is used to segment the fused features to obtain the seal area and background area (for anti-counterfeiting feature analysis); the classification discriminant head is used to recognize handwritten and printed characters based on the fused features and output the confidence score of uncommon characters.

[0020] In an exemplary embodiment, the optical character recognition model can be a convolutional recurrent neural network or a Transformer architecture. When training the optical character recognition model, it is first pre-trained using publicly available datasets (e.g., Chinese street view, printed documents) to give it basic text recognition capabilities. Subsequently, the model is trained using a dataset of land use approval documents. Specifically, the land use approval documents are augmented, including processes such as simulated occlusion, random distortion, and blur enhancement. Based on the augmented land use approval documents, the optical character recognition model is trained using a loss function (end-to-end backpropagation based on the total loss) to obtain the trained optical character recognition model. The loss function is calculated as follows: ; in, For the total loss, For connection-time classification loss coefficients, The connection-time classification loss is used for text file alignment. The sum of the Dice loss and the binary cross-entropy loss coefficients is used. The weighted sum of Dice loss and binary cross-entropy is used for segmentation and localization of stamps / background patterns. The cross-entropy loss coefficient is... Cross-entropy loss is used for classifying handwritten / uncommon characters.

[0021] The optical character recognition model in this application employs a Swing Transformer backbone network combined with a transfer learning strategy. Based on general pre-training, it is fine-tuned using a dataset of construction land approval materials. The loss function uses CTC loss, resolving the text alignment problem and forming a closed-loop solution integrating data, model, and rules for the approval scenario. Furthermore, the trained optical character recognition model can automatically extract stroke and edge features from construction land approval materials (including text and mixed text / image information) and convert them into structured text files. It also outputs stamp areas, background areas, handwritten markings, printed markings, and rare character markings. Further anti-counterfeiting feature analysis can be performed using the stamp and background areas, eliminating the need for manual rule design.

[0022] S3: Construct a land natural language processing model based on a text file dataset.

[0023] A text file dataset is constructed based on the application materials for construction land approval. These materials include land use requests, land use approvals, planning documents, various assessment reports, and land use agreements—all necessary documents for construction land approval. The land natural language processing model comprises an input layer, an encoding layer, a fusion layer, and an output layer. The input layer converts the text files into text vectors. The encoding layer encodes the text vectors to obtain text features. The fusion layer uses an attention mechanism to weightedly fuse text features with knowledge vectors to obtain land-enhanced semantic vectors. These knowledge vectors are obtained by vectorizing land resource databases, land standard and normative corpora, and historical land records. The output layer performs sequence labeling on the land-enhanced semantic vectors to obtain the approval information.

[0024] Named Entity Recognition (NER) modules in Natural Language Processing (NLP) models often employ Transformer-based pre-trained models (such as BERT and RoBERTa). These pre-trained models can understand complex contextual semantics and automatically capture long-distance dependencies without requiring tedious manual feature engineering. However, NLP models cannot understand land-related technical terms such as agricultural land, basic farmland, and land ownership. Therefore, this application embeds knowledge vectors (obtained by vectorizing land resource databases, land standard and normative corpora, and historical land records) into the NLP model to fine-tune it, resulting in a land-related NLP model. Specifically, the input embedding layer (input layer) of the land natural language processing model converts text files into vector representations; the context encoding layer (encoding layer) is based on the BERT / RoBERTa backbone network of Transformer, and captures the contextual semantic dependencies of the text through a multi-head self-attention mechanism to obtain text features; the land semantic fusion layer (fusion layer) integrates the land resource database, land standard and normative corpus, and land historical case files into knowledge data, vectorizes the knowledge data to obtain knowledge vectors, and aligns and weights the text features with the knowledge vectors through an attention mechanism to output land-enhanced semantic vectors; the task output layer (output layer) performs sequence labeling (BIO labeling strategy) based on the fused land-enhanced semantic vectors and outputs the named entity recognition results, i.e., the approval information.

[0025] In one exemplary embodiment, land natural language processing (NLP) models often employ Transformer-based pre-trained models (such as BERT and RoBERTa). This application employs a two-stage transfer learning strategy to train the land NLP model. First, pre-training is performed on a large-scale general corpus to give the model basic language understanding capabilities. Then, end-to-end fine-tuning is performed on a dataset of text files labeled with land-related entities (including multiple text files related to construction land). The cross-entropy loss function is used to optimize the model parameters, resulting in a well-trained land NLP model. The trained model can accurately distinguish and extract business fields in the land domain, resolving issues such as confusion of similar fields, identification of professional entities, and correlation of table items.

[0026] S4: Use the trained optical character recognition model to perform text recognition on the construction land approval materials to be processed, and obtain the text file to be processed.

[0027] Prior to this, the application materials for land use approval are preprocessed to obtain preprocessed application materials. The preprocessing process includes tilt correction (automatic tilt correction), filtering and noise reduction, and image enhancement. Preprocessing can improve the accuracy of subsequent text extraction. After preprocessing, the preprocessed application materials are uniformly structured and stored in the database for easy extraction and comparison later.

[0028] The pre-processed application materials (in image format, uniformly normalized to a fixed resolution, such as 1024×1024, RGB three channels) are input into a trained optical character recognition (OCR) model, and the output is a text file to be processed. The trained OCR model extracts text and image-text mixed information from the pre-processed application materials (unstructured data) (extracting text content and recognizing table information), and converts it into a processable structured text file with layout structure (the text file to be processed). This achieves the transition from images or unstructured content to pure text data, solving the text extraction challenges of land application paper scans, PDF scans, photographs with blurry text, misaligned tables, obscured seals, and mixed formats (approval documents, planning map appendices, ownership certificates, preliminary review opinions, land survey and demarcation tables). Furthermore, the trained OCR model can also perform layout analysis on the pre-processed application materials, that is, recognize the title, body text, tables, seals, and other areas of the pre-processed application materials. After practical application, the results show that the recognition accuracy of the trained optical character recognition model reaches over 99% for approval documents with fixed formats.

[0029] S5: Use the trained land natural language processing model to extract information from the text file to be processed, and obtain the approval information to be processed.

[0030] A well-trained land natural language processing model can accurately distinguish and extract business fields from text files, solving the pain points of natural language processing models that cannot distinguish similar land fields, cannot identify professional business entities, and cannot associate tabular data.

[0031] The trained land natural language processing model uses named entity recognition technology to accurately locate key information (keywords) with business value in the text file to be processed and extract the information to be processed, including project information, approval information, land use information, ownership information and planning information.

[0032] Specifically, project information includes project name, project code, full name of construction unit, construction nature, investment scale, etc.; approval information includes approval unit, approval date, approval authority, construction content, etc.; land use information includes land location (accurate to city, county, township, and village), total land area, agricultural land area, construction land area, unused land area, cultivated land area (including paddy fields / dry land), basic farmland area, etc.; ownership information includes land owner, land user, and ownership nature, etc.; planning information includes land spatial planning, planning map number, geological hazard assessment conclusions, and information on important mineral resources covered by land, etc.

[0033] Each approval entry corresponds to a specific project; that is, the approval information is the variable real business value (dynamic business data) of a single project. The approval information is used to provide the original business values ​​to be filled. An example table of approval information is shown in Table 1: Table 1. Example of Approval Information

[0034] This application combines an optical character recognition (OCR) model with a land natural language processing (NLP) model, enabling it to efficiently and accurately transform complex construction land approval materials into high-value structured data assets (approval information). The OCR model and the NLP model are not isolated entities, but rather deeply integrated through a seamlessly connected automated pipeline to achieve efficient and accurate extraction of approval information.

[0035] S6: Based on the preset rule configuration library, the approval information to be processed is mapped to the approval form, and the approval form is populated.

[0036] Specifically, this application pre-constructs a rule configuration library, also known as a special field mapping library for construction land approval. This library is uniformly configured according to natural resource approval specifications and land form standards. Its core content consists of a field mapping lookup table and a data dictionary, used to automatically translate and convert fields in the approval information to columns in the approval form. For example, data corresponding to the project name in the approval information is added to the project name column in the "Construction Land Application Form," and data corresponding to the total land area in the approval information is added to the total land area column in the "Construction Project Land Use Report."

[0037] In this application, the rule configuration library only stores the matching mapping relationship of approval information field name ↔ approval form name ↔ approval form column name, and does not contain any actual values ​​of the items. An example table is shown in Table 2: Table 2 Example table of rule configuration library

[0038] Based on the matching mapping relationships in the rule configuration library, the approval information to be processed is mapped to the approval forms, and the approval forms are populated. The approval forms include the construction land application form, the construction project land submission explanation, and the land classification area table, etc. The approval forms are available in Excel, Word, PDF forms, and standard XML formats. This application supports batch export of approval forms.

[0039] The rules in the rule configuration library are static rules, applicable across the entire project, and will not change due to changes in the application materials for construction land. The mapping rules are only updated when the application form is revised; that is, the rule configuration library updates the mapping rules according to the changes in the application form.

[0040] Compared to traditional methods, this application significantly improves the efficiency of filling out application forms through end-to-end automation. The time for filling out a single land parcel has been reduced from the traditional 4-6 hours to 10-15 minutes, an efficiency improvement of over 24 times. This application supports batch processing, enabling a daily processing volume of no less than 100 land parcels. Secondly, this application reduces the human error rate from over 8% to below 0.5%, effectively reducing returns for correction due to problems with construction land application materials and significantly improving the approval rate. Thirdly, this application supports one-stop parsing and fusion of multiple file formats such as PDF, scanned documents, and Excel, eliminating the need for separate manual processing and possessing powerful multi-source material unified processing capabilities. Furthermore, regarding compliance assurance, this application automatically completes key checks such as area balance, land category logic, and planning compliance, proactively mitigating approval risks.

[0041] In practical applications, this application can handle large files exceeding 500MB and multiple batches of materials daily without crashes or lags, effectively ensuring business continuity. Furthermore, this application conforms to approval form specifications and rules, supports adaptation to form templates from different regions, and can be directly integrated with approval systems without secondary modifications. Moreover, this application reduces the workload of manual data entry, verification, and organization, decreasing the approval process by more than 50% and lowering manpower requirements.

[0042] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores automatically populated data for construction land approval forms. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for automatically populating construction land approval forms.

[0043] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0044] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0045] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0046] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0047] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0048] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0049] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0050] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for automatically filling in a construction land approval form, characterized in that, The method includes: Obtain the application materials for construction land use approval and construct a dataset of such materials. Based on a dataset of land use approval documents, an optical character recognition (OCR) model is constructed. The OCR model includes a Swin converter, a feature pyramid network, and a prediction head unit. The Swin converter is used to extract features from the land use approval documents to obtain spatial and semantic features. The feature pyramid network is used to fuse the spatial and semantic features to obtain fused features. The prediction head unit is used to determine the text file of the land use approval documents based on the fused features. A land natural language processing model is constructed based on a text file dataset. The model includes an input layer, an encoding layer, a fusion layer, and an output layer. The input layer converts text files into text vectors. The encoding layer encodes the text vectors to obtain text features. The fusion layer uses an attention mechanism to weightedly fuse text features with knowledge vectors to obtain land-enhanced semantic vectors. The knowledge vectors are vectorized from land resource databases, land standard and normative corpora, and historical land records. The output layer performs sequence labeling on the land-enhanced semantic vectors to obtain approval information. The text file dataset is constructed based on construction land approval text materials. The trained optical character recognition model is used to perform text recognition on the construction land approval materials to be processed, and the text file to be processed is obtained. The trained land natural language processing model is used to extract information from the text file to be processed, and the approval information to be processed is obtained. Based on a pre-defined rule configuration library, the approval information to be processed is mapped to the approval form, and the approval form is populated.

2. The method for automatically filling in the construction land approval form according to claim 1, characterized in that, The process of using a trained optical character recognition model to perform text recognition on the construction land approval materials to be processed, resulting in a text file to be processed, also includes the following steps: Pre-processing of the application materials for construction land to be processed.

3. The method for automatically filling in the construction land approval form according to claim 2, characterized in that, The preprocessing process includes: tilt correction, filtering and noise reduction, and image enhancement.

4. The method for automatically filling in the construction land approval form according to claim 1, characterized in that, The information to be submitted for approval includes: project information, approval information, land use information, ownership information, and planning information.

5. The method for automatically filling in the construction land approval form according to claim 1, characterized in that, The approval forms can be in the following formats: Excel, Word, PDF, and XML standard forms.

6. A method for automatically filling in a construction land approval form, characterized in that, The process of using a trained optical character recognition model to perform text recognition on the construction land approval materials to be processed, resulting in a text file to be processed, also includes the following steps: Using formula Train the optical character recognition model; in, For the total loss, For connection-time classification loss coefficients, For connection-time classification loss, The sum of the Dice loss and the binary cross-entropy loss coefficients is used. The sum of Dice loss and binary cross-entropy is given. The cross-entropy loss coefficient is... This represents the cross-entropy loss.

7. The method for automatically filling in the construction land approval form according to claim 1, characterized in that, The method further includes: The pre-configured rule library updates the mapping rules based on changes in the approval forms.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the automatic filling method for construction land approval forms according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the automatic filling method for the construction land approval form as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the automatic filling method for the construction land approval form as described in any one of claims 1-7.

Images