A large model driven cadastral archive digitization processing method and system, electronic equipment and storage medium

Abstract

This application discloses a method, system, electronic device, and storage medium for digitizing cadastral archives driven by a large model. The method includes: acquiring cadastral archive raster images; extracting map element information using an OCR recognition model; performing structured parsing of boundary point description text and ownership description text in the map element information using a large language model, automatically constructing the spatial connection order and ordered boundary point topology; calculating the plane coordinates of boundary points based on the map frame coordinates and scale parameters in the map element information; generating a ownership redline vector map based on the boundary point plane coordinates; performing automated quality inspection on the ownership redline vector map graphics; and, after passing the quality inspection, integrating the boundary point plane coordinates, ownership redline vector map graphics, and standard ownership attribute information into a database. This application enables end-to-end automated conversion of cadastral archives from unstructured raster images to a standard vector spatial database.

Description

Technical Field

[0001] This application belongs to the interdisciplinary technical field of cadastral archive data processing and artificial intelligence, specifically involving a large model-driven method, system, electronic device and storage medium for digitizing cadastral archives, which is used to solve the problem of high-precision conversion of historical cadastral maps to standard vector space databases. Background Technology

[0002] Currently, with the deepening of smart city construction and the unified real estate registration system, massive amounts of historical paper-based cadastral archives urgently need to be transformed into digital and vectorized spatial data to achieve refined management and efficient retrieval of land resources. The digitization of cadastral archives is not merely about scanning paper maps into raster images; more importantly, it involves extracting boundary point coordinates, ownership red lines, and attribute information to construct a standard GIS (Geographic Information System) vector database.

[0003] However, existing land registration digitization technologies face severe challenges when processing complex historical archives, including low efficiency, insufficient accuracy, and difficulties in constructing topological logic. These challenges are specifically reflected in the following three aspects:

[0004] 1. Traditional manual digital processing is inefficient and highly subjective.

[0005] Currently, mainstream digitization methods in the industry still heavily rely on manual interaction. In practice, operators must manually digitize (trace) and input attributes on screens using CAD or GIS software, interpreting paper documents or scanned copies by eye. This model has significant limitations:

[0006] (1) Manual point-by-point drawing and entry is extremely slow and cannot meet the digitization needs of massive historical archives, resulting in long data production cycles and high costs;

[0007] (2) When manually judging the location of boundary points, there are visual biases and subjective arbitrariness. Different operators may produce inconsistent vector results when processing the same file, making it difficult to guarantee the standardization and consistency of data.

[0008] 2. Traditional OCR technology struggles to solve the semantic-spatial mapping problem.

[0009] Although Optical Character Recognition (OCR) technology has been attempted to be introduced into the archival digitization process, its application in the cadastral field has been limited. Traditional OCR technology mainly focuses on recognizing general text and cannot understand the complex semantic-spatial coupling relationships in cadastral maps.

[0010] (1) Traditional OCR can only recognize strings such as “Boundary point 1:X=100,Y=200”, but cannot distinguish whether the coordinates are text attributes of map annotations or geometric positions that need to be drawn precisely in space;

[0011] (2) The cadastral red line requires a strict closed topological structure. Traditional OCR cannot handle the logical order of the boundary point connection (such as clockwise or counterclockwise). When faced with missing, skipped, or chaotic boundary point numbers, it cannot automatically generate closed polygon vector graphics.

[0012] 3. Existing AI technologies lack deep integration with surveying and mapping geometry algorithms.

[0013] In recent years, some document processing systems based on deep learning (such as CNN or NLP) have begun to emerge. However, these systems mostly focus on document classification, retrieval, or simple summary generation, lacking in-depth support for mapping coordinate systems, scale conversion, and geometric topology construction.

[0014] (1) Existing AI models usually separate text recognition from graphics rendering, and cannot achieve automatic mapping from pixel coordinates to real plane coordinates;

[0015] (2) Faced with the common problems of deformation, fading and stains in historical archives, existing technologies lack the ability to use large models for contextual logic reasoning and error correction, resulting in a high error rate in digital results, which still requires a lot of manual review.

[0016] In summary, existing technologies lack an automated processing solution that deeply integrates visual perception, semantic understanding, and surveying calculations, making it difficult to achieve end-to-end conversion from unstructured raster archives to high-precision structured vector data. Therefore, there is an urgent need for a large-model-driven method for digitizing cadastral archives. This method should leverage the powerful logical reasoning capabilities of large language models to analyze the topological relationships of boundary points, combining OCR visual recognition and surveying coordinate calculation algorithms to achieve fully automated vectorization and data storage of cadastral archives, thereby addressing the aforementioned technical challenges. Summary of the Invention

[0017] The purpose of this application is to provide a method, system, electronic device and storage medium for large-model-driven digitization of cadastral archives, in order to solve the problems of low efficiency, difficulty in topology construction and insufficient automation in the existing technology of cadastral archive digitization.

[0018] The first objective of this application is to provide a large-model-driven method for digitizing cadastral archives.

[0019] The aforementioned objective of this application is achieved through the following technical solution:

[0020] A large-model-driven method for digitizing cadastral archives, the method comprising:

[0021] Obtain raster image data of cadastral archives;

[0022] The raster image data is used to perform image and text recognition using an OCR recognition model to extract map feature information, which includes map frame coordinates, scale parameters, boundary point description text, and ownership description text.

[0023] The boundary point description text and ownership description text in the map element information are structured and parsed using a large language model to obtain the spatial connection order and standard ownership attribute information of the boundary points, and an ordered boundary point topology is constructed based on the spatial connection order of the boundary points.

[0024] Based on the map frame coordinates and scale parameters in the map element information, and combined with the ordered boundary point topology, the plane coordinates of the boundary points are calculated using a surveying coordinate calculation algorithm.

[0025] Generate a red line vector graphic of ownership based on the plane coordinates of the boundary points;

[0026] The generated ownership red line vector graphics are subjected to automated quality inspection. If the quality inspection passes, a spatial dataset containing the plane coordinates of the boundary points, the ownership red line vector graphics, and standard ownership attribute information is constructed and stored in the spatial database.

[0027] Preferably, before performing image and text recognition on the raster image data using the OCR recognition model, the method further includes:

[0028] The raster image data is preprocessed, including image correction, noise reduction, and image enhancement, in order to improve the recall and accuracy of key image elements in subsequent OCR recognition.

[0029] The graphic enhancement is used to highlight the outline of the drawing frame, the coordinate grid, and the features of the boundary point symbols.

[0030] Preferably, the step of using an OCR recognition model to perform image and text recognition on the raster image data and extract map feature information includes:

[0031] The enhanced image is subjected to frame parameter recognition using an OCR recognition model. The geodetic coordinates of the four corners of the frame are extracted as the frame coordinates. The grid spacing and scale parameters are extracted. The grid spacing is used to correct the distortion of the map scale, and the scale parameters are used to calculate the correspondence between image pixels and actual distances.

[0032] The enhanced image is identified using an OCR recognition model, and ownership information is extracted to extract ownership attribute annotations as ownership description text. The ownership attribute annotations include the name of the right holder, the land parcel code, and the use category.

[0033] The enhanced image is identified using an OCR recognition model. If there are boundary points with specific coordinate labels in the image, the coordinate values ​​are directly extracted as the coordinate information of the boundary point description text, and the extraction of the boundary point outline pixel position is skipped. If there are no coordinate labels in the image, the pixel position of the boundary point outline is extracted and the pixel position is used as the positioning information of the boundary point description text.

[0034] Preferably, the step of using a large language model to perform structured parsing of the boundary point description text and ownership description text in the map element information to obtain the spatial connection order and standard ownership attribute information of the boundary points includes:

[0035] The boundary point description text, ownership description text, and corresponding OCR recognition confidence information are combined with a preset cadastral surveying standard knowledge base fragment and encapsulated into a Prompt input command containing context constraints.

[0036] The pre-trained large language model is invoked to parse the Prompt input command, identify abnormal situations such as missing boundary point numbers, logical breakpoints, or coordinate contradictions, and automatically complete or correct the connection order of boundary points based on the connection rules and geometric closure principles in cadastral surveying specifications.

[0037] Output structured JSON data that conforms to the real estate registration database standard. The structured JSON data shall include at least: a unique identifier for the boundary point, a standard ownership attribute description of the land parcel to which the boundary point belongs, and an array of topological connection relationships arranged in the order of actual connection.

[0038] Preferably, the step of calculating the planar coordinates of the boundary points based on the map frame coordinates and scale parameters in the map element information, combined with the ordered boundary point topology, using a surveying coordinate calculation algorithm includes:

[0039] Using the coordinates of the four corners of the map frame as reference control points, calculate the scaling factor in the X / Y direction based on the image pixel size and the actual size of the map frame;

[0040] For boundary points that only have pixel positions on the map but no specific coordinates, their planar coordinates are calculated using a spatial interpolation algorithm combined with the scaling factor.

[0041] For boundary points with coordinate closure errors, the least squares method is used for adjustment to correct the coordinate calculation errors.

[0042] Preferably, the automated quality inspection of the generated ownership redline vector graphics includes:

[0043] Check the geometric closure of the vector graphics to ensure that the land parcel boundary lines are connected end to end;

[0044] Check the topology to ensure there are no self-intersections or dangling points;

[0045] If the area calculated from the vector graphic is compared with the area of ​​ownership recorded in the archive, and the error exceeds a preset threshold, it is marked as an abnormal graphic.

[0046] Preferably, the step of constructing a spatial dataset containing the planar coordinates of the boundary points, the vector graphics of the ownership red line, and standard ownership attribute information, and storing the spatial dataset in a spatial database, includes:

[0047] The planar coordinates of the boundary points are associated and bound with the ownership red line vector graphics to generate standard vector format data;

[0048] The standard ownership attribute information is written into the attribute table of the spatial database, wherein the spatial database includes PostGIS or ArcSDE.

[0049] The second objective of this application is to provide a large-scale model-driven cadastral archive digitization system.

[0050] The second objective of this application is achieved through the following technical solution:

[0051] A large-model-driven cadastral archive digitization system, the system comprising:

[0052] The data acquisition module is used to acquire raster image data from cadastral archives;

[0053] The information extraction module is used to perform image and text recognition on the raster image data using an OCR recognition model to extract map feature information, including map frame coordinates, scale parameters, boundary point description text, and ownership description text.

[0054] The large model parsing module is used to perform structured parsing of the boundary point description text and ownership description text in the map element information using a large language model, to obtain the spatial connection order and standard ownership attribute information of the boundary points, and to construct an ordered boundary point topology based on the spatial connection order of the boundary points.

[0055] The coordinate calculation module is used to calculate the planar coordinates of the boundary points based on the map frame coordinates and scale parameters in the map element information, combined with the ordered boundary point topology, and using a surveying coordinate calculation algorithm.

[0056] The vector graphic generation module is used to generate a vector graphic of the ownership red line based on the planar coordinates of the boundary points;

[0057] The data quality inspection and storage module is used to automatically inspect the generated ownership red line vector graphics. If the quality inspection passes, a spatial dataset containing the plane coordinates of the boundary points, the ownership red line vector graphics, and standard ownership attribute information is constructed and stored in the spatial database.

[0058] Preferably, the system further includes:

[0059] The data preprocessing module is used to preprocess the raster image data, and the preprocessing includes image correction, noise reduction and image enhancement.

[0060] The graphic enhancement is used to highlight the features of the map frame outline, coordinate grid and boundary point symbols, so as to improve the recall and accuracy of subsequent OCR recognition of key map elements.

[0061] Preferably, when the information extraction module performs image and text recognition on the raster image data using an OCR recognition model to extract map feature information, it is specifically used for:

[0062] The enhanced image is subjected to frame parameter recognition using an OCR recognition model. The geodetic coordinates of the four corners of the frame are extracted as the frame coordinates. The grid spacing and scale parameters are extracted. The grid spacing is used to correct the distortion of the map scale, and the scale parameters are used to calculate the correspondence between image pixels and actual distances.

[0063] The enhanced image is identified using an OCR recognition model, and ownership information is extracted to extract ownership attribute annotations as ownership description text. The ownership attribute annotations include the name of the right holder, the land parcel code, and the use category.

[0064] The enhanced image is identified using an OCR recognition model. If there are boundary points with specific coordinate labels in the image, the coordinate values ​​are directly extracted as the coordinate information of the boundary point description text, and the extraction of the boundary point outline pixel position is skipped. If there are no coordinate labels in the image, the pixel position of the boundary point outline is extracted and the pixel position is used as the positioning information of the boundary point description text.

[0065] Preferably, when the large model parsing module performs structured parsing of the boundary point description text and ownership description text in the map feature information using a large language model to obtain the spatial connection order and standard ownership attribute information of the boundary points, it is specifically used for:

[0066] The boundary point description text, ownership description text, and corresponding OCR recognition confidence information are combined with a preset cadastral surveying standard knowledge base fragment and encapsulated into a Prompt input command containing context constraints.

[0067] The pre-trained large language model is invoked to parse the Prompt input command, identify abnormal situations such as missing boundary point numbers, logical breakpoints, or coordinate contradictions, and automatically complete or correct the connection order of boundary points based on the connection rules and geometric closure principles in cadastral surveying specifications.

[0068] Output structured JSON data that conforms to the real estate registration database standard. The structured JSON data shall include at least: a unique identifier for the boundary point, a standard ownership attribute description of the land parcel to which the boundary point belongs, and an array of topological connection relationships arranged in the order of actual connection.

[0069] Preferably, when the coordinate calculation module performs the calculation of the plane coordinates of the boundary points based on the map frame coordinates and scale parameters in the map feature information, combined with the ordered boundary point topology, and using the surveying coordinate calculation algorithm, it is specifically used for:

[0070] Using the coordinates of the four corners of the map frame as reference control points, calculate the scaling factor in the X / Y direction based on the image pixel size and the actual size of the map frame;

[0071] For boundary points that only have pixel positions on the map but no specific coordinates, their planar coordinates are calculated using a spatial interpolation algorithm combined with the scaling factor.

[0072] For boundary points with coordinate closure errors, the least squares method is used for adjustment to correct the coordinate calculation errors.

[0073] Preferably, when the data quality inspection and warehousing module performs automated quality inspection on the generated ownership red line vector graphics, it is specifically used for:

[0074] Check the geometric closure of the vector graphics to ensure that the land parcel boundary lines are connected end to end;

[0075] Check the topology to ensure there are no self-intersections or dangling points;

[0076] If the area calculated from the vector graphic is compared with the area of ​​ownership recorded in the archive, and the error exceeds a preset threshold, it is marked as an abnormal graphic.

[0077] Preferably, when the data quality inspection and database entry module constructs a spatial dataset containing the plane coordinates of the boundary points, the vector graphics of the ownership red line, and the standard ownership attribute information, and enters the spatial dataset into the spatial database, it is specifically used for:

[0078] The planar coordinates of the boundary points are associated and bound with the ownership red line vector graphics to generate standard vector format data;

[0079] The standard ownership attribute information is written into the attribute table of the spatial database, wherein the spatial database includes PostGIS or ArcSDE.

[0080] The third objective of this application is to provide an electronic device.

[0081] The aforementioned objective three of this application is achieved through the following technical solution:

[0082] An electronic device, comprising:

[0083] The method includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the large model-driven cadastral archive digitization method described in any one of the first objectives of this application.

[0084] The fourth objective of this application is to provide a computer-readable storage medium.

[0085] The fourth objective of this application is achieved through the following technical solution:

[0086] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the large-model-driven cadastral archive digitization method described in any one of the first objectives of this application.

[0087] Compared with the prior art, this application has the following beneficial effects:

[0088] 1. This application introduces a large language model to perform semantic parsing and logical reasoning on the boundary point description text after OCR recognition, which can automatically construct the topological connection order of boundary points. This process effectively overcomes the shortcomings of traditional OCR technology, which can only perform text recognition and cannot understand the complex logic of cadastral maps. It solves the problem of not being able to accurately restore cadastral relationships when boundary point numbers are missing or skipped, and realizes accurate conversion from unstructured text to structured topological data.

[0089] 2. This application constructs an automated digital processing workflow that includes OCR visual perception, large language model semantic parsing, and surveying coordinate calculation, replacing the traditional work mode that relies on manual drawing and manual data entry. This technical solution significantly improves the efficiency of data processing when dealing with massive amounts of cadastral archives digitization, and greatly reduces the cost and error rate of manual operation.

[0090] 3. This application combines the reference coordinates of the map frame with surveying algorithms to accurately convert the pixel coordinates in the image into actual plane coordinates and automatically generate vector data that conforms to the real estate registration database standard. This solution not only ensures the geometric accuracy of the cadastral data, but also ensures the attribute standardization of the output data, enabling it to be directly connected to the subsequent real estate registration business system. Attached Figure Description

[0091] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0092] Figure 1 This is a flowchart illustrating a large-model-driven cadastral archive digitization method in one embodiment of this application;

[0093] Figure 2 This is a schematic diagram of the structure of a large model-driven cadastral archive digitization system in one embodiment of this application;

[0094] Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application. Detailed Implementation

[0095] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0096] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described below are merely illustrative. For example, the division of units and modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or modules can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, and can be electrical, mechanical, or other forms.

[0097] In addition, each functional unit in the various embodiments of this application can be integrated into a single processor, or each unit can be a separate device, or two or more units can be integrated into a single device; each functional unit in the various embodiments of this application can be implemented in hardware or in the form of hardware plus software functional units.

[0098] Those skilled in the art will understand that all or part of the steps of the following method embodiments can be implemented by program instructions and related hardware. The aforementioned program instructions can be stored in a computer-readable storage medium. When the program instructions are executed, they perform the steps of the following method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0099] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "a plurality of" or "several" means two or more, unless otherwise explicitly specified.

[0100] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. It should be noted that the core innovative ideas of this application can be implemented through a variety of specific technical means.

[0101] like Figure 1 As shown in the figure, this application provides a large model-driven method for digitizing cadastral archives, which may include the following steps:

[0102] S1, acquire raster image data of cadastral archives;

[0103] In practical land registration management, natural resources departments or real estate registration centers often maintain a large number of historical land registration archives (such as historical land registration maps and parcel sketches). These archives are mostly in the form of paper drawings, scanned copies, or old electronic raster maps, with formats including but not limited to TIFF, JPEG, and PDF with embedded images. To digitize them, it is first necessary to collect and integrate this unstructured raster image data into the processing system. For example, in a county's historical homestead land ownership confirmation project, the archivist converted thousands of paper land registration survey maps drawn in the 1990s into raster image files using a high-speed scanner at 300 DPI resolution and created an index catalog according to the archive numbers. The system reads these raster image files in batches through a standardized data access interface, providing raw input for subsequent image recognition and information extraction. This step is the data entry point for the entire digitization process, and its input quality directly affects the accuracy of all subsequent processing stages.

[0104] S2. Use the OCR recognition model to perform image and text recognition on raster image data and extract map feature information, including map frame coordinates, scale parameters, boundary point description text and ownership description text.

[0105] After obtaining raster image data, traditional methods usually rely on manual visual interpretation and manual input, which is not only inefficient but also prone to errors.

[0106] This step introduces an OCR (Optical Character Recognition) model to automatically recognize text and images from raster images.

[0107] Specifically, deep learning-based OCR recognition models (such as PaddleOCR, CRNN+CTC, or end-to-end recognition models based on the Transformer architecture) are used to parse map elements in the image, such as frame coordinates, scale parameters, boundary point description text, and ownership description text.

[0108] On the one hand, the corner coordinates of the frame area are located and extracted through the object detection network, while the scale text in the legend (such as "1:500" and "1:1000") is recognized.

[0109] On the other hand, text detection and recognition branches are used to extract descriptive text near the boundary points (such as "J1", "J2", "neighboring Zhang Moumou's north wall") and ownership information text (such as "land user: Li Moumou", "land type: residential land").

[0110] For example, when processing a rural homestead cadastral map, the OCR recognition model successfully identified the coordinates of the four corner points of the map frame in the pixel coordinate system, the scale label "1:300", and the location descriptions and adjacent relationship descriptions of the 12 boundary points.

[0111] This step transforms the seemingly chaotic raster image into structured map feature information with semantic meaning, laying a data foundation for deeper understanding in subsequent large language models.

[0112] S3. Using a large language model, the boundary point description text and ownership description text in the map element information are structured and parsed to obtain the spatial connection order and standard ownership attribute information of the boundary points, and an ordered boundary point topology is constructed based on the spatial connection order of the boundary points.

[0113] The boundary point description text extracted by the OCR recognition model often contains non-standard natural language expressions (such as "from J1 along the old ditch to J2, then turn to J3 to the roadside"), and the ownership description text may also have problems such as inconsistent format, abbreviations or typos, which are difficult to be directly processed by traditional rule parsers.

[0114] This step uses pre-trained large language models (such as LLaMA, ChatGLM, or Qwen-14B-Chat-Int4, which are vertical models fine-tuned with corpora in this field) to perform semantic understanding and structured extraction on these unstructured texts.

[0115] Specifically, the boundary point description texts identified by OCR are first concatenated into a paragraph containing sequential logic, input into a large language model, and corresponding prompts are designed to require the model to output: a unique identifier for each boundary point, the connection order between points (e.g., the output is the sequence J1→J2, J2→J3, ..., Jn→J1), and the corner type (e.g., "right angle" or "arc").

[0116] Meanwhile, the large language model performs entity extraction and standardization on the ownership description text, such as mapping "User: Zhang Laosan" to the standard field "Rights Holder: Zhang Moumou", and mapping "Land Category: Residential" to "Land Use Code: 0701".

[0117] Finally, based on the spatial connection order output by the large language model, a directed or undirected ordered boundary point topology is constructed, which is a node linked list arranged in a clockwise or counterclockwise closed manner.

[0118] For example, in one case, the original text describes "starting from J1, going north along the field ridge to J2, turning east to J3, then turning south to J4, and finally closing to J1 in the west". After being parsed by the large language model, an ordered topological structure of J1, J2, J3, J4, J1 is generated, and the boundary type of each segment is marked.

[0119] This step fully leverages the powerful semantic understanding and reasoning capabilities of large language models to transform unstructured descriptive text into machine-executable spatial logical relationships, effectively solving the problem that traditional methods cannot handle natural language descriptions.

[0120] Specifically, in this embodiment, the standard ownership attribute information refers to attribute data that conforms to the "Real Estate Registration Database Standard" or the requirements of a preset business rule base. During the parsing process, the large language model will perform semantic matching and normalization processing on the extracted original text (such as 'residential house') and a preset standard dictionary (such as 'residential') to obtain the standard ownership attribute information.

[0121] S4. Based on the map frame coordinates and scale parameters in the map element information, combined with the ordered boundary point topology, the plane coordinates of the boundary points are calculated using the surveying coordinate solution algorithm.

[0122] After obtaining the ordered boundary point topology, it is necessary to convert the relative positional relationships at the pixel level into absolute planar coordinates in real geographic space.

[0123] This step first uses the frame coordinates and scale parameters extracted from S2 to establish an affine transformation model from the image pixel coordinate system to the geographic reference coordinate system (such as the local coordinate system or the 2000 National Geodetic Coordinate System).

[0124] Specifically, the six parameters of the affine transformation are first calculated based on the pixel coordinates of the four corner points of the map frame and their corresponding known geographic reference coordinates (which can be obtained from the map sheet number or control point database), and then scale correction is performed in conjunction with the scale parameters.

[0125] Subsequently, for each boundary point in the ordered boundary point topology, its planar coordinates are recursively derived using a mapping coordinate calculation algorithm. Commonly used algorithms include: when the boundary point description includes the azimuth and distance relationship with adjacent solved control points, the polar coordinate method is used; when the description is a corner, the forward intersection or traverse adjustment method is used.

[0126] For example, in an old cadastral map, the map frame is rectangular and its four corner geographic coordinates are known to be (50000, 60000) to (50200, 60200), with a scale of 1:500. First, the geographic offset corresponding to any pixel within the map frame is determined. Then, for the boundary point sequence arranged in topological order, starting from the reference point J1, based on descriptions in the text such as "move 12.5 meters from J1 in a direction 30° east of north," the true planar coordinates of J2, J3, ..., Jn are calculated sequentially. If a closure condition exists in the description (i.e., it should eventually return to J1), automatic closure error adjustment is performed, distributing small redundant errors by distance weighting to improve overall coordinate accuracy.

[0127] This step successfully transforms the graphic and textual descriptions on historical cadastral maps into a set of high-precision boundary point coordinates that can be directly processed by modern GIS software.

[0128] S5, Generate the ownership red line vector graphic based on the plane coordinates of the boundary points;

[0129] After calculating the planar coordinates of all boundary points, this step converts the discrete coordinate points into standard geographic information vector graphics, namely the ownership red line.

[0130] The specific process is as follows: According to the ordered boundary point topology structure (e.g., clockwise ordered sequence) constructed in S3, the coordinates of each boundary point are sequentially connected into line segments using a GIS underlying library (such as GDAL, Shapely, or ArcGIS Engine), and finally a closed polygon is constructed. This polygon is the ownership red line graphic of the land parcel, and its boundary accurately expresses the spatial range of the land parcel.

[0131] During the construction process, complex topological situations are automatically handled. For example, for ownership parcels containing cavities (such as ponds or public land within a parcel), an island-shaped polygon can be formed by constructing outer and inner rings. For multi-part parcels (such as those composed of several non-adjacent plots), a MultiPolygon structure is generated. Furthermore, to facilitate subsequent visualization and application, standardized attribute fields can be assigned to the vector graphic, including parcel code, right holder, land category code, area (automatically calculated), and metadata information (such as data source and digitization date).

[0132] For example, the boundary point coordinate sequence of a project site is (50123.45,60123.67),(50145.78,60145.89),…(50123.45,60123.67),(50145.78,60145.89),… The system generates a closed WKT (Well-Known Text) format string “POLYGON((...))” and exports it to Shapefile or GeoJSON format for further quality inspection and data entry.

[0133] This step completes the transformation from numerical coordinates to standard spatial graphics, forming a complete information-based representation of land parcels.

[0134] S6 performs automated quality checks on the generated ownership red line vector graphics. If the quality check passes, a spatial dataset containing the plane coordinates of boundary points, ownership red line vector graphics, and standard ownership attribute information is constructed and stored in the spatial database.

[0135] To ensure that the quality of digital results meets the requirements of real estate registration and cadastral management, the generated ownership red line vector graphics must undergo multi-dimensional automated quality checks before being formally entered into the database. The quality checks include geometric validity checks (e.g., whether polygons are self-intersecting and whether non-closed loops exist), topological consistency checks (e.g., whether shared edges of adjacent parcels match, and whether overlaps or gaps occur), attribute integrity checks (e.g., whether key fields such as rights holder and land type are non-empty), and coordinate accuracy checks (e.g., whether the positional error within the map exceeds the tolerance limit).

[0136] During quality inspection, a predefined quality inspection rule library is loaded, and each generated ownership red line graphic is automatically judged item by item. For example, if a self-intersection error is detected in a certain polygon, the graphic is marked as failing and a detailed quality inspection report is generated, including the error type, location coordinates, and repair suggestions, for manual review or to trigger an automatic repair process.

[0137] Once all quality control items pass, a complete spatial dataset is constructed. This dataset is closely linked to three core pieces of information: the boundary point plane coordinate table calculated in S4, the ownership red line vector graphics generated in S5 (stored in BLOB or vector field format), and the standard ownership attribute information extracted in S3. Subsequently, this dataset is batch-written into the target spatial database using a spatial database engine (such as PostGIS or Oracle Spatial), and spatial and attribute indexes are created to facilitate efficient subsequent queries and analysis. The entire data import process generates log records, including import time, data volume, and quality control result summaries, for audit traceability.

[0138] As described above, this embodiment provides a large-model-driven method for digitizing cadastral archives. First, raster image data is acquired. Then, map element information is extracted using OCR. Next, a large language model is used to perform structured parsing of boundary points and ownership text, constructing an ordered topological structure. Then, the planar coordinates of boundary points are calculated by combining map frame and scale parameters. Finally, ownership red line vector graphics are generated, and the data is stored in the database after automated quality inspection. This method, by introducing the semantic understanding capabilities of a large language model, overcomes the technical bottleneck of traditional OCR+rule engines' inability to parse natural language cadastral descriptions. It achieves end-to-end automated conversion from raster images to high-precision, structured, and computable ownership spatial datasets, significantly reducing the proportion of manual intervention and the risk of errors. Simultaneously, the integration of automated quality inspection and spatial databases ensures the standardization and usability of digital results, providing efficient and reliable technical support for unified real estate registration, revitalization of historical cadastral data, and smart land spatial planning.

[0139] In one embodiment, before performing image and text recognition on raster image data using an OCR recognition model, the method may further include:

[0140] The raster image data is preprocessed, including image correction, noise reduction, and image enhancement, in order to improve the recall and accuracy of key image elements in subsequent OCR recognition.

[0141] Among them, graphic enhancement is used to highlight the outline of the map frame, coordinate grid and boundary point symbols.

[0142] In real-world scenarios involving the digitization of cadastral archives, original raster images often suffer from various quality defects due to their age, differences in scanning equipment, or poor storage conditions.

[0143] For example, a batch of cadastral survey maps from the 1980s received by a county natural resources bureau showed that some of the maps were obviously tilted (the tilt angle of the scanned map was 3° to 5°), the map surface was covered with salt and pepper noise generated by the scan, and the map frame lines and boundary point symbols became blurred due to ink diffusion.

[0144] To address the aforementioned issues, this embodiment introduces a dedicated preprocessing procedure before performing OCR recognition:

[0145] First, based on the Hough transform, the line segments in the image are detected, the main direction of the frame edge or coordinate grid is identified, the tilt angle is calculated, and then the image is corrected by affine transformation to reduce the offset to within 0.1°.

[0146] Secondly, the image is denoised using a medium-frequency filtering algorithm, which effectively eliminates isolated salt-and-pepper noise points while preserving the edge information of the frame, grid, and boundary point symbols.

[0147] Finally, graphic enhancement operations are performed: Adaptive histogram equalization (CLAHE) is used to improve the local contrast of the frame outline, coordinate grid lines and boundary point symbols, and morphological closing operations are used to fill in minor breaks in the frame lines.

[0148] After the above preprocessing, the overlapping of multiple lines of text caused by tilting was eliminated, the detection rate of boundary point symbols under noise interference was significantly improved, and the continuity of the frame and grid was restored, providing high-quality input images for the high-precision recognition of subsequent OCR models.

[0149] In one embodiment, step S2 involves using an OCR recognition model to perform image and text recognition on the raster image data and extract map feature information. This specifically includes the following steps:

[0150] The OCR recognition model is used to identify the map frame parameters of the enhanced image. The geodetic coordinates of the four corners of the map frame are extracted as the map frame coordinates. The grid spacing and scale parameters are extracted. The grid spacing is used to correct the map scale distortion, and the scale parameters are used to calculate the correspondence between image pixels and actual distances.

[0151] OCR recognition model is used to identify ownership information in the enhanced image and extract ownership attribute annotations as ownership description text. Ownership attribute annotations include the name of the right holder, land parcel code and use category.

[0152] The enhanced image is identified using an OCR recognition model. If there are boundary points with specific coordinate labels in the image, the coordinate values ​​are directly extracted as the coordinate information of the boundary point description text, and the extraction of the boundary point outline pixel position is skipped. If there are no coordinate labels in the image, the pixel position of the boundary point outline is extracted and the pixel position is used as the positioning information of the boundary point description text.

[0153] In order to extract various map elements comprehensively and accurately from the preprocessed raster image, this embodiment has modularized the OCR recognition task and performed special recognition logic for the map frame parameters, ownership information and boundary point information respectively.

[0154] In the first sub-step, a deep learning-based object detection network (such as DBNet or PP-OCRv4) is used to locate the map frame region in the image, and then identify the geodetic coordinates of the four corners of the map frame (e.g., "X=2856743.21, Y=384521.67"). Simultaneously, grid spacing annotations (e.g., "grid spacing 50 meters") and scale text (e.g., "1:500" or "1:1000") are detected around the map frame or within the legend area. In practical applications, some older cadastral maps may have distortions caused by paper stretching. Simply combining the map frame coordinates with the scale is insufficient to completely eliminate distortion. Therefore, the actual grid spacing within the map is further extracted, and the comparison between the grid spacing and the theoretical spacing is used to correct scale distortion, improving the accuracy of subsequent coordinate calculations.

[0155] In the second sub-step, the focus is on the area for annotating ownership attributes, which is usually located inside or at the edge of the land parcel within the map frame. Key fields such as the name of the right holder (e.g., "Li Moumou"), the land parcel code (e.g., "120101001001"), and the land use category (e.g., "residential land" or "cultivated land") are extracted using text box detection and character recognition technology to form the ownership description text.

[0156] In the third sub-step, for the identification of boundary points, a differentiated strategy is adopted according to the labeling habits of different cadastral maps: For maps that use direct coordinate labeling (e.g., "J1 X=… Y=…" next to the boundary points), the OCR model directly extracts the value as the coordinate information of the boundary point description text, without having to infer the pixel position, thus avoiding secondary conversion errors; For traditional hand-drawn maps with only symbols and no numerical labels, the row and column positions of the boundary point symbols (such as solid circles, cross marks, or polygon vertex locators) in the pixel coordinate system are identified through object detection, and the pixel position is used as the positioning information of the boundary point description text for subsequent coordinate calculation.

[0157] Through this segmented identification strategy, this embodiment can adapt to various cadastral archive formats, ranging from modern digital survey maps to historical hand-drawn maps, maximizing the extraction of original information that can be used to reconstruct ownership boundaries.

[0158] In one embodiment, step S3 involves using a large language model to perform structured parsing of the boundary point description text and ownership description text in the map feature information to obtain the spatial connection order and standard ownership attribute information of the boundary points. This specifically includes the following steps:

[0159] The boundary point description text, ownership description text, and corresponding OCR recognition confidence information are combined with a pre-set cadastral surveying standard knowledge base fragment and encapsulated into a Prompt input command containing context constraints.

[0160] The pre-trained large language model is invoked to parse the Prompt input command, identify abnormal situations such as missing boundary point numbers, logical breakpoints, or coordinate contradictions, and automatically complete or correct the connection order of boundary points based on the connection rules and geometric closure principles in cadastral surveying specifications.

[0161] Output structured JSON data that conforms to the real estate registration database standard. The structured JSON data shall include at least: a unique identifier for the boundary point, a standard ownership attribute description of the land parcel to which the boundary point belongs, and an array of topological connection relationships arranged in the order of actual connection.

[0162] Because boundary point descriptions in historical cadastral archives often contain non-standard elements such as incomplete numbering, logical jumps, and even contradictory statements, traditional rule-based programs struggle to handle them effectively. This embodiment innovatively introduces a large language model, combining it with knowledge from the field of cadastral surveying for intelligent parsing.

[0163] Specifically, the boundary point description text identified by OCR (such as "from J1 north along the ditch to J2, then 30 degrees east of north to J3, then to the roadside stake, and finally back to the starting point"), the ownership description text (such as "right holder: Zhang Moumou, land type: dry land"), and the corresponding OCR confidence information (such as "J1" confidence 0.98, "roadside stake" is identified with low confidence, which may be an error) are first spliced ​​together.

[0164] Subsequently, relevant rule fragments are extracted from a pre-defined cadastral surveying standard knowledge base, such as "boundary lines should be closed to form polygons," "adjacent boundary points should be connected by straight line segments," and "unnumbered points (such as roadside stakes) should be virtually numbered according to their actual orientation and included in the sequence." These rules, along with the aforementioned text, are encapsulated into a Prompt input instruction containing explicit contextual constraints. The instruction explicitly requires the large language model to output in JSON format.

[0165] Next, a finely tuned vertical language model (such as a model further trained on the survey text based on LLaMA or ChatGLM-6B) is invoked to infer the prompt. During the inference process, the model can automatically identify anomalies: for example, it finds that "roadside stake" should actually be defined as virtual boundary point J4, and the implicit direction in the text "finally returning to the starting point" suggests that there should be a closed edge between J1 and J4; at the same time, the model also detects the logical breakpoint J2→J3 in the original description, and completes the direction type of the connection segment based on the descriptions of "along the ditch" and "30 degrees east of north".

[0166] Finally, the large language model outputs JSON structured data that conforms to the real estate registration database standard: it includes unique boundary point identifiers ("J1", "J2", "J3", "J4"), standardized ownership attributes ("ownerName": "Zhang Moumou", "landType": "dry land"), and an array of topological connections arranged in clockwise order on the ground ("connections": ["J1→J2", "J2→J3", "J3→J4", "J4→J1"]).

[0167] This embodiment introduces the semantic understanding and reasoning capabilities of large language models into the cadastral archive parsing process, which can automatically handle abnormal text patterns that are difficult to enumerate manually, significantly reducing the workload of manual proofreading. At the same time, the output standardized JSON data can be directly used for subsequent coordinate calculation and vector graphics construction.

[0168] In one embodiment, step S4 involves calculating the planar coordinates of the boundary points based on the map frame coordinates and scale parameters in the map feature information, combined with the ordered boundary point topology, using a surveying coordinate calculation algorithm. This specifically includes the following steps:

[0169] Using the coordinates of the four corners of the map frame as reference control points, calculate the scaling factor in the X / Y direction based on the image pixel size and the actual size of the map frame;

[0170] For boundary points that only have pixel locations on the map but no specific coordinates, spatial interpolation algorithms combined with scaling factors are used to calculate their planar coordinates.

[0171] For boundary points with coordinate closure errors, the least squares method is used for adjustment to correct the coordinate calculation errors.

[0172] The accuracy of the conversion from image pixel coordinates to real geographic planar coordinates directly determines the reliability of the ownership redline results. This embodiment designs a coordinate calculation algorithm that integrates benchmark control, spatial interpolation, and error adjustment.

[0173] First, using the geodetic coordinates of the four corners of the map frame extracted in step S2 as reference control points (e.g., lower left corner (50000, 60000), lower right corner (50200, 60000), upper left corner (50000, 60200), upper right corner (50200, 60200)). Based on the pixel coordinates of the four corners of the map frame in the pixel image (e.g., lower left corner (100, 800), lower right corner (1100, 800), upper left corner (100, 100), upper right corner (1100, 100)), and the corresponding real-world dimensions of the map frame (width 200 meters, height 200 meters), the scaling factors in the X and Y directions are calculated respectively (in this example, each pixel in the X direction corresponds to 0.2 meters, and each pixel in the Y direction corresponds to approximately 0.2 meters).

[0174] For the boundary points whose numerical coordinates are directly identified in step S2, the interpolation process is skipped, and the numerical coordinates are directly used as the final planar coordinates.

[0175] For boundary points that only have pixel locations (e.g., the boundary point symbol identified in step S2 is located at pixel coordinates (600, 450)), bilinear interpolation or thin plate spline interpolation algorithms are used to calculate the true planar coordinates of the pixel location by combining the pixel coordinates of the four corners of the frame and their corresponding geodetic coordinates.

[0176] It is worth emphasizing that, since historical drawings may have local deformations, relying solely on global linear interpolation at the four corners of the drawing frame may introduce errors. Therefore, this embodiment also introduces grid correction: using the grid intersection coordinates extracted by OCR as more control points, a segmented correction model is constructed to further improve the interpolation accuracy.

[0177] Finally, due to potential minor errors introduced by the text description or pixel extraction process, a non-closed "coordinate closure error" may occur between the final closed point and the starting point after calculating the boundary point coordinates sequentially according to the topological order. To address this, the least squares method is used for traverse adjustment: using the coordinates of all boundary points as observations and the coordinates of the control points of the map frame as constraints, the closure error is distributed to each intermediate point according to the side length ratio by minimizing the sum of squares of the coordinate errors, thereby obtaining the accurate plane coordinates of the boundary points after adjustment correction.

[0178] This embodiment uses a three-stage solution process of "benchmark control + spatial interpolation + adjustment optimization" to minimize the negative impact of original archive deformation and intermediate identification errors on the final coordinate accuracy, thus ensuring the surveying-grade reliability of the ownership red line vector results.

[0179] In one embodiment, step S6 involves automated quality inspection of the generated ownership red line vector graphics, specifically including the following steps:

[0180] Check the geometric closure of the vector graphics to ensure that the land parcel boundary lines are connected end to end;

[0181] Check the topology to ensure there are no self-intersections or dangling points;

[0182] If the area calculated from the vector graphic is compared with the area of ​​ownership recorded in the archive, and the error exceeds a preset threshold, it is marked as an abnormal graphic.

[0183] To ensure that the digital results meet the quality standards for real estate registration, this embodiment introduces a triple automated quality inspection mechanism after the ownership red line vector graphic is generated.

[0184] The first layer is a geometric closure check: This involves reading the sequence of line segments that constitute the land parcel boundary line from the vector graphics and verifying whether they form a strictly closed loop, meaning the starting point of the first line segment coincides with the ending point of the last line segment within the tolerance range. For example, for a land parcel boundary line, if the distance between the first and last points exceeds 0.01 meters (a configurable threshold) due to coordinate calculation errors, it will be judged as not closed, and a quality inspection report containing the distance and location of the non-closed loop will be generated.

[0185] The second layer is topology checking: A GIS topology engine (such as using the `is_simple` and `is_valid` methods from the Shapely library) is invoked to detect whether polygons have self-intersections (bow-shaped), whether there are dangling points between adjacent line segments (i.e., endpoints of a line segment are not shared by other line segments), and other non-manifold anomalies. Once the above topology errors are detected, the vector graphic is marked as a "topology anomaly," and the error location is visualized for manual fine-tuning or to trigger an automatic repair algorithm (such as Douglas-Peucker simplification and reconstruction).

[0186] The third step is area comparison checking: Based on the adjusted boundary point coordinates, the area of ​​the land parcel's boundary polygon is accurately calculated using the shoelace formula, and then compared with the ownership area recorded in the archives (e.g., the "registered area" field in the standard ownership attribute information parsed in step S3). A preset relative error threshold (e.g., 0.5% or 0.05%) is set. If the relative error between the calculated area and the archived area exceeds this threshold, it is marked as "area abnormal," indicating that there may be coordinate calculation errors, errors in the archive records, or drawing distortions exceeding the tolerance limit.

[0187] Through the aforementioned triple quality checks, only ownership red line graphics that simultaneously pass the geometric closure, topological validity, and area consistency checks will be deemed qualified and enter the subsequent data entry process. Otherwise, they will be returned to the corresponding stage for review or manual intervention. This implementation greatly improves the overall quality of the data entering the database and reduces the risk of ownership disputes caused by data errors.

[0188] In one embodiment, step S6 involves constructing a spatial dataset containing the planar coordinates of boundary points, vector graphics of ownership red lines, and standard ownership attribute information, and then storing the spatial dataset in a spatial database. This process specifically includes the following steps:

[0189] The planar coordinates of the boundary points are associated and bound with the vector graphics of the ownership red line to generate standard vector format data;

[0190] Standard ownership attribute information is written into the attribute table of the spatial database, which may include PostGIS or ArcSDE.

[0191] After completing coordinate calculation, vector graphics generation, and automated quality inspection, the resulting data needs to be persistently stored in a standard spatial data format for subsequent real estate registration, query analysis, and sharing. This embodiment designs a data entry scheme that tightly binds the three elements of "graphics-coordinates-attributes".

[0192] First, the set of plane coordinates of the boundary points after adjustment (e.g., the X and Y values ​​of each boundary point) is constructed into a standard geographic information geometric object (such as Polygon or MultiPolygon in the OGC specification) according to the ordered topological connection relationship determined in step S3. This geometric object is the ownership red line vector graphic.

[0193] Simultaneously, the geometric object is associated with and bound to the corresponding boundary point coordinate table. Each record in the boundary point coordinate table contains the boundary point number, the X and Y plane coordinates, and the foreign key identifier of the corresponding land parcel, ensuring that the coordinates of all boundary points constituting the boundary line of the land parcel can be retrieved using the land parcel ID. The bound graphic and coordinate data, together with the standardized ownership attribute information (rights holder, land parcel code, use category, registered area, etc.) parsed in step S3, together form a complete spatial dataset.

[0194] Subsequently, the spatial database interface is invoked to write the dataset into the target spatial database. Specifically, the ownership redline vector graphics are stored in the spatial data table as geometry columns; the boundary point coordinate table is stored independently as a subordinate table and linked to the main table via a foreign key; and the standard ownership attribute information is written into the attribute fields of the main table. The database used can be either the open-source spatial database PostgreSQL with PostGIS extensions or the commercial-grade ArcSDE geographic database.

[0195] For example, in the application of a provincial real estate registration center, thousands of land parcels processed daily are processed in batches. Using Shapefile as the intermediate exchange format, the vector graphics and attributes are then imported into the database in batches using PostGIS's shp2pgsql tool. Spatial indexes are automatically created to accelerate subsequent queries (such as "searching land parcels by coordinate range" or "analysis of adjacent land parcels").

[0196] This embodiment achieves the complete transformation and storage of raster archives into standard spatial data, forming high-quality, computable data assets that can directly support real estate registration and land spatial planning analysis.

[0197] As can be seen from the above embodiments, the large model-driven cadastral archive digitization method provided in this application organically introduces the semantic understanding and reasoning capabilities of a large language model, a phased and refined OCR recognition strategy, and a systematic coordinate calculation and quality inspection and warehousing process on the basis of traditional OCR and GIS technologies.

[0198] Specifically, by performing preprocessing such as correction, denoising, and enhancement on raster images, the detection rate of key elements in low-quality historical archives was significantly improved. Differentiated OCR recognition was implemented for map frame parameters, ownership information, and boundary point information (distinguishing between coordinate labels and pixel positions), achieving compatibility with multiple cadastral map formats and maximizing information extraction. By combining cadastral surveying standards with a large language model, unstructured anomalies such as missing numbers, logical breaks, and even contradictions in boundary point description texts were effectively resolved, and standardized topological connections were automatically generated. A multi-level coordinate calculation mechanism of "map frame benchmark + grid correction + spatial interpolation + adjustment optimization" significantly suppressed the impact of map deformation and recognition errors on coordinate accuracy. Finally, triple automated quality inspection and binding of graphics, coordinates, and attributes into the database ensured the standardization and high reliability of the final spatial data results. The aforementioned technical solutions work together to transform the digitization of cadastral archives, which previously relied heavily on manual interpretation, data entry, and repeated verification, into an end-to-end automated, standardized, and highly accurate intelligent processing workflow. This significantly reduces labor costs and error rates, shortens the data production cycle, and provides a solid technical foundation for unified real estate registration, revitalization of historical cadastral data, and smart land management.

[0199] It should be noted that the large-model-driven cadastral archive digitization method of this application is mainly applicable to scenarios involving the vectorization of existing cadastral archives that have a certain digitization foundation, clear map attributes, accurate content, and relatively consistent logical relationships. Specifically, this method can leverage its advantages of automation and high efficiency when processing batches of high-quality scanned or photographed image data.

[0200] In practical applications, this application, through the collaborative work of OCR technology and a large language model, can effectively parse and restore the boundary point description text and its topological relationships in cadastral maps. However, it should be understood that for complex situations where historical archives are severely blurred, damaged, or whose content is logically chaotic due to their age, or even beyond the scope of conventional understanding, requiring subjective interpretation and speculation by professionals based on specific experience, the automated processing capabilities of this application have certain limitations. Such extreme cases, heavily reliant on human expert experience for fuzzy judgment and deep semantic inference, represent advanced challenges in the field and are not the core technical problems addressed by this application, nor are they within the scope of the standardized processing primarily covered by this application.

[0201] Therefore, this application focuses more on replacing repetitive and labor-intensive manual operations in a standardized and regulated process of digitizing archives, thereby improving overall operational efficiency and data structuring level.

[0202] like Figure 2 As shown in the figure, this application provides a large model-driven cadastral archive digitization system, which may include:

[0203] Data acquisition module 201 is used to acquire raster image data of cadastral archives;

[0204] Information extraction module 202 is used to perform image and text recognition on raster image data using an OCR recognition model to extract map feature information, including map frame coordinates, scale parameters, boundary point description text and ownership description text.

[0205] The large model parsing module 203 is used to perform structured parsing of the boundary point description text and ownership description text in the map feature information using a large language model, to obtain the spatial connection order and standard ownership attribute information of the boundary points, and to construct an ordered boundary point topology based on the spatial connection order of the boundary points.

[0206] The coordinate calculation module 204 is used to calculate the plane coordinates of the boundary points based on the map frame coordinates and scale parameters in the map feature information, combined with the ordered boundary point topology, using a surveying coordinate calculation algorithm.

[0207] Vector graphics generation module 205 is used to generate ownership red line vector graphics based on the planar coordinates of boundary points;

[0208] The data quality inspection and storage module 206 is used to automatically inspect the generated ownership red line vector graphics. If the quality inspection passes, a spatial dataset containing the plane coordinates of boundary points, ownership red line vector graphics, and standard ownership attribute information is constructed and stored in the spatial database.

[0209] In one embodiment, the system may further include:

[0210] The data preprocessing module is used to preprocess raster image data, including image correction, noise reduction, and image enhancement.

[0211] Among them, graphic enhancement is used to highlight the features of the map frame outline, coordinate grid and boundary point symbols, so as to improve the recall and accuracy of subsequent OCR recognition of key map elements.

[0212] In one embodiment, when the information extraction module 202 performs image and text recognition on raster image data using an OCR recognition model to extract map feature information, it is specifically used for:

[0213] The OCR recognition model is used to identify the map frame parameters of the enhanced image. The geodetic coordinates of the four corners of the map frame are extracted as the map frame coordinates. The grid spacing and scale parameters are extracted. The grid spacing is used to correct the map scale distortion, and the scale parameters are used to calculate the correspondence between image pixels and actual distances.

[0214] OCR recognition model is used to identify ownership information in the enhanced image and extract ownership attribute annotations as ownership description text. Ownership attribute annotations include the name of the right holder, land parcel code and use category.

[0215] The enhanced image is identified using an OCR recognition model. If there are boundary points with specific coordinate labels in the image, the coordinate values ​​are directly extracted as the coordinate information of the boundary point description text, and the extraction of the boundary point outline pixel position is skipped. If there are no coordinate labels in the image, the pixel position of the boundary point outline is extracted and the pixel position is used as the positioning information of the boundary point description text.

[0216] In one embodiment, when the large model parsing module 203 performs structured parsing of the boundary point description text and ownership description text in the map feature information using a large language model to obtain the spatial connection order and standard ownership attribute information of the boundary points, it is specifically used for:

[0217] The boundary point description text, ownership description text, and corresponding OCR recognition confidence information are combined with a pre-set cadastral surveying standard knowledge base fragment and encapsulated into a Prompt input command containing context constraints.

[0218] The pre-trained large language model is invoked to parse the Prompt input command, identify abnormal situations such as missing boundary point numbers, logical breakpoints, or coordinate contradictions, and automatically complete or correct the connection order of boundary points based on the connection rules and geometric closure principles in cadastral surveying specifications.

[0219] Output structured JSON data that conforms to the real estate registration database standard. The structured JSON data shall include at least: a unique identifier for the boundary point, a standard ownership attribute description of the land parcel to which the boundary point belongs, and an array of topological connection relationships arranged in the order of actual connection.

[0220] In one embodiment, when the coordinate calculation module 204 calculates the planar coordinates of boundary points based on the map frame coordinates and scale parameters in the map feature information, combined with the ordered boundary point topology, using a surveying coordinate calculation algorithm, it is specifically used for:

[0221] Using the coordinates of the four corners of the map frame as reference control points, calculate the scaling factor in the X / Y direction based on the image pixel size and the actual size of the map frame;

[0222] For boundary points that only have pixel locations on the map but no specific coordinates, spatial interpolation algorithms combined with scaling factors are used to calculate their planar coordinates.

[0223] For boundary points with coordinate closure errors, the least squares method is used for adjustment to correct the coordinate calculation errors.

[0224] In one embodiment, when the data quality inspection and warehousing module 206 performs automated quality inspection on the generated ownership red line vector graphics, it is specifically used for:

[0225] Check the geometric closure of the vector graphics to ensure that the land parcel boundary lines are connected end to end;

[0226] Check the topology to ensure there are no self-intersections or dangling points;

[0227] If the area calculated from the vector graphic is compared with the area of ​​ownership recorded in the archive, and the error exceeds a preset threshold, it is marked as an abnormal graphic.

[0228] In one embodiment, when the data quality inspection and storage module 206 constructs a spatial dataset containing the planar coordinates of boundary points, vector graphics of ownership red lines, and standard ownership attribute information, and stores the spatial dataset in the spatial database, it is specifically used for:

[0229] The planar coordinates of the boundary points are associated and bound with the vector graphics of the ownership red line to generate standard vector format data;

[0230] Standard ownership attribute information is written into the attribute table of the spatial database, which may include PostGIS or ArcSDE.

[0231] like Figure 3 As shown, this application provides an electronic device 3, which includes a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and executable on the processor 302. The memory 301 and the processor 302 are connected via a bus 304. When the processor 302 executes the computer program 303, it implements the large model-driven cadastral archive digitization processing method as described in the above method embodiment of this application.

[0232] Specifically, the electronic device 3 may be a server, edge computing node, or high-performance workstation deployed in a data center. The memory 301 may be a high-speed RAM memory, or may include non-volatile memory (such as EEPROM, flash memory, etc.), and the processor 302 may be a central processing unit (CPU), a graphics processing unit (GPU), or a combination thereof (for accelerating large model inference).

[0233] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a large-model-driven cadastral archive digitization method as described in the above-described method embodiments of this application.

[0234] The computer-readable storage medium can be a physical storage medium such as an optical disc, hard disk, or USB flash drive, or it can be virtual storage space in a cloud storage service. This computer program enables the computer to execute large-scale model-driven cadastral digitization, solving the technical problem of low efficiency in traditional GIS data production.

[0235] It should be noted that the large model-driven cadastral archive digitization system, electronic equipment, and computer-readable storage medium in the above embodiments have the same working principle and technical effect as the large model-driven cadastral archive digitization method in the above embodiments, and will not be described again here.

[0236] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0237] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0238] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0239] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A large model driven cadastral archive digitization processing method, characterized in that, The method includes: Obtain raster image data of cadastral archives; The raster image data is used to perform image and text recognition using an OCR recognition model to extract map feature information, which includes map frame coordinates, scale parameters, boundary point description text, and ownership description text. The boundary point description text and ownership description text in the map element information are structured and parsed using a large language model to obtain the spatial connection order and standard ownership attribute information of the boundary points, and an ordered boundary point topology is constructed based on the spatial connection order of the boundary points. Based on the map frame coordinates and scale parameters in the map element information, and combined with the ordered boundary point topology, the plane coordinates of the boundary points are calculated using a surveying coordinate calculation algorithm. Generate a red line vector graphic of ownership based on the plane coordinates of the boundary points; The generated ownership red line vector graphics are subjected to automated quality inspection. If the quality inspection passes, a spatial dataset containing the plane coordinates of the boundary points, the ownership red line vector graphics, and standard ownership attribute information is constructed and stored in the spatial database.

2. The large model driven cadastre digitization processing method according to claim 1, characterized in that, Before performing image and text recognition on the raster image data using the OCR recognition model, the method further includes: The raster image data is preprocessed, including image correction, noise reduction, and image enhancement. The graphic enhancement is used to highlight the features of the map frame outline, coordinate grid and boundary point symbols, so as to improve the recall and accuracy of subsequent OCR recognition of key map elements.

3. The large model driven cadastre digitization processing method according to claim 2, characterized in that, The step of using an OCR recognition model to perform image and text recognition on the raster image data and extract image feature information includes: The enhanced image is subjected to frame parameter recognition using an OCR recognition model. The geodetic coordinates of the four corners of the frame are extracted as the frame coordinates. The grid spacing and scale parameters are extracted. The grid spacing is used to correct the distortion of the map scale, and the scale parameters are used to calculate the correspondence between image pixels and actual distances. The enhanced image is identified using an OCR recognition model, and ownership information is extracted to extract ownership attribute annotations as ownership description text. The ownership attribute annotations include the name of the right holder, the land parcel code, and the use category. The enhanced image is identified using an OCR recognition model. If there are boundary points with specific coordinate labels in the image, the coordinate values ​​are directly extracted as the coordinate information of the boundary point description text, and the extraction of the boundary point outline pixel position is skipped. If there are no coordinate labels in the image, the pixel position of the boundary point outline is extracted and the pixel position is used as the positioning information of the boundary point description text. 4.The large model driven cadastre digitization processing method according to claim 1, wherein, The method involves using a large language model to perform structured parsing of the boundary point description text and ownership description text in the map feature information to obtain the spatial connection order and standard ownership attribute information of the boundary points, including: The boundary point description text, ownership description text, and corresponding OCR recognition confidence information are combined with a preset cadastral surveying standard knowledge base fragment and encapsulated into a Prompt input command containing context constraints. The pre-trained large language model is invoked to parse the Prompt input command, identify abnormal situations such as missing boundary point numbers, logical breakpoints, or coordinate contradictions, and automatically complete or correct the connection order of boundary points based on the connection rules and geometric closure principles in cadastral surveying specifications. Output structured JSON data that conforms to the real estate registration database standard. The structured JSON data shall include at least: a unique identifier for the boundary point, a standard ownership attribute description of the land parcel to which the boundary point belongs, and an array of topological connection relationships arranged in the order of actual connection.

5. The method for large-model-driven digitization of cadastral archives according to claim 1, characterized in that, The step of calculating the planar coordinates of boundary points based on the map frame coordinates and scale parameters in the map element information, combined with the ordered boundary point topology, and using a surveying coordinate calculation algorithm includes: Using the coordinates of the four corners of the map frame as reference control points, calculate the scaling factor in the X / Y direction based on the image pixel size and the actual size of the map frame; For boundary points that only have pixel locations on the map but no specific coordinates, their planar coordinates are calculated using a spatial interpolation algorithm combined with the aforementioned scaling factor. For boundary points with coordinate closure errors, the least squares method is used for adjustment to correct the coordinate calculation errors. 6.The large model driven cadastre digitization processing method according to claim 1, wherein, The automated quality inspection of the generated ownership redline vector graphics includes: Check the geometric closure of the vector graphics to ensure that the land parcel boundary lines are connected end to end; Check the topology to ensure there are no self-intersections or dangling points; If the area calculated from the vector graphic is compared with the area of ​​ownership recorded in the archive, and the error exceeds a preset threshold, it is marked as an abnormal graphic. 7.The large model driven cadastre digitization processing method according to claim 1, wherein, The construction of a spatial dataset containing the planar coordinates of the boundary points, the vector graphics of the ownership red line, and standard ownership attribute information, and the importation of the spatial dataset into a spatial database, includes: The planar coordinates of the boundary points are associated and bound with the ownership red line vector graphics to generate standard vector format data; The standard ownership attribute information is written into the attribute table of the spatial database, wherein the spatial database includes PostGIS or ArcSDE.

8. A large model-driven cadastral archive digitization processing system, characterized by, The system includes: The data acquisition module is used to acquire raster image data from cadastral archives; The information extraction module is used to perform image and text recognition on the raster image data using an OCR recognition model to extract map feature information, including map frame coordinates, scale parameters, boundary point description text, and ownership description text. The large model parsing module is used to perform structured parsing of the boundary point description text and ownership description text in the map element information using a large language model, to obtain the spatial connection order and standard ownership attribute information of the boundary points, and to construct an ordered boundary point topology based on the spatial connection order of the boundary points. The coordinate calculation module is used to calculate the planar coordinates of the boundary points based on the map frame coordinates and scale parameters in the map element information, combined with the ordered boundary point topology, and using a surveying coordinate calculation algorithm. The vector graphic generation module is used to generate a vector graphic of the ownership red line based on the planar coordinates of the boundary points; The data quality inspection and storage module is used to automatically inspect the generated ownership red line vector graphics. If the quality inspection passes, a spatial dataset containing the plane coordinates of the boundary points, the ownership red line vector graphics, and standard ownership attribute information is constructed and stored in the spatial database.

9. An electronic device, comprising: It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the large model-driven cadastral archive digitization method as described in any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the large model-driven cadastral archive digitization method as described in any one of claims 1-7.

Images