Joint review and judgment method for land space planning compilation and review whole chain business

Abstract

The present application belongs to the technical field of data analysis and processing, and specifically relates to a joint review and judgment method for land space planning compilation and review whole-chain business, comprising the following steps: step one: reading project declaration text, reply text, planning layer and control layer, reading planning polygon elements and control polygon elements from the planning layer and the control layer and generating the corresponding enclosing rectangle; step two: inputting the project declaration text mark sequence into a pointer network model, performing polygon overlay operation on the project land polygon and the search result to form a spatial overlay unit set; step three: reading the land space planning control rule library to generate a project joint review opinion sheet containing a list of modification points and a review conclusion. The present application can significantly improve the review efficiency, reduce the amount of manual operation, improve the review consistency, and provide intelligent support with high precision and high reliability for land space planning management.

Description

Technical Field

[0001] This invention belongs to the field of data analysis and processing technology, specifically involving a joint review and judgment method for the entire chain of land and space planning compilation and review. Background Technology

[0002] In the context of rapid urbanization, territorial spatial planning plays a crucial role in coordinating regional resources and ecological protection with development. As local planning systems gradually improve, the volume of tasks such as land use approvals, project compliance reviews, and planning scheme optimization is constantly increasing, leading to significant bottlenecks in efficiency and accuracy for manual review methods. In current industry practice, planning review typically relies on multiple types of data, including project application texts, approval documents, planning layers, and control layers, requiring manual verification against textual descriptions and spatial boundaries. For example, reviewers need to find land coordinates in project application texts, manually input the coordinates into the map system, draw the plot boundaries, and then determine whether the plot exceeds ecological protection red lines, falls within permanent basic farmland, or infringes on urban development boundaries. Due to significant differences in the expression methods of different projects, coordinate content may appear in various forms such as latitude and longitude lists, side length and angle descriptions, and paragraph descriptions. Manual interpretation is time-consuming and susceptible to subjective influence, lacking sufficient support for large-scale, batch review processes. Some information systems have attempted to extract coordinate content from project application texts using rule matching, regular expressions, or keyword extraction techniques. However, due to the lack of a standardized template for application texts, the text often contains numerous explanatory statements, variations in spacing and formatting, irregular characters, and differing writing habits among various project units. This results in poor stability for parsing methods based on fixed rules. For example, if extra punctuation marks appear before or after coordinates, or if there are spaces or line breaks between numbers, the regular expression method may fail to recognize the complete coordinate sequence, leading to inaccurate review results. For actual review processes requiring batch processing, unstable coordinate extraction can severely impact the overall system's usability. Summary of the Invention

[0003] Therefore, the main objective of this invention is to provide a joint review and judgment method for the entire process of land spatial planning compilation and approval. By constructing an integrated technical link that runs through text parsing, spatial retrieval, and spatial overlay, the land spatial planning compilation and approval process is transformed from a large amount of repetitive work that previously relied on manual judgment into a structured process that is automatically executed by the system, resulting in significant efficiency and accuracy improvements. The pointer network model can reliably extract coordinate content even when project application texts have inconsistent writing styles, large content spans, and multiple format expressions, avoiding extraction failures caused by text differences in traditional rule-based methods, thus making the review process more robust. At the same time, the parsed coordinate content can be accurately reconstructed into project plot polygons, maintaining a one-to-one correspondence with the specific expressions in the text, ensuring the traceability of spatial analysis results from the source.

[0004] The technical solution adopted in this invention is as follows: The joint review and judgment method for the entire chain of land and space planning compilation and review includes the following steps: Step 1: Read the project application text, approval text, planning layer, and control layer; break down the project application text into text tag units to construct a project application text tag sequence; read the planning polygon elements and control polygon elements from the planning layer and control layer and generate the corresponding enclosing rectangles. Step 2: Input the project application text tag sequence into the pointer network model, extract the coordinate content from the project application text tag sequence and parse it into project plot polygons, insert the bounding rectangles of the planning polygon elements and control polygon elements into the R-tree index structure, retrieve the planning polygon elements and control polygon elements that intersect with the bounding rectangles of the project plot polygons in the R-tree index structure through the bounding rectangles of the project plot polygons, perform polygon overlay operation on the project plot polygons and the retrieval results to form a set of spatial overlay units; Step 3: Read the land and space planning control rule library, determine the compliance category of the spatial overlay unit based on the attribute information of the spatial overlay unit in the planning layer and control layer, extract the boundary points adjacent to the boundary of the control polygon element of the spatial overlay unit whose compliance category is not the category, and generate a joint review opinion form for the project containing a list of modification points and review conclusions.

[0005] Furthermore, the pointer network model in step two includes an encoding part and a decoding part; wherein, the encoding part sequentially reads each text tag unit in the project declaration text tag sequence, maps each text tag unit to a fixed-length vector representation, and generates an encoding vector based on the vector representation of the current text tag unit and the state vector of the previous position, and assembles all the encoding vectors into an encoding vector sequence according to the order of the text tag units.

[0006] Furthermore, the decoding part pre-inserts coordinate segment start indicator markers, coordinate segment end indicator markers, and decoding end indicator markers into the project declaration text marker sequence. At the initial moment of decoding, the initial decoding state is set. In each decoding iteration, the similarity calculation module is used to compare the current decoding state with each encoded vector in the encoded vector sequence to generate a score list that corresponds one-to-one with all text marker units. The position of the text marker unit with the highest score is selected from the score list, and the index of the text marker unit corresponding to the position is used as the pointer output of the current decoding iteration. The pointer output is added to the pointer sequence, and the decoding state is updated at the same time.

[0007] Furthermore, the decoding part in step two is further executed as follows: when the pointer of the current decoding iteration outputs the starting indicator mark of the corresponding coordinate segment, in the next and subsequent decoding iterations, the text marker unit positions corresponding to the specific coordinate content are selected from the encoded vector sequence, and the text marker unit numbers corresponding to these positions are added to the pointer sequence in sequence until the current pointer outputs the ending indicator mark of the corresponding coordinate segment. The text marker unit numbers located between the starting indicator mark and the ending indicator mark of the coordinate segment are determined as a pointer sequence corresponding to the coordinate content. When the pointer of the current decoding iteration outputs the decoding ending indicator mark, the entire decoding process ends, resulting in one or more pointer sequences corresponding to the coordinate content. Each pointer in each pointer sequence corresponds to a text marker unit in the project declaration text marker sequence.

[0008] Furthermore, step two also includes: according to the text marker unit number recorded in each pointer sequence, sequentially reading the coordinate content from the project application text marker sequence; parsing the continuous text fragments containing longitude values, latitude values, and separators into a set of ordered coordinate points; connecting the ordered coordinate points belonging to the same coordinate content in pairs according to their order in the coordinate content to form a project plot polyline; connecting the starting coordinate point and the ending coordinate point of the project plot polyline to form a closed boundary, thus obtaining the project plot polygon; establishing an association between the project plot polygon and the pointer sequence of the corresponding coordinate content; recording the corresponding pointer sequence and the corresponding text marker unit number range in the project plot polygon, thus forming a correspondence between the project plot polygon and the coordinate content in the project application text.

[0009] Furthermore, the construction and querying of the R-tree index structure in step two includes: sequentially inserting the bounding rectangles of each planned polygon feature and each controlled polygon feature into the R-tree index structure; selecting the insertion path based on the center coordinates of the bounding rectangles during the insertion process; when the number of bounding rectangles in a node exceeds a preset capacity threshold, splitting the current node and re-dividing the set of bounding rectangles in the node during the splitting process, generating new internal nodes and leaf nodes, so that the R-tree index structure forms a multi-level hierarchy containing root nodes, internal nodes, and leaf nodes; and storing the planned polygon feature number and the controlled polygon feature associated with the bounding rectangle in each leaf node. Shape element numbering; for each project plot polygon, calculate the minimum and maximum values ​​of the planar coordinate points of the project plot polygon in the horizontal and vertical directions, generate the bounding rectangle of the project plot polygon, use the bounding rectangle of the project plot polygon as the query rectangle, start from the root node of the R-tree index structure and traverse downwards, only visiting nodes that have a planar intersection with the query rectangle during the traversal, until the leaf node is reached, read the planning polygon element number and control polygon element number corresponding to the bounding rectangle that has a planar intersection with the query rectangle in the leaf node, and form the planning candidate set and control candidate set corresponding to the project plot polygon.

[0010] Furthermore, the polygon overlay operation in step two includes: for the polygon boundaries of the project plot polygon and each planning polygon element in the planning candidate set, calculating all intersection points between the project plot polygon boundary and the planning polygon element boundary, and inserting each intersection point into the boundary point sequence of the corresponding edge to form an extended boundary point sequence containing the original vertices and intersection points; for the polygon boundaries of the project plot polygon and each control polygon element in the control candidate set, calculating all intersection points between the project plot polygon boundary and the control polygon element boundary, and inserting each intersection point into the boundary point sequence of the corresponding edge to form an extended boundary point sequence containing the original vertices and intersection points. The sequence of extended boundary points at the intersection points; a starting point is selected from the extended boundary point sequence, and the edges are traversed sequentially along the polygon boundary in a predetermined direction. When the intersection point is reached, the traversal path is switched between the project plot polygon, the planning polygon element, and the control polygon element according to the predetermined switching rules. At the end of the traversal, a closed polyline representing the boundary of different overlay areas is formed. The area enclosed by each closed polyline is written as a spatial overlay unit into the spatial overlay unit set. In each spatial overlay unit, the corresponding project plot polygon number, planning polygon element number, or control polygon element number, as well as the sequence of boundary points constituting the boundary of the spatial overlay unit are recorded.

[0011] Furthermore, step two also includes: based on the correspondence between the project plot polygons and the coordinate content in the project application text, mapping the project plot polygon numbers recorded in the spatial overlay unit to the coordinate content positions in the project application text; mapping the planning polygon element numbers and control polygon element numbers recorded in the spatial overlay unit to the element numbers in the planning layer and control layer, forming a three-way mapping relationship between the spatial overlay unit and the coordinate content of the project application text, the planning layer elements, and the control layer elements; when storing the spatial overlay unit, writing the three-way mapping relationship as an attribute field into the spatial overlay unit, so that each spatial overlay unit simultaneously corresponds to a specific geographical area and the specific coordinate content position in the project application text.

[0012] Furthermore, step three includes: reading the land and space planning control rule base from the business system, representing each rule in the land and space planning control rule base as a structured rule entry containing the planning use type, control type, and permitted or prohibited overlay relationships; for each spatial overlay unit in the spatial overlay unit set, based on the project plot polygon number, planning polygon element number, and control polygon element number recorded in the spatial overlay unit, searching for the attribute information of the corresponding elements in the planning layer and control layer to determine the planning use type and control type to which the spatial overlay unit belongs; matching the planning use type and control type to which the spatial overlay unit belongs with the structured rule entries in the land and space planning control rule base to obtain structured rule entries that conform to the current spatial overlay unit, and classifying the current spatial overlay unit into one of the following compliance categories: approved category, optimization category, and adjustment category, based on the permitted or prohibited overlay relationships recorded in the structured rule entries.

[0013] By adopting the above technical solutions, this invention achieves the following beneficial effects: By inserting the enclosing rectangles of planning polygon elements and control polygon elements into the R-tree index structure, this invention enables rapid retrieval even with large amounts of planning data, significantly reducing the number of objects requiring precise overlay calculations, and ensuring efficient response of the spatial analysis component even in complex urban scenarios. In polygon overlay processing, by meticulously dividing spatial overlay units, each area possesses clear planning and control attributes, achieving a much more refined review than traditional intersection judgments. Furthermore, by integrating with the land and space planning control rule library, it can automatically match corresponding review conclusions for each spatial overlay unit, providing distinctions between pass, optimization, and adjustment categories, allowing the system to objectively present the severity of the problem. Based on this, boundary points adjacent to the boundaries of control polygon elements are extracted from the boundary point sequence of the spatial overlay unit as modification points, realizing the direct conversion of review conclusions into executable modification suggestions, enabling applicants to quickly adjust planning schemes based on system outputs and reducing repeated communication. The technical solution of this invention integrates three types of data: text, space, and rules, making planning compilation and review a standardized, intelligent, and traceable end-to-end process, providing a practical technical path for the digital reform of the national land space governance system. Attached Figure Description

[0014] Figure 1 This is a schematic diagram illustrating the principle of the splitting mechanism of the R-tree spatial index structure when the node data volume overflows, as provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the principle of the spatial index hierarchy enclosure structure based on the R-tree provided in an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the principle of automatic identification of boundary modification points and generation of yield vectors provided in an embodiment of the present invention. Detailed Implementation

[0015] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0016] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0017] The joint review and judgment method for the entire chain of land and space planning compilation and review includes the following steps: Step 1: Read the project application text, approval text, planning layer, and control layer; break down the project application text into text tag units to construct a project application text tag sequence; read the planning polygon elements and control polygon elements from the planning layer and control layer and generate the corresponding enclosing rectangles. Step 2: Input the project application text tag sequence into the pointer network model, extract the coordinate content from the project application text tag sequence and parse it into project plot polygons, insert the bounding rectangles of the planning polygon elements and control polygon elements into the R-tree index structure, retrieve the planning polygon elements and control polygon elements that intersect with the bounding rectangles of the project plot polygons in the R-tree index structure through the bounding rectangles of the project plot polygons, perform polygon overlay operation on the project plot polygons and the retrieval results to form a set of spatial overlay units; Step 3: Read the land and space planning control rule library, determine the compliance category of the spatial overlay unit based on the attribute information of the spatial overlay unit in the planning layer and control layer, extract the boundary points adjacent to the boundary of the control polygon element of the spatial overlay unit whose compliance category is not the category, and generate a joint review opinion form for the project containing a list of modification points and review conclusions.

[0018] In a preferred embodiment, step one is executed on the server of the urban planning information system. The server reads project application texts, approval texts, planning layers, and control layers from the business database through a network interface, and establishes an independent data structure for each project in memory so that the textual and spatial information of the same project can be jointly processed in subsequent steps.

[0019] For project application and approval texts, the system first retrieves the corresponding record from the business database based on the project number and reads the complete text content from the record. The text content is typically stored using a unified character encoding, such as a common multi-byte character encoding, to avoid garbled characters during subsequent splitting. After reading, the system performs preprocessing on the project application and approval texts, including removing extra blank lines at the beginning and end, standardizing newline characters generated by different operating systems to a single newline character, and normalizing multiple consecutive spaces into a single space. This preprocessing step ensures that the subsequently split text tagging units remain consistent across different deployment environments, thus providing a stable input sequence for the pointer network model.

[0020] After text preprocessing, the system begins constructing a sequence of text tags for the project application. Specifically, starting with the first character, the system reads each character sequentially, breaking them down into text tag units based on their category. Character categories include Chinese characters, Arabic numerals, punctuation marks, and spaces. Chinese characters are typically treated as a single text tag unit. Arabic numerals are processed based on their continuity; for example, consecutive occurrences of the numeral characters "123456" are considered a single text tag unit, allowing for the extraction of complete longitude or latitude values ​​during subsequent coordinate analysis. Punctuation marks are generally treated as separate text tag units, such as commas, periods, semicolons, and parentheses. Spaces are also treated as separate text tag units, which helps maintain their positions when reconstructing the original text layout. This method of segmentation preserves all text information at the character level while forming meaningful wholes at consecutive numerical segments, which is highly beneficial for coordinate analysis and numerical extraction.

[0021] In one specific implementation, latitude and longitude coordinates in project application text typically appear in the form of "120.123456 degrees East longitude, 31.234567 degrees North latitude". Using the aforementioned splitting rules, consecutive numbers and decimal points form a text marker unit, such as "120.123456". This allows the complete value to be read as a single unit during subsequent coordinate parsing, eliminating the need for internal string concatenation. This reduces the probability of coordinate parsing errors, especially when a text contains more than 50 sets of coordinate points, where the stability of continuous processing is even more crucial.

[0022] During the splitting process, the system assigns an incrementing text tag number to each text tag unit. The text tag unit numbers start from 1. For the same project application text, the text tag unit number is incremented by 1 each time a new text tag unit is read, until the entire project application text has been traversed. For each text tag unit, the system records the text content, text tag unit number, starting character position in the original text, line number, and paragraph number in a structured record. All structured records are then stored in a list structure in ascending order of text tag unit numbers. This list constitutes the project application text tag sequence. In actual deployment, the length of a typical project application text may be between 8,000 and 15,000 characters, and the number of text tag units split into is generally between 10,000 and 20,000. This granularity facilitates the training of the pointer network model while ensuring a one-to-one correspondence between the pointer output and the original text.

[0023] By creating a record containing line and paragraph numbers for each text tag unit, the coordinates corresponding to the spatial overlay unit can be quickly located back to the specific paragraph in the project application text in subsequent steps. For example, the pointer sequence returned by the pointer network model in subsequent steps contains text tag unit numbers from 100 to 120. This range can be immediately mapped to a paragraph in the project application text describing the land use scope, making it convenient to directly reference that paragraph when generating the joint review opinion form for the project. This design ensures that the association between spatial analysis results and textual descriptions no longer relies on string searches but on stable text tag unit numbers, improving retrieval speed and accuracy.

[0024] In another optional implementation, before the project application text enters the text tagging unit splitting process, it can be divided into blocks by chapter titles to facilitate subsequent location. For example, chapter titles such as "Project Overview," "Current Land Use," "Planning Scheme," and "Environmental Impact" can be identified, each chapter can be assigned a chapter number, and a chapter number field can be added to the record of each text tagging unit. In this way, when generating the joint review opinion form for the project later, the system can not only locate the specific line number, but also directly indicate that the issue appears in the "Planning Scheme" chapter, allowing reviewers to understand the context of the issue more quickly.

[0025] For the approval text, it can be split and recorded in the same way as the project application text, forming an approval text tag sequence. Although the key to step one is constructing the project application text tag sequence, in a practical system, a unified splitting rule facilitates the reuse of the same processing logic across different texts, reducing implementation complexity. The approval text tag sequence is mainly used for comparative review in subsequent steps, such as comparing whether the approved land area in the approval document is consistent with the land area description in the project application text.

[0026] After processing the project application and approval documents, the system begins processing the planning and control layers. The planning layer, typically stored in a geographic information database, provides various planning polygon elements for each urban planning project, such as residential land, commercial land, and road land. The control layer contains control polygon elements for various nature reserves, ecological red lines, permanent basic farmland, and other controlled areas. Through a spatial database interface, the system searches for all planning and control polygon elements covering the project area from both the planning and control layers, based on the administrative division number of the area, and loads these elements into memory.

[0027] For each planned polygon feature, the system reads its boundary geometry data. Boundary geometry data is typically represented by a set of planar coordinate points, each including an x-coordinate and a y-coordinate. The system arranges these planar coordinate points into a point sequence according to their storage order, with consecutive planar coordinate points in the sequence connected by a straight line to form an edge of the polygon boundary. In some data formats, the first and last planar coordinate points of the polygon are identical, used to represent a closed boundary. The system can standardize this situation during reading; for example, if the first and last planar coordinate points are identical and the point sequence length is greater than or equal to 4, only the first planar coordinate point can be retained, and the last duplicate point can be deleted, with the closure guaranteed by the drawing logic. The advantage of this approach is that it avoids repeatedly counting the same point when calculating the enclosing rectangle, reducing one redundant comparison.

[0028] For each controlled polygon feature, the system reads its boundary geometry data and constructs a sequence of planar coordinate points in the same manner. To ensure the accuracy of subsequent geometric calculations, the system can uniformly retain the numerical precision of the planar coordinate points to 3 or 6 decimal places. In one embodiment, when using 3 decimal places, the error of the planar coordinate points does not exceed 1 meter, which is suitable for land parcel-level review scenarios. In implementations requiring more detailed review, 6 decimal places can be used, corresponding to an error on the ground surface of approximately centimeters. This uniform precision handling avoids the problem of error accumulation caused by using different precisions for different layers of data.

[0029] After obtaining the planar coordinate point sequences of the planned polygon features and the controlled polygon features, the system begins generating a bounding rectangle for each feature. The specific process is as follows: For a given planned polygon feature, the system iterates through all its planar coordinate points, comparing the x-coordinate and y-coordinate values ​​of each point to identify the minimum, maximum, minimum, and maximum x-coordinate values. Using the minimum x-coordinate and minimum y-coordinate values ​​as the lower-left corner and the maximum x-coordinate and maximum y-coordinate values ​​as the upper-right corner, these two corner points are combined to form a rectangular area, which is the bounding rectangle of the planned polygon feature. The steps for generating bounding rectangles for controlled polygon features are exactly the same.

[0030] Generating bounding rectangles is not merely for recording an approximate range; more importantly, it provides coarse filtering capabilities for subsequent spatial retrieval. Directly performing precise polygon intersection operations on all planned and controlled polygon features results in a computational burden that increases linearly with the number of features. When there are tens of thousands of features in the planning and controlled layers, performing precise intersection checks one by one would exceed acceptable processing times. For example, in an area containing 5,000 planned polygon features and 3,000 controlled polygon features, performing precise intersection calculations on each project plot polygon with all 8,000 features could take tens of seconds to process a single project. By pre-generating bounding rectangles for each feature, rapid intersection checks can be performed at the bounding rectangle level during subsequent spatial retrieval, pre-excluding many features that are unlikely to intersect with the project plot polygons. This typically reduces the number of features requiring precise overlay operations to a few dozen or even a dozen, significantly reducing the overall computational load.

[0031] In one alternative implementation, the system can add a fixed-width buffer to each side when generating the bounding rectangle. For example, when the planar coordinates are in meters, 10 can be subtracted from the minimum x-coordinate value, 10 can be added to the maximum x-coordinate value, 10 can be subtracted from the minimum y-coordinate value, and 10 can be added to the maximum y-coordinate value to form an extended bounding rectangle slightly larger than the polygon boundary. This extended bounding rectangle can ensure that relevant features are included in the candidate set even with minor coordinate mapping errors or slight offsets between layers, avoiding misjudging intersections as non-intersections due to perfectly aligned boundaries. For very small planning polygon features, such as polygons representing small infrastructure land, the extended bounding rectangle can increase the probability of being retrieved, thereby improving the comprehensiveness of the review.

[0032] In another alternative implementation, the planning and control layers may contain multi-component elements, such as several non-contiguous polygons jointly describing a control area. For such elements, there are two possible processing methods. First, treat each component as an independent planning or control polygon element, generating a separate sequence of planar coordinate points and a bounding rectangle. Record their affiliation to the same planning or control object in the attribute information using group numbers. Second, generate a general bounding rectangle for the entire multi-component element, covering the coordinate range of all constituent polygons, while still retaining the individual planar coordinate point sequence for each constituent polygon for subsequent precise overlay calculations. These two methods have different focuses: the former is more refined at the bounding rectangle level, resulting in fewer candidates during subsequent retrieval; the latter is coarser at the bounding rectangle level, but with a simpler data structure. The specific method used can be selected based on the characteristics of the target area's planning data.

[0033] Through the above processing, step one completed the construction of the project application text mark sequence from the project application text and approval text, and read the planning polygon elements and control polygon elements from the planning layer and control layer and generated the corresponding enclosing rectangles. This provides unified and structured basic data for subsequent coordinate extraction using the pointer network model and joint review using spatial retrieval and polygon overlay.

[0034] In a preferred embodiment, step two begins execution after the data preparation in step one is completed. The system has already stored in memory the project application text tag sequence, planning polygon elements, control polygon elements, and the enclosing rectangle corresponding to each planning polygon element and each control polygon element. Based on this, the system sequentially completes coordinate content extraction, project plot polygon construction, R-tree index structure establishment and retrieval, and polygon overlay operations, ultimately forming a set of spatial overlay units.

[0035] First, the system inputs the project application text tag sequence into the pointer network model. Each text tag unit in the project application text tag sequence has a fixed sequential number, accurately reflecting its position in the original project application text. The system finds a corresponding vector representation for each text tag unit. For example, using a pre-trained vector table, common Chinese characters, numbers, punctuation marks, and spaces are mapped to vectors of fixed length. For rare characters in the project application text that do not appear in the vector table, the system can use a uniform placeholder vector to represent them, thus ensuring that each text tag unit in the project application text tag sequence receives a vector of the same length. Subsequently, the system inputs these vectors sequentially into the encoding part of the pointer network model according to the order of the project application text tag sequence. When processing each text tag unit, the encoding part combines the current text tag unit's vector with the internal state of the previous time step to generate a new encoded vector. In this way, from the first text tag unit at the beginning of the project application text to the last text tag unit at the end of the project application text, a sequence of encoded vectors is formed, where each encoded vector contains certain semantic information of its context.

[0036] To enable the pointer network model to not only recognize coordinate content but also its start and end positions, the system pre-inserts special text marker units into the project application text marker sequence to identify the start, end, and decoding ends of coordinate segments. These text marker units typically do not appear naturally in the project application text; for example, a set of special markers can be defined internally and inserted into appropriate positions when constructing the project application text marker sequence. During training, human annotators mark the start and end positions of each coordinate content segment in the project application text as coordinate segment start and end indicators. After all coordinate content is annotated, a decoding end indicator is inserted at the end of the sequence. In this way, the pointer network model can establish the correspondence between coordinate segment start indicators, coordinate content characters, and coordinate segment end indicators during the learning process, without requiring the model to determine the boundaries of coordinate segments itself.

[0037] During the decoding phase, the pointer network model starts from an initial decoding state. In each decoding iteration, the system compares the current decoding state with each encoded vector in the encoded vector sequence, calculating a score list. A higher score indicates that the current position is more likely to be the text tag unit that should be pointed to. Because the pointer network model directly selects positions from the project application text tag sequence instead of generating new characters, it ensures that each output pointer corresponds one-to-one with the original text, avoiding missing or incorrect characters when the coordinate content is long. In a typical scenario, a project application text may contain more than 30 segments of coordinate content, each containing 20 to 60 characters. Traditional regular expression-based solutions are prone to failure when the format changes slightly, while the pointer network model, through learning the contextual semantics, can reliably output the correct pointer sequence even if the applicant uses different expressions, as long as the overall structure is similar.

[0038] When the pointer output of a decoding iteration falls at the start marker of a coordinate segment, the system interprets this as the beginning of a new coordinate content segment. Subsequent decoding iterations tend to select specific coordinate content characters from the encoded vector sequence, such as numbers, decimal points, and commas separating longitude and latitude. The system sequentially adds these positions to the pointer sequence corresponding to the current coordinate content, according to the decoding order, until the decoding output falls at the end marker of the coordinate segment, at which point the current coordinate content segment ends. For the same project application text tag sequence, if there are multiple coordinate content segments, the pointer network model will sequentially output multiple start markers, multiple sequences of coordinate content characters, and multiple corresponding end markers during a single decoding process, until the decoding output falls at the end marker. In this way, the system obtains one or more pointer sequences corresponding to coordinate content. Each pointer sequence consists of a set of text tag unit indices and has a stable mapping relationship with specific positions in the project application text tag sequence.

[0039] The pointer network model was chosen instead of a simple sequence labeling method because coordinate content in project application texts often spans multiple line numbers and paragraphs, sometimes interspersed with descriptive text. Sequence labeling methods are sensitive to boundaries and are prone to missing segments when coordinate content length varies significantly or format is irregular. The pointer network model, by examining the encoded vector sequence during the decoding phase and selecting the most appropriate position based on the current decoding state, can flexibly switch between coordinate content of different lengths. Furthermore, the pointer network model does not require generating additional coordinate content characters; it directly references text marker units in the project application text's marker sequence, thus ensuring complete consistency between the coordinate content and the original text, facilitating retrospective referencing in subsequent joint review opinion forms.

[0040] After the pointer network model outputs the pointer sequence corresponding to the coordinate content, the system reads the coordinate content sequentially from the project application text tag sequence according to the text tag unit number recorded in each pointer sequence. The system concatenates continuous text segments containing longitude values, latitude values, and separators into several coordinate pairs. For example, in a common expression, the text may contain a string of longitude and latitude coordinates, arranged in the order of longitude first and latitude last, similar to "120.123456,31.234567120.223456,31.334567120.323456,31.434567", etc. After splitting and reading, the system parses these contents into 3 coordinate points, each coordinate point containing one longitude value and one latitude value. During the parsing process, the system can group data according to separators such as spaces and commas. If a mismatch is detected in the number of longitude and latitude values ​​in a coordinate content segment, for example, if there are 5 longitude values ​​and 4 latitude values, the system can make simple corrections based on the relationship between adjacent positions or mark the coordinate content segment as abnormal for subsequent manual review.

[0041] For a set of ordered coordinate points belonging to the same coordinate content, the system connects them pairwise according to their order of appearance in the coordinate content to form a polyline for the project plot. To form a closed boundary, the system connects the starting and ending coordinate points of the project plot polyline. If the starting and ending coordinate points are very close in value and the distance is less than a preset tolerance, such as less than 0.5 meters, the system considers the two points to be geographically equivalent. In this case, the ending coordinate point can be replaced with the starting coordinate point to avoid tiny gaps in value on the boundary of the project plot polygon. In one implementation, for multiple coordinate content segments in the same project application text, multiple project plot polygons can be constructed, each representing an independent construction land area in the project. The system establishes an association between the project plot polygons and the pointer sequences of the corresponding coordinate content, recording the pointer sequences and text marker unit number ranges in the data structure of the project plot polygons, thereby forming a two-way correspondence between the project plot polygons and the coordinate content in the project application text.

[0042] For some projects, the applicant may describe the land parcel outline using side lengths and angles, rather than directly listing all vertex coordinates. In another alternative implementation, the pointer network model not only extracts latitude and longitude coordinates from the project application text's tag sequence, but also simultaneously extracts the starting coordinates of the land parcel, the length of each side, and the angle between each side and the previous side. After parsing these geometric constraints, the system recovers all vertex coordinates in the planar coordinate system through simple geometric derivation, thus still being able to construct the project land parcel polygon. In this way, even if the project application text uses different descriptive conventions, as long as it contains sufficient geometric information, the system can recover the project land parcel polygon.

[0043] After obtaining the project site polygon, the system calculates the bounding rectangle of the project site polygon based on the sequence of planar coordinate points of the polygon. The method is consistent with generating the bounding rectangles of the planning polygon elements and control polygon elements in step one, i.e., traversing all planar coordinate points of the project site polygon to determine the minimum, maximum, minimum, and maximum x-coordinate values, and using these four values ​​to construct the bounding rectangle of the project site polygon. In one embodiment, to ensure a certain degree of redundancy, the system can extend a fixed-width buffer outside the boundary of the bounding rectangle of the project site polygon, for example, extending it outward by 5 meters. This can cover planning polygon elements and control polygon elements very close to the edge of the project site polygon.

[0044] Next, the system inserts the bounding rectangles of the planned and controlled polygon features into the R-tree index structure. The R-tree index structure is a hierarchical structure, with each level consisting of several nodes, each corresponding to a set of bounding rectangles. Bottom-level nodes store the bounding rectangles of the planned and controlled polygon features along with their corresponding feature numbers, while upper-level nodes store larger bounding rectangles formed by merging the bounding rectangles of lower-level nodes. During insertion, the system starts from the root node, selecting child nodes with a high degree of overlap with the bounding rectangle to be inserted or with a small area expansion, and adding the rectangle to be inserted to the bounding rectangle set of that child node. When the number of bounding rectangles in a node exceeds a preset capacity limit, for example, more than 50, the system splits that node, dividing the bounding rectangle set into two subsets according to spatial distribution, generating a new node for each subset, replacing the original node with the two new nodes, and adjusting the reference relationships of the upper-level nodes accordingly. Through this on-demand splitting method, the R-tree index structure can maintain a relatively balanced tree structure even as the number of planned and controlled polygon features increases. The access path length during queries is typically between several and a dozen levels, and does not grow linearly with the total number of features.

[0045] refer to Figure 1 , Figure 1This diagram illustrates the principle of the splitting mechanism of the R-tree spatial index structure when node data overflows, as described in this embodiment of the invention. The diagram shows how the system dynamically maintains the balance and query efficiency of the index structure when planning polygon features or control polygon features are inserted into the R-tree index structure during the spatial index creation process described in step two. The left side of the diagram shows the overflow state before node splitting, the right side shows the balanced state after node splitting, and the arrows in the middle indicate the state transition process. In actual land spatial planning and review operations, planning and control layers contain massive amounts of polygon features. As data is continuously loaded or updated, the leaf nodes or intermediate nodes in the R-tree index structure will continuously accept new bounding rectangle objects. Figure 1 The large rectangle on the left represents a parent node to be split. Inside this parent node are several densely packed smaller rectangles, which represent the bounding rectangles of the planned or controlled polygon features already inserted into the node. According to the invention, each index node has a preset capacity limit, for example, illustrated in the diagram as accommodating a certain number of features. When the system attempts to insert a new bounding rectangle into the node, and this causes the total number of bounding rectangles stored in the node to exceed the preset capacity threshold (e.g., exceeding 50 or the density illustrated in the diagram), the system triggers a node splitting algorithm to prevent the traversal query efficiency from degrading to a linear search due to excessive data volume in a single node.

[0046] Figure 1The right side shows the result after the split. The original parent node was replaced by two new nodes, labeled as new node A and new node B. The graph clearly distinguishes the assignment relationships after the split using different fill textures: one texture (such as diagonal fill) represents the set of features assigned to new node A, and another texture (such as intersecting fill) represents the set of features assigned to new node B. This assignment is not random but based on the principle of spatial clustering. The system calculates the geometric center or boundary range of all enclosing rectangles in space. Using a heuristic algorithm that minimizes the overlapping area or the growth of the total coverage area, it assigns a group of enclosing rectangles that are spatially close and concentrated to new node A, and another group of enclosing rectangles that are spatially independent or clustered on the other side to new node B. It is evident in the graph that new node A mainly covers the features in the lower left region, while new node B mainly covers the features in the upper right region. The enclosing rectangles of the two new nodes only cover the feature range within their respective jurisdictions, thus reducing the overall coverage of blank areas. Through this splitting mechanism, the R-tree index structure can always maintain the hierarchical balance of the tree structure, ensuring that the path length from the root node to any leaf node is basically consistent. This is crucial for this invention because subsequent steps require frequent retrieval within the R-tree using the bounding rectangles of the project plot polygons. Only by maintaining the balance of the index structure and the spatial compactness of the nodes can the retrieval time complexity under large-scale data be kept at the logarithmic level, thereby achieving millisecond-level rapid initial screening of tens of thousands of planning and control elements.

[0047] After inserting the bounding rectangles of the planning polygon elements and control polygon elements, the system uses the bounding rectangles of the project plot polygons to perform a search in the R-tree index structure. The search process begins at the root node of the R-tree index structure. The system compares the bounding rectangles of the project plot polygons with the bounding rectangles of the root node's child nodes one by one. For child nodes that have a planar intersection with the bounding rectangles of the project plot polygons, the system continues to visit lower levels; for child nodes that do not intersect at all, the system skips them and does not visit their child nodes again. As the search expands deeper into the tree, the number of nodes visited gradually decreases until it reaches the leaf nodes. In the leaf nodes, the system again compares the bounding rectangles of the project plot polygons with the bounding rectangles stored in the leaf nodes, selecting all bounding rectangles that have a planar intersection with the bounding rectangles of the project plot polygons. Based on the planning polygon element numbers and control polygon element numbers recorded in these bounding rectangles, a planning candidate set and a control candidate set are formed.

[0048] By using the R-tree index structure, the system can preemptively eliminate a large number of planning and regulatory polygon elements that are unlikely to interact with the project site polygons. In a real-world scenario, a municipal planning database may store over 10,000 planning polygon elements and 6,000 regulatory polygon elements. However, for a specific project site polygon, after filtering using the R-tree index structure, the number of elements in the planning candidate set generally does not exceed 100, and the number of elements in the regulatory candidate set generally does not exceed 50, significantly reducing the computational load of subsequent polygon overlay operations.

[0049] In another optional implementation, the system can establish multiple R-tree index structures for different types of controlled polygon features. For example, separate R-tree index structures can be established for ecological protection red lines, permanent basic farmland, and river blue lines. During retrieval, the system can select to query only a portion of the R-tree index structures based on the nature of the project. For instance, for industrial park projects, the R-tree index structures for ecological protection red lines and permanent basic farmland can be queried, while for waterfront landscape projects, the R-tree index structure for river blue lines can be queried. This can further reduce the number of nodes and candidate features that need to be accessed.

[0050] After obtaining the planning candidate set and the control candidate set, the system begins to perform polygon overlay operations on the project site polygons and these candidate elements. The goal of polygon overlay operations is to subdivide the geometric overlap relationship between the project site polygons and the planning and control polygon elements into several non-overlapping spatial overlay units. Each spatial overlay unit has the same planning attributes and the same control attributes, which facilitates accurate compliance judgment for each spatial overlay unit in the subsequent process.

[0051] The polygon overlay operation can be divided into two stages. The first stage is to calculate the intersection points between the project plot polygon boundary and the boundaries of each planned polygon feature and each controlled polygon feature. The system sequentially traverses each edge of the project plot polygon. For each edge, it performs an intersection test with each edge of each planned polygon feature in the planning candidate set. If an intersection point exists, the planar coordinates of the intersection point are recorded, and the intersection point is inserted into the boundary point sequence of the corresponding two edges. The same method is applied to the project plot polygon and each controlled polygon feature in the control candidate set. Through this process, the original boundary point sequence containing only polygon vertices is expanded into an extended boundary point sequence containing both the original vertices and intersection points.

[0052] The advantage of this approach is that, when constructing spatial overlay units later, the traversal path can be changed at the intersection points to jump from the boundary of the project site polygon to the boundary of the planning polygon feature or the boundary of the control polygon feature. The intersection points are essentially connecting nodes between different polygons, dividing the entire space into smaller, more refined regions. From a two-dimensional geometric perspective, the boundary of each closed region can be composed of several edge segments, all of which originate from the extended boundary point sequence of the project site polygon, the planning polygon feature, and the control polygon feature.

[0053] The second stage involves generating closed boundaries by traversing the extended boundary point sequence, thus forming spatial overlay units. The system can randomly select an unused boundary point from the extended boundary point sequence as a starting point and proceed along the boundary of the current polygon in a predetermined direction, such as always traversing clockwise. When an intersection is encountered, the traversal path is switched between different polygons based on the association relationship recorded at the intersection point, for example, switching from the boundary of a project site polygon to the boundary of a planning polygon feature or a control polygon feature. During the traversal, the system adds the boundary points passed through to the boundary point list of the current area in sequence. When the traversal path returns to the starting point, it indicates that a closed boundary has been formed, corresponding to a complete spatial overlay unit. The system saves the boundary point list of this spatial overlay unit and records the project site polygon number, planning polygon feature number, and control polygon feature number associated with the boundary in the spatial overlay unit.

[0054] This traversal process continues until all boundary points and edges have been traversed at least once. At this point, the overlay relationship between the entire project site polygon, all planning polygon elements, and all control polygon elements is decomposed into a set of spatial overlay units. Each spatial overlay unit in the set corresponds to a specific area inside or adjacent to the project site polygon. This area has a single and definite planned land use attribute and control attribute. For example, a spatial overlay unit may represent an area inside the project site polygon that is residential land in the planning layer and falls within the ecological protection red line in the control layer. In subsequent steps, the system can directly apply relevant rules from the land spatial planning control rule base to perform compliance judgments for this spatial overlay unit.

[0055] In one optional implementation, to avoid generating too many extremely small spatial overlay units, the system can calculate the area of ​​each spatial overlay unit immediately after its formation. When the area is less than a preset lower limit, such as less than 1 square meter, these spatial overlay units can be merged into adjacent spatial overlay units. During merging, spatial overlay units that are completely identical in both planning and control attributes are prioritized to ensure that the same rule can still be used for judgment after merging. In this way, the number of spatial overlay units can be reduced while maintaining the accuracy of the review, thereby reducing the storage and computational burden on subsequent steps.

[0056] After the above processing, step two completes the extraction of coordinate content from the project application text mark sequence and parsing it into project plot polygons. The bounding rectangles of the planning polygon elements and the control polygon elements are inserted into the R-tree index structure. The planning polygon elements and control polygon elements that intersect with the bounding rectangles of the project plot polygons are retrieved in the R-tree index structure through the bounding rectangles of the project plot polygons. Polygon overlay operation is performed on the project plot polygons and the retrieval results to finally form a set of spatial overlay units, providing a detailed spatial analysis basis for subsequent joint review and judgment.

[0057] refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the principle of the spatial index hierarchy enclosure structure based on the R-tree in this embodiment of the invention. The diagram visually demonstrates the three-dimensional hierarchical relationship of the R-tree index structure constructed in step two, revealing how the system organizes and manages complex spatial data through progressively enclosing rectangles. The diagram shows different levels of node structures from bottom to top, corresponding to the leaf node layer, intermediate node layer, and root node layer in the R-tree, respectively. The lowest level (Level 0) shows several specific geometric shapes, representing actual planning polygon elements (such as residential land parcels) and control polygon elements (such as ecological protection red line areas). At this level, each polygon element is enclosed by a minimum bounding rectangle (MBR) that is immediately adjacent to its edge. These MBRs are the basic data units actually stored in the index structure. The diagram clearly depicts the distribution of these bottom-level rectangles in the spatial coordinate system; there may be gaps between them, or local overlaps.

[0058] Figure 2The intermediate level (Level 1) shows the parent node hierarchy. Several lower-level bounding rectangles are enclosed by a larger dashed rectangle, representing an intermediate node in the R-tree. The extent of this intermediate node is determined by the union of the bounding rectangles of all its child nodes; that is, the minimum x-coordinate of this node equals the minimum of the minimum x-coordinates of all its child nodes, the maximum x-coordinate equals the maximum of the maximum x-coordinates of all its child nodes, and so on. This hierarchical enclosure relationship is visually represented as "containment" in the diagram, indicating that if the query engine determines that the query window does not intersect with the dashed rectangle of an intermediate node, it can directly exclude all lower-level elements contained within that node without traversing it step by step. This is the core principle behind this invention's ability to significantly improve the speed of collaborative review. The topmost layer in the diagram shows the bounding rectangle of the root node, which has the widest coverage and contains all intermediate nodes. When the system uses project plot polygons for retrieval, the search path starts from the root node and performs pruning and filtering from top to bottom. For example, if the project site is located in the lower left corner of the coordinate system shown in the figure, after the system determines the intersection at the root node and enters the next layer, it will find that it does not intersect with the middle node in the upper right corner, and will immediately abandon the search of all data in the upper right branch, thereby greatly reducing the amount of computation. Figure 2 This clear nested structure diagram illustrates how the present invention transforms complex intersection operations of millions of polygons into an efficient rectangular tree traversal process, providing an efficient data access foundation for the subsequent generation of spatial overlay units and review opinions.

[0059] In a preferred embodiment, step three is initiated after the spatial overlay unit set calculation is completed. At this point, the system has obtained the project plot polygon number, planning polygon element number, control polygon element number, and complete boundary point sequence corresponding to each spatial overlay unit in the spatial overlay unit set. It also saves the correspondence between the coordinate content in the project application text and the project plot polygons. Based on this information, the system classifies the spatial overlay units through the land spatial planning control rule base and further generates a joint project review opinion form.

[0060] First, the system reads the land and space planning control rule base from the business system. The land and space planning control rule base is typically stored in the form of database tables, configuration files, or dedicated rule files, with each rule recorded in a structured manner. Each rule includes at least the following fields: planning use type, control type, overlay relationship type, and suggested processing method. For example, the planning use type may include several enumerated values ​​such as residential land, industrial land, and public management and public service land; the control type may include categories such as ecological protection red line, permanent basic farmland, and urban development boundary; the overlay relationship type indicates whether overlay is allowed between a certain planning use type and a certain control type, and can take three states: allowed overlay, restricted overlay, or prohibited overlay. The suggested processing method provides handling suggestions when restricted or prohibited overlay occurs, such as partial setback, adjustment of functional attributes, or splitting of land parcels. Through this structured representation, the system can directly match and judge spatial overlay units without interpreting the text rules.

[0061] In a typical configuration, the land use planning and control rule base contains more than 100 rule records. For example, for the combination of residential land and ecological protection red lines, there may be a rule record prohibiting their overlap; for the combination of public green space and ecological protection red lines, there may be a rule record allowing their overlap; and for the combination of industrial land and permanent basic farmland, there may be a rule record prohibiting their overlap. By explicitly listing these combinations in the rules, misunderstandings during manual review can be avoided, and it is also convenient to expand and adjust them in different regions according to local policies.

[0062] The system then determines the attribute information of each spatial overlay unit in the spatial overlay unit set within the planning and control layers. To do this, the system accesses the planning layer data based on the planning polygon element number recorded by the spatial overlay unit, reading the planning use type field of that planning polygon element, such as residential land or industrial land. Similarly, the system accesses the control layer data based on the control polygon element number recorded by the spatial overlay unit, reading the control type field of that control polygon element, such as ecological protection red line or permanent basic farmland. For some spatial overlay units, they may only be overlaid with planning polygon elements without any control polygon elements; in this case, the control type can be recorded as "no control category." For spatial overlay units falling within the overlapping area of ​​multiple control polygon elements, the control type can be sorted according to the overlay area, with the control type covering the largest area being the primary control type, while other control types are recorded as supplementary information and presented together when generating the joint project review opinion form.

[0063] After obtaining the planned use type and control type of the spatial overlay unit, the system begins matching them with the land and space planning control rule base. The matching process typically involves two steps. First, the system selects a subset of rules from the rule base that match the planned use type of the spatial overlay unit. For example, if the planned use type of the spatial overlay unit is residential land, the system first filters out all records with the planned use type of residential land from the rule base. Second, the system continues filtering within this rule subset based on the control type of the spatial overlay unit, finding rule records with the same control type as the spatial overlay unit. If multiple rule records match the current spatial overlay unit, they can be sorted according to the rule priority field, and the rule with the highest priority is selected as the final rule. If no completely matching rule record is found, a default rule can be used; for example, if the planned use type is general construction land and the control type is empty, the default is "pass category".

[0064] In a preferred embodiment, the system categorizes spatial overlay units into three compliance categories: permitted, requiring optimization, and requiring adjustment. In the matched rule records, if the overlay relationship type is permitted, the system classifies the spatial overlay unit as permitted; if the overlay relationship type is restricted, the system classifies the spatial overlay unit as requiring optimization, generally indicating that the area can retain its existing planned use after minor adjustments to boundaries or building layout; if the overlay relationship type is prohibited, the system classifies the spatial overlay unit as requiring adjustment, indicating that the land use boundaries or land use nature of the area require substantial changes. Using three compliance categories, instead of a simple two-category system of compliant and non-compliant, introduces a transitional layer in the review process, allowing the system to automatically identify which areas only require minor adjustments and which areas must undergo strict modifications, thereby reducing the judgment burden on manual reviewers.

[0065] For example, in a municipal project, suppose a spatial overlay unit has an area of ​​500 square meters, a planned use type of residential land, and a control type of ecological protection red line. After matching the rule base, if the corresponding rule record configures this combination as prohibited from overlaying, the spatial overlay unit will be classified as requiring adjustment. Another spatial overlay unit has an area of ​​200 square meters, a planned use type of park green space, and a control type of ecological protection red line. If the corresponding rule record is configured as allowing overlaying, this spatial overlay unit will be classified as approved and will not require special handling when generating the modification point list later.

[0066] After classifying spatial overlay units into compliance categories, the system focuses on handling spatial overlay units whose compliance category does not meet the passing criteria, i.e., those requiring optimization or adjustment. For these two types of spatial overlay units, simply marking them as problematic is insufficient; the system needs to provide specific modification points to guide subsequent adjustments to the project site's polygon boundaries or planning scheme. To minimize modifications while eliminating illegal overlays, the system searches for boundary points adjacent to the boundaries of controlled polygon elements within the spatial overlay unit's boundary point sequence, using these boundary points as a candidate set of modification points.

[0067] In practice, the system first identifies boundary segments on the spatial overlay unit boundary that belong to the project site polygon and those that belong to the control polygon element. When constructing the spatial overlay unit, each boundary segment can record a source type field to distinguish whether it originates from the project site polygon, planning polygon element, or control polygon element. The system traverses the entire boundary of the spatial overlay unit. If it finds a boundary segment with a project site polygon on one side and a control polygon element on the other, it can determine that the project site polygon and the control polygon element are adjacent or even partially overlapped at that location. Spatially, these boundary segments are located at the intersection of the project land and the control area. By adjusting the boundaries of the project site polygon at these locations, it is possible to completely avoid the control area while minimizing changes to the main development scope.

[0068] In one specific embodiment, for spatial overlay units whose compliance category requires adjustment, the system traverses the boundary point sequence of the spatial overlay unit, identifies all boundary segments whose source types combine to form the project plot polygon and the control polygon elements, and adds all boundary points at both ends of these boundary segments to the modification point candidate set. If the length of a boundary segment is much greater than that of adjacent segments, for example, exceeding 50 meters, the system can insert several intermediate points at fixed intervals along this boundary line, such as inserting an intermediate point every 10 meters, and add these intermediate points to the modification point candidate set as well. In this way, when generating modification schemes, planners can not only make adjustments at both ends of the boundary segments, but also select more suitable turning points in the middle, achieving avoidance with minimal movement. This approach helps to reduce the impact on the overall plot shape.

[0069] For spatial overlay units categorized as requiring optimization, the system can employ a more lenient strategy. In one alternative implementation, the system only includes the boundary points at both ends of the boundary segment in the candidate set of modification points, without interpolating additional points on the longer side. This is because the category requiring optimization typically indicates slight boundary overlap or minor contact within the buffer zone. Small adjustments at both ends can often completely remove the spatial overlay unit from the sensitive area, avoiding complex modifications to the entire boundary. This reduces subsequent processing workload and avoids introducing new vertices in unnecessary areas, maintaining the simplicity of the project site's polygonal boundary.

[0070] After identifying the candidate set of modification points, the system will filter and sort them. During the filtering phase, certain thresholds can be set based on the area of ​​the spatial overlay unit and the length of the corresponding boundary segment. For example, if the area of ​​the spatial overlay unit is less than 5 square meters and the corresponding boundary segment length is less than 3 meters, the corresponding modification point can be classified as low priority and submitted to manual review if necessary. During the sorting phase, groups can be formed based on the type of control and the severity of the violation. For example, modification points involving ecological protection red lines and permanent basic farmland are given priority, followed by modification points involving general control lines. This way, when presenting modification points in the joint project review opinion form, the most critical modification locations can be listed first, making it easier for reviewers to quickly locate key issues.

[0071] Modifying points requires not only spatial location but also establishing a connection with the coordinates in the project application text. Since the project site polygons have already established a correspondence with the coordinates in the project application text during previous steps, the system can use the project site polygon numbers to find the corresponding pointer sequences and text marker unit number ranges. For each modified point, the system finds the relevant coordinates based on its project site polygon number, corresponding to the original coordinate point's location in the project application text. Thus, when generating the list of modified points, it can not only list the specific values ​​of the modified points in geographic coordinates but also indicate the specific line numbers and paragraph positions of these points in the project application text, making it easy for the applicant to quickly locate the parts that need adjustment in the original text and update the coordinate descriptions.

[0072] In one embodiment, the joint review opinion form for the project is generated according to a standardized format. The opinion form typically includes a basic project information section, an overall review conclusion section, a spatial overlay unit classification summary section, and a list of modification points. The basic project information section records information such as the project name, project number, review time, and reviewing unit. The overall review conclusion section describes the overall status of the entire project site polygon after matching it with the planning and control layers in natural language. For example, it explains whether the project land use completely avoids the ecological protection red line and whether it partially encroaches on permanent basic farmland, and provides statistics on the number of spatial overlay units in the approved, optimized, and adjusted categories. The spatial overlay unit classification summary section can be in tabular form, listing the total area and quantity of each type of spatial overlay unit according to planning use type and control type.

[0073] The modified site list is a key component for seamlessly integrating automated review with manual adjustments. For each spatial overlay unit whose compliance category does not fall under the "passed" category, the system generates one or more records in the list. Each record includes at least the spatial overlay unit number, planned use type, control type, administrative region, area involved, modified site number, modified site planar coordinates, corresponding project plot polygon number, and line and paragraph numbers from the corresponding project application text. In practical deployment, visual link fields can be further added, such as generating a map view link for each modified site. Reviewers and staff from the applying unit can click on this link to directly view the modified site's location on the map within the geographic information system interface, allowing for comprehensive judgment in conjunction with the base map and other layers.

[0074] In another optional implementation, the joint review opinion form can also generate suggested modification directions. When calculating modification points, the system can record the directional relationship between the project site polygon boundary and the control polygon element boundary, such as determining whether the project land boundary overlaps northward or southward. Based on this spatial relationship, textual prompts can be added to the modification point list, such as suggesting that the project site polygon be moved back 5 meters to the south along one side. In this way, the applicant does not need to devise new boundary positions from scratch when adjusting the plan; they only need to make appropriate translations along the suggested direction, effectively eliminating the overlap of violations.

[0075] refer to Figure 3 , Figure 3This is a schematic diagram illustrating the principle of automatic identification of boundary modification points and generation of yield vectors in this embodiment of the invention. The diagram details how, in step three, when the system determines that the project plot polygon and the control polygon elements have an illegal overlap (e.g., the compliance category is an adjustment category or an optimization category), it automatically calculates the specific modification location and generates a suggested adjustment direction. The core elements in the diagram include the project plot boundary (usually represented by a solid line), representing the declared land use area; the control polygon boundary, representing the mandatory binding force (usually represented by a dashed line or different colors, such as an ecological red line); and the area where the two spatially conflict. The diagram clearly marks the "illegally encroached area," that is, the part of the project plot boundary that intrudes into the control polygon, which is the direct cause of the review failure. To guide the applicant in making rectifications, Figure 3 The diagram showcases key modification points identified by the system. These points include not only the geometric intersections of project site boundaries and control boundaries, but also the vertices of project sites located within violation areas or midpoints inserted according to the long-side subdivision strategy. These locations are marked with prominent dots in the diagram, indicating that they are key points requiring coordinate adjustments.

[0076] Furthermore, Figure 3 The diagram illustrates the generation of the setback vectors according to the present invention. At each identified modification point, arrows pointing outwards from the controlled area are drawn, representing the suggested setback direction and distance calculated by the system. The system analyzes the geometry of the overlapping area, calculates the shortest distance vector from the modification point to the nearest boundary of the controlled polygon, and adds a safety buffer distance (e.g., 0.5 meters or 1 meter) to this, thus forming the green setback vector in the diagram. For example, the diagram shows that the upper edge of the project plot encroaches on the controlled area above. The system generates a downward arrow at the corresponding modification point, visually indicating that shifting the boundary downwards by the length indicated by the arrow at that point will eliminate the illegal overlap. This visual representation transforms the abstract "no overlap" rule into a concrete geometric operational guide. In the generated joint project review opinion form, this graphical information is converted into specific textual descriptions (e.g., "It is recommended to set back 5 meters south at coordinate point P") or provided directly as an illustration to the user. Figure 3 The system also demonstrated how it handles boundary conflicts on long straight lines by generating multiple suggested points through interpolation on the long straight lines. This guides users to achieve flexible local avoidance by adding fold points, rather than drastically cutting up plots to meet compliance requirements.

[0077] In another optional implementation, after generating the joint review opinion form for the project, the system can return the results to the process management section of the business system. Spatial overlay units that pass the review are directly marked as approved, while only spatial overlay units requiring category optimization or adjustment are subject to a review process. The review process can require the project entity to adjust the planning scheme according to the list of modified locations within 30 days and resubmit the updated project application text and planning layer. Upon receiving the updated data, the system can repeat the aforementioned steps to regenerate the spatial overlay unit set and the joint review opinion form for the project. Through this cycle, project entities can gradually align their project land use with various control requirements under the system's guidance, while the review department can significantly reduce repetitive workload.

[0078] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result according to substantially the same method falls within the scope of the present invention. Therefore, the scope of the present invention is defined only by the appended claims.

Claims

1. A joint review and judgment method for the entire chain of land and space planning compilation and review, characterized in that: Includes the following steps: Step 1: Read the project application text, approval text, planning layer, and control layer; break down the project application text into text tag units to construct a project application text tag sequence; read the planning polygon elements and control polygon elements from the planning layer and control layer and generate the corresponding enclosing rectangles. Step 2: Input the project application text tag sequence into the pointer network model, extract the coordinate content from the project application text tag sequence and parse it into project plot polygons, insert the bounding rectangles of the planning polygon elements and control polygon elements into the R-tree index structure, retrieve the planning polygon elements and control polygon elements that intersect with the bounding rectangles of the project plot polygons in the R-tree index structure through the bounding rectangles of the project plot polygons, perform polygon overlay operation on the project plot polygons and the retrieval results to form a set of spatial overlay units; Step 3: Read the land and space planning control rule library, determine the compliance category of the spatial overlay unit based on the attribute information of the spatial overlay unit in the planning layer and control layer, extract the boundary points adjacent to the boundary of the control polygon element of the spatial overlay unit whose compliance category is not the category, and generate a joint review opinion form for the project containing a list of modification points and review conclusions.

2. The method according to claim 1, characterized in that the pointer network model in step two includes an encoding part and a decoding part; wherein, The encoding part sequentially reads each text tag unit in the project application text tag sequence, maps each text tag unit to a fixed-length vector representation, and generates an encoding vector based on the vector representation of the current text tag unit and the state vector of the previous position. All encoding vectors are then assembled into an encoding vector sequence according to the order of the text tag units.

3. The method according to claim 2, characterized in that, the decoding part pre-inserts a coordinate segment start indicator for identifying the start of the coordinate segment, a coordinate segment end indicator for identifying the end of the coordinate segment, and a decoding end indicator for identifying the end of decoding in the project declaration text mark sequence; sets an initial decoding state at the initial moment of decoding; in each decoding iteration, the similarity calculation module is used to compare the current decoding state with each encoding vector in the encoding vector sequence to generate a score list corresponding one-to-one with all text mark units; the position of the text mark unit with the highest score is selected from the score list; the index of the text mark unit corresponding to the position is used as the pointer output of the current decoding iteration; the pointer output is added to the pointer sequence; and the decoding state is updated at the same time.

4. The joint review and judgment method for the entire chain of land and space planning compilation and review as described in claim 3 is characterized in that the decoding part in step two is further executed as follows: when the pointer output of the current decoding iteration corresponds to the start indicator mark of the coordinate segment, in the next and subsequent multiple decoding iterations, the text mark unit positions corresponding to the specific coordinate content are selected from the encoded vector sequence, and the text mark unit numbers corresponding to these positions are added to the pointer sequence in sequence until the current pointer outputs the end indicator mark of the corresponding coordinate segment. The text mark unit numbers located between the start indicator mark and the end indicator mark of the coordinate segment are determined as a pointer sequence corresponding to a coordinate content. When the pointer outputs the end indicator mark of the current decoding iteration, the entire decoding process is ended, and one or more pointer sequences corresponding to coordinate content are obtained. Each pointer in each pointer sequence corresponds to a text mark unit in the project application text mark sequence.

5. The joint review and judgment method for the entire chain of land spatial planning compilation and review as described in claim 4, characterized in that step two further includes: According to the text marker unit number recorded in each pointer sequence, the coordinate content is read sequentially from the project application text marker sequence. The continuous text fragment containing longitude values, latitude values, and separators is parsed into a set of ordered coordinate points. The ordered coordinate points belonging to the same coordinate content are connected in pairs according to the order in the coordinate content to form a project plot polyline. The starting coordinate point and the ending coordinate point of the project plot polyline are connected to form a closed boundary, resulting in a project plot polygon. The project plot polygon is associated with the pointer sequence of the corresponding coordinate content. The corresponding pointer sequence and the corresponding text marker unit number range are recorded in the project plot polygon, forming a correspondence between the project plot polygon and the coordinate content in the project application text.

6. The method according to claim 5, characterized in that the construction and querying of the R-tree index structure in step two includes: The bounding rectangles of each planned polygon feature and each controlled polygon feature are sequentially inserted into the R-tree index structure. During insertion, the insertion path is selected based on the center coordinates of the bounding rectangles. When the number of bounding rectangles in a node exceeds a preset capacity threshold, the current node is split, and the set of bounding rectangles in the node is re-divided during the splitting process, generating new internal nodes and leaf nodes. This makes the R-tree index structure form a multi-level hierarchy containing root nodes, internal nodes, and leaf nodes. Each leaf node stores the planning polygon feature number and the controlled polygon feature number associated with the bounding rectangle. For each project plot polygon, the minimum and maximum values ​​of the planar coordinate points of the project plot polygon in the horizontal and vertical directions are calculated to generate the bounding rectangle of the project plot polygon. The bounding rectangle of the project plot polygon is used as the query rectangle. Starting from the root node of the R-tree index structure, the system traverses downwards. During the traversal, only nodes that have a planar intersection with the query rectangle are visited downwards until a leaf node is reached. The planning polygon feature number and the controlled polygon feature number corresponding to the bounding rectangle that has a planar intersection with the query rectangle are read from the leaf node, forming the planning candidate set and the controlled candidate set corresponding to the project plot polygon.

7. The method according to claim 6, characterized in that the polygon overlay operation in step two includes: For the polygon boundaries of the project site polygon and each planning polygon element in the planning candidate set, calculate all intersection points between the project site polygon boundary and the planning polygon element boundary, and insert each intersection point into the boundary point sequence of the corresponding edge to form an extended boundary point sequence containing the original vertices and intersection points. For the polygon boundaries of the project site polygon and each control polygon element in the control candidate set, calculate all intersection points between the project site polygon boundary and the control polygon element boundary, and insert each intersection point into the boundary point sequence of the corresponding edge to form an extended boundary point sequence containing the original vertices and intersection points. Select a starting point from the extended boundary point sequence, and traverse the edges along the polygon boundary in a predetermined direction. When traversing to an intersection point, switch the traversal path between the project site polygon, planning polygon element, and control polygon element according to a predetermined switching rule. At the end of the traversal, form closed polylines representing the boundaries of different overlay areas. Write the area enclosed by each closed polyline as a spatial overlay unit into a spatial overlay unit set. Record the corresponding project site polygon number, planning polygon element number, or control polygon element number, as well as the boundary point sequence constituting the boundary of the spatial overlay unit in each spatial overlay unit.

8. The method according to claim 7, characterized in that step two further comprises: Based on the correspondence between the project plot polygons and the coordinate content in the project application text, the project plot polygon numbers recorded in the spatial overlay unit are mapped to the coordinate content positions in the project application text. The planning polygon element numbers and control polygon element numbers recorded in the spatial overlay unit are mapped to the element numbers in the planning layer and control layer, forming a three-way mapping relationship between the spatial overlay unit and the coordinate content of the project application text, the planning layer elements and the control layer elements. When storing spatial overlay units, the three-party mapping relationship is written as an attribute field into the spatial overlay unit, so that each spatial overlay unit corresponds to both the specific geographical area and the specific coordinate content location in the project application text.

9. The method according to claim 8, characterized in that, Step three includes: reading the land and space planning control rule base from the business system, representing each rule in the land and space planning control rule base as a structured rule entry containing the planning use type, control type, and permitted or prohibited overlay relationships; for each spatial overlay unit in the spatial overlay unit set, based on the project plot polygon number, planning polygon element number, and control polygon element number recorded in the spatial overlay unit, searching for the attribute information of the corresponding elements in the planning layer and control layer to determine the planning use type and control type to which the spatial overlay unit belongs; matching the planning use type and control type to which the spatial overlay unit belongs with the structured rule entries in the land and space planning control rule base to obtain the structured rule entries that match the current spatial overlay unit, and classifying the current spatial overlay unit into one of the following compliance categories: approved category, optimization category, and adjustment category, based on the permitted or prohibited overlay relationships recorded in the structured rule entries.

Images