A method and system for automatically generating and marking boundary point numbers

Abstract

The application discloses a kind of boundary point automatic generation and boundary point number annotation method and system, it is related to cadastral survey and graphics processing field, technical scheme is as follows: obtaining to be processed land parcel polyline set, existing boundary point position set and configuration parameter;Extract coordinate node output candidate boundary point set;Candidate position coordinate and existing position coordinate are compared in space, and after eliminating coincident position, generate boundary point object set;Extract the boundary node order of target boundary point object in to-be-processed land parcel polyline to generate order number;Calculate the space azimuth angle formed by previous adjacent boundary point object, target boundary point object and next adjacent boundary point object, input coordinate translation operator with configuration parameter, output annotation position and generate corresponding boundary point number annotation.The application solves the problems of low extraction efficiency of boundary point in cadastral mapping, boundary number confusion and mutual overlapping conflict of drawing annotation.

Description

Technical Field

[0001] This invention relates to the field of cadastral surveying and graphic processing technology, and more specifically, to a method and system for automatically generating boundary points and annotating boundary point numbers. Background Technology

[0002] Cadastral surveying, real estate registration, and parcel map compilation are fundamental tasks in land resource management and land ownership registration. In these processes, boundary points are used to identify the turning points, intersections, or boundary control points of land parcel ownership boundaries. The accuracy of their coordinates, the uniqueness of their numbers, and the clarity of their map annotations directly affect the accuracy of the land parcel ownership boundary representation and also influence the review, storage, and use of subsequent registration results. Typically, office staff need to extract boundary points one by one from the multiple lines of the parcel, assign each boundary point a boundary point number and a sequence number within the parcel, and then annotate the boundary point numbers near the boundary points on the parcel map. Due to the complex shape of land parcel boundaries, adjacent parcels often share boundaries or boundary points. Boundary point generation and annotation are not simply point-line extraction problems, but rather involve comprehensive processing issues such as graphic recognition, topological relationship maintenance, coordinate accuracy control, and text avoidance on the map.

[0003] Existing graphic generation and annotation avoidance technologies are typically geared towards general mapmaking, common spatial entity annotation, or engineering icon annotation optimization. Their focus is primarily on graphic display effects, symbol drawing efficiency, or text collision avoidance, without establishing a comprehensive processing mechanism for boundary points in cadastral maps. Boundary points are not ordinary graphic points; they carry both geometric coordinates and the legal meaning of land parcel ownership boundaries. Boundary point numbers are not ordinary text annotations either; they need to be consistent with the internal sequence number of the parcel, the sharing relationship with adjacent parcels, and the parcel map expression standards. Therefore, directly using existing graphic generation or annotation avoidance methods still cannot meet the requirements for automatic generation and compliant storage of cadastral results.

[0004] Specifically, the existing technology has at least the following problems.

[0005] In the automatic boundary point generation stage, traditional methods typically generate point objects directly based on polyline nodes or line segment turning points of a parcel. For a single parcel, this method can obtain basic boundary point locations; however, in scenarios where adjacent parcels share boundaries or multiple parcels intersect at the same boundary marker location, the same physical boundary point may belong to multiple parcels simultaneously. Existing methods lack a unified maintenance mechanism for the subordinate relationships between the same physical boundary point and multiple parcels, which easily leads to problems such as the same boundary marker being repeatedly generated as multiple point objects, or the sequential number relationship of the same boundary point in different parcels being covered, omitted, or confused. This results in point numbers appearing on the map, but the underlying ownership numbering logic is inconsistent, affecting subsequent result review and the expression of ownership relationships.

[0006] In the candidate boundary point extraction stage, existing graphic processing methods often only focus on the geometric nodes themselves, making it difficult to distinguish between truly significant inflection points and auxiliary nodes generated by line segment densification, drawing errors, and historical editing. Treating all polyline nodes as boundary points results in an excessive number of boundary points, redundant numbering, and crowded annotations; while filtering based solely on simple angle thresholds may miss important control points related to main building corners, building corners, or actual boundary marker locations. Therefore, existing technologies lack a candidate boundary point extraction method that combines the order of land parcel boundary nodes, the interior angle characteristics of consecutive nodes, and the spatial overlap of main building corners.

[0007] In the coordinate accuracy processing stage, boundary point coordinates typically need to be rounded according to specifications before cadastral data is entered into the database. While coordinate rounding may seem to only change the decimal places of coordinates, simultaneous minor changes in the coordinates of multiple boundary points can alter the inverse area of ​​the land parcel polygon, leading to inconsistencies between the inverse area and the legally owned area. Current general cartographic techniques usually only focus on map display and graphic output, without actively assessing the impact of coordinate rounding on the ownership area. They also lack a constraint optimization mechanism that allows for flexible node fine-tuning to absorb area errors while keeping rigid boundary points stationary. Therefore, there is a difficulty in automatically coordinating coordinate standardization and area legitimacy.

[0008] In the process of arranging boundary point numbers, existing annotation avoidance methods mostly rely on text bounding box collision detection, local movement, or view window rasterization to achieve avoidance. While these methods can reduce text overlap to some extent, they typically do not fully utilize the topological hierarchical relationships of land parcel boundaries. In scenarios where multiple land parcels share the same boundary point, if avoidance is performed only from a single annotation object, the boundary point number of one parcel may be placed inside an adjacent parcel, or the point numbers of multiple parcels may crowd each other around the same shared point, overlapping and affecting the accurate representation of ownership boundaries on the land parcel map. Therefore, an automatic annotation method is needed that can use the shared boundary point as the center, divide spatial areas according to the intersecting boundary lines, and constrain the boundary point numbers of different parcels to be arranged within their respective corresponding areas. Summary of the Invention

[0009] The technical problems to be solved by this invention are low efficiency in extracting boundary points in cadastral mapping, chaotic numbering of cross-parcel boundaries, and overlapping and conflicting map annotations.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: A method for automatically generating boundary points and annotating boundary point numbers includes: Obtain the set of polylines for the land parcel to be processed, the set of existing boundary point locations, and the boundary point number annotation configuration parameters; Extract the polyline coordinate nodes from the polyline set of the land parcel to be processed, and output the candidate boundary point set; The candidate position coordinates in the candidate boundary point set are compared with the existing position coordinates in the existing boundary point position set. The candidate position coordinates that overlap are eliminated, and a set of boundary point objects is generated at the remaining candidate position coordinates. For each target boundary point object in the set of boundary point objects, extract the boundary node order of the target boundary point object in the corresponding polyline of the parcel to be processed, and perform a sorting operation to generate the sequence number of the target boundary point object; Extract the preceding and following adjacent boundary point objects of the target boundary point object in the corresponding polyline of the parcel to be processed. Calculate the spatial azimuth angle formed by the preceding adjacent boundary point object, the target boundary point object, and the following adjacent boundary point object. Input the spatial azimuth angle and the boundary point number annotation configuration parameters into the coordinate translation operator, output the annotation position of the target boundary point object, and generate a boundary point number annotation associated with the target boundary point object at the annotation position.

[0011] Preferably, the step of extracting polyline coordinate nodes from the polyline set of the land parcel to be processed and outputting a candidate boundary point set includes: obtaining a set of main house corner points; for each polyline of the land parcel to be processed in the set of polylines of the land parcel to be processed, extracting the upper left corner vertex of the polyline of the land parcel to be processed as the starting node, and sequentially extracting all polyline nodes of the polyline of the land parcel to be processed in a clockwise direction to construct a standardized vertex sequence; extracting the interior angle feature formed by three consecutive polyline nodes in the standardized vertex sequence, and inputting the interior angle feature into the inflection point determination template library; if the inflection point determination template library outputs a matching success identifier, then the middle polyline node among the three consecutive polyline nodes is identified as an important inflection point; merging all the important inflection points and the main house corner points in the set of main house corner points that have a spatial overlap with the corresponding polyline of the land parcel to be processed, and outputting the candidate boundary point set.

[0012] Preferably, after generating the sequence number of the target boundary point object by performing sorting operations, the method further includes: obtaining the identification information of the polyline to be processed to which the target boundary point object belongs; combining the identification information and the sequence number of the target boundary point object in the corresponding polyline to be processed using a connector to generate a parcel sequence number unit; when it is determined that the target boundary point object belongs to multiple different polylines to be processed, concatenating the multiple parcel sequence number units using a separator to output a spliced ​​structure, and marking the boundary point object with the spliced ​​structure as a globally shared boundary point object.

[0013] Preferably, after generating a set of boundary point objects at the remaining candidate location coordinates and before performing sorting operations to generate the sequence number of the target boundary point objects, the method further includes: for each polyline to be processed in the set of polylines to be processed, extracting each boundary point object belonging to that polyline to be processed, performing precision rounding on the coordinates of each boundary point object, and outputting the rounded coordinate sequence corresponding to the polyline to be processed; obtaining the legal ownership area data associated with the polyline to be processed; inputting the rounded coordinate sequence and the legal ownership area data into an area conservation constraint optimization model, and outputting the fine-tuned position coordinates of each coordinate in the rounded coordinate sequence through the area conservation constraint optimization model; and updating the coordinates of each boundary point object to the corresponding fine-tuned position coordinates.

[0014] Preferably, the area conservation constraint optimization model is a classification and regression joint model obtained through supervised learning training using qualified historical cadastral survey data and unqualified historical cadastral survey data. The area conservation constraint optimization model includes a rigid node decision subnetwork and a coordinate optimization subnetwork. The step of inputting the rounded coordinate sequence and the legal ownership area data into the area conservation constraint optimization model, and outputting the fine-tuning position coordinates of each coordinate in the rounded coordinate sequence through the area conservation constraint optimization model, includes inputting the rounded coordinate sequence into the rigid node decision subnetwork, and the rigid node decision subnetwork outputting the rounded coordinate sequence. The node classification labels corresponding to each coordinate in the input coordinate sequence are provided, including rigid node labels and flexible node labels. The rounded coordinate sequence with the node classification labels and the legal ownership area data are input into the coordinate optimization sub-network. The coordinate optimization sub-network locks the coordinates with the rigid node labels in the rounded coordinate sequence to keep their positions fixed. Using the legal ownership area data as a constraint, and minimizing the total displacement of the coordinates with the flexible node labels in the rounded coordinate sequence as the objective function, the sub-network performs a sequential quadratic programming optimization calculation and outputs the fine-tuned position coordinates.

[0015] Preferably, the training process of the rigid node decision sub-network includes: acquiring sample attribute data of sample boundary point objects and the real node classification labels corresponding to the sample boundary point objects, wherein the sample attribute data includes boundary type attributes and the number of polyline shares of the parcel to be processed; inputting the sample attribute data into the untrained rigid node decision sub-network and outputting predicted node classification labels; calculating the cross-entropy loss value between the predicted node classification labels and the real node classification labels, and updating the network parameters of the untrained rigid node decision sub-network based on the cross-entropy loss value, continuously performing parameter update operations until the cross-entropy loss value converges, and outputting the trained rigid node decision sub-network.

[0016] Preferably, for the target boundary point object marked as a globally shared boundary point object, the step of outputting the annotation position of the target boundary point object includes: extracting all boundary line segments intersecting the globally shared boundary point object; inputting the coordinates of all boundary line segments and the globally shared boundary point object into a topological sector spatial allocation model; the topological sector spatial allocation model, centered on the coordinates of the globally shared boundary point object, using all boundary line segments to divide the two-dimensional plane space into multiple interior corner sectors enclosed by adjacent boundary line segments; obtaining the spatial subordinate mapping relationship between each to-be-processed land parcel polyline associated with the globally shared boundary point object and each of the interior corner sectors; for each target to-be-processed land parcel polyline among the various to-be-processed land parcel polylines, the topological sector spatial allocation model constrains the text bounding box of the boundary point number annotation belonging to the target to-be-processed land parcel polyline within the corresponding interior corner sector for arrangement calculation, and outputs the annotation position independent of other to-be-processed land parcel polylines.

[0017] Preferably, the topological sector spatial allocation model constrains the text bounding boxes belonging to the boundary point number annotations of the target parcel polyline within the corresponding interior corner sector for arrangement calculation, outputting the annotation position independent of other parcel polylines, including: extracting the two edge rays and the angle bisector of the interior corner sector; placing the center point of the text bounding box on the angle bisector; calculating the vertical distance from the vertex of the text bounding box to the two edge rays within the interior corner sector; performing particle swarm optimization iterative calculation, using the maximum vertical distance as the optimization fitness function, changing the position vector of the center point on the angle bisector and the scaling parameter of the text bounding box until the optimization fitness function converges, and outputting the position vector when the maximum fitness function value is obtained as the annotation position.

[0018] Preferably, before inputting the coordinates of all boundary segments and the globally shared boundary point object into the topological sector spatial allocation model, the method further includes: extracting a polygonal topology tree composed of polylines of the parcel to be processed containing the globally shared boundary point object; identifying the inner ring boundary segments and outer ring boundary segments among all boundary segments through the polygonal topology tree; performing a flipping process on the direction vector of the inner ring boundary segment to obtain an adjusted boundary segment with a unified direction; and outputting the coordinate combination of the outer ring boundary segment, the adjusted boundary segment, and the globally shared boundary point object as the input variable of the topological sector spatial allocation model.

[0019] This invention also discloses an automatic boundary point generation and boundary point number annotation system, comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the above-described method for automatically generating and annotating boundary points.

[0020] The advantages of this invention over the prior art are: This invention obtains the set of polylines for the land parcel to be processed, the set of existing boundary point locations, and the boundary point number annotation configuration parameters. It then performs unified extraction and spatial distance comparison of candidate boundary points, generating new boundary point objects only at locations that do not overlap with existing boundary points. This avoids duplicate creation of the same physical boundary marker, reduces manual point-by-point verification and correction, and improves the batch and standardization of boundary point generation.

[0021] This invention, when extracting candidate boundary points, combines the sequence of polyline nodes on the land parcel, the interior angle features of consecutive nodes, and the spatial overlap of the main building corner points to identify important turning points with boundary significance. Compared to simply extracting all polyline nodes, this invention can reduce invalid boundary points caused by line segment densification and drawing errors, while retaining important control points related to land parcel turning points and building corner points, making the boundary point set more accurate.

[0022] This invention extracts the sequence of boundary nodes of a target boundary point object within a polyline of a land parcel and generates a corresponding sequence number. For physical boundary points shared by multiple land parcels, the sequence numbers within each parcel are recorded and combined into a concatenated structure. This ensures the uniqueness of the same physical boundary point object while clearly expressing its numbering relationship within the boundaries of different land parcels, avoiding confusion, omissions, or overlaps in the numbering of shared boundary points.

[0023] This invention, when generating boundary point annotations, utilizes the spatial azimuth angle formed by the target boundary point object and its adjacent boundary point objects, combined with annotation configuration parameters, to automatically determine the annotation position through a coordinate translation operator. This method enables the annotation's outward offset direction to adapt to the local angular trend of the land parcel boundary, reducing issues such as annotations overlapping lines, crossing boundaries, and inconsistent directions, thereby improving the clarity and consistency of land parcel point annotations.

[0024] This invention addresses area discrepancies caused by coordinate rounding using an area conservation constraint optimization model. The system locks rigid nodes in place, making only minor adjustments to permissible flexible nodes, and optimizes the process using the legally owned area as a constraint and minimizing the total displacement of flexible nodes as the objective. This reduces the risk of discrepancies between the calculated area and the legally owned area while meeting coordinate accuracy requirements, thus improving the compliance of cadastral data entry and review.

[0025] This invention addresses the annotation congestion problem when multiple land parcels share the same boundary point by employing a topological sector spatial allocation model. Centered on the shared boundary point, multiple internal sector divisions are created based on the intersecting boundary segments, and the boundary point annotations of different land parcels are constrained to be arranged within their respective sectors. This method avoids cross-boundary overlap and mutual compression of annotations, ensuring that multiple annotations at the shared boundary point are displayed separately and do not cross boundaries, thus improving the readability of complex land parcel maps and the rigor of ownership representation. Attached Figure Description

[0026] Figure 1 is a schematic diagram of the overall process of the method of the present invention, which macroscopically shows the complete execution process from obtaining the coordinates of the polyline to generating the set of boundary points, and finally outputting the annotation position by calculating the spatial azimuth.

[0027] Figure 2 is a physical image of the extraction of the corner points of the land parcel and the main house in this invention, showing the geometric operation process of extracting multiple line nodes from the upper left corner in a clockwise direction and generating candidate boundary points from the corner points of the main house.

[0028] Figure 3 is a screenshot of the inflection point determination feature of the present invention. It magnifies a local segment of the polyline and shows how to calculate the interior angle feature by the included angle of three adjacent nodes and input it into the template library to identify it as an important inflection point.

[0029] Figure 4 This is a physical diagram of the globally shared boundary points and splicing structure of the present invention, showing the situation where multiple land parcel polygons intersect at a common point, and how to use multiple land parcel identifiers to generate a sequential number splicing structure with separators.

[0030] Figure 5 This is a topological sector spatial allocation diagram of the present invention, which shows the macroscopic distribution of dividing a two-dimensional space into multiple interior sector sectors by radial boundary segments at a global shared point, and constraining the text bounding boxes of the corresponding parcels within the corresponding sectors.

[0031] Figure 6 is a hardware structure diagram of the automatic boundary point generation and annotation system of the present invention. The actual operating state of the system is simulated by a display and a host chassis, and the processor and memory communication and coordination mechanism that supports the execution of the above algorithm is clearly marked. Detailed Implementation

[0032] The embodiments of the present invention will be described below with reference to the accompanying drawings. The following embodiments are intended to enable those skilled in the art to implement the present invention. Equivalent substitutions can be made according to cadastral surveying specifications, coordinate system types, database formats, and map scales without departing from the technical concept of the present invention.

[0033] In one embodiment, the present invention provides a method for automatically generating boundary points and annotating boundary point numbers. This method can run on cadastral mapping software, GIS platforms, CAD secondary development plugins, parcel map batch production systems, or real estate registration result quality inspection systems. The set of polylines to be processed for a parcel can come from parcel boundary layers in DWG, DXF, SHP, GeoJSON, GDB, or a cadastral database. Each polyline to be processed represents a closed boundary line of a parcel, or it can represent a boundary line that can be closed into a surface after preprocessing. The set of existing boundary point locations can come from historical boundary point layers, database boundary point tables, or manually collected boundary point results. Boundary point number annotation configuration parameters are used to control parameters such as font size, font width, font height, rotation angle, minimum offset distance, maximum offset distance, text bounding box size, connection style between annotation and boundary point, and annotation avoidance level.

[0034] In one embodiment, the minimum offset distance can be set to 0.5mm to 2mm on the map, and the maximum offset distance can be set to 2mm to 10mm on the map. If the system calculates using real-world coordinate units, the distance on the map is converted to the real-world distance according to the map scale. For example, at a scale of 1:500, 1mm on the map corresponds to 0.5m on the real ground. The text scaling parameter can be set between 0.8 and 1.2; when map space is very limited, the scaling ratio can be allowed to drop to 0.6, but generally not lower than 0.6, to avoid affecting the readability of boundary point numbers. The distance tolerance for spatial overlap comparison can be set to 0.001m to 0.05m, preferably 0.01m in general urban cadastral scenarios; in rural large-scale mapping or scenarios with low historical accuracy, it can be relaxed to 0.02m to 0.05m. By setting this distance tolerance, the duplicate generation of the same physical boundary point can be avoided due to floating-point calculation errors, different coordinate decimal places, or slight offsets during graphic editing.

[0035] like Figure 1 As shown, the system first acquires the set of polylines for the parcel to be processed, the set of existing boundary point locations, and the boundary point number annotation configuration parameters. After acquisition, basic checks are performed on the polylines for the parcel to be processed, including determining whether the polylines are closed, whether there are duplicate nodes, whether there are zero-length edges, whether there are self-intersections, and whether the coordinate systems are consistent. For polylines where the distance between the first and last nodes is less than the closure tolerance, the first and last nodes can be unified to the same coordinate system; for duplicate nodes, the first occurrence of the node can be retained and subsequent duplicate nodes can be deleted; for zero-length edges, they can be deleted without changing the topological meaning of the parcel boundary. The purpose of this processing is to ensure that subsequent boundary point extraction, sequential numbering, and annotation offset are all based on a stable boundary sequence.

[0036] Subsequently, the system extracts the polyline coordinate nodes from the polyline set of the parcels to be processed and outputs a candidate boundary point set. These candidate boundary points are not the final boundary points to be entered into the database, but rather a set of locations where boundary point objects may need to be generated or reused. Each candidate record in the candidate boundary point set can contain attributes such as candidate location coordinates, source parcel identifier, source polyline node sequence number, node front and rear edge directions, node type, and whether it was obtained by merging main building corner points. The purpose of retaining these source attributes is that subsequent sorting, shared identification, area conservation optimization, and annotation layout all require knowing which parcel the point originates from and its location on the parcel boundary.

[0037] After the candidate boundary point set is generated, the system compares the coordinates of the candidate locations with the coordinates of existing locations in the existing boundary point location set. To improve comparison efficiency, an R-tree, KD-tree, grid index, or quadtree index can be built based on the existing location coordinates. For each candidate location coordinate, the system queries the index for existing location coordinates whose distance is not greater than the overlap tolerance. If a query result exists, it is determined that the candidate location already has a corresponding boundary point, and a new boundary point object is not generated at that location. At the same time, a reuse relationship is established between the candidate location and the existing boundary point object. If no query result exists, a new boundary point object is generated at the candidate location. This ensures that historical boundary point results are inherited and used, and also avoids multiple overlapping point numbers of the same physical boundary marker on the map.

[0038] To ensure that existing boundary points are not overlooked during subsequent sorting and annotation processes, the system can combine newly generated boundary point objects with reused existing boundary point objects to form a set of boundary point objects to be processed. In other words, the elimination in spatial distance comparison refers to eliminating duplicate generation actions, not excluding existing boundary points from the numbering and annotation process. After this processing, both new and historical points on the parcel boundary can participate in sequence number generation, sharing relationship determination, and boundary point number annotation generation, avoiding the problem of interrupted sequence at a particular parcel boundary.

[0039] When generating boundary point objects, these objects can include fields such as coordinates, a unique object identifier, a list of parcel identifiers, a list of boundary node numbers, boundary marker type, whether shared, whether movable, annotation object identifier, creation time, and data source. The unique object identifier can be automatically generated by the system or uniformly assigned by the cadastral database during the results entry stage. If a candidate location spatially overlaps with multiple polylines of parcels to be processed, the system can generate a single physical boundary point object or reuse an existing physical boundary point object, recording the multiple parcel affiliations within that object. This design aims to distinguish a single boundary marker in physical space from the numbering relationships within multiple parcels, avoiding point object duplication while preserving the independent boundary point order within each parcel.

[0040] For each target boundary point object in the set of boundary point objects to be processed, the system extracts the boundary node order of the target boundary point object in the corresponding polyline of the parcel to be processed, and performs a sorting operation to generate a sequence number. Specifically, the system uses the standardized vertex sequence of the parcel polyline as a basis to find the matching node position of the target boundary point object's coordinates in the standardized vertex sequence. If the target boundary point object is obtained by merging the main house corner points and is not exactly equal to the original polyline node, the main house corner point can be projected onto the corresponding parcel boundary line, and the insertion order is determined according to the starting point number and the distance along the edge of the edge segment where the projection position is located. A closed circular sequence is used for sorting, which increases sequentially in a clockwise direction starting from the starting node. The sequence number can be in the form of natural numbers 1, 2, 3, etc., or it can be converted into display text such as J1, J2, J3, etc. according to local cartographic standards.

[0041] After generating the sequence number, the system continues to generate boundary point number annotations for each target boundary point object. To do this, the system extracts the preceding and following adjacent boundary point objects of the target boundary point object within the corresponding polyline of the parcel to be processed. For closed parcel polylines, if the target boundary point object is the first boundary point in the standardized vertex sequence, then the preceding adjacent boundary point object is the last boundary point object in the sequence; if the target boundary point object is the last boundary point object in the sequence, then the following adjacent boundary point object is the first boundary point object. This circular adjacency relationship processing ensures that annotation breaks do not occur at the start and end positions of the parcel boundaries.

[0042] The system calculates the spatial azimuth angle based on the preceding adjacent boundary point object, the target boundary point object, and the following adjacent boundary point object. The spatial azimuth angle can be understood as the local directional characteristic formed by the two boundary line segments near the target boundary point. Specifically, it can be determined by the direction from the preceding boundary point to the target boundary point, the direction from the target boundary point to the following boundary point, the size of the local interior angle, and the clockwise direction of the polyline of the parcel. For convex corner nodes, the annotation is preferably offset towards the direction bisecting the outer corner of the parcel; for concave corner nodes, the annotation is preferably offset away from the dense boundary line area; for approximately straight line nodes, the annotation can be offset along the boundary normal direction. Fixed up, down, left, and right offsets are not used here; instead, the offset direction is automatically determined based on the local spatial azimuth angle. This is because the shape of parcel boundaries is often irregular, and fixed offsets can easily cause annotations to overlap the line, cross boundaries, or conflict with annotations of adjacent parcels.

[0043] Subsequently, the system inputs the spatial azimuth and boundary point number annotation configuration parameters into the coordinate translation operator, outputting the annotation position of the target boundary point object. The coordinate translation operator can be implemented using vector offset. In specific implementation, the offset direction vector is first determined based on the spatial azimuth, then the offset length is determined based on the outward offset distance in the configuration parameters. Next, the system moves along the offset direction with the coordinates of the target boundary point object as the base point to obtain the candidate annotation position. The system can calculate the text bounding box based on the text font size, font width, font height, and scale, and detect whether the text bounding box conflicts with the land parcel boundary line, other boundary points, or existing annotations. If a conflict exists, the offset distance can be increased incrementally within the range of minimum to maximum outward offset distance, or the offset direction can be slightly rotated in angular steps of 5° to 15° near the spatial azimuth until a position that meets the avoidance conditions is found. If complete avoidance is still not possible, the position with the lowest conflict cost can be retained, and this annotation can be marked as requiring manual review. After generating the annotation, the system establishes a relationship between the annotation object and the target boundary point object, so that the annotation can be updated synchronously when the boundary point is moved, the number is updated, or the land parcel boundary is modified.

[0044] In a further embodiment, to improve the accuracy of candidate boundary point extraction, the system obtains a set of main building corner points and merges important turning points of the land parcel boundary with the main building corner points to generate a candidate boundary point set, such as... Figure 2 As shown. The set of main building corner points can come from building outline layers, building survey results, or manual plotting results. Main building corner points are included as candidate boundary points because in some cadastral transactions, main building corner points may coincide with or nearly coincide with the land parcel boundary, and this location often has practical boundary identification significance. Relying solely on land parcel polyline nodes may miss boundary control points related to building corner points.

[0045] For each polyline in the set of polylines to be processed, the system extracts the top-left vertex of the polyline as the starting node. The top-left vertex can be determined by coordinate sorting; for example, in a plane coordinate system, first select several vertices with the largest ordinates, and then select the vertex with the smallest abscissa from among them. If the coordinate system orientation differs from the map display orientation, it can be converted according to the current GIS or CAD coordinate system. The purpose of selecting the top-left vertex as the starting node is to unify the differences in the starting points when different cartographers draw polylines, so that the boundary of the same parcel can generate a consistent order of boundary points in different batches of processing.

[0046] After the starting node is determined, the system determines the direction of the polyline node sequence. If the current node sequence is counterclockwise, the sequence is reversed; if it is already clockwise, the original order is maintained. Direction determination can be achieved through the directed area of ​​the polygon, the boundary loop direction function, or the loop direction detection interface in the GIS geometry library. After unifying the starting point and direction, the system extracts all polyline nodes sequentially in a clockwise direction to construct a standardized vertex sequence. This standardized vertex sequence is the basis for subsequent inflection point determination, sequence number generation, and adjacent boundary point extraction.

[0047] The system sequentially selects three consecutive polyline nodes from the standardized vertex sequence and extracts the interior angle features formed by these three nodes, such as... Figure 3 As shown. Interior angle features can include the included angle at the intermediate node, turning direction, front-to-back side length ratio, front-to-back side direction difference, and deviation distance from the node to the extension of the adjacent line segment. The included angle is used to determine if the node has a significant turning point, the turning direction is used to distinguish between convex and concave angles, and the front-to-back side length ratio is used to avoid misclassifying very short burrs as valid boundary points. The inflection point determination template library can pre-set various templates, such as right-angle turning templates, acute-angle turning templates, obtuse-angle turning templates, polygonal concave templates, and building edge templates. The angle matching tolerance can be set to 3° to 15°, preferably 5° to 10°; the side length ratio tolerance can be set to 0.1 to 10, and excessively short side segments can be filtered using a minimum side length threshold, which can be set to 0.02m to 0.2m.

[0048] If the inflection point determination template library outputs a successful match flag, the system will identify the middle polyline node among three consecutive polyline nodes as an important inflection point. The purpose of important inflection point determination is to distinguish between nodes that truly represent the turning points of land parcel boundaries and auxiliary nodes generated during the mapping process due to line segment densification, curve fitting, or historical editing. This method can reduce the number of invalid boundary points, making boundary point number annotations more concise and standardized.

[0049] When a main building corner point spatially overlaps with a polyline of the land parcel to be processed, the system incorporates this main building corner point into the candidate boundary point set. Spatial overlap can include the main building corner point's vertical distance to the land parcel boundary line not exceeding the main building corner point's adsorption tolerance, or the main building corner point's distance to a polyline node of the land parcel not exceeding the same-point tolerance. The main building corner point's adsorption tolerance can be set to 0.01m to 0.1m, preferably 0.03m. If the main building corner point falls inside a certain edge segment, the system can split that edge segment into two segments and insert the corresponding node position of the main building corner point into the standardized vertex sequence. This allows building corner points with actual boundary significance to be included in the subsequent numbering and annotation process without disrupting the land parcel boundary topology.

[0050] In a further embodiment, after the target boundary point object generates a sequence number, the system obtains the identification information of the polyline of the parcel to be processed to which the target boundary point object belongs. The identification information can be a parcel code, a unified parcel number, a plot number, a patch number, or a polygon number within the system. The system combines the identification information with the sequence number of the target boundary point object within the corresponding parcel polyline using a connector to generate a parcel sequence number unit. For example, if the parcel identifier is ZD001 and the sequence number is 5, the parcel sequence number unit can be generated as ZD001-5. The connector can be configured as -, _, or / , preferably using a connector that does not conflict with the characters of the parcel code itself.

[0051] When the system determines that a target boundary point object belongs to multiple polylines of different pending land parcels, it indicates that the target boundary point object is a physical boundary point shared by multiple land parcels. In this case, the system generates multiple land parcel sequence number units for this physical boundary point object and concatenates these units using a separator to output a spliced ​​structure, such as... Figure 4 As shown. For example, if the same physical point belongs to both ZD001 and ZD002, and its sequence numbers within the two parcels are 5 and 1 respectively, then the concatenation structure can be ZD001-5|ZD002-1. The separator can be configured as |, ; or . Through this concatenation structure, the system retains the uniqueness of the same physical boundary point while clearly expressing the different sequence relationships of the point within different parcels, avoiding the overwriting or confusion of cross-parcel numbering relationships. Boundary point objects with concatenation structures are marked as globally shared boundary point objects for subsequent entry into the topology sector space allocation process.

[0052] In a further embodiment, after generating the set of boundary point objects and before performing sorting operations to generate sequence numbers, the system can also perform coordinate precision rounding and area conservation constraint optimization. This step is used to solve the problem of inconsistency between the area calculated from the map after coordinate rounding and the legally owned area. Specifically, for each polyline to be processed in the set of polylines to be processed, the system extracts each boundary point object belonging to that polyline and performs precision rounding on the coordinates of each boundary point object, outputting the rounded coordinate sequence corresponding to that polyline. Here, rounding does not only refer to taking integer coordinates, but also to grid rounding according to the target precision unit. For example, when the target precision is 0.01m, the coordinate components are rounded to the centimeter level; when the target precision is 0.001m, the coordinate components are rounded to the millimeter level. Precision rounding can be set to precision levels such as 0.001m, 0.01m, 0.1m, or 1m according to business specifications. In general cadastral surveying and mapping results entering the database, it is preferred to process them according to 0.01m or 0.001m.

[0053] The system acquires the legal ownership area data associated with the polyline of the land parcel. This legal ownership area data can come from the real estate register, cadastral survey forms, historical land parcel attribute tables, ownership survey results database, or manually imported tables. The system inputs the rounded coordinate sequence and the legal ownership area data into the area conservation constraint optimization model. The model then outputs the fine-tuned position coordinates of each coordinate in the rounded coordinate sequence and updates the coordinates of each boundary point object to the corresponding fine-tuned position coordinates.

[0054] The purpose of the area conservation constraint optimization model is not to arbitrarily change the shape of the land parcel, but rather to make minimal adjustments to flexible nodes while maintaining the stability of rigid boundary points and the overall boundary shape, so that the inverse area of ​​the polygon returns to the allowable error range of the legally owned area. Rigid nodes can be understood as nodes that are not suitable for or prohibited from being moved, such as points with physical boundary stakes, boundary nails, or permanent boundary markers, or points shared by multiple land parcels whose movement would affect the consistency of adjacent land parcels. Flexible nodes can be understood as nodes that can be finely adjusted within the allowable error range, such as ordinary vertices generated by graphic rounding, non-physical boundary marker nodes, or auxiliary boundary points that have been approved by quality inspection to absorb errors. This distinction between rigid and flexible nodes avoids destroying the actual boundary marker positions in order to achieve area consistency.

[0055] The area conservation constraint optimization model can employ a classification and regression joint model trained using supervised learning with historical cadastral survey data that meets and does not meet the requirements. The model includes a feature encoding layer, a rigid node decision subnetwork, and a coordinate optimization subnetwork. The feature encoding layer converts coordinate sequences, node attributes, side lengths, included angles, shared quantity, coordinate differences before and after rounding, legally owned area, and inversely calculated area differences into feature vectors that the model can process. The rigid node decision subnetwork determines whether each coordinate corresponds to a rigid or flexible node. The coordinate optimization subnetwork can be implemented using a combination of a regression network and a constraint optimization layer. The regression network predicts the initial fine-tuning direction and magnitude of flexible nodes, while the constraint optimization layer performs sequential quadratic programming optimization calculations based on the initial predictions. This approach allows the coordinate optimization subnetwork to both learn from historical samples and satisfy the legally owned area constraint in the actual output.

[0056] In one embodiment, the rigid node determination subnetwork can be implemented using a multilayer perceptron, gradient boosting tree, random forest, or lightweight graph neural network. If the sample attributes are mainly structured fields, such as boundary marker type attributes and the number of polylines shared by the parcel to be processed, then a multilayer perceptron or gradient boosting tree can achieve high efficiency. The boundary marker type attribute can include boundary stakes, boundary nails, wall corners, building corners, intersections of road edges, intersections of ditch edges, and ordinary vertices. The number of shared polylines indicates the number of parcel polylines to which the sample boundary point object simultaneously belongs, such as 1, 2, 3, or more. Generally, the larger the number of shared polylines, the more significant the impact of node movement on surrounding parcels, and therefore the higher the probability of being determined as a rigid node.

[0057] The training process of the rigid node decision sub-network can be implemented as follows: The system first acquires the sample attribute data of the sample boundary point objects and the corresponding real node classification labels. Real node classification labels can be obtained from historical qualified results, manual review results, quality inspection rules, or annotations by surveying experts; these labels include rigid node labels and flexible node labels. The system inputs the sample attribute data into the untrained rigid node decision sub-network to obtain predicted node classification labels. Then, the cross-entropy loss value between the predicted node classification labels and the real node classification labels is calculated, and the network parameters are updated based on the cross-entropy loss value. Parameter updates can employ existing optimization algorithms such as stochastic gradient descent, Adam, or RMSProp. Training continues until the cross-entropy loss value converges. The convergence condition can be set to a loss decrease of less than 0.001 within 5 to 20 consecutive training epochs, or the validation set classification accuracy stabilizes within a preset range. After training, the rigid node decision sub-network can be used for node classification of new land parcel data.

[0058] The training of the coordinate optimization sub-network can be completed using both qualified and unqualified historical cadastral survey data. For qualified historical data, the fine-tuned coordinates that have passed manual review or database acceptance can be used as the regression learning target. For unqualified historical data, states such as area exceeding limits, erroneous rigid node movement, and excessive displacement of flexible nodes can be used as back samples, allowing the model to learn which adjustment methods should not be output. During training, the coordinate optimization sub-network receives rounded coordinate sequences with node classification labels and legal ownership area data, outputs predicted fine-tuned coordinates, and updates network parameters based on factors such as the difference between the predicted fine-tuned coordinates and historical qualified coordinates, the difference between the predicted back-calculated area and the legal ownership area, and whether rigid nodes have been erroneously moved. Through this training method, the model can learn from the experience of area correction in historical results, while avoiding the solidification of erroneous adjustment methods from unqualified results into the processing of new data.

[0059] In actual reasoning, the coordinate optimization subnetwork receives rounded coordinate sequences with attached node classification labels and legal ownership area data. For coordinates with rigid node labels, the coordinate optimization subnetwork locks them, keeping their positions fixed; for coordinates with flexible node labels, the coordinate optimization subnetwork allows them to move within a small range. The upper limit of single-point displacement of flexible nodes can be set to 0.001m to 0.05m, preferably not exceeding 0.5 times the rounding accuracy; when the accuracy of historical data is low, it can be relaxed to 0.1m, but a verification mark needs to be output. The coordinate optimization subnetwork uses the legal ownership area data as a constraint and performs sequential quadratic programming optimization calculations with the goal of minimizing the total displacement of flexible nodes. Sequential quadratic programming can be implemented by calling existing numerical optimization libraries, such as optimization modules in GIS backends, scientific computing libraries in Python, numerical optimization libraries in C++, or built-in optimizers in commercial surveying software. During the optimization process, the system adjusts the coordinates of flexible nodes round by round, calculates the difference between the land parcel area composed of the fine-tuned coordinate sequence and the legal ownership area, and minimizes coordinate movement while satisfying area constraints. The convergence condition for optimization can be set to an area difference of no more than 0.001 square meters to 0.05 square meters, or a target change of less than 0.0001 in 10 to 50 consecutive iterations. If all movable nodes reach their displacement limits but still cannot meet the area constraint, the system will not forcibly move the rigid nodes, but will output the current optimal fine-tuning position coordinates and generate a verification mark to ensure the traceability of the result processing.

[0060] In a further embodiment, for a target boundary point object marked as a globally shared boundary point object, the annotation location output no longer relies solely on the adjacent points of a single parcel, but is processed using a topological sector spatial allocation model. The system first extracts all boundary segments intersecting at the globally shared boundary point object. Boundary segments can include adjacent edges from polylines of different parcels, or edge segments formed at that point by inner or outer loops within the same parcel. The system then inputs the coordinates of all boundary segments and the globally shared boundary point object into the topological sector spatial allocation model.

[0061] like Figure 5 As shown, the topological sector spatial allocation model uses the coordinates of the globally shared boundary point object as the center and divides the two-dimensional plane space into multiple interior corner sectors enclosed by adjacent boundary segments using all intersecting boundary line segments. These interior corner sectors can be understood as the usable map space formed between two adjacent boundary directions radiating outward from the shared boundary point. For nodes where multiple parcels intersect, if the ordinary point number annotation method is still used, it is easy for the boundary point numbers of multiple parcels to overlap in the same position. The topological sector spatial allocation model, through sector division, restricts the annotations of different parcels to their respective spatial ranges, fundamentally reducing cross-parcel overlap.

[0062] The system obtains the spatial hierarchy mapping relationship between each unprocessed parcel polyline associated with the globally shared boundary point object and each interior corner sector. This spatial hierarchy mapping relationship can be determined by the coverage relationship between the internal regions of the parcel polygons and the sectors. Specifically, the system can select a test point near the shared boundary point within each interior corner sector, determine which parcel polygon the test point falls into, and thus establish the correspondence between the parcel and the sector. For scenarios with overlapping boundaries or narrow sectors, the system can combine the direction of the boundary line segments, the direction of the parcel ring, and the topological relationship of adjacent polygons for joint determination. For each target unprocessed parcel polyline, the topological sector spatial allocation model constrains the text bounding boxes of the boundary point annotations belonging to that parcel within the corresponding interior corner sector for arrangement calculations, and outputs the annotation positions independent of other parcels.

[0063] In the layout calculation, the system extracts the two edge rays and the angle bisector of the interior corner sector. The two edge rays point from the shared boundary point coordinates to the directions of adjacent boundary segments, and the angle bisector represents the direction of relative centering within the sector. The system first places the center point of the text bounding box on the angle bisector, so that the annotation is naturally located in the center of the corresponding sector of the land parcel. Then, the system calculates the perpendicular distance from each vertex of the text bounding box to the two edge rays within the interior corner sector. To avoid considering only one vertex and causing other vertices to exceed the boundary, the system can use the minimum perpendicular distance from each vertex to the two edge rays as the safe distance of the current text bounding box. The larger this safe distance, the farther the annotation is from the sector boundary, and the less likely it is to overlap the boundary of adjacent land parcels or fall into other land parcel sectors.

[0064] To obtain optimal annotation locations, the system performs particle swarm optimization iterative calculations. The particle position can represent the position vector of the text bounding box center point on the angle bisector and the text bounding box scaling parameter. The position vector can be restricted to a map distance of 0.5mm to 10mm outside the shared boundary point, and the scaling parameter can be restricted to a range of 0.6 to 1.2. Particle swarm optimization uses maximizing the safe distance as the fitness objective, while simultaneously checking whether the text bounding box is completely within the interior corner sector. If any vertex of the text bounding box crosses the edge ray, the particle's fitness is reduced or it is directly determined as infeasible. The particle swarm size can be set to 10 to 50 particles, and the number of iterations can be set to 20 to 100; when the optimal fitness change is less than 0.001 after 5 to 15 consecutive iterations, the optimization is considered converged. The system outputs the position vector at which the maximum fitness value is obtained as the annotation location. In this way, multiple annotations at the same shared boundary point can be radially distributed and each constrained within the sector corresponding to its respective parcel.

[0065] In a further embodiment, before inputting the coordinates of all boundary segments and globally shared boundary point objects into the topological sector spatial allocation model, the system can also extract a polygon topology tree composed of polylines of the parcels to be processed, containing the globally shared boundary point objects. The polygon topology tree is used to represent outer rings, inner rings, adjacent parcels, containment relationships, and hole relationships. During construction, the system can convert each parcel polyline into a polygon object and establish parent-child node relationships based on spatial containment, intersection, adjacency, and shared edge relationships. The parcel boundary located on the outer layer can be used as an outer ring node, and the boundary located inside it, representing deducted areas, courtyards, public passages, or other hole-like areas, can be used as an inner ring node.

[0066] Using a polygon topology tree, the system identifies inner and outer ring boundary segments among all boundary segments. Since the outer and inner rings are typically geometrically opposite, directly mixing these two types of segments in the sector partitioning model could lead to a reverse interpretation of sector affiliation relationships. To avoid this, the system flips the direction vectors of the inner ring boundary segments, ensuring topological consistency between them. After achieving this consistency, the system outputs the coordinates of the outer ring boundary segments, the adjusted boundary segments, and the globally shared boundary point objects as input variables for the topological sector spatial allocation model. This ensures that sector partitioning is based not only on geometric angles but also on the true topological orientation of the land parcel.

[0067] In one embodiment, the present invention also provides an automatic boundary point generation and boundary point number annotation system, such as... Figure 6 As shown. The system includes at least one processor and a memory communicatively connected to the at least one processor. The memory can be a hard disk, solid-state drive, RAM, network storage, or a storage unit in a database server. The memory stores a computer program that can be executed by the processor. When the processor executes the computer program, it can complete the above-mentioned method for automatically generating and annotating boundary points.

[0068] The system can include a data acquisition module, a candidate point extraction module, a spatial overlap elimination module, a boundary point object generation module, a sequence number generation module, an area conservation optimization module, a shared point identification module, a topological sector layout module, an annotation generation module, and a results output module. The data acquisition module reads the polyline data of the land parcel, the locations of existing boundary points, the corner points of main buildings, the legal ownership area data, and annotation configuration parameters. The candidate point extraction module extracts standardized vertex sequences, identifies important inflection points, and merges the corner points of main buildings. The spatial overlap elimination module eliminates duplicate candidate points using spatial indexing and distance tolerance, and establishes a reuse relationship between candidate locations and existing boundary point objects. The boundary point object generation module creates new boundary point objects and combines these new boundary point objects with reused existing boundary point objects to form a set of boundary point objects to be processed. The sequence number generation module generates point numbers based on the order of the land parcel boundary nodes. The area conservation optimization module performs rigid node determination and flexible node fine-tuning after coordinate rounding. The shared point identification module identifies shared boundary points across land parcels and generates a spliced ​​structure. The topology sector layout module is used to sectorize and arrange multiple annotations at shared boundary points. The annotation generation module is used to generate boundary point number annotations based on spatial azimuth, coordinate translation operators, and sector optimization results. The output module is used to output the boundary point layer, annotation layer, quality inspection log, parcel sequence number table, and shared point relationship table.

[0069] In actual operation, the system can first batch read all polylines of parcels within a cadastral survey area, automatically completing candidate boundary point extraction, duplicate location removal, new boundary point generation, reuse of existing boundary points, and sequence number assignment. Subsequently, the system performs coordinate rounding and fine-tuning on parcels requiring area conservation, and writes the adjusted coordinates back to the boundary point object. Next, the system identifies shared boundary points across parcels, generating annotations for ordinary boundary points using a coordinate translation operator based on local spatial azimuth, and for globally shared boundary points using a topological sector spatial allocation model. Finally, the system outputs boundary points and boundary point number annotations that can be directly used for parcel map compilation, real estate registration, or manual verification. Through the above process, this invention can simultaneously solve problems such as automatic boundary point extraction, shared numbering across parcels, area conservation after coordinate rounding, and annotation overlay of densely shared points.

[0070] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for automatically generating boundary points and annotating boundary point numbers, characterized in that, include: Obtain the set of polylines for the land parcel to be processed, the set of existing boundary point locations, and the boundary point number annotation configuration parameters; Extract the polyline coordinate nodes from the polyline set of the land parcel to be processed, and output a candidate boundary point set; compare the spatial distance between the candidate position coordinates in the candidate boundary point set and the existing position coordinates in the existing boundary point position set, eliminate the candidate position coordinates that overlap, and generate a set of boundary point objects at the remaining candidate position coordinates. For each target boundary point object in the set of boundary point objects, extract the boundary node order of the target boundary point object in the corresponding polyline of the parcel to be processed, and perform a sorting operation to generate the sequence number of the target boundary point object; Extract the preceding and following adjacent boundary point objects of the target boundary point object in the corresponding polyline of the parcel to be processed. Calculate the spatial azimuth angle formed by the preceding adjacent boundary point object, the target boundary point object, and the following adjacent boundary point object. Input the spatial azimuth angle and the boundary point number annotation configuration parameters into the coordinate translation operator, output the annotation position of the target boundary point object, and generate a boundary point number annotation associated with the target boundary point object at the annotation position. The step of extracting polyline coordinate nodes from the polyline set of the land parcel to be processed and outputting a candidate boundary point set includes: obtaining the set of main building corner points; For each polyline to be processed in the set of polylines to be processed, the top left vertex of the polyline to be processed is extracted as the starting node, and all polyline nodes of the polyline to be processed are extracted in a clockwise direction to construct a standardized vertex sequence. Extract the interior angle features formed by three consecutive multi-segment line nodes in the standardized vertex sequence, and input the interior angle features into the inflection point determination template library; If the inflection point determination template library outputs a matching success flag, then the middle multi-segment line node among the three consecutive multi-segment line nodes is identified as an important inflection point. Merge all the important inflection points and the main house corner points in the set of main house corner points that have a spatial overlap with the corresponding polyline of the parcel to be processed, and output them as the candidate boundary point set. After generating the sequence number of the target boundary point object by performing sorting operations, the method further includes: obtaining the identification information of the polyline of the land parcel to be processed to which the target boundary point object belongs; The identification information and the sequence number of the target boundary point object within the corresponding polyline of the land parcel to be processed are combined using a connector to generate a land parcel sequence number unit. When it is determined that the target boundary point object belongs to multiple different land parcel polylines to be processed, the multiple land parcel sequence number units are concatenated with a separator to output a spliced ​​structure, and the boundary point object with the spliced ​​structure is marked as a globally shared boundary point object; For the target boundary point object that is marked as a global shared boundary point object, the step of outputting the annotation position of the target boundary point object includes: extracting all boundary line segments that intersect the global shared boundary point object; Input the coordinates of all the boundary segments and the globally shared boundary point objects into the topology sector space allocation model; The topological sector space allocation model uses the coordinates of the globally shared boundary point object as the center and uses all the boundary line segments to divide the two-dimensional plane space into multiple interior corner sectors surrounded by adjacent boundary line segments. Obtain the spatial dependency mapping relationship between each unprocessed land parcel polyline associated with the globally shared boundary point object and each of the interior corner sectors. For each target unprocessed land parcel polyline, the topological sector spatial allocation model constrains the text bounding box of the boundary point number annotation belonging to the target unprocessed land parcel polyline within the corresponding interior corner sector for arrangement calculation, and outputs the annotation position independent of other unprocessed land parcel polylines.

2. The method according to claim 1, characterized in that, After generating a set of boundary point objects at the remaining candidate location coordinates, and before performing sorting operations to generate the sequence number of the target boundary point objects, the method further includes: for each polyline to be processed in the set of polylines to be processed, extracting each boundary point object belonging to the polyline to be processed, performing precision rounding on the coordinates of each boundary point object, and outputting the rounded coordinate sequence corresponding to the polyline to be processed; Obtain the legal ownership area data of the land parcel to be processed, which is associated with multiple segments. The rounded coordinate sequence and the legal ownership area data are input into the area conservation constraint optimization model, and the fine-tuning position coordinates of each coordinate in the rounded coordinate sequence are output through the area conservation constraint optimization model. Update the coordinates of each boundary point object to the corresponding fine-tuning position coordinates.

3. The method according to claim 2, characterized in that, The area conservation constraint optimization model is a classification and regression joint model obtained by supervised learning training using qualified historical cadastral survey data and unqualified historical cadastral survey data. The area conservation constraint optimization model includes a rigid node decision subnetwork and a coordinate optimization subnetwork. The step of inputting the rounded coordinate sequence and the legal ownership area data into the area conservation constraint optimization model, and outputting the fine-tuning position coordinates of each coordinate in the rounded coordinate sequence through the area conservation constraint optimization model, includes inputting the rounded coordinate sequence into the rigid node decision sub-network, and the rigid node decision sub-network outputting the node classification label corresponding to each coordinate in the rounded coordinate sequence, wherein the node classification label includes rigid node label and flexible node label; The rounded coordinate sequence with the node classification label and the legal ownership area data are input into the coordinate optimization subnetwork; The coordinate optimization subnetwork locks the coordinates with the rigid node labels in the rounded coordinate sequence to keep their positions fixed. Using the legal ownership area data as a constraint, it performs a sequential quadratic programming optimization calculation with the objective function of minimizing the total displacement of the coordinates with the flexible node labels in the rounded coordinate sequence, and outputs the fine-tuned position coordinates.

4. The method according to claim 3, characterized in that, The training process of the rigid node determination subnetwork includes: acquiring sample attribute data of sample boundary point objects and the real node classification labels corresponding to the sample boundary point objects. The sample attribute data includes boundary type attributes and the number of shared segments of the parcel to be processed. The sample attribute data is input into an untrained rigid node decision subnetwork, which outputs the predicted node classification label. Calculate the cross-entropy loss value between the predicted node classification label and the real node classification label, and update the network parameters of the untrained rigid node decision subnetwork based on the cross-entropy loss value. Continue to perform parameter update operations until the cross-entropy loss value converges, and output the trained rigid node decision subnetwork.

5. The method according to claim 1, characterized in that, The topological sector spatial allocation model constrains the text bounding boxes of the boundary point number annotations of the target land parcel polyline to be processed within the corresponding interior corner sector for arrangement calculation, and outputs the annotation position independent of other land parcel polylines to be processed, including: extracting the two edge rays and the angle bisector of the interior corner sector. Place the center point of the text bounding box on the angle bisector; Within the inner corner sector, calculate the vertical distance from the vertex of the text bounding box to the two edge rays; Perform particle swarm optimization iterative calculation, using the maximum vertical distance as the optimization fitness function, and change the position vector of the center point on the angle bisector and the scaling parameter of the text bounding box until the optimization fitness function converges. The position vector at which the maximum fitness function value is obtained is output as the annotation position.

6. The method according to claim 1, characterized in that, Before inputting the coordinates of all boundary segments and the globally shared boundary point objects into the topological sector spatial allocation model, the method further includes: extracting a polygon topology tree composed of polylines of the parcel to be processed containing the globally shared boundary point objects; The inner and outer ring boundary segments among all boundary segments are identified using the polygon topology tree. The direction vector of the inner ring boundary line segment is flipped to obtain an adjusted boundary line segment with a unified direction; The coordinates of the outer ring boundary segment, the adjusted boundary segment, and the globally shared boundary point object are combined and output as the input variables of the topological sector spatial allocation model.

7. A system for automatically generating boundary points and annotating boundary point numbers, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the automatic generation and annotation method of boundary points as described in any one of claims 1 to 6.

Images