Building function recognition method and device based on POI outlier spatial action range

Abstract

This invention discloses a method and apparatus for identifying building functions based on the spatial influence range of Points of Interest (POIs). The method includes: acquiring POI data for a target area; comparing the distribution characteristics of POI point sets at different spatial neighborhood scales with the theoretical random distribution characteristics; adaptively determining the abrupt change scale of clustering features and extracting a set of outliers; constructing a distance anomaly judgment benchmark based on these distribution characteristics; filtering out scale-related outlier categories with spatial expansion attributes; determining multi-dimensional spatial control points based on an anisotropic neighborhood search strategy; constructing a spatial network topology using the envelope structure formed by the control points as constraints; and extracting irregular polygons containing hollow structures as the initial spatial influence range through specific topology pruning; and constructing building function association features based on this range and inputting them into a classification model for identification. This invention achieves accurate characterization of irregular land occupation and hollow structures of large geographic entities, significantly improving the accuracy of building function identification.

Description

Technical Field

[0001] This invention relates to the field of geographic information systems and spatial data processing technology, and in particular to a method and apparatus for identifying building functions based on the spatial range of POI outliers. Background Technology

[0002] With the advancement of urban digital transformation, building function identification plays a crucial role in urban planning, commercial site selection, and traffic simulation. Points of Interest (POI) data, as an important data source representing the spatial distribution and semantic attributes of geographic entities, are widely used in building function identification.

[0003] Existing POI-based building function identification methods typically employ fixed-distance buffers, convex hulls, or simple spatial clustering algorithms to characterize the spatial extent of POIs. However, these existing techniques have the following limitations:

[0004] Unable to distinguish between “remote and isolated POIs” and “spatially expanded large-scale POIs”: Existing methods often use fixed thresholds when detecting outliers, which cannot adapt to different spatial distribution densities. Furthermore, they do not combine the spatial proximity relationship between POIs and road networks, which leads to scattered small shops in remote areas and large commercial complexes in city centers being misclassified as the same type of outlier.

[0005] Inaccurate spatial characterization: Large geographic entities (such as commercial complexes, campuses, etc.) often contain plazas or open spaces. Existing convex hull algorithms or Delaunay triangulation clipping methods based on fixed rules (such as deleting only the long-side triangles on the edges) can only generate "solid" polygonal regions, which will incorrectly include the internal open spaces within the occupied area, and cannot accurately represent the true topological structure of the entity.

[0006] Spatial computation is detached from real physical constraints: existing spatial range descriptions mostly remain at the level of pure geometric calculations, without considering the blocking effect of insurmountable physical barriers such as roads and rivers on the actual spatial radiation range of POI.

[0007] Therefore, there is an urgent need for a technical solution that can adaptively identify large-scale POIs and accurately characterize their spatial range, including hollow structures and physically constrained functions, in order to improve the accuracy of building function identification. Summary of the Invention

[0008] In view of the aforementioned existing problems, the present invention is proposed.

[0009] Therefore, the present invention provides a method and apparatus for identifying building functions based on the spatial range of POI outliers, solving the technical problem that the prior art cannot accurately distinguish large-scale POIs and the spatial range of their functions is not accurately characterized.

[0010] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0011] In a first aspect, the present invention provides a method for identifying building functions based on the spatial range of outliers (POIs), the method comprising the following steps:

[0012] S1. Obtain POI data for the target area, and by comparing the distribution characteristics of POI point sets at different spatial neighborhood scales with the theoretical random distribution characteristics, adaptively determine the abrupt change scale of clustering characteristics, and extract the set of outliers at this scale;

[0013] S2. Statistically analyze the spatial proximity relationship between the outlier set and the road network, construct a distance anomaly judgment benchmark based on the distribution characteristics of the spatial proximity relationship, and select scale-related outlier categories with spatial expansion attributes from the outlier set according to the distance anomaly judgment benchmark.

[0014] S3. For scale-related outlier categories, under the constraint of abrupt change in clustering feature scale, multidimensional spatial control points are determined based on an anisotropic neighborhood search strategy; the spatial network topology is constructed with the envelope structure formed by the multidimensional spatial control points as constraints, and irregular polygons containing hollow structures are extracted through specific topology clipping operations, which are used as the initial spatial scope.

[0015] S4. Construct building function association features based on the initial spatial scope and input them into the classification model for building function identification.

[0016] Preferably, step S1 includes the following specific steps:

[0017] Constructing a Delaunay triangulation based on POI data;

[0018] Calculate the cross-K nearest neighbor distance and corresponding variance at different neighborhood scales K, and construct a variance ratio index. Variance ratio index The calculation formula is:

[0019] ;

[0020] in:

[0021] This represents the variance of the cross-K nearest neighbor distance;

[0022] This represents the variance of the cross (K+1) nearest neighbor distance;

[0023] This represents the theoretical variance ratio under homogeneous Poisson distribution conditions;

[0024] Starting with K=1, gradually increase the variance ratio index. When the value is greater than 1 for the first time, the minimum K value at this time is determined as the abrupt change scale of the cluster feature. Candidate outliers are determined by combining the local triangular network structure under the abrupt change scale of the cluster feature, and the set of outliers is obtained by using adaptive side length constraints.

[0025] Preferably, the specific steps for obtaining the set of outliers using adaptive side length constraints include:

[0026] Calculate the graph diameter for the connected subgraph formed by candidate outliers;

[0027] The outlier cluster determination threshold is adaptively calculated based on the mean, median, and standard deviation of the diameters of all connected subgraphs, where the central tendency statistic of the diameter distribution is the smaller of the mean and the median.

[0028] When the diameter of a connected subgraph is less than the outlier cluster determination threshold, it is determined to be an outlier cluster, and isolated outliers are integrated with outlier clusters to form the final set of outliers.

[0029] Preferably, step S2 includes the following specific steps:

[0030] We statistically analyzed the distance samples from outliers to the nearest roads, and used a two-component Gaussian mixture model to fit the probability distribution of the distance samples, dividing them into near-road and far-road components.

[0031] Extract the mean of the near-road component. and standard deviation Establish distance anomaly judgment benchmark ;

[0032] ;

[0033] in:

[0034] The mean of the component near the road;

[0035] The standard deviation of the component near the road;

[0036] k is an adjustment parameter;

[0037] When the median distance to outliers corresponding to a certain POI category exceeds the distance anomaly judgment benchmark. When the time is right, it is determined to be a candidate scale-related outlier category;

[0038] The above process is performed on multiple city datasets, and a majority voting mechanism is used to include the candidate categories determined in no less than two-thirds of the cities into the final scale-related outlier category.

[0039] Preferably, step S3 includes the following specific steps:

[0040] Based on the number of neighboring points, 6 or 12 directions are adaptively selected to establish a polar sector structure. K-nearest neighbor search is performed independently in each direction to obtain the nearest neighbor distance of non-target objects. Based on this, the radius of influence in each direction is calculated to generate multi-dimensional spatial control points.

[0041] Connect adjacent control points to form an envelope structure, and use this envelope structure as the constraint boundary to construct a constrained Delaunay triangulation;

[0042] Remove the edges associated with the vertices of the envelope structure in the constrained Delaunay triangulation, and extract the polygon containing the hollow structure enclosed by the remaining triangular units as the initial spatial scope.

[0043] Preferably, after obtaining the initial spatial range of action, the method further includes:

[0044] By integrating POI category information, neighborhood semantic relationships, and spatial obstacle constraints such as roads and rivers, the boundary of the initial spatial scope is corrected.

[0045] When the radius of influence of the polar sector crosses the centerline of a road or river, the control point is truncated and projected onto the edge of the obstacle for boundary correction to obtain the final spatial range of influence.

[0046] Preferably, step S4 includes the following specific steps:

[0047] Based on building objects, a preset distance buffer is constructed. The spatial overlap area between the final spatial range of different categories and the buffer is calculated. The spatial range coverage feature is constructed based on the overlap ratio, and the dominance feature is obtained by normalization.

[0048] The XGBoost classification model integrates dominance features, POI statistical features, and building morphology features to output building function identification results.

[0049] Secondly, the present invention provides a building function identification device based on the spatial range of POI outliers, comprising:

[0050] The outlier detection module is used to acquire POI data of the target area. By comparing the distribution characteristics at different spatial neighborhood scales with the theoretical random distribution characteristics, it adaptively determines the abrupt change scale of the clustering characteristics and extracts the set of outliers at this scale.

[0051] The scale filtering module is used to statistically analyze the spatial proximity relationship between outlier sets and road networks, construct a distance anomaly judgment benchmark based on the distribution characteristics of spatial proximity relationship, and filter out scale-related outlier categories with spatial expansion attributes.

[0052] The range characterization module is used to determine multi-dimensional spatial control points based on an anisotropic neighborhood search strategy for scale-related outlier categories under the constraint of abrupt scale changes in clustering features. The spatial network topology is constructed with the envelope structure formed by the control points as a constraint, and irregular polygons containing hollow structures are extracted as the initial spatial range through topology pruning operations.

[0053] The function identification module is used to construct the functional association features of buildings based on the initial spatial scope and input them into the classification model for building function identification.

[0054] Thirdly, the present invention provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0055] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the above-described method.

[0056] The beneficial effects of this invention are as follows: The rationality of the strongly coupled parameter closed-loop enhancement scheme: This invention adaptively determines the abrupt change scale of clustering features by comparing the differences between the distribution characteristics and the theoretical random distribution characteristics, and directly uses this scale as the distance constraint threshold for subsequent anisotropic neighborhood searches. This parameter-level linkage makes outlier detection and spatial range characterization an inseparable whole, avoiding the errors caused by isolated parameter settings in existing technologies.

[0057] The reverse derivation logic enables objective screening of anomaly categories: This invention innovatively uses a two-component Gaussian mixture model to fit the distance distribution between outliers and roads, extracts the distribution characteristics of the "near-road component" representing ordinary POIs, and reverse derivations the benchmark threshold for judging "far-road anomalies (i.e., large-scale spatially extended POIs)". Combined with a multi-city voting mechanism, it effectively overcomes the pain point of existing technologies being unable to objectively distinguish between remote and isolated POIs and large-scale POIs.

[0058] Precise Generation of Hollow Structures Through Specific Topological Trimming: After constructing a constrained Delaunay triangulation, this invention does not simply delete edge triangles, but rather deletes edges associated with the vertices of the initial envelope structure, thereby extracting irregular polygons containing "hollow structures." This operation produces unexpected technical effects, enabling precise fitting and peeling of plazas or open spaces within large commercial complexes, campuses, and other entities, greatly improving the accuracy of spatial delineation.

[0059] Deep integration of physical constraints enhances geographic realism: When performing boundary correction, this invention uses insurmountable spatial barriers such as roads and rivers as hard constraints for geometric operations. When the radius of action crosses the barrier, the projection is truncated, so that the spatial range generated by the pure algorithm conforms to the real geographic laws, further improving the accuracy of extracting the functional association features of buildings. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart of a building function identification method based on the spatial range of POI outliers provided in an embodiment of the present invention;

[0062] Figure 2 This is a schematic diagram illustrating the principle of POI outlier detection provided in an embodiment of the present invention.

[0063] Figure 3 A schematic diagram illustrating the principle of scale-related outlier category screening provided in an embodiment of the present invention.

[0064] Figure 4 This is a schematic diagram illustrating the process of generating the spatial scope of POI outliers provided in an embodiment of the present invention.

[0065] Figure 5 This is a framework diagram for constructing and classifying building function identification features according to an embodiment of the present invention.

[0066] Figure 6 This is a diagram showing the POI outlier detection results in the example.

[0067] Figure 7 The following are the spatial range characterization results of different categories of scale-related POI outliers provided in the embodiments of the present invention, wherein: (a) Education and research category; (b) Commercial and residential category; (c) Tourist attraction category; (d) Company and enterprise category; (e) Government agency category; (f) Public infrastructure category.

[0068] Figure 8 The following is a comparison chart of the building function identification results of different methods provided in the embodiments of the present invention, wherein: (a) the actual function distribution (local area); (b) the results of the statistical method of POI with fixed radius buffer of building; (c) the results of the method of spatial range of fixed radius (100 m) based on scale-related POI outliers; and (d) the identification results of the method of the present invention. Detailed Implementation

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0070] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0072] Reference Figures 1-8 As one embodiment of the present invention, this embodiment provides a method for identifying building functions based on the spatial range of POI outliers, the method comprising the following steps:

[0073] S1. Obtain POI data for the target region. By comparing the distribution characteristics of POI point sets at different spatial neighborhood scales with the theoretical random distribution characteristics, adaptively determine the abrupt change scale of clustering characteristics and extract the set of outliers at this scale.

[0074] Specifically, after acquiring the POI dataset within the target region, a global Delaunay triangulation is first constructed. For each POI, its cross-K nearest neighbor distance and corresponding variance are calculated at different neighborhood scales K (K=1,2,3…). To measure the degree to which the POI distribution deviates from a random distribution, a variance ratio metric is constructed. This index is the ratio of the variance of the cross-K nearest neighbor distance to the theoretical variance of a homogeneous Poisson distribution (representing a completely random distribution). It is calculated by gradually increasing from K=1. Under the assumption of a completely random distribution, Theoretically, it is close to 1. When local clustering characteristics appear as K increases, Mutations can occur. Therefore, when When the value is first greater than 1, the minimum K value at this point is determined as the "clustering feature mutation scale," which reflects the optimal observation range for significant clustering of data in local space. At this mutation scale, candidate outliers are determined by combining the side length structure of the local triangulation. To distinguish between isolated points and outlier clusters, the graph diameter of the connected subgraphs formed by the candidate outliers is calculated. The outlier cluster determination threshold OC is adaptively calculated based on the mean, median, and standard deviation of the diameters of all connected subgraphs. To overcome the interference of the long-tailed distribution, the central tendency statistic of the diameter distribution is taken as the smaller of the mean and median. When the graph diameter of a connected subgraph is less than OC, it is determined to be an outlier cluster. Finally, isolated points and outlier clusters are integrated to form an outlier set.

[0075] S2. Statistically analyze the spatial proximity relationship between the outlier set and the road network, construct a distance anomaly judgment benchmark based on the distribution characteristics of the spatial proximity relationship, and select scale-related outlier categories with spatial expansion attributes from the outlier set according to the distance anomaly judgment benchmark.

[0076] We obtain distance samples from each point in the outlier set to the nearest road. Since most POIs are distributed along roads (near roads), while a few large-scale POIs (such as large parks and commercial complexes) are far from roads (far roads), this distance sample exhibits a bimodal distribution. A two-component Gaussian mixture model (GMM) is used to fit the probability distribution of the distance samples, dividing them into near-road and far-road components. The mean of the near-road component is extracted. and standard deviation A distance anomaly detection benchmark is constructed. When the median distance from outliers to roads for a certain POI category exceeds this benchmark, it indicates that POIs of that category are generally far from roads, and they are identified as candidate scale-related outlier categories. To enhance robustness, the above process is performed on multiple city datasets, using a majority voting mechanism to include POI categories identified as candidate categories in at least two-thirds of the cities into the final set of scale-related outlier categories.

[0077] S3. For scale-dependent outlier categories, under the constraint of abrupt change in clustering feature scale, multidimensional spatial control points are determined based on an anisotropic neighborhood search strategy; the spatial network topology is constructed with the envelope structure formed by the multidimensional spatial control points as constraints, and irregular polygons containing hollow structures are extracted through specific topology clipping operations, which are used as the initial spatial scope.

[0078] For POI objects of scale-related outlier categories selected through screening, a polar sector structure is established by adaptively selecting 6 or 12 directions based on the number of neighboring points. Under the distance constraint of the abrupt change scale of the clustering feature determined in S1, a K-nearest neighbor search is performed independently in each sector direction to obtain the nearest neighbor distance from the target outlier to non-target objects (i.e., POIs of other categories or obstacles). Based on this, the radius of influence in each direction is calculated, and a multi-dimensional spatial control point is generated at the ray endpoint.

[0079] The control points connecting adjacent sectors form an initial envelope structure. A constrained Delaunay triangulation network (CDTN) is then constructed using this envelope structure as the constraint boundary. At this point, the network may contain local triangulations due to the presence of non-target objects or empty spaces within the network.

[0080] To extract boundaries that conform to the actual land area shape, a specific topological pruning is performed: edges in the CDTN that are associated with the vertices of the initial envelope structure are deleted. This operation, while preserving the connectivity of the local CDTN network, disconnects the direct connection between the envelope boundary and the internal triangular network. Thus, the outer boundary enclosed by the remaining triangular units and the retained internal void boundary together form an "irregular polygon containing a hollow structure," which serves as the initial spatial scope. This structure can accurately represent internal open spaces such as hollow-enclosed plazas.

[0081] S4. Construct building function association features based on the initial spatial scope and input them into the classification model for building function identification.

[0082] Based on building objects, a preset distance buffer is constructed. The spatial overlap area between the final spatial scope and the buffer for different categories is calculated. The spatial scope coverage feature is constructed based on the overlap ratio, and normalization is performed to obtain the dominance feature. Finally, the dominance feature, POI conventional statistical features, and building morphology features are integrated into the XGBoost classification model to output the building function identification result (such as commercial, residential, industrial, etc.).

[0083] Preferably, step S1 includes the following specific steps:

[0084] Constructing a Delaunay triangulation based on POI data;

[0085] Calculate the cross-K nearest neighbor distance and corresponding variance at different neighborhood scales K, and construct a variance ratio index. Variance ratio index The calculation formula is:

[0086] ;

[0087] in:

[0088] This represents the variance of the cross-K nearest neighbor distance;

[0089] This represents the variance of the cross (K+1) nearest neighbor distance;

[0090] This represents the theoretical variance ratio under homogeneous Poisson distribution conditions;

[0091] As K increases from 1, the corresponding variance ratio is calculated step by step for the POI set. ,when When the value is greater than 1 for the first time, it is considered that the spatial distribution of POI has undergone significant clustering characteristics change at this scale. At this scale, the candidate outlier set and its local triangular network structure are determined by combining local neighborhood relationships, and the iteration of K is stopped.

[0092] Then, using a local triangulation ( Adaptive side length constraint CI is constructed using the statistical features of side lengths, and neighborhood relationships connected by long sides are removed to obtain a stable set of outliers.

[0093] Preferably, the specific steps for obtaining the set of outliers using adaptive side length constraints include:

[0094] Calculate the graph diameter for the connected subgraph formed by candidate outliers;

[0095] The outlier cluster determination threshold is adaptively calculated based on the mean, median, and standard deviation of the diameters of all connected subgraphs, where the central tendency statistic of the diameter distribution is the smaller of the mean and the median.

[0096] When the diameter of a connected subgraph is less than the outlier cluster determination threshold, it is determined to be an outlier cluster, and isolated outliers are integrated with outlier clusters to form the final set of outliers.

[0097] In this embodiment,

[0098]

[0099] in, Representing an edge The corresponding adaptive side length constraint threshold; Represents a local triangular network The i-th edge in; This represents a local triangular network structure constructed around candidate outliers; This represents the average length of all sides in a local triangulation. This represents the standard deviation of all side lengths in a local triangulation. This represents the adaptive adjustment coefficient; Let represent the length of the i-th edge in the local triangulation. The connected subgraph formed for each candidate outlier. Calculate the central tendency of its diameter distribution:

[0100]

[0101] Where OC represents the outlier cluster determination threshold; This represents a connected subgraph consisting of candidate outliers; Representing a connected subgraph Diameter of the figure; This represents the average diameter of all connected subgraphs; This represents the median of the diameters of all connected subgraphs; This represents the standard deviation of the diameters of all connected subgraphs; The central tendency statistic of the diameter distribution is taken as the smaller of the mean and the median. This represents the adjustment coefficient adaptively determined based on the diameter distribution characteristics; Let represent the diameter of the i-th connected subgraph. A threshold for outlier / cluster determination is constructed by combining the standard deviation of the diameter with the adaptive coefficient γ.

[0102] When the diameter of a subgraph is less than OC, the subgraph is determined to be an outlier cluster; after the isolated outlier points are integrated with the outlier cluster, the complete outlier detection results are obtained.

[0103] It should be noted that this step adaptively identifies local POI outlier clusters by statistically analyzing the changes in the diameter of local subgraphs, which is different from traditional density clustering or isolated outlier point processing.

[0104] The purpose of this step is to automatically identify POI entities with significant outlier distribution characteristics from complex urban spatial structures, providing input data for subsequent POI spatial scope analysis.

[0105] Preferably, step S2 includes the following specific steps:

[0106] The distance samples from outliers to the nearest road were statistically analyzed, and a two-component Gaussian Mixture Model (GMM) was used to fit the probability distribution of the distance samples, dividing them into near road component and far road component.

[0107] Extract the mean of the near-road component. and standard deviation Establish distance anomaly judgment benchmark ;

[0108] ;

[0109] in:

[0110] The mean of the component near the road;

[0111] The standard deviation of the component near the road;

[0112] k is an adjustment parameter;

[0113] When the median distance to outliers corresponding to a certain POI category exceeds the distance anomaly judgment benchmark. When the time is right, it is determined to be a candidate scale-related outlier category;

[0114] The above process is performed on multiple city datasets, and a majority voting mechanism is used to include the candidate categories determined in no less than two-thirds of the cities into the final scale-related outlier category.

[0115] This step is used to identify POI categories with obvious spatial extension properties, thereby avoiding interference from ordinary isolated POIs in subsequent range characterization.

[0116] Preferably, step S3 includes the following specific steps:

[0117] Based on the number of neighboring points, 6 or 12 directions are adaptively selected to establish a polar sector structure. K-nearest neighbor search is performed independently in each direction to obtain the nearest neighbor distance of non-target objects. Based on this, the radius of influence in each direction is calculated to generate multi-dimensional spatial control points.

[0118] Connect adjacent control points to form an envelope structure, and use this envelope structure as the constraint boundary to construct a constrained Delaunay triangulation;

[0119] Remove the edges associated with the vertices of the envelope structure in the constrained Delaunay triangulation, and extract the polygon containing the hollow structure enclosed by the remaining triangular units as the initial spatial scope.

[0120] In this embodiment, the spatial scope of POI objects in the scale-dependent outlier category is characterized.

[0121] First, establish the association between POIs and street semantic units, and remove objects whose boundaries are known or whose ranges can be directly obtained.

[0122] For each outlier at the target scale, a local star-shaped topology is constructed centered on that outlier. The local clustering feature distance of the outliers obtained in step S2 is used as the anomaly detection criterion. The nearest neighbor distance d between the target outlier and its non-target neighbor objects is obtained by K-nearest neighbor search. nn And calculate the radius of action in each direction accordingly. :

[0123] ;

[0124] Where, d nnThis represents the spatial distance from the target outlier to its Kth nearest non-target object. This distance is determined by gradually increasing the value of K until it exceeds a distance threshold.

[0125] Based on the number of neighboring points and the standard deviation of spatial distribution, 6 or 12 directions are adaptively selected to establish a polar sector structure, and a set of control points is generated in each direction based on the radius of action, thereby forming a local spatial control structure with directional constraint characteristics. In some implementations, Steiner points can be introduced to enhance local connections to improve topological connectivity.

[0126] Connect the control points in each direction to form an initial envelope structure, and use this envelope structure as the constraint boundary to construct a constrained Delaunay triangulation network (CDTN) to generate a local spatial network structure.

[0127] In CDTN, the edges associated with the initial envelope vertices are deleted, resulting in a hollow polygonal region enclosed by the remaining triangular units. This region serves as the initial spatial range of the target outlier.

[0128] By further integrating POI category information, neighborhood semantic relationships, and constraints from natural obstacles such as roads and rivers, the initial range is modified in terms of boundaries and semantic refinement, ultimately obtaining the POI spatial scope.

[0129] This step introduces an adaptive radius of influence based on K-nearest neighbor scale determination to dynamically characterize the local spatial influence range of outliers; combines the directional structure of polar sectors to express spatial heterogeneity; and utilizes constrained Delaunay triangulation and edge deletion mechanisms to generate a spatial range consistent with the real geographical morphology, providing structured spatial input for subsequent building function identification.

[0130] Preferably, step S4 includes the following specific steps:

[0131] Based on building objects, a preset distance buffer is constructed. The spatial overlap area between the final spatial range of different categories and the buffer is calculated. The spatial range coverage feature is constructed based on the overlap ratio, and the dominance feature is obtained by normalization.

[0132] The XGBoost classification model integrates dominance features, POI statistical features, and building morphology features to output building function identification results.

[0133] In this embodiment, the building function is identified as follows:

[0134] Obtain building data and POI data.

[0135] Extract the following features for each building:

[0136] (1) Spatial range characteristics, including the spatial relationship between the spatial range of different types of outliers and buildings.

[0137] (2) POI statistical characteristics, including the number, density and category proportion of POIs within the 500m buffer zone of the building.

[0138] (3) Architectural features, including building area, building height, building compactness, etc.

[0139] (4) Spatial relationship characteristics, including the shortest distance between buildings and the road network, and the spatial distance between buildings and their nearest neighboring buildings.

[0140] By fusing the above features, a building function identification model is constructed.

[0141] The trained model is used to classify building function types and output building function identification results.

[0142] This step enhances the ability to express the functional association between POIs and buildings by introducing POI spatial scope information, thereby improving the accuracy of building function identification.

[0143] Example:

[0144] The identification results were evaluated using Precision, Recall, and F1-Score indices, and compared with the POI density statistical method based on a fixed radius buffer of buildings (buffer radius of 500m and 1000m) and the fixed radius buffer statistical method based on scale-related POI outliers (non-adaptive spatial range method, with buffer radii set to 100m, 200m, 300m, and 500m respectively).

[0145]

[0146] Experimental results show that this invention can effectively identify scale-related outlier categories with spatial expansion attributes and achieve a reasonable representation of the spatial footprint of geographic entities by constructing the spatial scope of POIs. Compared with existing POI statistical methods based on fixed spatial neighborhoods (such as fixed buffer methods based on buildings or POI outliers), the adaptive spatial scope feature constructed in this invention can more accurately reflect the spatial relationship between POIs and buildings, thereby improving the accuracy of building function identification.

[0147] Secondly, the present invention provides a building function identification device based on the spatial range of POI outliers, comprising:

[0148] The outlier detection module is used to acquire POI data of the target area. By comparing the distribution characteristics at different spatial neighborhood scales with the theoretical random distribution characteristics, it adaptively determines the abrupt change scale of the clustering characteristics and extracts the set of outliers at this scale.

[0149] The scale filtering module is used to statistically analyze the spatial proximity relationship between outlier sets and road networks, construct a distance anomaly judgment benchmark based on the distribution characteristics of spatial proximity relationship, and filter out scale-related outlier categories with spatial expansion attributes.

[0150] The range characterization module is used to determine multi-dimensional spatial control points based on an anisotropic neighborhood search strategy for scale-related outlier categories under the constraint of abrupt scale changes in clustering features. The spatial network topology is constructed with the envelope structure formed by the control points as a constraint, and irregular polygons containing hollow structures are extracted as the initial spatial range through topology pruning operations.

[0151] The function identification module is used to construct the functional association features of buildings based on the initial spatial scope and input them into the classification model for building function identification.

[0152] This embodiment also provides a computer device applicable to the above-described method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the above-described method as proposed in the above embodiments.

[0153] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0154] This embodiment also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0155] In summary, this invention achieves automatic building function identification through scale-related outlier category screening, adaptive spatial scope construction, and spatial scope feature extraction. Compared with existing technologies, this invention has the following advantages:

[0156] (1) The distribution characteristics of outlier distances to roads are used to achieve automatic screening of scale-related outlier categories, which improves the objectivity and applicability of spatial extension attribute POI identification;

[0157] (2) Construct an adaptive spatial range of action based on the spatial structure characteristics of outliers, which can more reasonably represent the actual space occupied by POIs compared with the fixed spatial neighborhood method;

[0158] (3) By constructing spatial scope features, the ability to express the spatial relationship between POI and buildings is improved;

[0159] (4) By utilizing spatial range characteristics, POI statistical characteristics, architectural form characteristics and spatial relationship characteristics, a building function identification model was constructed, which improved the accuracy of building function identification.

[0160] Experimental results show that the present invention achieves better identification results than the POI density statistical method based on fixed radius buffer of buildings and the statistical method based on fixed radius buffer of scale-related POI outliers in the central urban area of ​​Hohhot, verifying the effectiveness and practicality of the present invention.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying building functions based on the spatial range of POI outliers, characterized in that, The method includes the following steps: S1. Obtain POI data for the target area, and by comparing the distribution characteristics of POI point sets at different spatial neighborhood scales with the theoretical random distribution characteristics, adaptively determine the abrupt change scale of clustering characteristics, and extract the set of outliers at this scale; S2. Statistically analyze the spatial proximity relationship between the outlier set and the road network, construct a distance anomaly determination benchmark based on the distribution characteristics of the spatial proximity relationship, and select scale-related outlier categories with spatial expansion attributes from the outlier set according to the distance anomaly determination benchmark. S3. For the scale-related outlier category, under the constraint of the abrupt change scale of the clustering feature, determine multidimensional spatial control points based on the anisotropic neighborhood search strategy; construct a spatial network topology with the envelope structure formed by the multidimensional spatial control points as a constraint, and extract irregular polygons containing hollow structures through a specific topology pruning operation, and use them as the initial spatial range. S4. Construct building function association features based on the initial spatial range and input them into the classification model to identify building functions.

2. The method according to claim 1, characterized in that, The specific steps of step S1 include: Construct a Delaunay triangulation based on the POI data; Calculate the cross-K nearest neighbor distance and corresponding variance at different neighborhood scales K, and construct a variance ratio index. The variance ratio index The calculation formula is: ； in: This represents the variance of the cross-K nearest neighbor distance; This represents the variance of the cross (K+1) nearest neighbor distance; This represents the theoretical variance ratio under homogeneous Poisson distribution conditions; Starting from K=1, gradually increase the variance ratio index. When the value is greater than 1 for the first time, the minimum K value at this time is determined as the abrupt change scale of the clustering characteristics; Candidate outliers are determined by combining the local triangular network structure under the abrupt change scale of the clustering characteristics, and the set of outliers is obtained by using adaptive side length constraints.

3. The method according to claim 2, characterized in that, The specific steps for obtaining the set of outliers using adaptive side length constraints include: Calculate the graph diameter for the connected subgraph formed by candidate outliers; The outlier cluster determination threshold is adaptively calculated based on the mean, median, and standard deviation of the diameters of all connected subgraphs, where the central tendency statistic of the diameter distribution is the smaller of the mean and the median. When the diameter of a connected subgraph is less than the outlier cluster determination threshold, it is determined to be an outlier cluster, and isolated outliers are integrated with outlier clusters to form the final set of outliers.

4. The method according to claim 1, characterized in that, The specific steps of step S2 include: The distance samples from the outliers to the nearest roads are statistically analyzed, and a two-component Gaussian mixture model is used to fit the probability distribution of the distance samples, dividing them into a near-road component and a far-road component. Extract the mean of the near-road component. and standard deviation Construct the distance anomaly determination benchmark ; ； in: The mean of the component near the road; The standard deviation of the component near the road; k is an adjustment parameter; When the median distance of outliers corresponding to a certain POI category exceeds the distance anomaly judgment benchmark When the time is right, it is determined to be a candidate scale-related outlier category; The above process is performed on multiple city datasets, and a majority voting mechanism is used to include the candidate categories determined in no less than two-thirds of the cities into the final scale-related outlier category.

5. The method according to claim 1, characterized in that, The specific steps of step S3 include: Based on the number of neighboring points, 6 or 12 directions are adaptively selected to establish a polar sector structure. K-nearest neighbor search is performed independently in each direction to obtain the nearest neighbor distance of non-target objects. Based on this, the radius of action in each direction is calculated to generate the multidimensional spatial control points. Connect adjacent control points to form an envelope structure, and use this envelope structure as the constraint boundary to construct a constrained Delaunay triangulation; Delete the edges associated with the vertices of the envelope structure in the constrained Delaunay triangulation, and extract the polygon containing the hollow structure enclosed by the remaining triangular units as the initial spatial scope.

6. The method according to claim 5, characterized in that, After obtaining the initial spatial range of action, the method further includes: By integrating POI category information, neighborhood semantic relationships, and spatial obstacle constraints such as roads and rivers, the boundary of the initial spatial scope is corrected. When the radius of influence of the polar sector crosses the centerline of a road or river, the control point is truncated and projected onto the edge of the obstacle for boundary correction to obtain the final spatial range of influence.

7. The method according to claim 6, characterized in that, The specific steps of step S4 include: Based on building objects, a preset distance buffer is constructed. The spatial overlap area between the final spatial range of different categories and the buffer is calculated. The spatial range coverage feature is constructed based on the overlap ratio, and the dominance feature is obtained by normalization. The dominant features, POI statistical features, and building morphology features are integrated into the XGBoost classification model to output the building function identification results.

8. A building function identification device based on the spatial range of POI outliers, characterized in that, include: The outlier detection module is used to acquire POI data of the target area. By comparing the distribution characteristics at different spatial neighborhood scales with the theoretical random distribution characteristics, it adaptively determines the abrupt change scale of the clustering characteristics and extracts the set of outliers at this scale. The scale filtering module is used to statistically analyze the spatial proximity relationship between the outlier set and the road network, construct a distance anomaly judgment benchmark based on the distribution characteristics of the spatial proximity relationship, and filter out scale-related outlier categories with spatial expansion attributes. The range characterization module is used to determine multidimensional spatial control points based on an anisotropic neighborhood search strategy, under the constraint of the abrupt change scale of the clustering feature, for scale-related outlier categories. A spatial network topology is constructed using the envelope structure formed by control points as constraints, and an irregular polygon containing a hollow structure is extracted as the initial spatial scope through topology clipping operations. The function identification module is used to construct building function association features based on the initial spatial range and input them into the classification model for building function identification.

9. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1-7.

Images