Building mode recognition method and system based on morphological coding

Abstract

The invention relates to the technical field of map making, in particular to a building pattern recognition method and system based on form coding, which are characterized in that grid description is carried out on buildings and adjacent relations thereof based on vector building data, and building forms are coded by adopting non-uniform Fourier transform. And learning the coding features through a spatial form coding network to obtain a mode classification result of the building. The method can better capture the local and global features of the building, overcomes the limitation of a traditional two-dimensional image recognition method under a complex building structure, improves the accuracy of building mode recognition, can more effectively process the complex spatial relationship between the buildings, can maintain higher recognition performance even in a high-density region of a city, and is suitable for popularization and application. Compared with a data processing method based on a high-resolution remote sensing image, the method has the advantages that the calculation complexity is remarkably reduced, and the rapid identification requirement of large-scale urban buildings can be met.

Description

Technical Field

[0001] The present invention relates to the technical field of map making, and in particular to a building pattern recognition method and system based on morphological coding. Background Art

[0002] With the acceleration of urbanization, the number and density of buildings in urban areas are increasing. How to effectively classify and identify buildings has become an important issue in urban planning, building management and smart city construction. Building pattern recognition can not only assist urban planners in optimizing land use, but also provide effective data support for areas such as automated building management, disaster warning and intelligent monitoring.

[0003] Traditional building pattern recognition methods mostly rely on image processing algorithms based on texture and shape features. These methods usually rely on high-quality remote sensing image data. However, in practical applications, the cost of obtaining high-resolution image data is high, and the image quality fluctuates greatly under different weather and lighting conditions, affecting the accuracy of recognition. In addition, recognition methods based on two-dimensional images are often unable to cope with complex building structures and cannot effectively capture the geometric features and spatial relationships of buildings.

[0004] In recent years, with the widespread application of geographic information systems (GIS), the research methods of pattern recognition based on vector building data have been continuously innovated and improved, and its application areas have been continuously expanding. Compared with image data, vector data can more accurately describe the geometric shape and spatial position of buildings, and has higher scalability and accuracy. However, most of the existing building pattern recognition methods based on vector data focus on simple geometric feature extraction, and fail to effectively utilize the spatial relationship between buildings and their morphological characteristics. In the field of building pattern recognition, how to combine the geometric shape and spatial adjacent relationship of buildings for more accurate pattern recognition has become an important technical problem. Summary of the invention

[0005] To this end, the present invention provides a building pattern recognition method and system based on morphological coding, which realizes more accurate and robust building classification and recognition by combining the geometric morphological features and spatial adjacent relationships of buildings.

[0006] According to the design scheme provided by the present invention, on the one hand, a building pattern recognition method based on morphological coding is provided, comprising:

[0007] Obtain open source vector building data, and construct a building pattern recognition sample dataset with building pattern labels based on the open source vector building data;

[0008] The spatial proximity relationship between buildings in the sample data set is extracted using the triangulated network of building outline vertices. Based on the spatial proximity relationship and using the second-order simplex grid, the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings are obtained.

[0009] The gridded spatial features of the building and the adjacent buildings are spatially morphologically encoded using non-uniform Fourier transform to obtain the spatial morphological encoding features of the building and the spatial morphological encoding features of the adjacent buildings;

[0010] The spatial morphology coding features of the building and the adjacent buildings are input into the building spatial morphology coding model, so as to iteratively train the building spatial morphology coding model using the building spatial morphology coding features and the adjacent building spatial morphology coding features in the sample data set to obtain a building pattern recognition target model;

[0011] For the vector building data to be identified, the building pattern recognition target model is used to identify the pattern to which the corresponding vector building data belongs;

[0012] Among them, the building spatial morphology coding model includes a relationship feature coding network for learning relationship features between buildings using spatial morphology coding features of adjacent buildings, a building feature coding network for learning building features using building spatial morphology coding features and relationship features between buildings, and a building pattern prediction network for predicting and identifying building patterns based on building features.

[0013] As the building pattern recognition method based on morphological coding of the present invention, further, a building pattern recognition sample data set with building pattern labels is constructed based on open source vector building data, including:

[0014] For the open source vector building data, calibrate the pattern labels of each building in the vector building data to obtain a sample data set of building pattern recognition;

[0015] The building pattern recognition sample data set is divided into a training data set and a test data set according to a preset ratio, so that the training data set is used to iteratively train the building space morphology coding model, and the test data set is used to tune the building space morphology coding model after iterative training.

[0016] As a building pattern recognition method based on morphological coding of the present invention, further, using the building outline vertex triangulation network to extract the spatial proximity relationship between buildings in the sample data set, it includes:

[0017] Interpolate and encrypt the building outlines in the specified area of ​​the sample data set and the polygon outlines in the area of ​​the same size;

[0018] The interpolated and encrypted contour points are used as triangle vertices, and the Delaunay triangulation is constructed with the overlap of the building contour edge and the polygon contour edge and triangle edge in the same size area as the constraint condition.

[0019] The triangles between two buildings in the Delaunay triangulation are retained, and other triangles are deleted to determine the buildings with triangle connections as adjacent buildings;

[0020] A building spatial proximity graph is constructed with the center of a building as a vertex and the lines connecting the centers of adjacent buildings as edges, so as to represent the spatial proximity relationship between buildings based on the building spatial proximity graph.

[0021] As a building pattern recognition method based on morphological coding of the present invention, further, based on the spatial proximity relationship and using the second-order simplex grid, the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings are obtained, including:

[0022] Construct a plane rectangular coordinate system, scale the building to the specified interval of the coordinate system based on the coordinate origin, number the vertices of the building's outer contour in counterclockwise order, and add the coordinates of the building's outer contour vertices and the origin to the point set;

[0023] Connect the outer contour vertices of two adjacent buildings and the coordinate origin to form a triangle, and add the triangle vertex index to the edge set;

[0024] Obtaining vectors for describing other attribute characteristics of the building, and constructing a second-order simplex network for gridding description of building characteristics based on the point set, edge set and other attribute characteristic vectors;

[0025] The second-order simplex network is used to obtain the grid description features of the building and the grid description features of the corresponding adjacent buildings.

[0026] As a building pattern recognition method based on morphological coding of the present invention, further, the gridded spatial features of the building and the adjacent buildings are morphologically coded using non-uniform Fourier transform, comprising:

[0027] A second-order simplex triangle density function is set according to a second-order simplex network, and the Fourier transform of the density function is obtained by density superposition;

[0028] The density function at each base frequency is taken as the weighted sum of the Fourier transforms of each second-order simplex triangle to obtain a non-uniform Fourier transform of the density function under the specified base frequency group, wherein the non-uniform Fourier transform of the second-order simplex triangle under the corresponding base frequency is calculated based on the sign content distortion factor in the weighted sum of the Fourier transforms of each second-order simplex triangle;

[0029] The non-uniform Fourier transform of the density function under the specified base frequency group is used to perform non-uniform Fourier coding on the gridded spatial characteristics of the building and the gridded spatial characteristics of the adjacent buildings, so as to obtain the spatial morphological coding characteristics of both the building and the adjacent buildings based on the coding results.

[0030] As a building pattern recognition method based on morphological coding of the present invention, further, using the building space morphological coding features and the adjacent building space morphological coding features in the sample data set to iteratively train the building space morphological coding model, comprising:

[0031] Inputting the spatial morphological coding features of neighboring buildings into a relational feature coding network to learn the relational features between buildings using the relational feature coding network, wherein the relational feature coding network is constructed based on Transformer;

[0032] The learned relationship features are used as a relationship matrix to measure the correlation between buildings. The relationship matrix and the building spatial morphology encoding features are input into the building feature encoding network to learn the building features using the building feature encoding network. The building feature encoding network consists of a graph convolutional network and a Transformer.

[0033] The building features are input into the building pattern prediction network, in which the multi-layer perceptron is used to reshape the building shape and the pattern category probability of each building is obtained through the softmax function, and the building pattern category corresponding to the maximum probability is used as the building pattern prediction output.

[0034] As the building pattern recognition method based on morphological coding of the present invention, further, the building space morphological coding model is iteratively trained, and further comprises:

[0035] The cross entropy function is used as the target loss function for model training, so as to iteratively train the building spatial morphology coding model using the target loss function.

[0036] In another aspect, the present invention further provides a building pattern recognition system based on morphological coding, comprising: a sample construction module, a feature gridding module, a feature morphological coding module, a model training module and a target recognition module, wherein:

[0037] A sample construction module is used to obtain open source vector building data and construct a building pattern recognition sample dataset with building pattern labels based on the open source vector building data;

[0038] A feature gridding module is used to extract the spatial proximity relationship between buildings in the sample data set using the triangulated network of building outline vertices, and obtain the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings based on the spatial proximity relationship and using a second-order simplex grid;

[0039] A feature morphology coding module is used to perform spatial morphology coding on the gridded spatial features of the building and the adjacent buildings by using non-uniform Fourier transform to obtain the spatial morphology coding features of the building and the spatial morphology coding features of the adjacent buildings;

[0040] A model training module is used to input the spatial morphology coding features of the building and the adjacent buildings into the building spatial morphology coding model, so as to iteratively train the building spatial morphology coding model using the building spatial morphology coding features and the adjacent building spatial morphology coding features in the sample data set to obtain a building pattern recognition target model;

[0041] The target recognition module is used to identify the pattern of the vector building data to be identified by using the building pattern recognition target model;

[0042] Among them, the building spatial morphology coding model includes a relationship feature coding network for learning relationship features between buildings using spatial morphology coding features of adjacent buildings, a building feature coding network for learning building features using building spatial morphology coding features and relationship features between buildings, and a building pattern prediction network for predicting and identifying building patterns based on building features.

[0043] Beneficial effects of the present invention:

[0044] The present invention uses non-uniform Fourier transform to encode the morphological features of buildings, which can better capture the local and global features of buildings, overcome the limitations of traditional two-dimensional image recognition methods under complex building structures, and thus significantly improve the accuracy of building pattern recognition; by incorporating the spatial adjacency of buildings into the feature learning process, the present invention can more effectively handle the complex spatial relationships between buildings, and maintain high recognition performance even in high-density urban areas; using vector building data for pattern recognition, compared with data processing methods based on high-resolution remote sensing images, the computational complexity is significantly reduced, and it can meet the needs of rapid recognition of large-scale urban buildings, and has good application prospects in the field of map making. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the building pattern recognition process based on morphological coding in the embodiment;

[0046] Figure 2Schematic diagram of the building space morphology coding network workflow in the embodiment. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention is further described in detail below in conjunction with the accompanying drawings and technical solutions.

[0048] In the field of building pattern recognition, how to combine the geometric form and spatial adjacent relationship of the building to perform more accurate pattern recognition, the embodiments of the present invention refer to Figure 1 As shown, a building pattern recognition method based on morphological coding is provided, comprising:

[0049] S101. Obtain open source vector building data, and construct a building pattern recognition sample data set with building pattern labels based on the open source vector building data.

[0050] Specifically, for the open source vector building data, the pattern label of each building in the vector building data is calibrated to obtain a building pattern recognition sample data set; the building pattern recognition sample data set is divided into a training data set and a test data set according to a preset ratio, so that the training data set is used to iteratively train the building space morphology coding model, and the test data set is used to tune the building space morphology coding model after iterative training.

[0051] The open source vector building data may be sourced from OpenStreetMap, comply with the requirements of its Open Database License (ODbL), and include necessary attribution information.

[0052] S102, extracting the spatial proximity relationship between buildings in the sample data set using the triangulated network of building outline vertices, and obtaining the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings based on the spatial proximity relationship and using a second-order simplex grid.

[0053] Specifically, the constraint-based Delaunay triangulation algorithm can be used to generate a triangulated network of building outline vertices to determine the spatial proximity relationship between buildings and describe the spatial proximity relationship between buildings. The spatial proximity relationship between buildings in the sample data set is extracted using the triangulated network of building outline vertices.

[0054] First, the building outlines in the study area and the polygon outlines of the same size as the study area are interpolated and encrypted; then, the encrypted points are used as triangle vertices to construct the Delaunay triangulation network, ensuring that the building outline edges and the polygon outline edges of the study range coincide with the triangle edges as constraints. Next, only the triangles between the two buildings are retained, and other triangles are deleted; finally, the buildings with triangle connections are judged as adjacent buildings, otherwise they are non-adjacent buildings; then, the building center is used as the vertex and the line connecting the centers of adjacent buildings as the edge to construct the building space proximity graph G.

[0055] The coordinates of the vector building data are standardized, and a second-order simplex grid is constructed to describe the spatial characteristics of the building and its neighboring buildings, and the grid description feature S is obtained. B and S A Specific steps may include:

[0056] Define the second-order simplex network structure as shown in formula (1):

[0057]

[0058] Construct a plane rectangular coordinate system with the origin x o =(0,0), scale the building to the interval [0.5,1.5], and number the vertices of the building's outer contour in counterclockwise order, and add the coordinates of these vertices and the origin to the point set V; connect two adjacent vertices and the coordinate origin to form a triangle, and add the vertex indexes of these triangles to the edge set E; vector D is used to describe other attribute characteristics of the building, and in this embodiment of the case, it can be set to a constant vector 1.

[0059] S103. Perform spatial morphology coding on the gridded spatial features of the building and the adjacent buildings by using non-uniform Fourier transform to obtain the spatial morphology coding features of the building and the spatial morphology coding features of the adjacent buildings.

[0060] The grid description feature S is described using non-uniform Fourier transform B and S A Perform spatial morphological coding to obtain the spatial morphological coding features F of the building and its neighboring buildings B and F A The specific steps may include the following:

[0061] First, for the second-order simplex grid S (2) The nth second-order simplex triangle in The density function is defined as shown in formula (2), and the Fourier transform of the density function is obtained by density superposition formula (3).

[0062] Density function As shown in formula (2):

[0063]

[0064] Density overlay function As shown in formula (3):

[0065]

[0066] In a group N w The fundamental frequency is The non-uniform Fourier transform under is shown in formula (4):

[0067]

[0068] Among them, each fundamental frequency w k ∈R 2 on It can be written as each second-order simplex triangle The weighted sum of the Fourier transform of is shown in formula (5):

[0069]

[0070] in is the nth second-order simplex triangle with base frequency w k The non-uniform Fourier transform of is shown in formula (6):

[0071]

[0072] Where μ(·):S (2) →{-1,1} is the decision Sign of the area function. is the distortion factor, Orthogonal to the unit second-order simplex ratio. is the symbol content distortion factor, and the calculation process is shown in formula (7):

[0073]

[0074] Through the above encoding process, non-uniform Fourier encoding of the building morphology is obtained.

[0075] S104, inputting the spatial morphology coding features of both the building and the neighboring buildings into the building spatial morphology coding model, so as to iteratively train the building spatial morphology coding model using the building spatial morphology coding features and the neighboring building spatial morphology coding features in the sample data set to obtain a building pattern recognition target model.

[0076] Among them, the building spatial morphology coding model includes a relationship feature coding network for learning relationship features between buildings using spatial morphology coding features of adjacent buildings, a building feature coding network for learning building features using building spatial morphology coding features and relationship features between buildings, and a building pattern prediction network for predicting and identifying building patterns based on building features.

[0077] The spatial morphology encoding feature F of the neighboring buildings A The input is a relational feature encoding network composed of Transformer modules to learn the relationship feature E between buildings; on the other hand, the spatial morphology encoding feature F of the building is B The building relationship feature E is input into the building feature encoding network composed of a graph convolutional network (GCN) and a Transformer module to learn the building feature H. The learned building feature H is input into the building pattern prediction model to learn the final building pattern prediction probability value. The building pattern prediction probability value and the actual building pattern category label are used to calculate the loss value. When the loss value converges, the network training is stopped, and the building space morphology encoding network in the current converged state is used as the building pattern recognition network to perform the building pattern recognition task; if the convergence state is not reached, the network is returned to be retrained.

[0078] The key role of the Transformer module is to capture the wide-area dependencies between input features. The specific steps can be summarized as follows:

[0079] Using the weight matrix W Q , W K and W V Map the input features to query, key, and value vectors; calculate the dot product between the query and the key to obtain the attention score and scale it; convert the attention score to a probability distribution through the Softmax function; and use the weighted sum of the feature values ​​as the final context information.

[0080] The attention score calculation process is shown in formula (8) and formula (9):

[0081]

[0082] Q=X×WQ, K=X×WK, V=X×WV (9)

[0083] The graph convolution module updates the features by aggregating the domain building features. The update calculation process is as shown in formula (10):

[0084] X c =σ(EHW+b) (10)

[0085] The building space data after graph structuring is represented as G = {H, E, A}, where H∈R m×N is the building characteristic value, E∈R P×m×m is the correlation eigenvalue of adjacent buildings or the proximity matrix of the graph, m is the number of buildings in the current building complex, N is the building eigenvalue dimension, and P is the correlation eigenvalue dimension of adjacent buildings. G ∈R P×m×Q is the output of the convolution operation, σ is the nonlinear activation function, W∈R P×N×Q is the model weight parameter, b∈R P×Q is the model bias parameter, and Q is the embedding dimension of the model output feature.

[0086] The building feature H is input into the Transformer module for feature encoding to further mine the building spatial features; the obtained building spatial features are input into the multi-layer perceptron for feature reshaping to obtain the feature vector of [number of buildings, number of pattern categories]; the feature vector is input into the Softmax function to calculate the pattern category probability of each building. The calculation formula of the Softmax function is shown in formula (11):

[0087]

[0088] In the above formula, p i represents the probability that the predicted building corresponds to the i-th category, K represents the total number of categories, and a represents the building characteristics.

[0089] The cross entropy loss function is used to control the gradient update during training. The calculation formula is as follows:

[0090]

[0091] In the above formula, M is the number of categories; y ic is a sign function, which takes 1 if the true category of sample i is equal to c, otherwise it takes 0; p ic is the output probability of predicting sample i belonging to category c.

[0092] S105 , for the vector building data to be identified, using the building pattern recognition target model to identify the pattern to which the corresponding vector building data belongs.

[0093] Using vector building data, the buildings and their adjacent relationships are described in a grid, the non-uniform Fourier transform is used to encode the building morphology, and the coding features are learned through the spatial morphology coding network to obtain the pattern classification results of the buildings. Compared with the traditional recognition method based on image data, in the embodiment of this case, feature mining through building morphology coding technology can enhance the objectivity of the model and effectively improve the accuracy and efficiency of building pattern recognition.

[0094] Furthermore, based on the above method, an embodiment of the present invention also provides a building pattern recognition system based on morphological coding, comprising: a sample construction module, a feature gridding module, a feature morphological coding module, a model training module and a target recognition module, wherein:

[0095] A sample construction module is used to obtain open source vector building data and construct a building pattern recognition sample dataset with building pattern labels based on the open source vector building data;

[0096] A feature gridding module is used to extract the spatial proximity relationship between buildings in the sample data set using the triangulated network of building outline vertices, and obtain the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings based on the spatial proximity relationship and using a second-order simplex grid;

[0097] A feature morphology coding module is used to perform spatial morphology coding on the gridded spatial features of the building and the adjacent buildings by using non-uniform Fourier transform to obtain the spatial morphology coding features of the building and the spatial morphology coding features of the adjacent buildings;

[0098] A model training module is used to input the spatial morphology coding features of the building and the adjacent buildings into the building spatial morphology coding model, so as to iteratively train the building spatial morphology coding model using the building spatial morphology coding features and the adjacent building spatial morphology coding features in the sample data set to obtain a building pattern recognition target model;

[0099] The target recognition module is used to identify the pattern of the vector building data to be identified by using the building pattern recognition target model;

[0100] Among them, the building spatial morphology coding model includes a relationship feature coding network for learning relationship features between buildings using spatial morphology coding features of adjacent buildings, a building feature coding network for learning building features using building spatial morphology coding features and relationship features between buildings, and a building pattern prediction network for predicting and identifying building patterns based on building features.

[0101] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0102] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0103] The units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each embodiment have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person of ordinary skill in the art may use different methods to implement the described functions for each specific application, but such implementation is not considered to be beyond the scope of the present invention.

[0104] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits, and accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present invention is not limited to any specific form of combination of hardware and software.

[0105] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A building pattern recognition method based on morphological coding, characterized in that: Include: Obtain open source vector building data, and construct a building pattern recognition sample dataset with building pattern labels based on the open source vector building data; The spatial proximity relationship between buildings in the sample data set is extracted using the triangulated network of building outline vertices. Based on the spatial proximity relationship and using the second-order simplex grid, the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings are obtained. The gridded spatial features of the building and the adjacent buildings are spatially morphologically encoded using non-uniform Fourier transform to obtain the spatial morphological encoding features of the building and the spatial morphological encoding features of the adjacent buildings; The spatial morphology coding features of the building and the adjacent buildings are input into the building spatial morphology coding model, so as to iteratively train the building spatial morphology coding model using the building spatial morphology coding features and the adjacent building spatial morphology coding features in the sample data set to obtain a building pattern recognition target model; For the vector building data to be identified, the building pattern recognition target model is used to identify the pattern to which the corresponding vector building data belongs; Among them, the building spatial morphology coding model includes a relationship feature coding network for learning relationship features between buildings using spatial morphology coding features of adjacent buildings, a building feature coding network for learning building features using building spatial morphology coding features and relationship features between buildings, and a building pattern prediction network for predicting and identifying building patterns based on building features.

2. The building pattern recognition method based on morphological coding according to claim 1 is characterized in that: Based on the open source vector building data, a building pattern recognition sample dataset with building pattern labels is constructed, including: For the open source vector building data, calibrate the pattern labels of each building in the vector building data to obtain a sample data set of building pattern recognition; The building pattern recognition sample data set is divided into a training data set and a test data set according to a preset ratio, so that the training data set is used to iteratively train the building space morphology coding model, and the test data set is used to tune the building space morphology coding model after iterative training.

3. The building pattern recognition method based on morphological coding according to claim 1 is characterized in that: The spatial proximity relationship between buildings in the sample dataset is extracted using the building outline vertex triangulation, including: Interpolate and encrypt the building outlines in the specified area of ​​the sample data set and the polygon outlines in the area of ​​the same size; The interpolated and encrypted contour points are used as triangle vertices, and the Delaunay triangulation is constructed with the overlap of the building contour edge and the polygon contour edge and the triangle edge in the same size area as the constraint condition; The triangles between two buildings in the Delaunay triangulation are retained, and other triangles are deleted to determine the buildings with triangle connections as adjacent buildings; A building spatial proximity graph is constructed with the center of a building as a vertex and the lines connecting the centers of adjacent buildings as edges, so as to represent the spatial proximity relationship between buildings based on the building spatial proximity graph.

4. The building pattern recognition method based on morphological coding according to claim 1 or 3, characterized in that: Based on the spatial proximity relationship and using the second-order simplex grid, the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings are obtained, including: Construct a plane rectangular coordinate system, scale the building to the specified interval of the coordinate system based on the coordinate origin, number the vertices of the building's outer contour in counterclockwise order, and add the coordinates of the building's outer contour vertices and the origin to the point set; Connect the outer contour vertices of two adjacent buildings and the coordinate origin to form a triangle, and add the triangle vertex index to the edge set; Obtaining vectors for describing other attribute characteristics of the building, and constructing a second-order simplex network for gridding description of building characteristics based on the point set, edge set and other attribute characteristic vectors; The second-order simplex network is used to obtain the grid description features of the building and the grid description features of the corresponding adjacent buildings.

5. The building pattern recognition method based on morphological coding according to claim 1 is characterized in that: The spatial morphology encoding of the gridded spatial features of both buildings and neighboring buildings is performed using non-uniform Fourier transform, including: A second-order simplex triangle density function is set according to a second-order simplex network, and the Fourier transform of the density function is obtained by density superposition; Taking the density function at each base frequency as the weighted sum of the Fourier transforms of each second-order simplex triangle to obtain the non-uniform Fourier transform of the density function under the specified base frequency group, wherein the non-uniform Fourier transform of the second-order simplex triangle under the corresponding base frequency is calculated based on the sign content distortion factor in the weighted sum of the Fourier transforms of each second-order simplex triangle; The non-uniform Fourier transform of the density function under the specified base frequency group is used to perform non-uniform Fourier coding on the gridded spatial characteristics of the building and the gridded spatial characteristics of the adjacent buildings, so as to obtain the spatial morphological coding characteristics of both the building and the adjacent buildings based on the coding results.

6. The building pattern recognition method based on morphological coding according to claim 1 is characterized in that: The building spatial morphology coding features and the adjacent building spatial morphology coding features in the sample data set are used to iteratively train the building spatial morphology coding model, including: Inputting the spatial morphological coding features of neighboring buildings into a relational feature coding network to learn the relational features between buildings using the relational feature coding network, wherein the relational feature coding network is constructed based on Transformer; The learned relationship features are used as a relationship matrix for measuring the correlation between buildings, and the relationship matrix and the building spatial morphology encoding features are input into a building feature encoding network to learn building features using the building feature encoding network, wherein the building feature encoding network is composed of a graph convolutional network and a Transformer; The building features are input into the building pattern prediction network, in which the multi-layer perceptron is used to reshape the building shape and the pattern category probability of each building is obtained through the softmax function, and the building pattern category corresponding to the maximum probability is used as the building pattern prediction output.

7. The building pattern recognition method based on morphological coding according to claim 1 or 6, characterized in that: The building space morphology encoding model is iteratively trained, and also includes: The cross entropy function is used as the target loss function for model training, so as to iteratively train the building spatial morphology coding model using the target loss function.

8. A building pattern recognition system based on morphological coding, characterized in that: It includes: sample construction module, feature gridding module, feature morphology encoding module, model training module and target recognition module, among which, A sample construction module is used to obtain open source vector building data and construct a building pattern recognition sample dataset with building pattern labels based on the open source vector building data; A feature gridding module is used to extract the spatial proximity relationship between buildings in the sample data set using the triangulated network of building outline vertices, and obtain the gridded spatial features of the buildings in the sample data set and the gridded spatial features of the corresponding adjacent buildings based on the spatial proximity relationship and using a second-order simplex grid; A feature morphology coding module is used to perform spatial morphology coding on the gridded spatial features of the building and the adjacent buildings by using non-uniform Fourier transform to obtain the spatial morphology coding features of the building and the spatial morphology coding features of the adjacent buildings; A model training module is used to input the spatial morphology coding features of the building and the adjacent buildings into the building spatial morphology coding model, so as to iteratively train the building spatial morphology coding model using the building spatial morphology coding features and the adjacent building spatial morphology coding features in the sample data set to obtain a building pattern recognition target model; The target recognition module is used to identify the pattern of the vector building data to be identified by using the building pattern recognition target model; Among them, the building spatial morphology coding model includes a relationship feature coding network for learning relationship features between buildings using spatial morphology coding features of adjacent buildings, a building feature coding network for learning building features using building spatial morphology coding features and relationship features between buildings, and a building pattern prediction network for predicting and identifying building patterns based on building features.

9. An electronic device, characterized in that: include: at least one processor, and a memory coupled to the at least one processor; The memory stores a computer program, and the computer program can be executed by the at least one processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 can be implemented.

Images