A method for architectural pattern recognition
By constructing a multi-view graph diffusion model and combining the spatial and non-spatial features of buildings, the accuracy and efficiency problems of building pattern recognition in existing technologies are solved, achieving more efficient building pattern recognition and map generalization.
Patent Information
- Application Number
- CN202211310369.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-25
- Publication Date
- 2025-09-16
- Estimated Expiration
- 2042-10-25
AI Technical Summary
Existing building pattern recognition methods struggle to effectively utilize complementary information from spatial proximity and non-spatial features, resulting in insufficient generalization ability. Furthermore, deep learning-based methods exhibit slow convergence speeds, strong dependence on initial weights, and require a large amount of manually labeled data.
A centroid distance view, a contour distance view, and a non-spatial attribute view are constructed. Graph diffusion is performed using a Gaussian kernel function and a cross-view graph diffusion equation. The resulting graph is then fused to obtain a joint graph. A normalized graph cutting algorithm is used for clustering to identify the spatial distribution patterns of buildings.
It improves the accuracy and efficiency of building pattern recognition, and can make full use of complementary information of spatial structure and non-spatial attributes in different views, making it suitable for map generalization and semantic segmentation of urban scenes.
Smart Images
Figure CN115620139B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of geospatial information processing technology, and specifically relates to a method for recognizing architectural patterns. Background Art
[0002] Buildings, as entities in cities, are important objects in maps. Building patterns refer to the arrangement of buildings displayed at a specific scale within a map space, and are recognizable and named through human visual perception. Recognizing building patterns in maps is crucial for understanding and modeling geographic space, as it simplifies the identification of patterns while preserving the spatial configuration of buildings within the scene. Building spatial patterns not only reflect human concepts regarding the design and utilization of space, but also contain a wealth of cultural and geological knowledge. Their temporal and spatial variations also reflect the evolution of the environment, economy, politics, and culture over time. Therefore, building patterns can be used in a wide range of applications, such as automatic map generalization, semantic classification of urban scenes, building attribute prediction, and visualization of 3D city models. In comprehensive evaluations, building patterns are also a key indicator of the preservation of original map features. If building patterns cannot be accurately identified, their preservation in subsequent map generalization processes is inherently difficult. Therefore, automatically identifying building patterns in maps is crucial.
[0003] With the significant improvement in the ease of access to crowdsourced data and the real-time nature of its updates, both academia and industry have easily accessed large amounts of geographic information. Voluntary geographic information (VGI) has become a crucial component of spatial databases due to its accessibility and global coverage. OpenStreetMap (OSM), one of the most successful VGI projects, provides freely accessible and editable maps, making it possible to identify spatial distribution patterns of buildings based on map vector data. Currently, there are three main approaches to building pattern recognition. The first is rule-based methods, which typically combine computational geometry, graph theory, and visual perception theory. In these methods, spatial proximity metrics and non-spatial attributes are often incorporated into hand-crafted rules, hindering algorithm generalization. The second category is machine learning-based methods, where common classifiers, including supervised learning methods such as support vector machines and random forests, are used to classify building clusters into underlying patterns. These methods combine spatial and non-spatial features by unifying them in a single vector space. However, they also encourage exclusivity in representation and ignore the complementary information between them. The third category is deep learning-based methods, which apply graph convolutional networks to the task of building grouping, a promising approach. However, using feedforward neural networks to build models has a slow convergence rate, is prone to falling into local extremes, and is highly dependent on initial weights. Furthermore, this approach is supervised learning, requiring manual annotation of large amounts of sample data and training of detection models to improve algorithm accuracy. Summary of the Invention
[0004] The purpose of this application is to overcome the shortcomings of the existing technology, utilize the complementary information between spatial proximity and non-spatial features, consider the underlying manifold geometry of the original data points in different views, and provide a new architectural pattern recognition method for identifying building complexes with complex nonlinear spatial distribution patterns.
[0005] In order to achieve the above application objectives, the present application provides a method for architectural pattern recognition, comprising the following steps:
[0006] Step 1): obtaining an original data set, wherein the original data set includes footprint vector data of buildings with different distribution patterns;
[0007] Step 2): Using the centroids of the buildings in the original data set as vertices, the lines between the vertices as edges, and the Euclidean distances between the vertices as weights, a Delaunay triangulation is constructed to represent the spatial structure of the buildings, and a centroid distance view is obtained;
[0008] Step 3): Using the centroids of the buildings in the original dataset as vertices, the lines between the vertices as edges, and the closest contour distances between the buildings as weights, a complete graph is constructed to obtain a contour distance view;
[0009] Step 4): Calculate the size, direction and shape indicators of the buildings in the original data set to obtain a non-spatial attribute view;
[0010] Step 5): using Gaussian kernel functions to construct a similarity matrix of the centroid distance view, a similarity matrix of the contour distance view, and a similarity matrix of the non-spatial attribute view for the centroid distance view, the contour distance view, and the non-spatial attribute view, respectively;
[0011] Step 6): using a cross-view diffusion equation to diffuse the similarity matrix of the centroid distance view, the similarity matrix of the contour distance view, and the similarity matrix of the non-spatial attribute view, respectively, to obtain an improved graph of the centroid distance view, an improved graph of the contour distance view, and an improved graph of the non-spatial attribute view;
[0012] Step 7): fusing the centroid distance view improvement map, the contour distance view improvement map, and the non-spatial attribute view improvement map to obtain a joint map;
[0013] Step 8): Use the normalized graph cutting algorithm to cut the joint graph to obtain the final clustering and building spatial distribution pattern results.
[0014] Furthermore, the different distribution modes include a linear distribution mode, a nonlinear distribution mode and a sparse distribution mode.
[0015] Furthermore, the linear distribution pattern includes a straight line arrangement and a curved line arrangement; the nonlinear distribution pattern includes a grid arrangement, a grid-like arrangement and an unstructured arrangement.
[0016] Furthermore, the size index includes the perimeter, area, and average radius of the building; the direction index is the direction indicated by the long side of the minimum circumscribed rectangle of the building; and the shape index includes compactness, fractal, elongation, concavity, and overlap index.
[0017] Furthermore, a Gaussian kernel function based on the k-nearest neighbor method is used to construct the similarity matrix of the centroid distance view, the similarity matrix of the contour distance view, and the similarity matrix of the non-spatial attribute view. The Gaussian kernel function based on the k-nearest neighbor method is shown in the following formula (1). If two points are k-neighbors of each other, the weight of the edge connecting the two points can be calculated by the following formula (1); on the contrary, if the two points are not k-neighbors of each other, the weight of the edge connecting the two points is 0;
[0018]
[0019] Among them, x i , x jRepresents building objects, which are multidimensional vectors, each dimension represents an attribute of the building, σ is the bandwidth of the Gaussian kernel function, W ij Represents the (i, j)th element of the similarity matrix W, W ji Represents the (j, i)th element of the similarity matrix W, W ji and W ji The values in represent the building object x i and x j similarity.
[0020] Furthermore, the use of the cross-view graph diffusion equation to diffuse the similarity matrix of the centroid distance view, the similarity matrix of the silhouette distance view, and the similarity matrix of the non-spatial attribute view respectively includes the following steps:
[0021] For the similarity matrix W of the v-th view v , v = 1, 2, 3, use the cross-view diffusion equation to update W v , the cross-view diffusion equation is:
[0022]
[0023]
[0024] Where D is a diagonal matrix whose diagonal elements satisfy N is The number of rows, p, q represent the pth row and qth column of the matrix respectively, t is the tth iteration number, Represents the result of the similarity matrix of the u-th view (u≠v) after the t-th iteration, represents the initial similarity matrix of the v-th view before iteration, and α represents the ratio of the initial similarity matrix before the iteration of the view to the improved graph obtained after iteration of the other two views at the previous moment.
[0025] Furthermore, fusing the centroid distance view improvement map, the contour distance view improvement map, and the non-spatial attribute view improvement map comprises the following steps:
[0026] The similarity matrix W of the three views v Perform normalization, v = 1, 2, 3, as shown in the following formula (2):
[0027]
[0028] After T iterations, the joint graph can be obtained by the following formula (3):
[0029]
[0030] Furthermore, after the joint graph is cut using the normalized graph cutting algorithm, a label is assigned to each building object. Buildings with the same label are considered as a class. Buildings of the same class have similar characteristics, and their arrangement is the spatial distribution pattern of the building cluster.
[0031] Furthermore, the centroid distance view includes a shortest distance matrix from each point to other points in the Delaunay triangulation graph calculated using the Floyd algorithm.
[0032] Furthermore, the silhouette distance view includes a shortest distance matrix from each point to other points in the complete graph calculated using the Floyd algorithm.
[0033] Compared with the existing technology, this application has the following technical effects:
[0034] An architectural pattern recognition method of the present application is a recognition method based on multi-view clustering. By mining and analyzing the geometric vector data of buildings, taking into account the characteristics of retaining the size, shape and other surface information of the buildings, compared with other methods, the recognition method of the present application uses two distance measurement methods, contour distance and centroid distance, to make full use of the complementary information of spatial structure and non-spatial attributes, and determine the architectural pattern by combining spatial proximity and non-spatial similarity, providing technical support for map synthesis, urban scene semantic segmentation and other aspects, and representing the geographical reality as realistically as possible within the limitation of map scale. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the specific implementation methods or the description of the prior art. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0036] Figure 1 A flowchart of a building pattern recognition method provided in an embodiment of the present application;
[0037] Figure 2 (a) is a schematic diagram of constructing a Delaunay triangulation to obtain a centroid distance view provided in an embodiment of the present application; Figure 2 (b) is a schematic diagram of constructing a complete graph to obtain a contour distance view according to an embodiment of the present application;
[0038] Figure 3 A schematic diagram of calculating a non-spatial attribute view provided in an embodiment of the present application; Figure 3 The SBR in (a) represents the minimum bounding rectangle, Ri Represents a vertex (x i ,y i ) to the center of mass (c x , c y ) distance; SBRO represents the orientation of the building, and the direction of the longer side (long axis) in the minimum circumscribed rectangle of the building is the direction of the building; Figure 3 (b) describes the shape index of the building, where CH and EAC represent the convex hull and the equal area circle, and L sbr and W sbr Represent the length and width of SBR respectively. DETAILED DESCRIPTION
[0039] In order to make the technical problems, technical solutions and beneficial effects to be solved by this application more clearly understood, the present application is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.
[0040] In this application, the term "and / or" describes the relationship between associated objects, indicating that three possible relationships exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. A and B can be singular or plural. The character " / " generally indicates that the associated objects are in an "or" relationship.
[0041] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, "at least one of a, b, or c", or "at least one of a, b, and c" can all mean: a, b, c, ab (i.e., a and b), ac, bc, or abc, where a, b, and c can be single or multiple.
[0042] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. Some or all of the steps can be executed in parallel or sequentially. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0043] The terms used in the embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. The singular forms "a", "an", "the" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.
[0044] The weights of the relevant components mentioned in the examples of this application may not only refer to the specific content of each component, but also represent the weight ratio between the components. Therefore, as long as the content of the relevant components is proportionally enlarged or reduced according to the examples of this application, it is within the scope disclosed in the examples of this application. Specifically, the mass described in the examples of this application may be a mass unit known in the chemical industry, such as μg, mg, g, kg, etc.
[0045] The terms "first" and "second" are used solely for descriptive purposes to distinguish objects, such as substances, from one another and should not be understood to indicate or imply relative importance or to implicitly specify the quantity of the technical features being referred to. For example, without departing from the scope of the embodiments of this application, a first XX may also be referred to as a second XX, and similarly, a second XX may also be referred to as a first XX. Thus, features defined as "first" or "second" may explicitly or implicitly include one or more of such features.
[0046] An embodiment of the present application provides a new architectural pattern recognition method for identifying building complexes with complex nonlinear spatial distribution patterns. The method is based on multiple spatial proximity graph distances and architectural attribute features, and is used to identify building complexes with complex nonlinear spatial distribution patterns. An embodiment of the present application constructs three graph sequences to capture the spatial adjacency structure of different feature views (i.e., center point distance view and nearest outlier distance view) and non-spatial feature attribute views (i.e., size, direction, and shape), and then obtains three improved graphs through a graph diffusion process, and fuses them into a joint graph shared by all views, which can depict the relationships between them and produce the final clustering results.
[0047] Specifically, the embodiment of the present application provides a method for identifying building patterns, the flow chart of which is as follows: Figure 1 As shown, the following steps are included:
[0048] Step 1): obtaining an original data set, the original data set including footprint vector data of buildings with different distribution patterns;
[0049] The embodiment of the present application first analyzes the distribution pattern of footprint vector data of multiple closed community buildings, and then selects multiple non-overlapping small areas with representative distribution patterns with different building distribution patterns as the original data set.
[0050] Specifically, the original data set of the embodiment of the present application is the footprint vector data of buildings composed of the bottom surface contour vector data of buildings in Wuhan and Chengdu, with a scale of 1:10000 and a shapefile file type. The building complexes selected in the two cities include linear distribution patterns (including straight line arrangement patterns and curved arrangement patterns), nonlinear distribution patterns (including grid-like, grid-like and unstructured arrangement patterns), and sparse distribution patterns.
[0051] Step 2): Take the centroids of the buildings in the original dataset as vertices, the lines between the vertices as edges, and the Euclidean distances between the vertices as weights, and construct a Delaunay triangulation to represent the spatial structure of the building, and obtain the centroid distance view V cd ,like Figure 2 (a); Specifically, the centroid distance view V cd The acquisition includes the shortest distance matrix from each point to other points in the Delaunay triangulation diagram calculated using the Floyd algorithm;
[0052] Step 3): Take the centroids of the buildings in the original dataset as vertices, the lines between the vertices as edges, and the nearest contour distances between the buildings as weights to construct a complete graph and obtain the contour distance view V od ,like Figure 2 (b) shows; specifically, the contour distance view V od The acquisition includes the shortest distance matrix from each point to other points in the complete graph calculated using the Floyd algorithm;
[0053] Step 4): Calculate the size, orientation and shape indicators of the buildings in the original dataset, such as Figure 3 (a) and 3(b), obtain the non-spatial attribute view V ns ;
[0054] In the embodiments of this application, building size indicators include perimeter, area, and average radius; building orientation indicators are the direction indicated by the long side of the building's minimum circumscribed rectangle; and building shape indicators include compactness, fractal shape, elongation, concavity, and overlap index. The following table lists the indicators and formulas for these indicators.
[0055]
[0056] In the above table, R i Represents the vertex of the building (x i ,y i ) to the center of mass (c x , c y ) distance; L sbr and W sbrThey represent the length and width of the minimum circumscribed rectangle of the building respectively; the concavity is the ratio of the area of the building to the area of the convex hull; the overlap index is the ratio of the area of the intersection and junction of the building and its equal-area circle.
[0057] Step 5): Use Gaussian kernel function to construct the similarity matrix of centroid distance view, the similarity matrix of contour distance view and the similarity matrix of non-spatial attribute view for the three views respectively;
[0058] Each row in each view matrix represents a building object, and each column of the matrix represents the attribute of the building (in the centroid distance view, each column of the matrix represents the shortest centroid distance between the building corresponding to the row and other buildings; in the silhouette distance view, each column of the matrix represents the shortest silhouette distance between the building corresponding to the row and other buildings; in the non-spatial attribute view, each column of the matrix represents the perimeter, area, average radius, orientation, compactness, fractal, elongation, concavity, and overlap index of the building). The Gaussian kernel function (as shown in formula (1) below) is used to calculate the similarity between each pair of building objects, thereby obtaining the similarity matrix.
[0059] The specific implementation steps are as follows:
[0060] The embodiment of the present application uses the k-nearest neighbor method based on the Gaussian kernel function. The k-nearest neighbor algorithm is used to traverse all sample data points and take the k points closest to each sample as the nearest neighbors. As shown in the following formula (1), if two points are k-neighbors of each other, the weight of the edge connecting the two points can be calculated by the Gaussian kernel function; on the contrary, if the two points are not k-neighbors of each other, the weight of the edge connecting the two points is 0. Where x i , x j Represents building objects, which are multidimensional vectors, each dimension represents an attribute of the building, σ is the bandwidth of the Gaussian kernel function, W ij Represents the (i, j)th element of the similarity matrix W, W ji Represents the (j, i)th element of the similarity matrix W, W ij and W ji The values in represent the building object x i and x j Similarities:
[0061]
[0062] Step 6): Use the cross-view diffusion equation to diffuse the similarity matrix of the centroid distance view, the similarity matrix of the silhouette distance view, and the similarity matrix of the non-spatial attribute view, respectively, to obtain the improved graph of the centroid distance view, the improved graph of the silhouette distance view, and the improved graph of the non-spatial attribute view. The specific processing method is as follows:
[0063] For the similarity matrix W of the v-th view v , v = 1, 2, 3, the embodiment of the present application designs a cross-view diffusion process to update W v , the cross-view diffusion equation is as follows:
[0064]
[0065]
[0066] Where D is a diagonal matrix whose diagonal elements satisfy N is The number of rows, p, q represent the pth row and qth column of the matrix respectively, t is the tth iteration number, Represents the result of the similarity matrix of the u-th view (u≠v) after the t-th iteration. represents the initial similarity matrix of the vth view before iteration. α represents the ratio of the initial similarity matrix before the iteration of the view to the improved graph obtained after iteration of the other two views at the previous moment.
[0067] In equation (2), the embodiment of the present application needs to set the value of α. Since α determines the proportion of information from the original graph that should be retained at each iteration, an adaptive method is used to determine α for each data set. For the original similarity matrix of each view, if the value of an edge is large in all different graphs, the edge is called a strong edge, and it should be clearly retained in each graph during the diffusion process. The more strong edges in the graph, the more original graph information needs to be retained. To this end, V can be used. cd 、V od and V ns The Hadamard product H of the similarity matrices of the three views is used to calculate the number of strong edges. The edges between the elements corresponding to the non-zero values in H are called strong edges, as shown below:
[0068]
[0069] Where Π represents the Hadamard product of the sequence. By using formula (3), if W v (i, j) has a larger value in the similarity matrix of all views, then H ij Will be larger. Otherwise, H ij will be close to 0. If the number of non-zero edges in H is Then set:
[0070]
[0071] That is, the larger A smaller α will be produced, and more strong edges can be preserved in the original graph.
[0072] Step 7): Fuse the centroid distance view improvement map, the contour distance view improvement map, and the non-spatial attribute view improvement map to obtain a joint map. The specific processing method is as follows:
[0073] Since the measurement units of the data in the similarity matrices of different views are different, in order to eliminate the influence of dimension, it is necessary to apply a normalization step to the similarity matrices of all views to ensure that they have the same scale. cd 、V od and V ns The similarity matrix W of the three views v Perform normalization, v = 1, 2, 3, as follows:
[0074]
[0075] By this normalization, the self-similarity in each row is set to The sum of the remaining elements is set to So the whole line and Still works.
[0076] After T iterations, the final joint graph can be obtained by the following formula:
[0077]
[0078] Step 8): Use the normalized graph cut algorithm to cut the joint graph to obtain the final clusters and the spatial distribution pattern of the buildings. Specifically, after performing the normalized graph cut on the joint graph, each building object is assigned a label. Buildings with the same label are considered to be a class. Buildings in the same class have similar characteristics, and their arrangement is the spatial distribution pattern of the building cluster.
[0079] An architectural pattern recognition method in an embodiment of the present application is a recognition method based on multi-view clustering. By mining and analyzing the geometric vector data of the building, taking into account the characteristics of retaining the size, shape and other surface information of the building, compared with other methods, the recognition method in the embodiment of the present application uses two distance measurement methods, namely contour distance and centroid distance, to make full use of the complementary information of spatial structure and non-spatial attributes, and determine the architectural pattern by combining spatial proximity and non-spatial similarity, providing technical support for map synthesis, urban scene semantic segmentation and other aspects, and representing the geographical reality as realistically as possible within the limitation of map scale.
Claims
1. A method for recognizing architectural patterns, characterized in that: The following steps are involved: Step 1): obtaining an original data set, wherein the original data set includes footprint vector data of buildings with different distribution patterns; Step 2): Using the centroids of the buildings in the original data set as vertices, the lines between the vertices as edges, and the Euclidean distances between the vertices as weights, a Delaunay triangulation is constructed to represent the spatial structure of the buildings, and a centroid distance view is obtained; Step 3): Using the centroids of the buildings in the original dataset as vertices, the lines between the vertices as edges, and the closest contour distances between the buildings as weights, a complete graph is constructed to obtain a contour distance view; Step 4): Calculate the size, direction and shape indicators of the buildings in the original data set to obtain a non-spatial attribute view; Step 5): using Gaussian kernel functions to construct a similarity matrix of the centroid distance view, a similarity matrix of the contour distance view, and a similarity matrix of the non-spatial attribute view for the centroid distance view, the contour distance view, and the non-spatial attribute view, respectively; Step 6): using a cross-view diffusion equation to diffuse the similarity matrix of the centroid distance view, the similarity matrix of the contour distance view, and the similarity matrix of the non-spatial attribute view, respectively, to obtain an improved graph of the centroid distance view, an improved graph of the contour distance view, and an improved graph of the non-spatial attribute view; Step 7): fusing the centroid distance view improvement map, the contour distance view improvement map, and the non-spatial attribute view improvement map to obtain a joint map; Step 8): Use the normalized graph cutting algorithm to cut the joint graph to obtain the final clustering and spatial distribution pattern of buildings.
2. A building pattern recognition method according to claim 1, characterized in that: The different distribution modes include a linear distribution mode, a nonlinear distribution mode and a sparse distribution mode.
3. A building pattern recognition method according to claim 2, characterized in that: The linear distribution pattern includes a straight line arrangement and a curved line arrangement; the nonlinear distribution pattern includes a grid arrangement, a grid-like arrangement and an unstructured arrangement.
4. The architectural pattern recognition method according to claim 1, wherein: The size index includes the perimeter, area, and average radius of the building; the direction index is the direction indicated by the long side of the minimum circumscribed rectangle of the building; and the shape index includes compactness, fractal, elongation, concavity, and overlap index.
5. The architectural pattern recognition method according to claim 1, wherein: A Gaussian kernel function based on the k-nearest neighbor method is used to construct the similarity matrix of the centroid distance view, the similarity matrix of the contour distance view, and the similarity matrix of the non-spatial attribute view. The Gaussian kernel function based on the k-nearest neighbor method is shown in the following formula (1). If two points are k-neighbors of each other, the weight of the edge connecting the two points can be calculated by the following formula (1); on the contrary, if the two points are not k-neighbors of each other, the weight of the edge connecting the two points is 0; Among them, x i , x j Represents building objects, which are multidimensional vectors, each dimension represents an attribute of the building, σ is the bandwidth of the Gaussian kernel function, W ij Represents the (i, j)th element of the similarity matrix w, w ji Represents the (j, i)th element of the similarity matrix W, W ij and W ji The values in represent the building object x i and x j similarity.
6. The architectural pattern recognition method according to claim 1, wherein: The method of using the cross-view graph diffusion equation to diffuse the similarity matrix of the centroid distance view, the similarity matrix of the silhouette distance view, and the similarity matrix of the non-spatial attribute view respectively comprises the following steps: For the similarity matrix W of the v-th view v , v = 1, 2, 3, use the cross-view diffusion equation to update W v , the cross-view diffusion equation is: Where D is a diagonal matrix whose diagonal elements satisfy N is The number of rows, p, q represent the pth row and qth column of the matrix respectively, t is the tth iteration number, Represents the result of the similarity matrix of the u-th view (u≠v) after the t-th iteration, represents the initial similarity matrix of the v-th view before iteration, and α represents the ratio of the initial similarity matrix before the iteration of the view to the improved graph obtained after iteration of the other two views at the previous moment.
7. The architectural pattern recognition method according to claim 1, wherein: The fusing of the centroid distance view improvement map, the contour distance view improvement map and the non-spatial attribute view improvement map comprises the following steps: The similarity matrix W of the three views v Perform normalization, v = 1, 2, 3, as shown in the following formula (2): After T iterations, the joint graph can be obtained by the following formula (3):
8. The architectural pattern recognition method according to claim 1, wherein: After the joint graph is cut using the normalized graph cutting algorithm, a label is assigned to each building object. Buildings with the same label are considered as a class. Buildings of the same class have similar characteristics, and their arrangement is the spatial distribution pattern of the building group.
9. A method for recognizing an architectural pattern according to any one of claims 1 to 8, characterized in that: The centroid distance view includes a shortest distance matrix from each point to other points in the Delaunay triangulation graph calculated using the Floyd algorithm.
10. The architectural pattern recognition method according to claim 9, wherein: The silhouette distance view includes a shortest distance matrix from each point to other points in the complete graph calculated using the Floyd algorithm.