An Automatic Modeling Method for Commercial Building Complexes Based on Image Learning

Abstract

This invention discloses an automatic modeling method for commercial building complexes based on image learning, belonging to the field of urban planning. It includes five steps: data acquisition, extraction of core morphological quantification indicators, training of a building complex morphology generation model based on the Pix2Pix algorithm, generation of building complex plan images, and generation of building complex morphology based on OpenCV. The purpose of this invention is to achieve the automatic generation of multiple layout schemes for commercial building complexes at the plot scale within a short time by constructing a database of building complex spatial morphology types and using an intelligent building complex morphology generation method based on Pix2Pix. This solves the problem of long, repetitive drawing cycles for designers and provides technical support for building complex morphology design practice.

Description

Technical Field

[0001] This invention relates to the field of urban planning, and more specifically to an automatic modeling method for commercial building complexes based on image learning. Background Technology

[0002] With economic development, the development of commercial land and the design of commercial building complexes have become an important part of urban design. However, the design process for such complexes is led by architects or planners, who must consider not only rigid conditions such as land use indicators and site functions, but also complex needs such as consumer flow organization, landmark image creation, and ensuring development benefits. Constrained by both the complexity of the site and the subjectivity of design, architects or planners rely on their own experience and logical organization to translate constraints into site form. Without established patterns to follow, the design process is time-consuming and labor-intensive, and the subjective nature of the design logic leads to vastly different results. This results in fragmented construction activities and may even contribute to the disordered development of urban space.

[0003] Existing methods for automatically generating building complex forms fall into two categories: rule-driven and reference-learning-based. The former includes methods based on mathematical models, shape grammars, metacellular automata models, and multi-agent systems. While this approach offers convenience to designers, it has limitations. To balance efficiency and representativeness, the generation process extracts only basic architectural knowledge such as functional topological relationships for rule transfer, and its application is primarily limited to residential buildings with strong constraints. This technology is not well-suited for generating commercial building complexes. The latter includes methods based on decision trees, support vector machines, Bayesian classification, reinforcement learning, and deep learning. This approach primarily targets building plans, facades, and individual 3D structures, lacking application in the more complex design of commercial building complexes. Furthermore, data processing relies on manual statistical analysis, limiting computational resources when dealing with large-scale urban data. Therefore, current technologies cannot achieve rapid, multi-scheme automatic generation of commercial building complexes. Summary of the Invention

[0004] The purpose of this invention is to extract quantitative indicators of commercial land parcels and building forms based on planning discipline theories, and then use clustering algorithms to classify typical building form types. By having computers learn from a large number of real building cluster sample cases of typical form categories, the system automatically analyzes the mapping rules between land parcel boundaries and building layout forms, thereby achieving intelligent scheme generation and assisting planners in making design decisions for commercial land parcels.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] An automatic modeling method for commercial building complexes based on image learning includes the following steps:

[0007] Acquire and clean information data on commercial buildings in the target area; establish core indicators of the form of commercial building complexes at the plot scale based on plot morphology, two-dimensional and three-dimensional building data; extract data composition I (I1~I2) from the building data corresponding to the core indicators. n );

[0008] Based on the core indicator library I (I1~I n A two-step clustering method is used to classify and merge morphologically similar samples, constructing a building cluster category library C(C1~C2). n ); For the core indicator library I (I1~I n The indicators in the database are used to divide the data into intervals and the attributes are then assigned to the building group category library C (C1~C2). n Define the network structure, iteratively train the Pix2pix deep convolutional neural network model, and construct a generative algorithm model library G(G1~G) for generating planar images of building complexes. n );

[0009] Based on the building group category library C(C1~C n Determine the building group category of the design plot outline and input the design plot outline data into the algorithm model library G(G1~G n Generate a plan view of the building complex on the plot;

[0010] Obtain land boundary vector data and building boundary vector data containing building height information from the site's building complex plan; construct a 3D model of the target area's commercial building complex design scheme based on the land boundary vector data and building boundary vector data.

[0011] Furthermore, the acquisition and cleaning of information data on commercial buildings in the target area includes the following steps:

[0012] Collect information data on commercial buildings in the target area, including building function data, building location data, and building height data;

[0013] Data cleaning: uniformly screen land use types B1 and plots with an area of ​​0.01-10 hectares; merge adjacent building elements with the same height; batch delete fragmented building volumes with an area of ​​less than 100 square meters and plots with a building density of less than 10%; and trim building volumes that cross plot boundaries.

[0014] Furthermore, the core indicators for establishing the form of commercial building complexes at the site scale based on site morphology, two-dimensional and three-dimensional architecture include the following steps:

[0015] In terms of land parcel morphology quantification, the land parcel perimeter (PER), land parcel area (BLA), and land parcel shape index (BLS) are selected as the core indicators for quantifying land parcel characteristics.

[0016] Among them, the plot perimeter (PER) and plot area (BLA) were statistically obtained using a geographic information platform as the data cleaning platform. The plot shape index (BLS) is the ratio of the plot perimeter to the perimeter of a square of the same area, and the calculation formula is as follows:

[0017]

[0018] In terms of two-dimensional morphological quantification, building density (BD), average building base area (ABA), building base area difference (DBA), average building base shape index (ASH), building base shape difference (DSH), number of buildings (BN), and dispersion (DR) are selected as metrics to characterize the two-dimensional planar morphology of building groups.

[0019] Wherein, building density BD is the ratio of the sum of the projected areas of buildings to the area of ​​the building site, and the calculation formula is as follows:

[0020]

[0021] in It is the sum of the building footprint areas within the plot;

[0022] The average building footprint area ABA is the average building footprint area across all plots of land, calculated using the following formula:

[0023]

[0024] The Building Footprint Area Difference (DBA) is the standard deviation of the footprint area of ​​all plots of land, calculated using the following formula:

[0025]

[0026] The dispersion ratio (DR) is the ratio of the product of the number of buildings to the product of the differences in land area and building volume. The calculation formula is as follows:

[0027]

[0028] DTBA refers to the building volume variation degree;

[0029] The Average Building Footprint Shape Index (ASH) is the average of the shape indices of all building footprints within the site, calculated using the following formula:

[0030]

[0031] in It is the building base shape index;

[0032] The building footprint shape variation degree DSH is the standard deviation of any building footprint shape index within a site, and is calculated using the following formula:

[0033]

[0034] In terms of three-dimensional morphology quantification, the floor area ratio (FAR), average building capacity (ATBA), building capacity difference (DTBA), average building height (ABH), and staggered height (DBH) are selected as metrics to characterize the three-dimensional morphology of the building complex.

[0035] Floor Area Ratio (FAR) refers to the ratio of the total building area within a plot to the plot area, and the calculation formula is as follows:

[0036]

[0037] in This represents the sum of the building footprint areas within the plot; where TBA is the building volume.

[0038] The Average Building Volume (ATBA) is the average volume of all buildings within the site, calculated using the following formula:

[0039]

[0040] The Building Volume Difference (DTBA) is the standard deviation of the volume of any building within the site, calculated using the following formula:

[0041]

[0042] The average building height ABH is the average height of all buildings within the plot, calculated using the following formula:

[0043]

[0044] The staggered height (DBH) is the standard deviation of the height of any building within the site, calculated using the following formula:

[0045]

[0046] Among them BH i This refers to the building's height.

[0047] Furthermore, the data structure I (I1~I) extracted from the building data corresponding to the core indicators constitutes the data. n This includes the following steps:

[0048] The building data is transformed by z-score standardization so that the transformed building data conforms to a standard normal distribution, i.e., the mean is 0 and the variance is 1.

[0049] To determine whether the building data is suitable for principal component analysis, the KMO and Bartlett's sphericity tests are used. A KMO value > 0.5 and a significance p-value < 0.001 indicate that the principal component analysis results are valid; a KMO value < 0.5 indicates that the building data is not suitable for principal component analysis.

[0050] In the total variance interpretation table, the core indicators corresponding to building data with eigenvalues ​​greater than 1 and cumulative percentages higher than 70% are selected as principal components. Data corresponding to the core indicators in the principal components are extracted to construct a representative core indicator library I (I1~I2). n ).

[0051] Furthermore, the core indicator library I (I1~I n A two-step clustering method is used to classify and merge morphologically similar samples, constructing a building cluster category library C(C1~C2). n This includes the following steps:

[0052] Pre-clustering: The core indicator library I (I1~I2) is divided into clusters using a sequential approach. n The data is divided into several subcategories, depending on the core indicator library I (I1~I2). n The data is categorized into one large class; the core indicator library I (I1~I) is read in. n After obtaining a data point, the system determines whether the sample should be derived into a new class or merged into an existing subclass based on its degree of familiarity. This process is repeated until L classes are formed.

[0053] Based on the pre-clustering, subclasses are merged according to their degree of affinity, ultimately forming class L'.

[0054] Furthermore, the core indicator library I (I1~I n The indicators in the database are used to divide the data into intervals and the attributes are then assigned to the building group category library C (C1~C2). n This includes the following steps:

[0055] Using the natural discontinuity grading method, the core indicator library I (I1~I) is divided into three level intervals: High / Large, Medium / Middle, and Low / Small. n The indicators in the table are divided into intervals;

[0056] Through the core indicator library I (I1~I n The core indicator in the natural discontinuity grading method describes different interval dimensions in the building group category library C (C1~C2). n ) Attributes are attached.

[0057] Furthermore, the defined network structure uses a Pix2pix deep convolutional neural network model for iterative training to construct a generative algorithm model library G(G1~G) for generating planar images of building complexes. n This includes the following steps:

[0058] For the building group category library C(C1~C n The data in the image is converted into an image format to obtain a sample library of building cluster categories S(S1~S2). n );

[0059] The network structure is defined using a Pix2pix deep convolutional neural network model. The generator in the model is based on the U-Net architecture, and the discriminator uses the PatchGAN classifier. The formula is as follows:

[0060]

[0061]

[0062]

[0063] The generator operates by generating images with similar feature distributions based on the characteristic patterns of the land parcel outline map and the real building texture map samples. The discriminator operates by taking the land parcel boundary and the generated image or the real building texture map as input to form a new sample pair, determining whether this sample pair is a correct mapping from the land parcel boundary to the real building complex, and outputting a probability value to identify the authenticity of the generated image. The network structure adopts the Patch GAN idea, dividing the generated result into multiple fixed-size patch images and inputting them into the discriminator network.

[0064] For the building cluster category sample library S(S1~S n All categories in the model are trained using gradient descent. The fluctuations in the generator and discriminator loss functions during training are observed for each parameter. The learning rate and iteration count are adjusted and optimized. The optimal values ​​for each category are determined by comparing training time and generation results. Finally, a generative algorithm model library G(G1~G2) for generating planar images of building complexes is constructed. n ); where the learning rate is the tuning parameter in the optimization algorithm, and the number of iterations is the number of loops during the iterative operation;

[0065] The gradient descent formula is: ;

[0066] Where η is the learning rate, i represents the i-th data point, and the weight parameter w represents the magnitude of change in each iteration.

[0067] Further, the process of obtaining plot boundary vector data and building boundary vector data containing building height information from the plot building complex plan view, and constructing a three-dimensional model of the plot building complex design scheme for commercial buildings in the target area based on the plot boundary vector data and building boundary vector data, includes the following steps:

[0068] Land parcel boundary extraction: The grayscale plan view of the building complex is read, and a threshold is set to classify the colored areas of the image into a single color. The area represented by the color is the location of the land parcel. The outlines of objects in the image are identified as land parcel boundaries and the land parcel data image is saved. The land parcel data image contains the outline information of the corresponding land parcel. By inversely mapping the pixels in the land parcel data image to the actual vector coordinates, the real latitude and longitude corresponding to any position in the image can be obtained. Based on this, the land parcel boundary vector data can be obtained.

[0069] Building outline extraction: Read the plan view of the building complex on the site, then extract buildings of different colors by analyzing the different color values ​​in the three channels. Find buildings of the same color and convert other buildings to white. Then convert the image to grayscale before detecting the building outlines. Encapsulate each detected curve into a polygon, and delete smaller polygons to reduce line impurities. At the same time, use the approxPolyDP function to approximate the curves with polygons. Finally, reverse map the improved building polygon outlines back to the real position, and determine the building height according to the correspondence between color values ​​and building floors. This gives us the building vector data including the building height.

[0070] Input the site boundary vector data and building vector data including building height into the 3D interactive display design. Preparation By stretching the building's floor information, a three-dimensional model of the building complex design for the site can be obtained.

[0071] Secondly, the present invention also provides an automatic modeling system for commercial building complexes based on image learning, comprising the following modules:

[0072] Data Acquisition and Cleaning Module: Acquires and cleans information data of commercial buildings in the target area;

[0073] The morphological quantification core indicator extraction module: establishes core indicators of the morphology of commercial building complexes at the site scale based on plot morphology, 2D and 3D building data; extracts data composition I (I1~I2) from building data corresponding to the core indicators. n );

[0074] Building morphology generation algorithm model training module: based on core indicator library I (I1~I n A two-step clustering method is used to classify and merge morphologically similar samples, constructing a building cluster category library C(C1~C2). n); For the core indicator library I (I1~I n The indicators in the database are used to divide the data into intervals and the attributes are then assigned to the building group category library C (C1~C2). n Define the network structure, iteratively train the Pix2pix deep convolutional neural network model, and construct a generative algorithm model library G(G1~G) for generating planar images of building complexes. n );

[0075] Building complex plan image generation module: based on building complex category library C(C1~C1) n Determine the building group category of the design plot outline and input the design plot outline data into the algorithm model library G(G1~G n Generate a plan view of the building complex on the plot;

[0076] The 3D visualization generation module for building complex morphology: obtains plot boundary vector data and building boundary vector data containing building height information from the plot building complex plan; and constructs a 3D model of the plot building complex design scheme for commercial buildings in the target area based on the plot boundary vector data and building boundary vector data.

[0077] Thirdly, the present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the memory stores the computer program capable of running on the processor, and when the processor loads and executes the computer program, it employs a large-scale automatic modeling method for commercial building complexes based on intelligent image learning as described above.

[0078] The beneficial effects of this invention are:

[0079] By quantifying land parcels and building forms, and using quantitative tools to understand the morphological characteristics of building clusters at the land parcel scale, we can enrich the research on the morphological types of commercial building clusters and provide target guidance and learning data support for the intelligent generation design of building cluster forms.

[0080] An algorithmic model for generating planar images of commercial land parcels and building complexes was built using the pix2pix deep convolutional neural network model. This algorithm can generate multiple schemes in a short time, helping designers reduce repetitive drawing work compared to traditional planning and design processes, and providing inspiration and decision support during the conceptual design phase.

[0081] By leveraging OpenCV to vectorize machine learning-generated images, a technical method was established to transform two-dimensional planar schemes learned from images into three-dimensional representations, providing technical support for the intelligent generation and design of building complex forms. Furthermore, a three-dimensional sand table of AI-generated schemes was created using a geographic information platform, providing a visualization tool for the planning and design practice of building complex forms. Attached Figure Description

[0082] The invention will now be further described with reference to the accompanying drawings.

[0083] Figure 1 This is a flowchart of a method for large-scale automatic modeling of commercial building complexes based on intelligent image learning, according to the present invention.

[0084] Figure 2 A graph showing the results of the KMO and Bartlett's test of sphericity.

[0085] Figure 3 Principal component analysis - explained variance table;

[0086] Figure 4 Principal component analysis - correlation interpretation table;

[0087] Figure 5 A chart showing the average levels of the morphological indicators C1-C4 for the architectural complex.

[0088] Figure 6 This is a chart showing the interval division of the core indicators I1-I4;

[0089] Figure 7 The C1-C4 building plan images and their 3D sand table models generated for the algorithm model G. Detailed Implementation

[0090] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0091] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. The technical solution of the present invention will be described in detail below with reference to a case study of an eco-technology city in a certain city and the accompanying drawings.

[0092] The flowchart of this method is as follows Figure 1 The following are the operation steps for the modules shown:

[0093] S1: The data acquisition and cleaning methods are as follows.

[0094] Step S1-1: Data Collection. In this embodiment of the invention, an open data platform and local planning department collection methods are comprehensively utilized to obtain building data and land use data for the central urban area. The data format can be Shapefile, DWG, DXF, etc. This embodiment uses Shapefile as the data format for description. It includes building information and land use information, wherein the building information is the building outline and the area it encloses, the area includes the building's floor area and the number of building floors; the land use information is the plot outline and the area it encloses, the area includes the land use nature and the land area.

[0095] Step S1-2: Data Cleaning. In this embodiment of the invention, a geographic information platform is used as the data cleaning platform. Plots with land use type B1 (commercial land) and area between 0.01 and 10 hectares in the data obtained in step S1-1 are uniformly screened. Building elements with adjacent heights are merged. Fragmented building volumes with an area less than 100 square meters and plots with a building density less than 10% are deleted in batches. Building volumes crossing plot boundaries are trimmed.

[0096] S2: The method for extracting core indicators for morphological quantification is as follows:

[0097] Step S2-1: Constructing Indicators. In this embodiment of the invention, the data cleaned in step S1-2 is calculated to obtain 15 indicators, including plot perimeter PER, plot area BLA, plot shape index BLS, building density BD, average building base area ABA, building base area variation degree DBA, average building base shape index ASH, building base shape variation degree DSH, number of buildings BN, dispersion degree DR, selected floor area ratio FAR, average building capacity ATBA, building capacity variation degree DTBA, average building height ABH, and staggered height DBH.

[0098] Among them, PER and BLA were obtained by using a geographic information platform as the data cleaning platform.

[0099] The Block Shape Index (BLS) is the ratio of the perimeter of a plot of land to the perimeter of a square of the same area. The calculation formula is as follows:

[0100]

[0101] Where PER is the perimeter of the plot; BLA is the area of ​​the plot.

[0102] The building density BD is the ratio of the sum of the projected areas of buildings to the building site area, and the calculation formula is as follows:

[0103]

[0104] in BLA is the sum of the building footprints within the plot; BLA is the plot area.

[0105] The average building footprint area ABA is the average of the building footprint areas across all plots of land, calculated using the following formula:

[0106]

[0107] in BN represents the sum of the building footprints within the plot; BN represents the number of buildings.

[0108] The Building Base Area Difference (DBA) is the standard deviation of the base area of ​​all plots, calculated using the following formula:

[0109]

[0110] in ABA represents the sum of the building footprint areas within the plot; ABA represents the average building footprint area.

[0111] The dispersion degree DR is the ratio of the product of the number of buildings and the difference in land area and building volume, and the calculation formula is as follows:

[0112]

[0113] Where DTBA is the building volume variation; BLA is the land area; and BN is the number of buildings within the plot.

[0114] The Average Building Base Shape Index (ASH) is the average of the shape indices of all building bases within the site, calculated using the following formula:

[0115]

[0116] in 1 is the building footprint shape index; ASH is the average footprint shape index of buildings within the plot; BN is the number of buildings within the plot.

[0117] The building footprint shape difference degree DSH is the standard deviation of any building footprint shape index within the plot, and is calculated using the following formula:

[0118]

[0119] Among them SH i 1 is the building footprint shape index; ASH is the average footprint shape index of buildings within the plot; BN is the number of buildings within the plot.

[0120] The Floor Area Ratio (FAR) refers to the ratio of the total building area within a plot to the plot area, and the calculation formula is as follows:

[0121]

[0122] in BLA is the sum of the building footprints within the plot; BLA is the plot area.

[0123] The Average Building Capacity (ATBA) is the average volume of all buildings within the site, calculated using the following formula:

[0124]

[0125] Where TBA is the building volume; ATBA is the average building volume within the plot; and BN is the number of buildings within the plot.

[0126] The Building Volume Variation DTBA is the standard deviation of the volume of any building within the plot, and is calculated using the following formula:

[0127]

[0128] Where TBA is the building volume; ATBA is the average building volume within the plot; and BN is the number of buildings within the plot.

[0129] The average building height ABH is the average height of all buildings within the plot, calculated using the following formula:

[0130]

[0131] in BN is the sum of the heights of all buildings within the plot, and BN is the number of buildings within the plot.

[0132] The stagger degree DBH is the standard deviation of any building height within the plot, and the calculation formula is as follows:

[0133]

[0134] Among them BH i ABH represents the building height; BN represents the average building height within the plot; and BN represents the number of buildings within the plot.

[0135] Step S2-2: Data standardization. In this embodiment of the invention, the data in step S2-1 is transformed by z-score standardization (standard deviation standardization) so that the transformed data conforms to a standard normal distribution, i.e., the mean is 0 and the variance is 1.

[0136] Step S2-3: In the embodiments of the present invention, the KMO and Bartlett's sphericity tests are used to input the data of each index in step S2-1 into the SPSS software platform to obtain the results. Figure 2A KMO value > 0.5 and a significance p-value < 0.001 indicate that the principal component analysis results are valid; a KMO value < 0.5 indicates that the data is not suitable for principal component analysis. The tests revealed that the KMO test coefficients for all functional samples were greater than 0.5, and the significance of the Barlett test values ​​was less than 0.001, making them suitable for principal component analysis.

[0137] Step S2-4: In embodiments of the present invention, the variance interpretation table ( Figure 3 The data shows four components with eigenvalues ​​greater than 1, whose cumulative variance contribution exceeds 70%, sufficient to quantitatively reflect the morphological characteristics of the building complex. Based on whether the values ​​are greater than 0.8, component 1 is strongly correlated with floor area ratio (FAR), plot perimeter (PER), and plot area (BLA); component 2 is strongly correlated with staggered elevation (D); component 3 is strongly correlated with average building height (ABH), staggered elevation (DBH), and number of buildings (BN); and component 4 is strongly correlated with building base shape difference (DSH) and average building shape coefficient (ASH). Figure 4 Therefore, the most relevant indicators for B1 type building complex morphology, namely FAR, ABH, DR, and DSH, are constructed as the core indicator library I (I1~I4). I1 is FAR, I2 is ABH, I3 is DR, and I4 is DSH.

[0138] S3: The training method for the building complex morphology generation algorithm model is as follows:

[0139] Step S3-1: In the embodiment of the present invention, based on the four extracted principal component factors, and using a two-step clustering method, four clustering results C1 to C4 are obtained through calculation and combined into a building group category library C(C1 to C4).

[0140] ( Figure 5 )

[0141] Step S3-2: In an embodiment of the present invention, the natural discontinuity grading method is used in the SPSS software platform to divide the indicators in the core indicator library I (I1~I4) into intervals according to three level intervals: high / Large, medium / Middle, and low / Small. Figure 6 )

[0142] Step S3-3: In the embodiment of the present invention, the core indicators I1-I4 in the core indicator library I (I1-I4) of step S2-4 and the different interval dimensions in S3-2 are used to add attributes to the building types C1-C4 in the building group category library C (C1-C4), such as "low FAR-low DR-medium ABH-low DSH"; the morphological indicators are divided into high, medium and low levels, and the number of indicators exceeding 70% of the total sample size is used as the evaluation criterion.

[0143] Therefore, the characteristics of the C1 group are "low plot perimeter PER, low average building footprint area ABA, low average building capacity ATBA, low average building height ABH, low floor area ratio FAR, low building footprint area variation DBA, low building capacity variation DTBA, and medium average building shape coefficient ASH".

[0144] Type C2 group is characterized by "low plot perimeter (PER), medium building density (BD), high density (DBH), and high floor area ratio (FAR)".

[0145] Type C3 group is characterized by “low average building volume (ATBA), low average building height (ABH), low floor area ratio (FAR), medium average building shape factor (ASH), medium building base shape difference (DSH), and high plot perimeter (PER).

[0146] The characteristics of the C4 group are: "low average building volume (ATBA), low average building height (ABH), low building footprint area variation (DBA), medium building density (BD), and high average building shape factor (ASH)".

[0147] Step S3-4: Image format conversion. This embodiment of the invention uses Python and Jupyter Notebook as tools to illustrate the process. Shapefile format files of the building group category library C (C1-C4) are labeled with outline tags and color data tags, integrated into a graphical visualization information platform, and output as 256*256 pixel JPG format images containing building height information, where the building height information is a color block containing the number of building floors with specified RGB values. The output JPG images are then used to construct the building group category sample library S (S1-S4).

[0148] Step S3-5: Define the network structure. In this embodiment of the invention, the open-source programming platform Anaconda, the deep learning framework TensorFlow, the Python language, and the Jupyter Notebook tool are used for illustration. A Pix2pix deep convolutional neural network model is built, in which the generator is based on the "U-Net" architecture, and the discriminator uses the "PatchGAN" classifier. The formula is as follows:

[0149]

[0150]

[0151]

[0152] Steps S3-6: In the embodiments of the present invention, gradient descent is used to train all categories in building group category C (C1 to C4), observe the fluctuation of the loss function of generator and discriminator corresponding to each parameter during training, adjust and optimize the learning rate and iteration number parameters, compare the training time and generation results to determine the optimal value of each category, and construct the generation algorithm model library G (G1 to G4) for generating planar images of building groups.

[0153] The learning rate is a tuning parameter in the optimization algorithm, which determines the step size in each iteration to minimize the loss function; the number of iterations is the number of iterations during the iterative operation.

[0154]

[0155] Number of times; the gradient descent formula is as follows:

[0156] Where η is the learning rate, and i represents the i-th data point. The weight parameter w changes by a certain amount in each iteration.

[0157] Step S4: The method for generating the planar image of the building complex is as follows.

[0158] In an embodiment of the present invention, a shapefile image format is used as an example. The shapefile file representing the design plot outline of the building complex category C (C1-C4) is input into the corresponding algorithm model library G (G1-G4) to generate a JPG file representing the plan view of the building complex on that plot. The plot outline file is a pure black area, 1:2000 scale, 100dpi resolution, 300mm*300mm JPG file.

[0159] Step S5: The method for generating a 3D visualization of the building complex is as follows.

[0160] Step S5-1: Extracting the land parcel boundaries. The image generated in step S4 is read in grayscale. A threshold is set to classify colored areas in the image into a single color; this color represents the location of the land parcel. The OpenCV `findContours` function is used to identify the contours of objects in the image as land parcel boundaries, and the resulting land parcel data image is saved. This image should contain the contour information corresponding to the land parcel, such as minimum longitude, minimum latitude, maximum longitude, and maximum latitude (Min-x, Min-y, Max-x, Max-y). By inversely mapping the pixels in the image to the actual vector coordinates, the true latitude and longitude corresponding to any location in the image can be obtained. Based on this, a vector Shapefile format file of the land parcel can be obtained.

[0161] Step S5-2: Building outline extraction. The image is read in RGB format. Buildings of different colors are extracted based on their color values ​​across the three channels. Buildings of the same color are identified, and other buildings are converted to white. The image is then converted to grayscale. The `findContours` function is used to detect building outlines. Each detected curve is encapsulated as a polygon, and smaller polygons are removed to reduce line clutter. The `approxPolyDP` function is used to approximate the curves as closely as possible with polygons. Finally, the improved building polygon outlines are mapped back to their true locations. The building height is determined based on the correspondence between color values ​​and building floor numbers, thus obtaining building vector data including building height.

[0162] Step S5-3: Input the obtained plot shapefile and building vector data into the 3D geographic information platform, and stretch the model based on the number of building floors with a floor height of 3m to obtain the 3D model of the building complex design scheme for the plot. Figure 7 )

[0163] This application also discloses an automatic modeling system for commercial building complexes based on image learning, including the following modules:

[0164] Data Acquisition and Cleaning Module: Acquires and cleans information data of commercial buildings in the target area;

[0165] The morphological quantification core indicator extraction module: establishes core indicators of the morphology of commercial building complexes at the site scale based on plot morphology, 2D and 3D building data; extracts data composition I (I1~I2) from building data corresponding to the core indicators. n );

[0166] Building morphology generation algorithm model training module: based on core indicator library I (I1~I n A two-step clustering method is used to classify and merge morphologically similar samples, constructing a building cluster category library C(C1~C2). n ); For the core indicator library I (I1~I n The indicators in the database are used to divide the data into intervals and the attributes are then assigned to the building group category library C (C1~C2). n Define the network structure, iteratively train the Pix2pix deep convolutional neural network model, and construct a generative algorithm model library G(G1~G) for generating planar images of building complexes. n );

[0167] Building complex plan image generation module: based on building complex category library C(C1~C1) n Determine the building group category of the design plot outline and input the design plot outline data into the algorithm model library G(G1~G n Generate a plan view of the building complex on the plot;

[0168] The 3D visualization generation module for building complex morphology: obtains plot boundary vector data and building boundary vector data containing building height information from the plot building complex plan; and constructs a 3D model of the plot building complex design scheme for commercial buildings in the target area based on the plot boundary vector data and building boundary vector data.

[0169] This application also discloses a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it employs any of the large-scale automatic modeling methods for commercial building complexes based on intelligent image learning described in the above embodiments.

[0170] The terminal device can be a computer device such as a desktop computer, a laptop computer, or a cloud server. The terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices, and buses.

[0171] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it in this regard.

[0172] The memory can be an internal storage unit of the terminal device, such as a hard disk or RAM of the terminal device, or an external storage device of the terminal device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) equipped on the terminal device. Furthermore, the memory can be a combination of internal storage units and external storage devices of the terminal device. The memory is used to store computer programs and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0173] In this terminal device, any one of the large-scale automatic modeling methods for commercial building complexes based on intelligent image learning in the above embodiments can be stored in the memory of the terminal device and loaded and executed on the processor of the terminal device for convenient use.

[0174] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for automatic modeling of commercial building complexes based on image learning, characterized in that, Includes the following steps: Acquire and clean up information data on commercial buildings in the target area; establish core indicators of the form of commercial building complexes at the plot scale based on plot morphology, two-dimensional and three-dimensional building structures; Data composition I (from building data) consists of core indicators extracted from building data. ~ ); Based on the core indicator library I ( ~ A two-step clustering method is used to classify and merge morphologically similar samples, constructing a building cluster category library C( ~ ); for the core indicator library I ( ~ The indicators in ) are used to divide the data into intervals and the attributes are attached to the building group category library C ( ~ Define the network structure, iteratively train the Pix2pix deep convolutional neural network model, and build a generative algorithm model library G( for generating planar images of building complexes). ~ ); Based on the building group category library C ( ~ Determine the building complex category of the design site outline and input the design site outline data into the algorithm model library G( ~ Generate a plan view of the building complex on the plot; Obtain land boundary vector data and building boundary vector data containing building height information from the site plan of the building complex; construct a 3D model of the building complex design scheme for commercial buildings in the target area based on the land boundary vector data and building boundary vector data; The core indicators for establishing the form of commercial building complexes at the site scale based on site morphology, two-dimensional and three-dimensional building dimensions include the following steps: In terms of land parcel morphology quantification, the land parcel perimeter (PER), land parcel area (BLA), and land parcel shape index (BLS) are selected as the core indicators for quantifying land parcel characteristics. Among them, the plot perimeter (PER) and plot area (BLA) were statistically obtained using a geographic information platform as the data cleaning platform. The plot shape index (BLS) is the ratio of the plot perimeter to the perimeter of a square of the same area, and the calculation formula is as follows: In terms of two-dimensional morphological quantification, building density (BD), average building base area (ABA), building base area difference (DBA), average building base shape index (ASH), building base shape difference (DSH), number of buildings (BN), and dispersion (DR) are selected as metrics to characterize the two-dimensional planar morphology of building groups. Wherein, building density BD is the ratio of the sum of the projected areas of buildings to the area of ​​the building site, and the calculation formula is as follows: in It is the sum of the building footprint areas within the plot; The average building footprint area ABA is the average building footprint area across all plots of land, calculated using the following formula: The Building Footprint Area Difference (DBA) is the standard deviation of the footprint area of ​​all plots of land, calculated using the following formula: The dispersion ratio (DR) is the ratio of the product of the number of buildings to the product of the differences in land area and building volume. The calculation formula is as follows: DTBA refers to the building volume variation degree; The Average Building Footprint Shape Index (ASH) is the average of the shape indices of all building footprints within the site, calculated using the following formula: in It is the building base shape index; The building footprint shape variation degree DSH is the standard deviation of any building footprint shape index within a site, and is calculated using the following formula: In terms of three-dimensional morphology quantification, the floor area ratio (FAR), average building capacity (ATBA), building capacity difference (DTBA), average building height (ABH), and staggered height (DBH) are selected as metrics to characterize the three-dimensional morphology of the building complex. Floor Area Ratio (FAR) refers to the ratio of the total building area within a plot to the plot area, calculated using the following formula: in This represents the sum of the building footprint areas within the plot; where TBA is the building volume. The Average Building Volume (ATBA) is the average volume of all buildings within the site, calculated using the following formula: The Building Volume Difference (DTBA) is the standard deviation of the volume of any building within the site, calculated using the following formula: The average building height ABH is the average height of all buildings within the plot, calculated using the following formula: The staggered height (DBH) is the standard deviation of the height of any building within the site, calculated using the following formula: Where BHi is the building height.

2. The automatic modeling method for commercial building complexes based on image learning according to claim 1, characterized in that, The process of acquiring and cleaning up information data on commercial buildings in the target area includes the following steps: Collect information data on commercial buildings in the target area, including building function data, building location data, and building height data; Data cleaning: uniformly screen target land use type B1 and plots with an area of ​​0.01-10 hectares; merge adjacent building elements with the same height, batch delete fragmented building volumes with an area of ​​less than 100 square meters and plots with a building density of less than 10%, and trim building volumes that cross plot boundaries.

3. The automatic modeling method for commercial building complexes based on image learning according to claim 1, characterized in that, The data structure I is composed of extracting corresponding core indicators from building data. ~ This includes the following steps: The building data is transformed by z-score standardization so that the transformed building data conforms to a standard normal distribution, i.e., the mean is 0 and the variance is 1. To determine whether the building data is suitable for principal component analysis, the KMO and Bartlett's sphericity tests are used. A KMO value > 0.5 and a significance p-value < 0.001 indicate that the principal component analysis results are valid; a KMO value < 0.5 indicates that the building data is not suitable for principal component analysis. In the total variance interpretation table, the core indicators corresponding to building data with eigenvalues ​​greater than 1 and cumulative percentages higher than 70% are selected as principal components. Data corresponding to the core indicators in the principal components are extracted to construct a representative core indicator library I. ~ ).

4. The automatic modeling method for commercial building complexes based on image learning according to claim 3, characterized in that, The core indicator library I ( ~ A two-step clustering method is used to classify and merge morphologically similar samples, constructing a building cluster category library C( ~ This includes the following steps: Pre-clustering: The core indicator library I (…) is divided into clusters using a sequential approach. ~ The data is divided into several subcategories, depending on the core indicator library I ( ~ The data is categorized into one large category; the core indicator library I is read in. ~ After obtaining a data point, the system determines whether the sample should be derived into a new class or merged into an existing subclass based on its degree of familiarity. This process is repeated until L classes are formed. Based on the pre-clustering, subclasses are merged according to their degree of affinity, ultimately forming class L'.

5. The automatic modeling method for commercial building complexes based on image learning according to claim 4, characterized in that, The core indicator library I ( ~ The indicators in ) are used to divide the data into intervals and the attributes are attached to the building group category library C ( ~ This includes the following steps: Using the natural discontinuity grading method, the core indicator library I is divided into three level intervals: High / Large, Medium / Middle, and Low / Small. ~ The indicators in the table are divided into intervals; Through the core indicator library I ( ~ The core indicator in the natural discontinuity grading method describes different interval dimensions in the building group category library C. ~ ) Attributes are attached.

6. The automatic modeling method for commercial building complexes based on image learning according to claim 1, characterized in that, The defined network structure uses a Pix2pix deep convolutional neural network model for iterative training to construct a generative algorithm model library G for generating planar images of building complexes. ~ This includes the following steps: For building group category library C ( ~ The data in ) is converted into image format to obtain the building group category sample library S( ~ ); The network structure is defined using a Pix2pix deep convolutional neural network model. The generator in the model is based on the U-Net architecture, and the discriminator uses the PatchGAN classifier. The formula is as follows: The generator operates by generating images with similar feature distributions based on the characteristic patterns of the land parcel outline map and the real building texture map samples. The discriminator operates by taking the land parcel boundary and the generated image or the real building texture map as input to form a new sample pair, determining whether this sample pair is a correct mapping from the land parcel boundary to the real building complex, and outputting a probability value to identify the authenticity of the generated image. The network structure adopts the Patch GAN idea, dividing the generated result into multiple fixed-size patch images and inputting them into the discriminator network. For the building cluster category sample library S( ~ All categories in the model are trained using gradient descent. The fluctuations in the generator and discriminator loss functions during training are observed for each parameter. The learning rate and iteration count are adjusted and optimized. The optimal values ​​for each category are determined by comparing training time and generation results. Finally, a generative algorithm model library G([…]) for generating planar images of building complexes is constructed. ~ ); where the learning rate is the tuning parameter in the optimization algorithm, and the number of iterations is the number of loops during the iterative operation; The gradient descent formula is: ; Where η is the learning rate, i represents the i-th data point, and the weight parameter w represents the magnitude of change in each iteration.

7. The automatic modeling method for commercial building complexes based on image learning according to claim 1, characterized in that... The process of obtaining land boundary vector data and building boundary vector data containing building height information from the site's building complex plan, and constructing a 3D model of the target area's commercial building complex design scheme based on the land boundary vector data and building boundary vector data, includes the following steps: Land parcel boundary extraction: The grayscale plan view of the building complex is read, and a threshold is set to classify the colored areas of the image into a single color. The area represented by the color is the location of the land parcel. The outlines of objects in the image are identified as land parcel boundaries and the land parcel data image is saved. The land parcel data image contains the outline information of the corresponding land parcel. By inversely mapping the pixels in the land parcel data image to the actual vector coordinates, the real latitude and longitude corresponding to any position in the image can be obtained. Based on this, the land parcel boundary vector data can be obtained. Building outline extraction: Read the site's building layout plan, then extract buildings of different colors based on their color values ​​across three channels. Identify buildings of the same color and convert other buildings to white. Then, convert the image to grayscale before detecting the building outlines. Encapsulate each detected curve into a polygon, removing smaller polygons to reduce line clutter. Simultaneously, use the approxPolyDP function to approximate the curves with polygons. Finally, reverse-map the improved building polygon outlines back to their true locations. Based on the correspondence between color values ​​and building floors, determine the building height to obtain building vector data including the building height. Input the site boundary vector data and building vector data including building height into the 3D interactive display design. Preparation By stretching the building's floor information, a three-dimensional model of the building complex design for the site can be obtained.

8. An automatic modeling system for commercial building complexes based on image learning, characterized in that... It includes the following modules: Data Acquisition and Cleaning Module: Acquires and cleans information data of commercial buildings in the target area; Morphological Quantification Core Indicator Extraction Module: Establishes core indicators of the morphology of commercial building complexes at the plot scale based on plot morphology, 2D and 3D building structures; Data composition I (from building data) consists of core indicators extracted from building data. ~ ); Building morphology generation algorithm model training module: based on core indicator library I ( ~ A two-step clustering method is used to classify and merge morphologically similar samples, constructing a building cluster category library C( ~ ); for the core indicator library I ( ~ The indicators in ) are used to divide the data into intervals and the attributes are attached to the building group category library C ( ~ Define the network structure, iteratively train the Pix2pix deep convolutional neural network model, and build a generative algorithm model library G( for generating planar images of building complexes). ~ ); Module for generating planar images of building complexes: based on a building complex category library C ( ~ Determine the building complex category of the design site outline and input the design site outline data into the algorithm model library G( ~ Generate a plan view of the building complex on the plot; The 3D visualization generation module for building complex morphology: obtains plot boundary vector data and building boundary vector data containing building height information from the plot building complex plan; and constructs a 3D model of the plot building complex design scheme for commercial buildings in the target area based on the plot boundary vector data and building boundary vector data. The core indicators for establishing the form of commercial building complexes at the site scale based on site morphology, two-dimensional and three-dimensional building dimensions include the following steps: In terms of land parcel morphology quantification, the land parcel perimeter (PER), land parcel area (BLA), and land parcel shape index (BLS) are selected as the core indicators for quantifying land parcel characteristics. Among them, the plot perimeter (PER) and plot area (BLA) were statistically obtained using a geographic information platform as the data cleaning platform. The plot shape index (BLS) is the ratio of the plot perimeter to the perimeter of a square of the same area, and the calculation formula is as follows: In terms of two-dimensional morphological quantification, building density (BD), average building base area (ABA), building base area difference (DBA), average building base shape index (ASH), building base shape difference (DSH), number of buildings (BN), and dispersion (DR) are selected as metrics to characterize the two-dimensional planar morphology of building groups. Wherein, building density BD is the ratio of the sum of the projected areas of buildings to the area of ​​the building site, and the calculation formula is as follows: in It is the sum of the building footprint areas within the plot; The average building footprint area ABA is the average building footprint area across all plots of land, calculated using the following formula: The Building Footprint Area Difference (DBA) is the standard deviation of the footprint area of ​​all plots of land, calculated using the following formula: The dispersion ratio (DR) is the ratio of the product of the number of buildings to the product of the differences in land area and building volume. The calculation formula is as follows: DTBA refers to the building volume variation degree; The Average Building Footprint Shape Index (ASH) is the average of the shape indices of all building footprints within the site, calculated using the following formula: in It is the building base shape index; The building footprint shape variation degree DSH is the standard deviation of any building footprint shape index within a site, and is calculated using the following formula: In terms of three-dimensional morphology quantification, the floor area ratio (FAR), average building capacity (ATBA), building capacity difference (DTBA), average building height (ABH), and staggered height (DBH) are selected as metrics to characterize the three-dimensional morphology of the building complex. Floor Area Ratio (FAR) refers to the ratio of the total building area within a plot to the plot area, calculated using the following formula: in This represents the sum of the building footprint areas within the plot; where TBA is the building volume. The Average Building Volume (ATBA) is the average volume of all buildings within the site, calculated using the following formula: The Building Volume Difference (DTBA) is the standard deviation of the volume of any building within the site, calculated using the following formula: The average building height ABH is the average height of all buildings within the plot, calculated using the following formula: The staggered height (DBH) is the standard deviation of the height of any building within the site, calculated using the following formula: Where BHi is the building height.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program that can run on a processor. When the processor loads and executes the computer program, it employs a method for large-scale automatic modeling of commercial building complexes based on intelligent image learning, as described in any one of claims 1 to 7.

Images