CN-115713605-B - Automatic modeling method for commercial building group based on image learning

CN115713605BCN 115713605 BCN115713605 BCN 115713605BCN-115713605-B

Abstract

The invention discloses an automatic modeling method of a commercial building group based on image learning, which belongs to the field of urban planning and comprises five steps of data acquisition, form quantization core index extraction, building group form generation model training based on a Pix2Pix algorithm, building group plane image generation and building group form generation based on OpenCV. The invention aims to realize the automatic generation of a lot of land scale commercial building group layout forms and schemes in a short time by constructing a building group space form type database and a building group form intelligent generation method based on Pix2 Pix. The problem of designer drawing labor machinery repetition period is long is solved, and technical support is provided for building group form design practice.

Inventors

YANG JUNYAN
WANG YICHONG
ZHU XIAO
XUE QIN
CAI JIYAO
SHI YI
SUN HAOCHENG
SHENG HUAXING
YANG XIAOFANG
ZHOU JINGLING

Assignees

东南大学

Dates

Publication Date: 20260505
Application Date: 20221128

Claims (9)

1. An automatic modeling method for commercial building groups based on image learning is characterized by comprising the following steps: Acquiring and cleaning information data of commercial buildings in a target area, establishing core indexes of the commercial building group forms of the land scale based on land form, two-dimensional building and three-dimensional building, and extracting data corresponding to the core indexes from the building data to form the I # ~ ); Based on core index library I ~ ) Dividing and merging morphological similar samples by using a two-step clustering method, and constructing a building group class library C # ~ ) The core index library I is% ~ ) Dividing the intervals of the indexes in the building group and adding the attributes to the building group class library C # ~ ) Defining a network structure, performing iterative training by using a Pix2Pix deep convolutional neural network model, and constructing a generation algorithm model library G of planar image generation of a building group ~ ); Based on building group class library C ~ ) Determining the building group category of the designed land parcel profile, and inputting the designed land parcel profile data into an algorithm model library G # ~ ) Generating a land block building group plane form map; The three-dimensional model of the design block building group scheme of the commercial building in the target area is constructed based on the block boundary vector data and the building boundary vector data; The core index for establishing the form of the commercial building group of the land parcel based on the land parcel form, the two-dimensional building and the three-dimensional building comprises the following steps: in the aspect of plot morphology quantization, selecting plot perimeter PER, plot area BLA and plot shape index BLS as core indexes of quantized plot features; the plot perimeter PER and plot area BLA are obtained by statistics by using a geographic information platform as a working platform for data cleaning, and the plot shape index BLS is the perimeter ratio of the plot perimeter to the square with the same area, and the calculation formula is as follows: In the aspect of two-dimensional morphology quantification, building density BD, building substrate average area ABA, building substrate area difference degree DBA, building substrate average shape index ASH, building substrate shape difference degree DSH, building number BN and dispersity DR are selected as measurement indexes to describe two-dimensional plane morphology of building groups; wherein, the building density BD is the ratio of the sum of the building projection areas to the building land area, and the calculation formula is as follows: Wherein the method comprises the steps of Is the sum of the areas of all building substrates in the land block; The average area ABA of the building base is the average value of the area of the building base in all plots, and the calculation formula is as follows: The building foundation area difference DBA is the standard deviation of foundation areas of all land plots, and the calculation formula is as follows: The dispersion DR is the ratio of the number of buildings to the product of the difference between the floor area and the building volume, and is calculated as follows: wherein DTBA is building volume variability; The building base average shape index ASH is the average value of all building base shape indexes in the land, and the calculation formula is as follows: Wherein the method comprises the steps of Is a building base shape index; The building foundation shape difference DSH is the standard deviation of any building foundation shape index in the land, and the calculation formula is as follows: In the aspect of three-dimensional form quantization, the volume ratio FAR, the building average capacity ATBA, the building capacity difference degree DTBA, the building average height ABH and the staggering degree DBH are selected as measurement indexes to describe three-dimensional forms of building groups; the volume ratio FAR refers to the ratio of the total building area in the land to the land area, and the calculation formula is as follows: Wherein the method comprises the steps of Is the sum of the areas of all building substrates in the land block; wherein TBA is the building volume; the average building capacity ATBA is the average of all building volumes in the plot and the calculation formula is as follows: The building volume difference DTBA is the standard deviation of any building volume in the land, and the calculation formula is as follows: the building average height ABH is the average value of all building heights in the land, and the calculation formula is as follows: the stagger degree DBH is the standard deviation of any building height in the land, and the calculation formula is as follows: Where BHi is building height.
2. The method for automatically modeling a commercial class building group based on image learning according to claim 1, wherein the step of acquiring and cleaning up the information data of the commercial class building in the target area comprises the steps of: collecting information data of commercial buildings in a target area, wherein the information data comprise building function data, building position data and building height data; And cleaning the information data, namely uniformly screening land parcels with the target land property of class B1 and the area within 0.01-10 hectares, fusing adjacent building elements with the same height, deleting the small building volume with the area smaller than 100 square meters and the land parcels with the building density lower than 10 percent in batches, and cutting the building volume crossing the boundaries of the land parcels.
3. The automatic modeling method for commercial building group based on image learning according to claim 1, wherein the data of the corresponding core index extracted from the building data constitutes ~ ) The method comprises the following steps: Transforming the building data through z-score standardization, so that the transformed building data accords with standard normal distribution, namely, the mean value is 0, and the variance is 1; the building data is checked to determine whether the building data is suitable for principal component analysis or not by using KMO and Bartlett sphericity check, wherein KMO value is more than 0.5, significance p value is less than 0.001, which indicates that the principal component analysis result is effective, and when KMO value is less than 0.5, the building data is not suitable for principal component analysis; Selecting a core index corresponding to building data with a characteristic value greater than 1 and an accumulated percent higher than 70% as a main component in a total variance interpretation table, extracting data of the corresponding core index in the main component, and constructing a representative core index library I # ~ )。
4. The automatic modeling method for commercial building group based on image learning according to claim 3, wherein the core index base I # ~ ) Dividing and merging morphological similar samples by using a two-step clustering method, and constructing a building group class library C # ~ ) The method comprises the following steps: performing pre-clustering division, and adopting a sequential mode to perform core index library I # ~ ) The data of (1) are divided into a plurality of subclasses according to all core index libraries I # ~ ) The data of the index is a large class, and the read-in core index library I is a large class ~ ) After the data in (2) is obtained, determining whether the sample should be derived into a new class or combined into an existing subclass according to the affinity and sparseness degree, and repeating to finally form L classes; Based on pre-clustering, the subclasses are combined according to the affinity and hydrophobicity degree, and finally the L' class is formed.
5. The automatic modeling method for commercial building group based on image learning according to claim 4, wherein the core index library I # ~ ) Dividing the intervals of the indexes in the building group and adding the attributes to the building group class library C # ~ ) The method comprises the following steps: the core index library I is classified according to three class intervals of high/Large, medium/Middle and low/Small by a natural break point classification method ~ ) Dividing intervals by the indexes in the process; Through the core index library I ~ ) Different intermediate dimension descriptions in the core index natural break point grading method of the building group class library C ~ ) Attribute attachment is performed.
6. The automatic modeling method of commercial building group based on image learning according to claim 1, wherein the defining network structure uses Pix2Pix depth convolution neural network model iterative training to construct a generation algorithm model library G of planar image generation of building group ~ ) The method comprises the following steps: For building group class library C ~ ) The data in the building group category sample library S is converted into a picture format to obtain the building group category sample library S # ~ ); Defining a network structure, using a Pix2Pix deep convolutional neural network model, wherein a generator in the model is based on a U-Net architecture, and a PatchGAN classifier is used by a discriminator, and the formula is as follows: The generator operation steps are that an image with similar characteristic distribution is generated according to the characteristic rules of the block outline map and the real building texture map sample; the operation steps of the discriminator are that a new sample pair is formed by a land block boundary and a generated image or a real building texture map and is input, whether the sample pair is a correct mapping from the land block boundary to a real building group form is judged, and probability values are output to discriminate the authenticity of image generation; Sample library S for building group category ~ ) All the categories in the building group are trained by adopting a gradient descent method respectively, fluctuation conditions of a generator and a discriminator loss function of the corresponding model of each parameter during training are observed, parameters of optimal learning rate and iteration times are adjusted and optimized, optimal values of each category are determined by comparing training time and a generating result, and finally a generating algorithm model library G of plane image generation of the building group is constructed ~ ) Wherein, the learning rate is tuning parameter in the optimization algorithm, and the iteration times are times of loops in the iterative operation process; The gradient descent method formula is as follows: ; Where η is a learning rate, i represents the ith data, and the weight parameter w represents the magnitude of each iteration change.
7. The method for automatically modeling a commercial building group based on image learning according to claim 1, wherein the obtaining of the block boundary vector data and the building boundary vector data containing the building height information from the block building group plan map, the constructing of the three-dimensional model of the design block building group scheme of the commercial building of the target area based on the block boundary vector data and the building boundary vector data, comprises the steps of: The method comprises the steps of extracting a block boundary, namely reading a block building group plane form graph by gray scale, setting a threshold value, planning a region with colors of a picture into a color, wherein the region represented by the color is the position of a block, identifying the object contour in the picture as the block boundary, and storing a block data picture which contains contour information corresponding to the block; The construction outline extraction comprises the steps of reading a land block construction group plane form graph, extracting constructions with different colors according to different values of colors among three channels, finding out the constructions with the same colors, converting other constructions into white, converting pictures into gray, detecting construction outlines, packaging each detected curve into a Polygon, deleting the Polygon with smaller area to reduce line impurities, adopting a approxPolyDP function to approximate the curve by a Polygon, reversely mapping the construction Polygon outlines with improved precision to real positions, and determining the construction heights according to the corresponding principle of color values and the number of layers of the constructions, so as to obtain construction vector data containing the construction heights; And inputting the block boundary vector data and the building vector data containing the building height into three-dimensional interactive display equipment, and stretching based on the information of the number of building layers to obtain a three-dimensional model for designing a block building group scheme.
8. An automatic modeling system of a commercial building group based on image learning is characterized by comprising the following modules: the data acquisition and cleaning module is used for acquiring and cleaning information data of commercial buildings in the target area; the morphological quantization core index extraction module is used for establishing core indexes of the morphological forms of the commercial building group of the block scale based on the block morphology, the two-dimensional building and the three-dimensional building, and extracting data corresponding to the core indexes from the building data to form the I (degree of the Chinese character) of the block scale ~ ); Building group morphology generation algorithm model training module based on core index library I ~ ) Dividing and merging morphological similar samples by using a two-step clustering method, and constructing a building group class library C # ~ ) The core index library I is% ~ ) Dividing the intervals of the indexes in the building group and adding the attributes to the building group class library C # ~ ) Defining a network structure, performing iterative training by using a Pix2Pix deep convolutional neural network model, and constructing a generation algorithm model library G of planar image generation of a building group ~ ); Building group plane image generation module based on building group class library C # ~ ) Determining the building group category of the designed land parcel profile, and inputting the designed land parcel profile data into an algorithm model library G # ~ ) Generating a land block building group plane form map; The building group form three-dimensional visualization generation module is used for acquiring block boundary vector data and building boundary vector data containing building height information from a block building group plane form map; The core index for establishing the form of the commercial building group of the land parcel based on the land parcel form, the two-dimensional building and the three-dimensional building comprises the following steps: in the aspect of plot morphology quantization, selecting plot perimeter PER, plot area BLA and plot shape index BLS as core indexes of quantized plot features; the plot perimeter PER and plot area BLA are obtained by statistics by using a geographic information platform as a working platform for data cleaning, and the plot shape index BLS is the perimeter ratio of the plot perimeter to the square with the same area, and the calculation formula is as follows: In the aspect of two-dimensional morphology quantification, building density BD, building substrate average area ABA, building substrate area difference degree DBA, building substrate average shape index ASH, building substrate shape difference degree DSH, building number BN and dispersity DR are selected as measurement indexes to describe two-dimensional plane morphology of building groups; wherein, the building density BD is the ratio of the sum of the building projection areas to the building land area, and the calculation formula is as follows: Wherein the method comprises the steps of Is the sum of the areas of all building substrates in the land block; The average area ABA of the building base is the average value of the area of the building base in all plots, and the calculation formula is as follows: The building foundation area difference DBA is the standard deviation of foundation areas of all land plots, and the calculation formula is as follows: The dispersion DR is the ratio of the number of buildings to the product of the difference between the floor area and the building volume, and is calculated as follows: wherein DTBA is building volume variability; The building base average shape index ASH is the average value of all building base shape indexes in the land, and the calculation formula is as follows: Wherein the method comprises the steps of Is a building base shape index; The building foundation shape difference DSH is the standard deviation of any building foundation shape index in the land, and the calculation formula is as follows: In the aspect of three-dimensional form quantization, the volume ratio FAR, the building average capacity ATBA, the building capacity difference degree DTBA, the building average height ABH and the staggering degree DBH are selected as measurement indexes to describe three-dimensional forms of building groups; the volume ratio FAR refers to the ratio of the total building area in the land to the land area, and the calculation formula is as follows: Wherein the method comprises the steps of Is the sum of the areas of all building substrates in the land block; wherein TBA is the building volume; the average building capacity ATBA is the average of all building volumes in the plot and the calculation formula is as follows: The building volume difference DTBA is the standard deviation of any building volume in the land, and the calculation formula is as follows: the building average height ABH is the average value of all building heights in the land, and the calculation formula is as follows: the stagger degree DBH is the standard deviation of any building height in the land, and the calculation formula is as follows: Where BHi is 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 the computer program capable of running on the processor, and that the processor, when loading and executing the computer program, employs a large-scale automatic modeling method for commercial building groups based on intelligent image learning according to any one of claims 1 to 7.

Description

Automatic modeling method for commercial building group based on image learning Technical Field The invention relates to the field of urban planning, in particular to an automatic modeling method for commercial building groups based on image learning. Background With economic development, the development of commercial plots and the design of commercial building groups are important contents of urban design work. The process of designing the building group is led by an architect or a planner, not only the rigid conditions such as the land indexes, the land functions and the like are considered, but also the complex requirements such as consumer traffic organization, landmark image modeling, development benefit guarantee and the like are met, and under the double constraints of land complexity and design subjectivity, the architect or the planner converts constraint conditions into land forms by virtue of own experience and logic organization. On the premise of lacking rule reference, the design process is time-consuming and labor-consuming, and the display results are greatly different due to subjectivity of design logic. Thus, the construction activities are administrative, and even the urban space disorder development can be assisted. The existing automatic building group form generation method is a generation method based on rule driving and a generation method based on reference learning. The former has been a generation method based on mathematical model, shape grammar, cellular automaton model, multi-agent system. This approach is technically convenient for a wide range of designers, but its technology has limitations. In order to achieve efficiency and representativeness, the generation process only extracts basic building knowledge such as functional topological relations and the like to carry out rule transfer, and most practical objects are residential buildings with strong constraint rules. The technology cannot be well applied to the generation of commercial building groups, and the latter has a generation method based on decision trees, support vector machines, bayesian classification, reinforcement learning and deep learning. The technical object of the method is mainly building planes, building vertical surfaces and building three-dimensional monomers, and the method is lack of application to more complex commercial building group morphological generation designs. And the data processing relies on human-machine statistics, so that the computational effort is limited when processing large-scale city data. Therefore, the prior art cannot achieve the automatic generation of a quick multi-scheme for commercial building groups. Disclosure of Invention The invention aims to extract quantitative indexes of commercial plots and building forms according to theoretical knowledge of planning disciplines, and then divide the typical form types of the buildings by using a clustering algorithm. By making a computer learn a large number of real building group sample cases of typical form types, the mapping rule between the land parcel boundaries and the building layout forms is automatically analyzed, so that intelligent generation of a scheme is realized, and the aim of assisting a planner in making business land parcel design decisions is fulfilled. The aim of the invention can be achieved by the following technical scheme: An automatic modeling method of a commercial building group based on image learning comprises the following steps: The method comprises the steps of acquiring and cleaning information data of commercial buildings in a target area, establishing core indexes of the forms of commercial building groups of the land scale based on land forms, two dimensions of the buildings and three dimensions of the buildings, extracting data corresponding to the core indexes from the building data to form I (I 1～In); Dividing and merging morphological similar samples by using a two-step clustering method based on a core index library I (I 1～In) to construct a building group class library C (C 1～Cn), dividing intervals of indexes in the core index library I (I 1～In) and adding attributes to the building group class library C (C 1～Cn), defining a network structure, performing iterative training by using a Pix2Pix deep convolutional neural network model, and constructing a generation algorithm model library G (G 1～Gn) for generating a planar image of a building group; Determining the building group type of the designed land parcel outline based on the building group type library C (C 1～Cn), inputting the designed land parcel outline data into the algorithm model library G (G 1～Gn), and generating a land parcel building group plane morphology map; And constructing a three-dimensional model of a design block building group scheme of the commercial building in the target region based on the block boundary vector data and the building boundary vector data. Further, the step of acquiring and cleaning the informati