A method for automatically generating three-dimensional form of air-ground integrated urban neighborhood buildings based on functional attributes and form constraints

Abstract

This invention proposes an automatic 3D form generation method for integrated urban block architecture based on functional attributes and morphological constraints. This method establishes a deep mapping mechanism between function and form, and utilizes a graph-model complementarity strategy to achieve deep collaboration: generative adversarial networks (GANs) are used to achieve high-precision generation of 2D functional layouts, graph neural networks are used to capture the topological logic of vertical space, and a diffusion model is used to achieve refined reconstruction of 3D voxels. This method aims to provide a computationally achievable, interpretable, and optimizable automated generation framework for next-generation three-dimensional urban design, filling the gap in integrated urban block architecture generation technology.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of generative design, architectural form design, and aerospace information technology, and in particular to a method for automatically generating three-dimensional architectural forms of urban blocks based on functional attributes and form constraints. Background Technology

[0002] With the acceleration of global urbanization and the increasing scarcity of land resources, the development model of urban space is undergoing a profound transformation from "two-dimensional planar expansion" to "three-dimensional interconnection." In recent years, the rapid rise of the low-altitude economy has made low-altitude airspace (typically referring to the space 5 to 300 meters above the ground) within the urban built environment a highly promising economic carrier and transportation medium. The concept of Air-Ground Integration (AGI) has emerged, its core being the planning and design of urban airspace and ground space as an interconnected, organically integrated, and collaborative whole. From this perspective, buildings are no longer merely extensions of the ground, but key physical nodes connecting ground transportation networks and air traffic networks, carrying new functions such as drone logistics delivery, urban air transportation, and environmental monitoring.

[0003] However, existing methods for generating urban block building forms are mainly based on traditional two-dimensional figure-ground relationship theory or simple floor area ratio stacking logic, which are insufficient to effectively address the complex three-dimensional spatial requirements in the context of integrated air-ground architecture. Currently, the field of integrated air-ground building form design faces the following challenges: First, there is a lack of vertical functional analysis and morphological responsiveness. Traditional urban design often treats buildings as stretched volumes on a two-dimensional foundation, neglecting the differentiated needs of different height levels for building form, openings, and topological connections. For example, low-altitude logistics requires building facades to be "porous" to accommodate drones, while manned flight requires large rooftop platforms and aerodynamically optimized forms to meet flight safety and noise control requirements. Existing design tools cannot automatically translate these complex functional requirements of vertical layering into specific geometric constraints.

[0004] Secondly, there is a disconnect between morphology generation technology and functional attributes. While existing parametric design tools (such as Grasshopper) can generate complex geometric shapes, they often lack built-in mechanisms for complex functional attributes such as airspace safety, logistics routes, and microclimate environments (wind, light, and sound). Designers need to manually adjust a large number of parameters to adapt to constantly changing low-altitude flight path planning, resulting in long design cycles and high trial-and-error costs.

[0005] Therefore, this invention proposes an automatic 3D form generation method for integrated urban block architecture based on functional attributes and morphological constraints. This method establishes a deep mapping mechanism between function and form, and utilizes a graph-model complementarity strategy to achieve deep collaboration: Generative Adversarial Networks (GANs) are used to achieve high-precision generation of 2D functional layouts, Graph Neural Networks (GNNs) are used to capture the topological logic of vertical space, and a diffusion model is used to achieve refined reconstruction of 3D voxels. This method aims to provide a computationally achievable, interpretable, and optimizable automated generation framework for next-generation three-dimensional urban design, filling the gap in integrated urban block architecture generation technology. Summary of the Invention

[0006] The purpose of this invention is to address the problem in existing technologies where the generation of integrated air-ground building forms struggles to simultaneously consider functional attribute constraints, vertical spatial topological coherence, and 3D geometric precision. This invention proposes an automatic 3D form generation method for integrated air-ground urban block buildings based on functional attributes and form constraints. This method is applicable to transforming abstract urban design indicators and airspace management requirements into concrete and compliant 3D building forms using computer algorithms.

[0007] This invention is achieved through the following technical solution: This invention proposes an automatic method for generating the three-dimensional form of urban blocks based on functional attributes and morphological constraints, the method comprising: Step 1: Segmentation and Functional Attribute Definition of the Air-Ground Integrated City Model: Based on the theoretical framework of air-ground integration and different traffic modes and activity characteristics, the vertical space of the city is divided into four segments with specific functional attributes and morphological constraints: ground level, low-altitude logistics level, low-altitude residence level, and air traffic level. Step 2, Collection and Standardization of Neighborhood Building Group Morphology Data: Collect and study the morphological model data of the integrated air-land neighborhood building group in the corresponding area, perform voxelization and semantic annotation; draw the planar composition diagram of each vertical segment, and construct a standardized training dataset containing semantic labels. Step 3, segment planar generation based on conditional generative adversarial network: Construct and train a planar generation model based on spatial adaptive normalization SPADE architecture; use semantic segmentation map as constraint condition to train generators for different height segments, and output high-resolution, functional and logically clear two-dimensional building planar slices. Step 4, 3D morphology reconstruction based on graph neural network and diffusion model: The graph neural network is used to extract the topological features of each layer of planar slices to capture the spatial correlation in the vertical direction; then a 3D voxel diffusion model is introduced, and the generated 2D slice sequence and the topological features extracted by the graph neural network are used as conditions to gradually recover the continuous and complete 3D building voxel model through an iterative denoising process, realizing the leap from 2D cross-section to 3D entity.

[0008] Furthermore, the ground layer is defined as follows: its functions focus on pedestrian flow organization, urban interface interaction, underground space connection, and open space creation; its morphological constraints include the permeability of the base, the elevation ratio, and path compatibility with ground micro-logistics robots.

[0009] Furthermore, the definition of the low-altitude dwelling layer is as follows: its function focuses on the flight path of small and medium-sized UAVs for ground take-off and landing, the connection of building corridors, and the hovering of small and medium-sized UAVs; the form constraints must follow a strict "avoidance model", reserving clearance for the aerial take-off and landing operation area through building setbacks or volume twists, while using building corridors to construct a secondary ground transportation network.

[0010] Furthermore, the definition of the low-altitude logistics layer is as follows: its functions focus on unmanned delivery, external attachment and maintenance of building equipment; its form requires the building facade to be "porous", with standardized drone docking ports and horizontal logistics corridors, and a reserved safe flight corridor for micro drones.

[0011] Furthermore, the air traffic layer is defined as follows: its functions focus on heavy-duty eVTOL takeoffs and landings, vertical airport operations, and rooftop transfers; its morphological constraints include a large-span flat roof area, a guide design to prevent downwash airflow interference, and a clearance sector that complies with aviation regulations.

[0012] Furthermore, step two specifically involves: Step 2.1, Data Collection: Select typical neighborhood models from the pilot city of air-ground integration, as well as some rule-based synthetic data, to ensure that the data covers the basic types of various building forms; Step 2.2, Drawing and semantic annotation of planar composition diagram: Slice the 3D model at equal intervals along the vertical axis; perform semantic segmentation on each slice, annotate the functional elements, and convert them into computer-recognizable RGB color code diagrams or single-channel label diagrams; Step 2.3, Dataset Construction: Pair the sliced ​​images with their corresponding semantic masks, divide them into training and test sets, and classify and store them according to the four segments in Step 1 for subsequent hierarchical training.

[0013] Furthermore, step three specifically includes: Step 3.1, Model Selection: SPADE is selected as the core module of the generator; Step 3.2, Generator Training: Train an independent SPADE generator for each segment; the input is the semantic label map of the segment, and the output is the corresponding building floor plan; during the training process, adversarial loss and feature matching loss are combined to enable the generator to learn the spatial layout rules of specific functional layers.

[0014] Furthermore, step four specifically involves: Step 4.1, Vertical Topology Encoding: Construct a vertical topology graph, treating the planar functional areas generated in Step 3 as nodes and the vertical connections between layers as edges; use a graph neural network to aggregate information between layers and generate a global feature vector containing the vertical logic of the entire space; Step 4.2, Voxel Diffusion Generation: Construct a diffusion model based on the 3D U-Net architecture; this model uses the aforementioned generated two-dimensional planar slice stacking and the global feature vector extracted by the graph neural network as conditions to gradually remove noise from Gaussian noise and recover a three-dimensional voxel model with fine geometric details; Step 4.3, Voxel to Mesh: Use the Marching Cubes algorithm to convert the generated voxel model into a triangular mesh format so that it can be imported into architectural design software for subsequent editing and rendering.

[0015] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for automatically generating three-dimensional forms of urban blocks based on functional attributes and morphological constraints.

[0016] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for automatically generating the three-dimensional form of urban blocks based on functional attributes and morphological constraints.

[0017] The beneficial effects of this invention are: 1. Achieved quantification and automation of integrated air-ground design: This invention is the first to transform the theoretical framework of integrated air-ground design into a computable generation logic. Through four-level segmentation and functional attribute mapping, the computer can automatically generate neighborhood building forms that meet the requirements of airspace safety, logistics efficiency and environmental conditions, greatly improving design efficiency and scientificity.

[0018] 2. Solves the structural coherence problem in 3D generation: Compared with the traditional "black box" 3D generation, this invention adopts the technical route of "2D planar generation + 3D topological reconstruction", which innovatively combines GNN and diffusion model. GNN effectively captures the logical relationship in vertical space, while diffusion model ensures the geometric quality of voxel generation, effectively avoiding the interlayer breakage and functional misalignment problems common in generation models.

[0019] 3. Enhanced functional adaptability and interpretability of morphological generation: This invention is based on explicit functional semantics-driven generation, ensuring that each generated spatial unit (such as logistics interfaces and aerial corridors) has a clear functional definition, rather than being a simple random shape. This allows the generated solutions to be directly integrated with subsequent wind environment simulation, solar radiation analysis, and airspace clearance calculation, providing designers with closed-loop decision support for "generation-evaluation-optimization".

[0020] 4. Optimized resource utilization in high-density environments: By generating refined forms of the low-altitude logistics layer and the passenger layer, this invention can effectively utilize urban vertical space resources without increasing ground congestion, achieving a win-win situation for development intensity and environmental quality. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the overall process of an automatic three-dimensional form generation method for integrated urban block architecture based on functional attributes and morphological constraints, as described in this invention.

[0023] Figure 2 This is a technical roadmap for an automatic three-dimensional form generation method for integrated urban block buildings based on functional attributes and morphological constraints, as described in this invention.

[0024] Figure 3 This is a schematic diagram of the two-dimensional planar generative network structure based on the SPADE architecture in this invention.

[0025] Figure 4 This is a schematic diagram of the graph-voxel hybrid generation architecture based on GNN and diffusion model in this invention. Detailed Implementation

[0026] 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.

[0027] Specifically, in combination Figures 1-4This invention proposes an automatic method for generating the three-dimensional form of urban blocks based on functional attributes and morphological constraints, the method comprising the following steps: Step 1: Segmentation and Functional Attribute Definition of the Integrated Air-Ground City Model. Based on the theoretical framework of integrated air-ground transportation, and according to different traffic modes (traditional ground type, low-altitude logistics type, air traffic type, and composite type) and activity characteristics, the vertical space of the city is divided into four segments with specific functional attributes and morphological constraints: ground layer, low-altitude logistics layer, low-altitude residence layer, and air traffic layer.

[0028] Step two: Data collection and standardization of neighborhood building morphology. Collect morphological model data of the integrated air-land neighborhood building complex in the corresponding area, and perform voxelization and semantic annotation. Draw planar composition diagrams of each vertical segment, and construct a standardized training dataset containing semantic labels (such as building entities, horizontal corridors, open space interfaces, transition spaces, and vertical corridors). This step aims to establish a machine-readable "morphological-semantic" correspondence.

[0029] Step 3: Segment planar generation based on Conditional Generative Adversarial Network (cGAN). A planar generation model based on the SPADE (Spatially-Adaptive Normalization) architecture is constructed and trained. Using semantic segmentation maps as constraints, generators are trained separately for different height segments, outputting high-resolution, functionally logical, and textured 2D architectural planar slices. The SPADE architecture effectively prevents the loss of semantic information in deep networks, ensuring that the generated planar images strictly adhere to the input functional layout constraints.

[0030] Step four involves 3D morphological reconstruction based on Graph Neural Networks (GNNs) and a diffusion model. GNNs are used to extract topological features from each layer of planar slices, capturing spatial relationships in the vertical direction. Subsequently, a 3D voxel diffusion model is introduced. Using the generated 2D slice sequence and the topological features extracted by GNNs as conditions, a continuous and complete 3D architectural voxel model is gradually recovered through an iterative denoising process, achieving the transition from 2D slices to 3D entities.

[0031] Furthermore, step one specifically includes: Step 1.1, Ground Level (0-5m) Definition: Functionally focused on pedestrian flow organization, urban interface interaction, underground space connection, and open space creation. Form constraints include the base's permeability, elevation ratio, and path compatibility with ground-based micro-logistics robots.

[0032] Step 1.2, Definition of Low-Altitude Dwelling Layer (5-30m): Functionally focused on airways for small and medium-sized UAVs taking off and landing on the ground, connecting building corridors, and allowing small and medium-sized UAVs to hover. Form constraints must adhere to a strict "avoidance model," reserving clearance for aerial take-off and landing operations through building setbacks or volumetric twists, while simultaneously utilizing building corridors to construct a secondary ground transportation network.

[0033] Step 1.3, Definition of Low-Altitude Logistics Layer (30-120m): Functionally focused on unmanned delivery, external attachment and maintenance of building equipment. Form constraints require the building facade to be "porous", with standardized drone docking ports and horizontal logistics corridors, and a safe flight corridor for micro unmanned aerial vehicles (UAVs) should be reserved.

[0034] Step 1.4, Definition of the Air Navigation Layer (120-300m): Functionally focused on heavy-duty eVTOL takeoffs and landings, vertical airport operations, and rooftop transfers. Form constraints include a large-span flat roof area, a guide design to prevent downwash interference, and a clearance sector compliant with aviation regulations.

[0035] Step two specifically involves: Step 2.1, Data Collection: Select typical neighborhood models from pilot cities such as Hong Kong and Shenzhen that integrate air and land, as well as some rule-based synthetic data, to ensure that the data covers a variety of basic building types (point type, slab type, enclosed type).

[0036] Step 2.2, Drawing and Semantic Annotation of the Plane Composition: Slice the 3D model at equal intervals along the vertical axis (Z-axis) (e.g., 3-4 meters per layer). Perform semantic segmentation on each layer slice, annotating functional elements such as building entities, horizontal corridors, open space interfaces, transition spaces, and vertical corridors, and convert them into computer-recognizable RGB color-coded diagrams or single-channel label diagrams.

[0037] Step 2.3, Dataset Construction: Pair the sliced ​​images with their corresponding semantic masks, divide them into training and test sets, and classify and store them according to the four segments in Step 1 for subsequent hierarchical training.

[0038] Step three specifically involves: Step 3.1, Model Selection: SPADE (Spatially-Adaptive Normalization) was selected as the core module of the generator. Compared to Pix2pixHD, SPADE, by introducing spatially adaptive affine transformations in the normalization layer, can better preserve the spatial structural information of the input semantic mask.

[0039] Step 3.2, Generator Training: For each segment (e.g., the low-altitude logistics layer), train an independent SPADE generator. The input is the semantic label map of that segment, and the output is the corresponding building floor plan. During training, adversarial loss and feature matching loss are combined to enable the generator to learn the spatial layout patterns of specific functional layers.

[0040] Step four specifically involves: Step 4.1, Vertical Topological Encoding (GNN Encoder): Construct a vertical topological graph, treating the planar functional areas generated in Step 3 as nodes and the vertical connections between layers as edges. Utilize a Graph Neural Network (GNN) to aggregate inter-layer information and generate a Global Context Vector containing the vertical logic of the entire space.

[0041] Step 4.2, Voxel Diffusion: Construct a diffusion model based on the 3D U-Net architecture. This model uses the previously generated stacked 2D planar slices and the global feature vectors extracted by GNN as conditions to progressively denoise from Gaussian noise, recovering a 3D voxel model with fine geometric details. This process employs a two-stage generation strategy of "coarse-fine" or a hierarchical upsampling strategy to balance computational efficiency and model accuracy.

[0042] Step 4.3, Voxel to Mesh: Use the Marching Cubes algorithm to convert the generated voxel model into a triangular mesh format, so that it can be imported into common architectural design software (such as Rhino, Revit) for subsequent editing and rendering.

[0043] Example This invention proposes an automatic 3D form generation method for integrated urban block architecture based on functional attributes and morphological constraints. Its core lies in constructing a generation chain from "functional semantics" to "two-dimensional plane" and then to "three-dimensional entity." This embodiment selects an NVIDIA 5090 GPU computing cluster as the running platform, and uses the PyTorch deep learning framework for modeling and training. The chosen technical approach combines the SPADE model, which excels in semantic generation in the current field of computer vision, with the diffusion model and graph neural network (GNN), which have high potential in 3D structure generation.

[0044] The method includes the following specific steps: Step 1: Segmentation and Functional Attribute Definition of the Integrated Air-Ground City Model. This step aims to establish a machine-understandable rule base for urban vertical space. Based on the low-altitude economic operation model and architectural principles, the urban space is divided into four vertical levels, and the functional attributes and morphological constraints of each level are defined.

[0045] Step two: Collection and standardization of neighborhood building morphology data. To train the deep generative model, a high-quality integrated air-ground building dataset needs to be constructed. Since there are few real-world examples that perfectly match AGI characteristics, this embodiment adopts a hybrid data acquisition strategy of "real-world case collection + parametric generation".

[0046] Step 3: Segment Plane Generation Based on SPADE. This step utilizes Conditional Generative Adversarial Networks (cGANs) to solve the mapping problem from abstract functional layout to specific building planes. Considering that each plane layer involves complex functional labels and requires high-resolution output, this invention selects the SPADE architecture. Existing research shows that SPADE, compared to the traditional Pix2pixHD, better preserves the geometric shape of semantic labels when processing semantic image synthesis tasks, avoiding the loss of semantic information caused by normalization operations.

[0047] Step four: 3D morphological reconstruction based on GNN and diffusion model. This step is crucial for achieving the transition from 2D to 3D. Traditional direct stacking of 2D planes leads to interlayer misalignment and structural discontinuities. This invention utilizes graph neural networks (GNNs) to maintain topological consistency and uses a 3D diffusion model to generate high-quality voxel entities.

[0048] Furthermore, step one specifically includes: Step 1.1: Spatial Layering Model Construction. Based on the operating height of different modes of transportation and the building interaction requirements, a vertical layering system as shown in Table 1 is established.

[0049] Table 1

[0050] Step 1.2, Functional Attribute Encoding. Convert the above functional attributes into computer-readable labels. Define a set of functional labels corresponding to each function. Each label corresponds to a specific RGB color value, used as semantic input for the subsequent generation model.

[0051] Step two specifically involves: Step 2.1, Data Collection. Typical neighborhood models were established in pilot cities for air-ground integration, such as Hong Kong and Shenzhen. The high-density urban landscape of these cities provides a solid physical foundation for air-ground integration.

[0052] Step 2.2, Data Standardization and Expansion. Using the Rhino / Grasshopper parametric platform, the original model is batch-modified according to the morphological constraint rules in Step 1. In this way, thousands of "synthetic data" samples that conform to the characteristics of air-ground integration are generated.

[0053] Step 2.3: Extraction of Planar Composition Graphs. Slice the model along the vertical axis to extract a series of two-dimensional semantic planar graphs. Each planar graph corresponds to the functional layout of a specific height layer. Scale all planar graphs to a uniform resolution of 256 or 512 pixels and normalize the pixel values. Simultaneously, extract the key topological nodes of each layer to construct vertical topological relationships.

[0054] Step three specifically involves: Step 3.1, Model Architecture Design. The generator adopts an encoder-decoder structure. Since the input is a sparse semantic mask, traditional normalization layers would erase semantic information. This invention uses a SPADE module, which learns a set of affine transformation parameters after adjusting the size of the semantic mask to perform spatially adaptive modulation of the feature map. A multi-scale discriminator is used to distinguish between the generated planar image and the real planar image at three different resolution scales to ensure the authenticity of the global structure and local details.

[0055] Step 3.2 Layered Training Strategy. Four independent SPADE generative models are trained for each of the four height segments.

[0056] Input: The semantic segmentation mask at this height level.

[0057] Output: RGB floor plan with details of specific building components.

[0058] Loss function: It combines adversarial loss to ensure realism, feature matching loss to stabilize training, and perceptual loss to improve the edge sharpness and structural accuracy of the generated image.

[0059] Step four specifically involves: Step 4.1: Vertical Topology Graph Construction and Encoding. Treat the planes generated in Step 3 as a sequence of nodes. Construct a vertical model using a CNN, where: node v_i: represents the planar feature vector of the i-th layer; edge e_{i,j}: represents the connection relationship between adjacent layers. Use a Graph Transformer or Graph Attention Network to extract features from this graph, obtaining a global embedding vector Z_{global} containing inter-layer contextual information. This vector ensures that the generated building entities are aligned with the vertical corridors.

[0060] Step 4.2, 3D Voxel Diffusion Generation. A 3D Voxel Diffusion Model is chosen as the core for generation. Compared to GANs, the diffusion model has significant advantages in generation diversity and pattern coverage, and the training process is more stable. Forward Process: Gaussian noise is progressively added to the real, complete 3D voxel model X_0 until it becomes isotropic Gaussian noise X_T. Backward Process: A 3D denoising network is trained to attempt to recover a clear voxel model from the noise X_t. Conditional Input: This denoising process is driven by two strongly conditional components: Z_{global}: Global topological embeddings from the GNN, providing macroscopic structural constraints. S_{stack}: Rough 3D volumes formed by stacking the 2D planar graphs generated in Step 3, providing geometric contour constraints.

[0061] Step 4.3: Meshing. X_T is sampled from standard Gaussian noise and combined with a semantic mask generated from the user-input functional layout sketch. This mask is then iteratively denoised using a trained diffusion model, ultimately outputting a probabilistic voxel mesh. Finally, the Marching Cubes algorithm is used to convert the voxel mesh into a triangular mesh and smooth it, resulting in a smooth, topologically correct 3D building solid model that can be directly exported as an OBJ or IFC file for use in BIM software.

[0062] This invention proposes an automatic 3D form generation method for integrated air-ground urban blocks based on functional attributes and morphological constraints. Through the innovative application of deep learning technology, it successfully realizes the automated translation from abstract functional requirements to specific 3D forms, providing strong technical support for the planning and design of integrated air-ground cities.

[0063] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for automatically generating three-dimensional forms of urban blocks based on functional attributes and morphological constraints.

[0064] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for automatically generating the three-dimensional form of urban blocks based on functional attributes and morphological constraints.

[0065] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0066] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0067] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0068] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as execution by a hardware decoding processor, or as a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0069] The above provides a detailed description of the automatic generation method for three-dimensional urban block building forms based on functional attributes and morphological constraints proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for automatically generating the three-dimensional form of urban blocks with integrated air and land use based on functional attributes and morphological constraints, characterized in that, The method includes: Step 1: Segmentation and Functional Attribute Definition of the Air-Ground Integrated City Model: Based on the theoretical framework of air-ground integration and different traffic modes and activity characteristics, the vertical space of the city is divided into four segments with specific functional attributes and morphological constraints: ground level, low-altitude logistics level, low-altitude residence level, and air traffic level. Step 2, Collection and Standardization of Neighborhood Building Group Morphology Data: Collect and study the morphological model data of the integrated air-land neighborhood building group in the corresponding area, perform voxelization and semantic annotation; draw the planar composition diagram of each vertical segment, and construct a standardized training dataset containing semantic labels. Step 3, segment planar generation based on conditional generative adversarial network: Construct and train a planar generation model based on spatial adaptive normalization SPADE architecture; use semantic segmentation map as constraint condition to train generators for different height segments, and output high-resolution, functional and logically clear two-dimensional building planar slices. Step 4, 3D morphology reconstruction based on graph neural network and diffusion model: The graph neural network is used to extract the topological features of each layer of planar slices to capture the spatial correlation in the vertical direction; then a 3D voxel diffusion model is introduced, and the generated 2D slice sequence and the topological features extracted by the graph neural network are used as conditions to gradually recover the continuous and complete 3D building voxel model through an iterative denoising process, realizing the leap from 2D cross-section to 3D entity.

2. The method according to claim 1, characterized in that, The ground layer is defined as follows: its functions focus on pedestrian flow organization, urban interface interaction, underground space connection and open space creation; its morphological constraints include the permeability of the base, the elevation ratio and path compatibility with ground micro-logistics robots.

3. The method according to claim 1, characterized in that, The definition of the low-altitude dwelling layer is: its function focuses on the airway for small and medium-sized UAVs to take off and land on the ground, the connection of building corridors, and the hovering of small and medium-sized UAVs; the form constraints must follow a strict "avoidance model", reserving clearance for the aerial take-off and landing operation area by building setbacks or volume twists, while using building corridors to build a secondary ground transportation network.

4. The method according to claim 1, characterized in that, The definition of the low-altitude logistics layer is as follows: its functions focus on unmanned delivery, external attachment and maintenance of building equipment; its form requires the building facade to be "porous", with standardized drone docking ports and horizontal logistics corridors, and a safe flight corridor for micro drones is reserved.

5. The method according to claim 1, characterized in that, The air traffic layer is defined as follows: its functions focus on heavy-duty eVTOL takeoffs and landings, vertical airport operations, and rooftop transfers; its morphological constraints include large-span roof flat areas, airflow guidance designs to prevent downwash interference, and clear airstrips that comply with aviation regulations.

6. The method according to claim 1, characterized in that, Step two specifically involves: Step 2.1, Data Collection: Select typical neighborhood models from the pilot city of air-ground integration, as well as some rule-based synthetic data, to ensure that the data covers the basic types of various building forms; Step 2.2, Drawing and semantic annotation of planar composition diagram: Slice the 3D model at equal intervals along the vertical axis; perform semantic segmentation on each slice, annotate the functional elements, and convert them into computer-recognizable RGB color code diagrams or single-channel label diagrams; Step 2.3, Dataset Construction: Pair the sliced ​​images with their corresponding semantic masks, divide them into training and test sets, and classify and store them according to the four segments in Step 1 for subsequent hierarchical training.

7. The method according to claim 1, characterized in that, Step three specifically involves: Step 3.1, Model Selection: SPADE is selected as the core module of the generator; Step 3.2, Generator Training: Train an independent SPADE generator for each segment; the input is the semantic label map of the segment, and the output is the corresponding building floor plan; during the training process, adversarial loss and feature matching loss are combined to enable the generator to learn the spatial layout rules of specific functional layers.

8. The method according to claim 1, characterized in that, Step four specifically involves: Step 4.1, Vertical Topology Encoding: Construct a vertical topology graph, treating the planar functional areas generated in Step 3 as nodes and the vertical connections between layers as edges; use a graph neural network to aggregate information between layers and generate a global feature vector containing the vertical logic of the entire space; Step 4.2, Voxel Diffusion Generation: Construct a diffusion model based on the 3D U-Net architecture; this model uses the aforementioned generated two-dimensional planar slice stacking and the global feature vector extracted by the graph neural network as conditions to gradually remove noise from Gaussian noise and recover a three-dimensional voxel model with fine geometric details; Step 4.3, Voxel to Mesh: Use the Marching Cubes algorithm to convert the generated voxel model into a triangular mesh format so that it can be imported into architectural design software for subsequent editing and rendering.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-8.

Images