A city renewal optimization method and system based on spatial form recognition

Abstract

The application discloses a kind of urban renewal optimization method and system based on spatial form identification, it is related to urban renewal optimization technical field.The method includes: obtaining the three-dimensional point cloud data of target area and processing, output each point carries the first point cloud data of building entity and open space category label, constructs three-dimensional space field, wherein the space unit of building entity category point is marked as first state, the rest space unit is marked as second state, carries out space quality diagnosis, first state space unit instance that will influence space quality is edited into and is given editable attribute range, generates the object list of optimization component to be, each optimization component to be is optimized and is solved, obtains the form parameter modification value of each optimization component to be, it is acted on the corresponding first state space unit in three-dimensional space field, generates the building form model after modification.The present application can provide a calculable, reproducible, interactive technical path for the fine management of stock space.

Description

Technical Field

[0001] This invention relates to the field of urban renewal and optimization technology, and in particular to an urban renewal and optimization method and system based on spatial morphology recognition. Background Technology

[0002] Urban renewal has shifted from large-scale demolition and reconstruction to micro-updates and refined management of existing spaces. Utilizing spatial data such as 3D point clouds, oblique photography, and street view images to perform morphological diagnosis of the existing built environment and generate optimization solutions is a current research hotspot in the field of smart cities. The technical approaches in this field are mainly divided into two relatively independent stages: The first step is spatial morphology identification and evaluation. This step uses methods such as deep learning to perform semantic segmentation on three-dimensional spatial data, identifying elements such as buildings, walls, ground, and vegetation. Based on this, it calculates quality indicators such as spatial openness, sense of enclosure, and green view rate, ultimately outputting a quality evaluation heatmap or marking problem areas. For example, existing technology can identify "poor sense of enclosure in a certain street" and mark it with a red area on the map.

[0003] Secondly, there is the urban renewal scheme optimization stage. This stage uses factors such as plot ratio, sunshine duration, and space utilization efficiency as objective functions, and existing building outlines and plot boundaries as constraints. It employs multi-objective optimization methods such as genetic algorithms and particle swarm optimization to search for the optimal reconstruction or renovation scheme. In this type of method, spatial form is treated as a holistic and static input condition.

[0004] There is a significant technological gap between the two stages mentioned above. Specifically, the upper limit of granularity achievable by existing 3D point cloud semantic segmentation technology is scene-level semantic categories (such as "building", "wall", "ground"), which cannot isolate a specific solid wall that causes spatial quality problems (e.g., "the 30-meter-long, 3.5-meter-high enclosed solid wall on the ground floor of the third building on the east side facing the street") from the scene as an independent object. In the data structure, this solid wall is only represented as a set of meaningless discrete points or triangular facets belonging to the "wall" category.

[0005] This technological gap creates insurmountable obstacles: even if morphological diagnosis reveals that "the solid wall is the main cause of the poor sense of enclosure in the street," this conclusion remains only a semantic judgment and cannot be automatically translated into an object that the optimization algorithm can manipulate. The data format required by the optimization algorithm is a component-level object carrying attributes—that is, the solid wall should be represented as an independent parameterized entity, carrying editable attributes such as "transparency," "upper limit of modifiable range," and "contribution weight to the sense of enclosure," and its mathematical form is a structured object containing variable domains and constraint relationships. However, there is no automatic conversion path between "a set of points in a scene-level semantic category" and "a component-level semantic object carrying design parameters."

[0006] In summary, the fundamental flaw in existing technologies lies in the fact that the processing paradigm for 3D spatial data remains at the scene-level semantic understanding level, lacking a mechanism that can automatically decompose the overall spatial morphology diagnostic conclusions downwards and create component-level objects carrying design semantics that can be directly manipulated by the optimizer. This deficiency constitutes the technical bottleneck preventing current urban renewal optimization methods based on spatial morphology recognition from forming an end-to-end closed loop of "diagnosis-solution". Summary of the Invention

[0007] This invention provides an urban renewal optimization method based on spatial morphology recognition, comprising: Step S1: Obtain the 3D point cloud data of the target area, process the 3D point cloud data through a scene-level semantic segmentation network, and output the first point cloud data of each point carrying the building entity and open space category labels; Step S2: Construct a three-dimensional spatial field based on the first point cloud data, in which spatial units containing building entity category points are marked as the first state, and the remaining spatial units are marked as the second state; Step S3: Perform spatial quality diagnosis based on the spatial units of the first and second states. Instantiate the spatial units of the first state that affect the spatial quality as components to be optimized and assign them an editable attribute range to generate a list of components to be optimized. Step S4: Based on the list of components to be optimized, optimize each component to be optimized and obtain the modified values ​​of the morphological parameters of each component to be optimized. Step S5: Apply the modified morphological parameters of each component to be optimized to the corresponding first state spatial unit in the three-dimensional spatial field to generate the modified building morphological model.

[0008] The urban renewal optimization method based on spatial morphology recognition described above includes the following sub-steps: acquiring 3D point cloud data of the target area, processing the 3D point cloud data through a scene-level semantic segmentation network, and outputting first point cloud data for each point carrying building entity and open space category labels. Step S11: Obtain the 3D point cloud data of the target area, perform voxel downsampling, extract multi-scale features, and obtain the feature vector of each point carrying multi-scale geometric features and color features; Step S12: Process the feature vectors through a scene-level semantic segmentation network to obtain initial point cloud data in which each point carries building entity and open space category labels and is accompanied by uncertain labels; Step S13: Perform label propagation correction on the initial point cloud data and filter out fragment entities to obtain the first point cloud data with building entity and open space category labels for each point.

[0009] The urban renewal optimization method based on spatial morphology recognition, as described above, involves constructing a three-dimensional spatial field based on first point cloud data. Spatial units containing building entity category points are labeled as a first state, and the remaining spatial units are labeled as a second state. This includes the following sub-steps: Step S21: Construct a three-dimensional spatial field framework based on the first point cloud data and divide it into spatial units to obtain the initialized three-dimensional spatial field; Step S22: Determine the initial three-dimensional spatial field occupancy state, fill empty cells, and obtain the preliminary spatial field marked as the first state or the second state for each spatial cell; Step S23: Perform morphological closing operations and legal open space restoration on the preliminary spatial field to obtain the final three-dimensional spatial field.

[0010] The urban renewal optimization method based on spatial morphology recognition, as described above, includes spatial quality diagnosis based on spatial units in a first and second state. The first-state spatial units affecting spatial quality are instantiated as components to be optimized and assigned editable attribute ranges, generating a list of components to be optimized. This process includes the following sub-steps: Step S31: Construct a spatial dual graph based on the adjacency relationship between the first and second state units in the three-dimensional spatial field, calculate the enclosing betweenness of the boundary unit corresponding to each adjacency surface, and obtain the boundary unit carrying the enclosing betweenness. Step S32: Based on the enclosing betweenness of each boundary unit, perform weighted Thiessen polygon subdivision based on spatial competition growth on the boundary units to obtain the transformation independence index of each boundary unit. Step S33: Extract the components to be optimized based on the transformation independent index of each boundary unit, calculate the value range of the editable morphological attributes of the components to be optimized, and generate a list of components to be optimized.

[0011] The urban renewal optimization method based on spatial morphology recognition, as described above, includes the following sub-steps: Based on a list of components to be optimized, each component is optimized and its morphological parameter modification values ​​are obtained. Step S41: Construct a decision vector based on the list of components to be optimized to obtain the variable space of the optimization problem; Step S42: Construct a spatial quality improvement objective function and a regulatory constraint check function based on the decision vector; Step S43: Perform multi-objective optimization on the objective function for improving spatial quality and the regulatory constraint check function to obtain the modified values ​​of the morphological parameters of each component to be optimized.

[0012] The urban renewal optimization method based on spatial morphology recognition, as described above, involves applying the morphological parameter modification values ​​of each component to be optimized to the corresponding first-state spatial unit in the three-dimensional spatial field to generate a model of the renovated building morphology. This includes the following sub-steps: Step S51: Based on the morphological parameter modification values ​​of each component to be optimized, the three-dimensional spatial field is modified to improve its transparency, and the spatial field after the transparency modification is obtained. Step S52: Based on the morphological parameter modification value, perform retractable distance modification on the spatial field after the transparency modification to obtain the retractable spatial field. Step S53: Eliminate component boundary conflicts and verify global compliance of the space field after the setback and modification, and generate a model of the modified building form.

[0013] This invention also provides an urban renewal optimization system based on spatial morphology recognition, comprising: The first point cloud data acquisition module acquires the 3D point cloud data of the target area, processes the 3D point cloud data through a scene-level semantic segmentation network, and outputs the first point cloud data of each point carrying the building entity and open space category labels. The state labeling module constructs a three-dimensional spatial field based on the first point cloud data. The spatial units containing building entity category points are labeled as the first state, and the remaining spatial units are labeled as the second state. The module for generating a list of components to be optimized performs spatial quality diagnosis based on spatial units in the first and second states. It instantiates the first-state spatial units that affect spatial quality into components to be optimized and assigns them an editable attribute range, thereby generating a list of components to be optimized. The morphological parameter modification value solving module optimizes each component based on the list of component objects to be optimized, and obtains the morphological parameter modification value of each component to be optimized. The building form transformation module applies the modified form parameters of each component to be optimized to the corresponding first state spatial unit in the three-dimensional spatial field to generate the transformed building form model.

[0014] The beneficial effects of this invention are as follows: This solution can transform abstract spatial quality issues such as "poor sense of enclosure" into quantifiable enclosure betweenness indices, making spatial diagnostic conclusions mathematically operable. Based on this, without requiring finer-grained component-level annotation of the point cloud, it can automatically extract independent components to be optimized from scene-level binary classification results and assign upper limits to design parameters such as transparency and setback distance for each component, significantly reducing data acquisition and annotation costs. In the optimization solution stage, the diagnostic conclusions can directly drive a multi-objective evolutionary algorithm to search for the optimal renovation solution under realistic constraints such as structural safety and fire safety regulations, avoiding the manual conversion from diagnostic results to renovation solutions, and greatly improving the automation level and solution generation efficiency of existing building micro-renewal. The final output of the renovated building form model can be directly used for visualization and detailed design, providing a calculable, reproducible, and interactive technical path for the refined management of existing spaces. Attached Figure Description

[0015] 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 some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of an urban renewal optimization method based on spatial morphology recognition provided in Embodiment 1 of this application; Figure 2 This is a schematic diagram of an urban renewal optimization system based on spatial morphology recognition provided in Embodiment 2 of this application. Detailed Implementation

[0017] 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, not all, of the embodiments of the present invention. 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.

[0018] Example 1 like Figure 1 As shown in Embodiment 1 of this application, an urban renewal optimization method based on spatial morphology recognition is provided. The method includes the following steps: Step S1: Obtain the 3D point cloud data of the target area, process the 3D point cloud data through a scene-level semantic segmentation network, and output the first point cloud data of each point carrying the building entity and open space category labels; Furthermore, the 3D point cloud data of the target area is acquired, and the 3D point cloud data is processed through a scene-level semantic segmentation network to output the first point cloud data for each point, carrying the building entity and open space category labels. This includes the following sub-steps: Step S11: Obtain the 3D point cloud data of the target area, perform voxel downsampling, extract multi-scale features, and obtain the feature vector of each point carrying multi-scale geometric features and color features; Specifically, the original 3D point cloud data of the target area is acquired through UAV-borne LiDAR scanning or oblique photogrammetry 3D reconstruction. Each point contains XYZ coordinates and RGB color values. The original point cloud is downsampled using a preset voxel edge length. The voxel edge length is determined by performing vertical plane detection based on the geometric prior that building facades are typically large-area vertical planes. Points within the detected vertical planes are used as the initial estimate of the building facade area. The nearest neighbor distance between local surface curvature extrema points is calculated, and the 5th quantile of this distance distribution is taken as the estimate of the minimum resolvable component size. The voxel edge length is taken as an equal or slightly smaller value of this estimate. If curvature extrema points cannot be reliably extracted, twice the global K-nearest neighbor average distance of the original point cloud is taken as the voxel edge length. The mean coordinates and mean color of each voxel are used as representative points to obtain a point cloud with uniform density. For each downsampled point, three spherical neighborhoods of different scales are constructed with preset small, medium, and large radii, corresponding to the architectural detail scale, architectural component scale, and architectural volume fragment scale, respectively. The small radius is 2 to 3 times the voxel side length, the medium radius is 5 to 8 times the voxel side length, and the large radius is 15 to 20 times the voxel side length. Within each neighborhood, the elevation standard deviation, the information entropy of the normal vector direction distribution, and the point density are extracted as geometric features, and the mean and variance of the hue histogram, the texture contrast based on the gray-level co-occurrence matrix, and the energy value are extracted as color features. All features from the three scales are concatenated into a single feature vector.

[0019] Step S12: Process the feature vectors through a scene-level semantic segmentation network to obtain initial point cloud data in which each point carries building entity and open space category labels and is accompanied by uncertain labels; Specifically, the feature vectors of all points are batch-input into a pre-trained scene-level semantic segmentation network for forward inference. The network architecture employs a deep learning network capable of directly processing unstructured point clouds, such as any one of PointNet++, KPConv, or RandLA-Net, outputting the probability value of each point belonging to the building entity category and the open space category. Pre-training refers to training on a publicly available point cloud dataset containing annotations of building and urban scenes (such as Semantic3D, SensatUrban, or a self-built drone-scanned urban scene dataset) using preset network structure parameters and training hyperparameters until the mean intersection-over-union (MIU) metric converges on the validation set, enabling the network to distinguish between building entities and open spaces in scene-level semantic segmentation. For each point, the category with the larger probability value is selected as the initial label. If the difference between the two probability values ​​is less than a preset uncertainty threshold, the point is additionally marked as an uncertain point. The uncertainty threshold is determined by extracting the probability difference of all points located at the category boundary in the validation set of the pre-trained network, and taking the 10th percentile of the difference distribution as the uncertainty threshold to ensure that about 10% of the boundary points are marked as uncertain points. The category boundary point is defined as a point with a different label in a spherical neighborhood centered at the point and with a radius of twice the point spacing. This yields initial point cloud data in which each point carries a building entity or open space label and is marked as uncertain.

[0020] Step S13: Perform label propagation correction on the initial point cloud data and filter out fragment entities to obtain the first point cloud data with building entity and open space category labels for each point; Specifically, for points marked as uncertain, a preset initial search radius is used to find points with determined labels in their neighborhood. The uncertain point is then assigned the category label that is more numerous. The initial search radius is determined based on the average point spacing of the downsampled point cloud, and is taken as two to three times the average point spacing. If there are no points with determined labels in the neighborhood, the search radius is gradually expanded until the assignment is completed, resulting in point cloud data with determined labels for all points. All building entity label points are extracted, and connected component analysis based on Euclidean distance is performed using a preset proximity distance threshold. The proximity distance threshold is determined based on the average point spacing of the downsampled point cloud, and is taken as 1.5 to 2 times the average point spacing. The volume of each connected component is calculated, and the labels of points in connected components with volumes smaller than the preset minimum building volume threshold are modified to open space. The minimum building volume threshold is set based on the lower limit of the volume of the smallest permanent building component (such as independent structural columns or stairwells) to be retained in the target area, with a default value of 1.0 cubic meters. Fragmented entities that do not have renovation value are filtered out, and the first point cloud data with each point carrying only one of the two labels, building entity or open space, is output.

[0021] Step S2: Construct a three-dimensional spatial field based on the first point cloud data, in which spatial units containing building entity category points are marked as the first state, and the remaining spatial units are marked as the second state; Furthermore, a three-dimensional spatial field is constructed based on the first point cloud data. Spatial units containing building entity category points are labeled as the first state, and the remaining spatial units are labeled as the second state. This includes the following sub-steps: Step S21: Construct a three-dimensional spatial field framework based on the first point cloud data and divide it into spatial units to obtain the initialized three-dimensional spatial field; Specifically, the minimum and maximum coordinates of the first point cloud data along the three coordinate axes are calculated to form a minimum spatial bounding box that exactly encloses the entire point cloud. This minimum spatial bounding box is then extended outwards along the X, Y, and Z directions by a preset buffer distance. This buffer distance is 20% of the diagonal length of the minimum bounding box on the horizontal plane, and not less than 3 meters. The extended bounding box is then equidistantly divided into cubic spatial units along the three coordinate axes. The side length of each spatial unit is determined based on the average thickness of the building walls in the target area. The average thickness is determined by extracting all building entity points from the first point cloud data, performing a ray search along the local normal vector of each point, and counting the distances from which the rays encounter open space points within 1.5 meters. The median of this distance is taken as an estimate; if a reliable estimate cannot be made, a default value of 0.25 meters is used. Half of this average thickness is taken as the side length of each spatial unit, ensuring that a single wall contains at least two spatial units in the thickness direction. This results in a three-dimensional spatial field framework composed of regular cubic units. Each spatial unit has a unique three-dimensional index coordinate, and its initial state is empty.

[0022] Step S22: Determine the initial three-dimensional spatial field occupancy state, fill empty cells, and obtain the preliminary spatial field marked as the first state or the second state for each spatial cell; Specifically, for each spatial unit, the number of points labeled as building entities in the first point cloud data falling within the unit's spatial range is counted, along with the total number of points falling within that unit. The ratio of building entity points to the total number of points is calculated. If this ratio is greater than or equal to a preset occupancy threshold, the spatial unit is marked as being in the first state, indicating that the unit is occupied by building entities; otherwise, it is marked as being in the second state, indicating that the unit is an open space. The preset occupancy threshold is set based on the point density and noise level of the first point cloud data, taking a value between 0.5 and 0.7 to achieve a balance between under-segmentation and over-segmentation. For spatial units with a total number of points of zero, no state marking is performed for the time being. For unmarked spatial units, a marked spatial unit is searched within a preset filling search radius with the unit as the center. The empty unit is assigned a state mark based on the number of units in its neighborhood. The filling search radius is set according to the side length of the spatial unit and is taken as an integer multiple of the side length of the unit. If there is no marked unit within the current search radius, the radius is gradually expanded until a marked unit is found and the value is assigned, or when the search radius reaches the upper limit of the diagonal length of the entire spatial field, the unit is uniformly marked as the second state. This process continues until all units are assigned a state mark, resulting in a preliminary spatial field in which each spatial unit is marked as either the first state or the second state.

[0023] Step S23: Perform morphological closing operations and legal open space recovery on the preliminary spatial field to obtain the final three-dimensional spatial field; Specifically, a three-dimensional mathematical morphology closing operation is performed on the set of spatial units marked as being in the first state in the initial spatial field: first, an expansion operation is performed, expanding the boundary of the first-state region outward by a preset expansion radius; then, an erosion operation is performed, shrinking the expanded boundary inward by the same radius. The expansion radius is twice the side length of the spatial unit. After performing the expansion operation, the rate of change of the projected area of ​​the first-state region on the horizontal plane is calculated. If the rate of change exceeds 5%, the expansion radius is reduced by one unit and the operation is repeated until the rate of change condition is met. After the closing operation is completed, continuous spatial unit regions that were changed from the second state to the first state during the closing operation are identified. For each such region, its projected area on the horizontal plane is calculated. If the projected area is greater than or equal to a preset connectivity threshold, all spatial units in that region are restored to the second state. The preset connectivity threshold is 2.0 square meters, which corresponds to the minimum projected area of ​​typical doorways, elevated floors, and other effective connectivity openings, and can be adjusted according to the actual situation of the project. The final output is a three-dimensional spatial field composed of all spatial units and their state markers, where the first state corresponds to the area occupied by the building entity, and the second state corresponds to the open space area.

[0024] Step S3: Perform spatial quality diagnosis based on the spatial units of the first and second states. Instantiate the spatial units of the first state that affect the spatial quality as components to be optimized and assign them an editable attribute range to generate a list of components to be optimized. Furthermore, spatial quality diagnosis is performed based on the spatial units in the first and second states. The spatial units in the first state that affect spatial quality are instantiated as components to be optimized and assigned an editable attribute range, generating a list of component objects to be optimized, including the following sub-steps: Step S31: Construct a spatial dual graph based on the adjacency relationship between the first and second state units in the three-dimensional spatial field, calculate the enclosing betweenness of the boundary unit corresponding to each adjacency surface, and obtain the boundary unit carrying the enclosing betweenness. Specifically, all adjacent surfaces between first-state units and second-state units are extracted from the three-dimensional spatial field. Each adjacent surface is uniquely determined by its first-state unit index and the index of its adjacent second-state unit. A spatial dual graph is constructed with each adjacent surface as a node. An edge exists between two nodes in the spatial dual graph if and only if their corresponding adjacent surfaces share the same edge in three-dimensional space. That is, the first-state units to which the two adjacent surfaces belong are the same unit or the two surfaces are adjacent first-state units, and the second-state units to which the two adjacent surfaces are adjacent are the same unit or the two surfaces are adjacent second-state units. For each node in the spatial dual graph, through... Calculate its enclosing betweenness, where, The enclosure betweenness number indicates the obstructive effect of an adjacent surface on the spatial flow path between all surrounding open space units. A higher enclosure betweenness number means that the adjacent surface is located on more essential spatial flow paths, and the greater the potential value of removing the surface in improving spatial flow. Let be a node in the spatial dual graph corresponding to a certain adjacent surface. The set of boundary open spaces comprises all second-state units located on the outermost perimeter of the building entity. A boundary open space unit is a second-state unit that is adjacent to at least one external boundary of the building entity and connected to the external space. The extraction method involves traversing all spatial units on the six boundary surfaces of the three-dimensional spatial field, marking units in the second state as externally connected source units. Starting from all externally connected source units, a three-dimensional region growth is performed throughout the entire three-dimensional spatial field. The propagation rule is that it only propagates between adjacent spatial units that are both in the second state. All second-state units grown to this point are marked as "connected to the outside." Spatial units that simultaneously satisfy the following two conditions are extracted to form the spatial region. It has been marked as "connected to the outside" and is adjacent to at least one first state unit. and For the set of open spaces at the boundary The elements in the text represent two different boundary open space units. To open space units at the boundaries arrive The number of all shortest paths in the spatial dual graph. For the nodes it passes through The number of paths, for and The Euclidean distance between nodes. The first state element of the corresponding adjacent surface is the boundary element. ,Will Assign the value to the boundary cell as its enclosing betweenness, and output the enclosing betweenness of each boundary cell. The set of boundary units.

[0025] Step S32: Based on the enclosing betweenness of each boundary unit, perform weighted Thiessen polygon subdivision based on spatial competition growth on the boundary units to obtain the transformation independence index of each boundary unit. Specifically, for each boundary unit Let it be a seed point, and let its enclosing betweenness be... For growth weights, all seed points are subjected to weighted Voronoi (Thieson polygon) partitioning within the second-state region of the 3D spatial field. This involves assigning each second-state unit in the open space to a specific seed point. The partitioning rule is as follows: Second-state unit... Belonging to seed point If and only if ,in, For the second state unit To the boundary unit Euclidean distance, Boundary unit Its enclosing betweenness, By traversing all seed points, this partitioning rule allows boundary units with higher betweenness numbers to obtain a larger spatial influence domain.

[0026] After the partitioning is completed, for each boundary element The number of second-state units contained within its Voronoi influence domain is counted and denoted as . ,pass Calculate boundary elements The independent index of transformation, among which, Boundary unit The transformation independence index, with a value range of (0,1), characterizes the boundary unit. The degree of independence of renovation within the building entity is indicated by an index closer to 1. This means that the impact of renovation on this unit is highly concentrated in the open space surrounding it, and it can be renovated independently without affecting the renovation effects of other boundary units on the same building. Conversely, an index closer to 0 indicates that the spatial impact of this unit on other boundary units highly overlaps, requiring collaborative renovation with adjacent boundary units. Boundary unit Enclosing betweenness, To belong to the boundary unit The total number of second state space units, For boundary unit The set of all boundary units belonging to the same connected building entity. The same connected building entity refers to the sum of the boundary units shared by all first-state spatial units that are interconnected through the adjacency relationship of first-state spatial units in a three-dimensional spatial field and constitute a whole building. Boundary unit Enclosing betweenness, To belong to the boundary unit The total number of second state space units, For boundary unit The sum of the comprehensive transformation impacts of all boundary units belonging to the same connected building entity.

[0027] Step S33: Extract the components to be optimized based on the transformation independent index of each boundary unit, calculate the value range of the editable morphological attributes of the components to be optimized, and generate a list of components to be optimized. Specifically, boundary units with a transformation independence index greater than a preset independence threshold are marked as units that can be transformed independently. The preset independence threshold is determined based on the spatial complexity and transformation refinement requirements of the buildings within the target area. The median or upper quartile of the transformation independence index of all boundary units is taken. Connectivity analysis based on spatial adjacency is performed on the units that can be transformed independently, and spatially adjacent units that can be transformed independently are clustered into independent sub-regions. Each sub-region is a set of units for a component to be optimized. A unique identifier is generated for each component to be optimized. ,pass Calculate the range of values ​​for its editable morphological attributes, where, For the first The upper limit of the range of transparency values ​​for each component to be optimized. To predetermine the absolute upper limit of the permeability, it is determined based on the type of the main building structure. Preferably, it is 0.8 for frame structures, 0.6 for brick-concrete load-bearing wall structures, and 0.7 for shear wall structures. For the first The upper limit of the range of retractable distance values ​​for each component to be optimized. For the first The set of indices of all boundary elements contained in the component to be optimized. For the first The total number of boundary elements contained in the component to be optimized. For components The single boundary unit contained therein, To predetermine the absolute upper limit of the retractable distance, it is set according to building structural safety codes and minimum functional space depth requirements. Preferably, it is the smaller value between 40% of the average depth of the first floor and 5.0 meters, but not less than 1.5 meters. Boundary unit The higher the average modification independence index of the boundary units within a component, the smaller the spatial impact of modifying that component on other components, and the greater the extent to which it can be modified independently.

[0028] Set the range of permeability values ​​as follows: The range of values ​​for the retreat distance is: Each component to be optimized is encapsulated into a component object, which includes the set of cell indices, component identifier, range of transparency values, and range of retraction distance values. All component objects are then combined into a list to generate a list of component objects to be optimized.

[0029] Step S4: Based on the list of components to be optimized, optimize each component to be optimized and obtain the modified values ​​of the morphological parameters of each component to be optimized. Furthermore, based on the list of components to be optimized, optimization solutions are performed on each component to obtain the modified morphological parameters of each component, including the following sub-steps: Step S41: Construct a decision vector based on the list of components to be optimized to obtain the variable space of the optimization problem; Specifically, iterate through each component in the list of components to be optimized, and create a transparency variable and a retreat distance variable for each component. The value range of the transparency variable is... The range of values ​​for the retreat distance variable is: Arrange the decision variables of all components in order of component identification number and assemble them into a decision vector. ,in The total number of components to be optimized. For the first The permeability variable of each component, For the first The retractable distance variable of each component, and the decision vector. Each set of values ​​corresponds to a candidate modification scheme.

[0030] Step S42: Construct a spatial quality improvement objective function and a regulatory constraint check function based on the decision vector; Specifically, according to decision vector The rules for conducting the modification simulation are as follows: For the first... A component to be optimized, based on its permeability variable. The boundary element set contained in the component is sorted by spatial Morton code and then sampled at equal intervals, with a selection ratio of 1:1. The unit's state is modified from the first state to the second state to simulate opening or permeation, based on its retractable distance variable. The boundary elements contained in the component are recessed along their inward normal direction. Distance is used to modify the original occupied unit to the second state to simulate the wall moving backward. After the simulation is modified, the spatial dual graph of S31 is reconstructed and the enclosing betweenness is calculated to obtain the updated spatial dual graph and the enclosing betweenness of each adjacent surface. At the same time, the updated three-dimensional spatial field is directly used for geometric checks of the constraints.

[0031] With decision vector For the domain, construct the objective function for improving spatial quality. and regulatory constraint check function ,in, For the set of open space units at the step boundary, for and The Euclidean distance between them To transform the open space units at the former boundary In a spatial dual graph, the enclosure cost of the shortest path is defined as the sum of the enclosure betweenness factors of all nodes (adjacent faces) traversed by the shortest path. According to the decision vector After performing the transformation simulation, the shortest path enclosure cost of the same element pair in the updated spatial dual graph is... The larger the value, the more significant the improvement in spatial circulation after the renovation; The structural remaining bearing capacity constraint means that the vertical bearing capacity of the remaining part of the building where each component to be optimized is located after the modification should not be lower than the design value of the load combination it bears. The bearing capacity calculation can be approximated based on the structural type of the target building and the conservative lower limit of material strength, or the data can be directly quoted when there is a structural inspection report. To constrain fire lane access, the fire truck access route is automatically generated by extracting the connection path of the second state unit with the largest width that connects to the external space from the first working copy. That is, the net width of all positions on the modified fire truck access route is not less than the preset fire lane width limit. The fire lane width limit is determined according to the provisions of the "Code for Fire Protection Design of Buildings" regarding the minimum net width of fire truck access roads. For load-bearing structural protection constraints, the markings of load-bearing structural units (such as load-bearing walls, columns, etc.) in the three-dimensional space field are pre-specified by the user based on the original structural construction drawings or structural inspection and appraisal reports, and are imported as known fixed constraints. In the modification simulation, these marked units always maintain their first state unchanged. The logical AND operation is valid if and only if all three sub-constraints are satisfied. .

[0032] Step S43: Perform multi-objective optimization on the objective function for improving spatial quality and the regulatory constraint check function to obtain the modified values ​​of the morphological parameters of each component to be optimized; Specifically, using decision vectors To optimize variables, the objective function is to improve spatial quality. To optimize the objective, the function is checked under regulatory constraints. To constrain the process, we perform optimization.

[0033] In satisfying Within the feasible region, maximize At the same time, by modifying the cost function For secondary optimization objectives, Defined as the weighted sum of the permeability variable and the retractable distance variable of each component, representing the total amount of work to be done. The weight coefficient of the weighted sum is determined by equal weighting after normalizing the permeability variable and the retractable distance variable to the [0,1] interval, or by assigning different weights according to the importance of the components.

[0034] A multi-objective evolutionary algorithm is used to solve the problem. The algorithm type is either NSGA-II or MOEA / D. The population size is set to 100 to 200, the maximum number of iterations is 500 to 1000, the crossover probability is 0.9, and the mutation probability is the reciprocal of the total number of decision variables. For each individual in the initial population, the values ​​of each component of its decision vector are randomly generated within the upper and lower bounds of its corresponding variable.

[0035] In each generation of evolution, the spatial field and spatial duality graph are updated for each individual in the population according to the modification simulation rules, and the regulatory constraint check function is invoked. Individuals that do not satisfy any constraint are marked as infeasible solutions and assigned a penalty fitness value. For individuals that satisfy all constraints, the objective function is called. and The two objective function values ​​are calculated, and the feasible individuals are ranked and crowded based on the Pareto dominance relationship. Individuals are selected to enter the next generation based on the non-dominated level and the crowding.

[0036] The algorithm terminates after reaching the maximum number of iterations. It then extracts all individuals from the first level of the Pareto front in the final population. From the Pareto front, a preference-based selection strategy is used to determine the final solution: When prioritizing spatial quality improvement, the algorithm selects... For the individual with the largest value, when the minimum intervention priority strategy is preset, select... The individual with the smallest value will have the values ​​of each decision variable in the final selected solution written back to the corresponding component to be optimized according to the component identifier number. This will be recorded as the transparency modification value and the retraction distance modification value of each component to be optimized. A list of morphological parameter modification values ​​will be output. The transparency modification value is the proportion of the boundary unit of the corresponding component to be optimized facing the enclosed cavity that should be transformed into a transparent interface (i.e., the first state is modified to the second state).

[0037] Step S5: Apply the modified morphological parameters of each component to be optimized to the corresponding first state spatial unit in the three-dimensional spatial field to generate the modified building morphological model. Furthermore, the modified morphological parameters of each component to be optimized are applied to the corresponding first-state spatial unit in the three-dimensional spatial field to generate the modified building morphological model, including the following sub-steps: Step S51: Based on the morphological parameter modification values ​​of each component to be optimized, the three-dimensional spatial field is modified to improve its transparency, and the spatial field after the transparency modification is obtained. Specifically, a complete deep copy of the original three-dimensional spatial field is made to create a first working copy, and all subsequent modification operations are performed on the first working copy.

[0038] Iterate through each item in the list of morphological parameter modification values ​​to obtain the component identifier and transparency modification value of the component to be optimized. Based on the component identifier, search for the corresponding component object in the list of component objects to be optimized, and obtain its spatial unit index set. In the spatial unit index set, identify the boundary units facing the enclosed void, that is, spatial units that are in the first state and have at least one adjacent unit in the second state. Count the total number of boundary units. Calculate the number of units that need to be modified, where, For the first The number of units that need to be modified for each component. The total number of boundary units. For the first The transparency modification value of each component. This indicates rounding up to the nearest integer.

[0039] Select in boundary elements Each unit is selected to change its state from the first state to the second state. The selection method is as follows: the boundary units are spatially sorted along three axes according to their center coordinates and then sampled at equal intervals to make the modified opening distribution uniformly dispersed and avoid excessive concentration. A pseudo-random sequence with a fixed random seed is used to ensure that the results are repeatable.

[0040] Step S52: Based on the morphological parameter modification value, perform retractable distance modification on the spatial field after the transparency modification to obtain the retractable spatial field. Specifically, the process iterates through each item in the list of morphological parameter modification values ​​to obtain the component identifier and retraction distance modification value of the component to be optimized. Based on the component identifier, the corresponding component object is searched in the list of components to be optimized to obtain its spatial unit index set. In the spatial unit index set, the boundary unit facing the enclosed cavity is identified, that is, the spatial unit that is in the first state and has at least one adjacent unit in the second state. On the boundary of the component facing the enclosed cavity, the inner normal direction of the boundary is determined, that is, the direction from the first state unit of the boundary to the first state unit inside the component. For each boundary unit, the inner normal direction is obtained by calculating the average of the direction vectors of the unit pointing to all its adjacent first state units. If the average value is zero, the average outer normal of all boundary faces of the component to which the boundary unit belongs is obtained, and its reverse direction is taken as the inner normal direction. If it still cannot be determined, the principal axis direction of the set of center points of all first state units of the component is taken, and the projection of the reverse of the principal axis direction onto the outer normal direction of the boundary is taken as the inner normal direction (that is, the direction pointing to the inside of the component). The first state unit is modified to the second state layer by layer along this direction. The depth of each layer is the side length of a spatial unit. The modification stops when the cumulative modification depth reaches the retraction distance modification value.

[0041] During the retreat process, after each layer of modification is completed, a 3D connectivity analysis is performed on the remaining first-state units of the component. Using the set of first-state units of the component before modification as a baseline, it is checked whether the remaining first-state units after modification remain connected. If a component is detected to be broken into two or more independent, unconnected parts, it indicates that the retreat is too large and has caused structural fracture. In this case, the modification operation for this layer is withdrawn, and the retreat processing of that component in that direction is stopped to ensure that the retreat modification does not damage the structural integrity of the building. After the retreat modification of all components is completed, the first working copy is output as the spatial field after the retreat modification.

[0042] Step S53: Eliminate component boundary conflicts and verify global compliance of the space field after the setback and renovation, and generate a model of the renovated building form. Specifically, the process iterates through the spatial units in the first working copy that are covered by the modification ranges of two or more components to be optimized. For each conflicting unit, if the modification operation of at least one component requires it to be set to the second state, then the unit is ultimately set to the second state. It checks whether the modification results in a thin-walled structure in the remaining part of a component that does not meet the minimum thickness requirement. If so, the conflicting unit is restored to the first state to ensure structural integrity. The preset minimum structural thickness is determined based on the main structural material type of the target building: 120mm to 240mm for brick-concrete structures and 150mm to 200mm for concrete structures, or an integer multiple of the spatial unit's side length as the converted value.

[0043] Three-dimensional connected component analysis is performed on all first-state spatial units in the first working copy. The state of isolated floating clumps with connected volumes less than the preset minimum building component volume threshold is modified to the second state. The local thickness distribution of each first-state connected component is calculated. Units in local areas with thickness less than the preset minimum structural thickness are modified to the second state. The modified boundary regions are smoothed. Protruding isolated first-state units and recessed isolated second-state units are detected and eliminated along the boundary.

[0044] The first working copy after verification is used as the modified building form model. This model can be stored directly in voxel format or converted into a triangular mesh model through isosurface extraction algorithm for 3D visualization and subsequent architectural detailing design.

[0045] Example 2 like Figure 2 As shown, Embodiment 2 of this application provides an urban renewal optimization system based on spatial morphology recognition, comprising: The first point cloud data acquisition module 21 acquires the three-dimensional point cloud data of the target area, processes the three-dimensional point cloud data through a scene-level semantic segmentation network, and outputs the first point cloud data of each point carrying the building entity and open space category labels. The state labeling module 22 constructs a three-dimensional spatial field based on the first point cloud data, in which spatial units containing building entity category points are labeled as the first state, and the remaining spatial units are labeled as the second state; The module 23 for generating a list of components to be optimized performs spatial quality diagnosis based on the spatial units of the first and second states, instantiates the first-state spatial units that affect spatial quality as components to be optimized and assigns them an editable attribute range, and generates a list of components to be optimized. The morphological parameter modification value solving module 24 optimizes each component based on the list of components to be optimized, and obtains the morphological parameter modification value of each component to be optimized. The building form transformation module 25 applies the modified form parameters of each component to be optimized to the corresponding first state space unit in the three-dimensional space field to generate the transformed building form model. The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for urban renewal optimization based on spatial morphology recognition, characterized in that, include: Step S1: Obtain the 3D point cloud data of the target area, process the 3D point cloud data through a scene-level semantic segmentation network, and output the first point cloud data of each point carrying the building entity and open space category labels; Step S2: Construct a three-dimensional spatial field based on the first point cloud data, in which spatial units containing building entity category points are marked as the first state, and the remaining spatial units are marked as the second state; Step S3: Perform spatial quality diagnosis based on the spatial units of the first and second states. Instantiate the spatial units of the first state that affect spatial quality as components to be optimized and assign them editable attribute ranges to generate a list of component objects to be optimized; including the following sub-steps: Step S31: Construct a spatial dual graph based on the adjacency relationship between the first and second state units in the three-dimensional spatial field, calculate the enclosing betweenness of the boundary unit corresponding to each adjacent surface, and obtain the boundary unit carrying the enclosing betweenness; specifically, for each node in the spatial dual graph, through... Calculate its enclosing betweenness, where, The enclosing betweenness factor represents the obstructive effect that the adjacent surface plays on the spatial flow path between all the surrounding open space units. Let be a node corresponding to a certain adjacent face in the spatial dual graph. This refers to the set of boundary open spaces comprised of all second-state units located on the outermost perimeter of the building entity. and For the set of open spaces at the boundary The elements in the text represent two different boundary open space units. To open space units at the boundaries arrive The number of all shortest paths in the spatial dual graph. For the nodes it passes through The number of paths, for and The Euclidean distance between them; Step S32: Based on the enclosing betweenness of each boundary unit, perform weighted Thiessen polygon subdivision based on spatial competition growth on the boundary units to obtain the transformation independence index of each boundary unit. Step S33: Extract the components to be optimized based on the transformation independent index of each boundary unit, calculate the value range of the editable morphological attributes of the components to be optimized, and generate a list of components to be optimized. Step S4: Based on the list of components to be optimized, optimize each component to be optimized and obtain the modified values ​​of the morphological parameters of each component to be optimized. Step S5: Apply the modified morphological parameters of each component to be optimized to the corresponding first state spatial unit in the three-dimensional spatial field to generate the modified building morphological model.

2. The urban renewal optimization method based on spatial morphology recognition as described in claim 1, characterized in that, Acquire 3D point cloud data of the target area, process the 3D point cloud data through a scene-level semantic segmentation network, and output the first point cloud data of each point carrying building entity and open space category labels, including the following sub-steps: Step S11: Obtain the 3D point cloud data of the target area, perform voxel downsampling, extract multi-scale features, and obtain the feature vector of each point carrying multi-scale geometric features and color features; Step S12: Process the feature vectors through a scene-level semantic segmentation network to obtain initial point cloud data in which each point carries building entity and open space category labels and is accompanied by uncertain labels; Step S13: Perform label propagation correction on the initial point cloud data and filter out fragment entities to obtain the first point cloud data with building entity and open space category labels for each point.

3. The urban renewal optimization method based on spatial morphology recognition as described in claim 1, characterized in that, A three-dimensional spatial field is constructed based on the first point cloud data. Spatial units containing building entity category points are labeled as the first state, and the remaining spatial units are labeled as the second state. The process includes the following sub-steps: Step S21: Construct a three-dimensional spatial field framework based on the first point cloud data and divide it into spatial units to obtain the initialized three-dimensional spatial field; Step S22: Determine the initial three-dimensional spatial field occupancy state, fill empty cells, and obtain the preliminary spatial field marked as the first state or the second state for each spatial cell; Step S23: Perform morphological closing operations and legal open space restoration on the preliminary spatial field to obtain the final three-dimensional spatial field.

4. The urban renewal optimization method based on spatial morphology recognition as described in claim 1, characterized in that, Based on the list of components to be optimized, each component is optimized and solved to obtain the modified morphological parameters of each component. This includes the following sub-steps: Step S41: Construct a decision vector based on the list of components to be optimized to obtain the variable space of the optimization problem; Step S42: Construct a spatial quality improvement objective function and a regulatory constraint check function based on the decision vector; Step S43: Perform multi-objective optimization on the objective function for improving spatial quality and the regulatory constraint check function to obtain the modified values ​​of the morphological parameters of each component to be optimized.

5. The urban renewal optimization method based on spatial morphology recognition as described in claim 1, characterized in that, The modification values ​​of the morphological parameters of each component to be optimized are applied to the corresponding first state spatial unit in the three-dimensional spatial field to generate the modified building morphological model, including the following sub-steps: Step S51: Based on the morphological parameter modification values ​​of each component to be optimized, the three-dimensional spatial field is modified to improve its transparency, and the spatial field after the transparency modification is obtained. Step S52: Based on the morphological parameter modification value, perform retractable distance modification on the spatial field after the transparency modification to obtain the retractable spatial field. Step S53: Eliminate component boundary conflicts and verify global compliance of the space field after the setback and modification, and generate a model of the modified building form.

6. An urban renewal optimization system based on spatial morphology recognition, comprising performing the method as described in any one of claims 1-5, characterized in that, The system includes: The first point cloud data acquisition module acquires the 3D point cloud data of the target area, processes the 3D point cloud data through a scene-level semantic segmentation network, and outputs the first point cloud data of each point carrying the building entity and open space category labels. The state labeling module constructs a three-dimensional spatial field based on the first point cloud data. The spatial units containing building entity category points are labeled as the first state, and the remaining spatial units are labeled as the second state. The module for generating a list of components to be optimized performs spatial quality diagnosis based on spatial units in the first and second states. It instantiates the first-state spatial units that affect spatial quality into components to be optimized and assigns them an editable attribute range, thereby generating a list of components to be optimized. The morphological parameter modification value solving module optimizes each component based on the list of component objects to be optimized, and obtains the morphological parameter modification value of each component to be optimized. The building form transformation module applies the modified form parameters of each component to be optimized to the corresponding first state spatial unit in the three-dimensional spatial field to generate the transformed building form model.

Images