Method for identifying to-be-optimized space of built environment based on morphological hierarchical model

Abstract

The invention discloses a method for identifying a to-be-optimized space of a built environment based on a form hierarchical model. The method comprises the steps of collecting and preprocessing data of the built environment, preliminarily performing form hierarchical operation of the built environment, correcting the form hierarchical operation of the built environment, preliminarily identifying the to-be-optimized space, identifying the hierarchical operation of the to-be-optimized space, and interacting and feeding back identification results. According to the method, data acquisition is carried out through a laser radar system, an aerial photography unmanned aerial vehicle, a street scene acquisition vehicle and a near-earth satellite, and recognition and display of a to-be-optimized space of a built environment are carried out in combination with multi-task Bayesian federal learning, an XGBoost model, a Vision Transformer and a convolutional neural network. According to the method, the rapid identification of the to-be-optimized space of the built environment in the urban planning field can be assisted, and the rationality and accuracy of the identification result of the to-be-optimized space of the built environment are improved through form classification.

Description

Technical Field

[0001] The present invention relates to the field of urban planning, and in particular to a method for identifying spaces to be optimized in the built environment. Background Art

[0002] The urban built environment includes constituent elements such as buildings, green spaces, and roads, and is an important carrier for the development of human society and life activities. The rationality of its spatial layout and the level of quality are directly related to the comfort of residents' lives, the efficiency of urban operation, and the sustainable development ability. At present, with the increasing densification and complexity of urban space, as land resources become increasingly tense and residents' demands become more diverse, there are many spaces to be optimized in the built environment. These spaces to be optimized are manifested as problems such as inconsistent styles, low space utilization rates, and poor environmental quality. Their existence not only affects the overall image and operation efficiency of the city, but also reduces the quality of residents' lives. Therefore, the accurate identification of spaces to be optimized in the built environment is the main prerequisite for formulating scientific and reasonable optimization strategies and enhancing the competitiveness and sustainable development ability of the city. Its significance lies in enabling urban planners and decision-makers to promote urban renewal actions targeted, optimize resource allocation, and effectively improve the quality of the urban built environment.

[0003] Currently, common methods for identifying spaces to be optimized in the built environment have certain limitations. On the one hand, the traditional method that relies on urban planning personnel to identify based on experience has defects such as low efficiency and strong subjectivity, and it is difficult to comprehensively consider the complexity and dynamics of the built environment. On the other hand, the identification method that horizontally divides the research object as a homogeneous whole lacks the hierarchical delineation of the spatial structure level, ignores the interaction and synergy of the morphological characteristics of the built environment at different levels, resulting in a lack of systematicness, thus being unable to accurately locate the spaces to be optimized, and greatly reducing the accuracy and reliability of the identification results, making it difficult to provide effective support for precise policy implementation. Summary of the Invention

[0004] To solve the deficiencies mentioned in the above background art, the purpose of the present invention is to provide a method for identifying spaces to be optimized in the built environment based on a morphological hierarchical model, which can automatically identify the spaces to be optimized in the built environment with high precision.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A method for identifying spaces to be optimized in the built environment based on a morphological hierarchical model includes the following steps:

[0007] Step 1: Collection and preprocessing of built environment data

[0008] Use a drone with a storage space of more than 5T to carry a multi-spectral camera and a lidar system with an accuracy of more than 100mm to collect and identify the three-dimensional morphological data of the target built environment, specifically including building morphological data, green space morphological data, and road morphological data; clean the data to remove outliers; unify the data format into a voxel network with a voxel volume of 1 cubic decimeter, and each voxel stores the three-dimensional morphological information of the building, green space, or road at the corresponding spatial position to generate a voxel network dataset;

[0009] Step two: Initially conduct morphological classification of the built environment

[0010] Collect the block boundary vector data of the target built environment and obtain the voxel network dataset within its coordinate range; based on the voxel network dataset, use multi-task Bayesian federated learning (BFL) to divide the block boundary vector data of the target built environment into a central area, a transition area, and an edge area;

[0011] Specifically, first define three subtasks: identify the central area, transition area, and edge area of the target built environment; on local devices, jointly model the three subtasks through a multi-output Gaussian process (MOGP), use the training set to optimize the model parameters, introduce a prior distribution, capture the three-dimensional combined features of building voxels, green space voxels, and road voxels in the voxel network data, and then upload the posterior distributions on different devices to the global processor for aggregation to form a global MOGP prior, which is distributed back to each local device for the next round of training. After no less than 50 iterations, the model converges, and the block boundary vector data of the target built environment is classified into a central area, a transition area, and an edge area;

[0012] Step three: Correct the morphological classification of the built environment

[0013] According to the classification result of the block boundary vector data of the target built environment, connect the blocks that belong to the same partition and are located in the spatial neighborhood to form several block sets; when the total area of the blocks in the block set is not less than 1 square kilometer, the verification passes; if it is less than 1 square kilometer, correct the belonging partition; during the correction process, if the adjacent block sets of this block set all belong to the same partition, then this block set also belongs to this partition; if the surrounding blocks of this block set belong to different partitions, compare and select which adjacent block set the three-dimensional combined features of the voxel network in the target block set are closest to, and classify this block set into the partition to which the most similar adjacent block set belongs;

[0014] Step four: Initially identify the space to be optimized

[0015] Use a near-earth satellite with a storage space of more than 100T to carry a multi-spectral camera and a lidar system with an accuracy of more than 10m to collect the night light data of the target built environment within a natural month; according to the land use function attributes, extract the commercial blocks, residential blocks and industrial blocks in the target built environment in the central area, transition area and edge area respectively, and match the corresponding average night light data according to the coordinates; for the blocks in the same classification and the same land use function attribute, according to the XGBoost model, divide each functional block in each classification into three intervals according to the average night light data, and the blocks in the lowest interval are the spaces to be optimized; according to the land use status information, verify and remove the blocks with the land use status of under construction and completed but not in operation, and obtain the preliminary dataset of the spaces to be optimized for commercial blocks, residential blocks and industrial blocks in the central area, transition area and edge area.

[0016] Step Five: Hierarchical Identification of Spaces to be Optimized

[0017] For the preliminary spaces to be optimized in the central area, use a drone equipped with a multi-spectral camera to collect the image data of the corresponding coordinate blocks, and conduct a circumferential collection every 10 meters in height. Use Vision Transformer (ViT) to determine whether the street-facing features of various blocks in the central area need to be optimized according to the central area case database; for the preliminary spaces to be optimized in the transition area, use a street view collection vehicle equipped with a 360° panoramic camera to collect the street-facing image data of the corresponding coordinate blocks, and use a convolutional neural network (CNN) to determine whether the street-facing features of various blocks in the transition area need to be optimized according to the transition area case database; for the preliminary spaces to be optimized in the edge area, obtain the satellite image data of the corresponding coordinate blocks through a near-earth satellite, take pictures of the target every 60° through a stereo pair to obtain overlapping stereo images, perform matching and calculation and extract the three-dimensional information of the graphic elements of the target area, and use CNN to determine whether the features of various blocks in the edge area need to be optimized according to the edge area case database; obtain the hierarchical identification dataset of the spaces to be optimized for commercial blocks, residential blocks and industrial blocks in the central area, transition area and edge area.

[0018] Step Six: Interaction and Feedback of Identification Results

[0019] Import the hierarchical identification dataset of the spaces to be optimized for the target built environment output in Step Five into a holographic sand table display device equipped with a wearable three-dimensional motion capture system and a display screen with a resolution of more than 8K to display the distribution of the commercial blocks, residential blocks and industrial blocks to be optimized in the central area, transition area and edge area of the target built environment, conduct human-computer virtual interaction and record the feedback information; feedback the block information considered not to be a space to be optimized in the feedback to the corresponding hierarchical case database to update the identification results of the spaces to be optimized in the corresponding hierarchy.

[0020] Furthermore, step 2 captures the three-dimensional combined features of building voxels, green space voxels, and road voxels in the voxel network data. Specifically, this is achieved by: using a multi-output Gaussian process (MOGP) to jointly model three subtasks to capture the three-dimensional combined features; first, based on the voxel network dataset, a prior distribution is defined, and the spatial distribution patterns of building, green space, and road voxels are embedded in a covariance function. The function is then used to quantify the spatial correlation of different voxel types, covering average building height, building height staggering, green space coverage, green space height staggering, green space plant morphological richness, road network density, and road network connectivity, as shown in the table below;

[0021]

[0022]

[0023] A multidimensional covariance matrix is constructed in this way to reflect the statistical laws of three-dimensional combination characteristics; during the local equipment training stage, when using the training set to optimize the MOGP model parameters, a category weight factor is introduced to assign dynamic weights to building, green space, and road voxels respectively. By maximizing the edge likelihood function, the weights are adaptively adjusted to strengthen the dominant characteristics; the density of building and road voxels in the central area is higher, and their weight accounts for an increased proportion in the covariance calculation, while that in the transition area and the edge area decreases in turn; the density of green space voxels in the edge area is higher, and its weight accounts for an increased proportion in the covariance calculation, while that in the edge area and the central area decreases in turn; during the training process, the posterior distribution of each subtask is updated through variational inference, and the weighted average method is used to fuse the parameters of each device to generate a global prior distribution; after no less than 50 iterations, when the model converges, the global prior is used to characterize the differences in three-dimensional combination characteristics of different partitions.

[0024] Furthermore, in step 3, the target block set is compared to determine which adjacent block set has the closest three-dimensional combination characteristics of the voxel network. Specifically, the following indicators of the voxel network in the block set are extracted: three-dimensional morphological fractal dimension, sky openness, floor area ratio, building density, green patch density, and road intersection density, as shown in the table below.

[0025]

[0026]

[0027] A feature matching algorithm is used to compare the extracted three-dimensional combined features with the features of the adjacent block sets one by one to obtain the similarity score between each adjacent block set and the target block set; the hierarchical type of the block set with the highest score is selected as the hierarchical type of the target block set.

[0028] Further, in the fourth step, each sub - level and functional block is divided into three intervals according to the average value of night - time lighting data. Specifically, the division work for different sub - levels and different functional blocks is carried out independently; the average value of night - time lighting data for each sub - level and each functional block is calculated, outliers are removed, and data normalization is performed; rules are set for model training, and the average night - time lighting of blocks belonging to the same functional type in the central area, transition area, and edge area decreases in sequence; for commercial land, night - time lighting data from 18:00 to 22:00 is counted, for residential land, night - time lighting data from 18:00 to 24:00 is counted, and for industrial land, night - time lighting data from 18:00 to 6:00 is counted; the pre - processed dataset is used to train the XGBoost model to learn how to map blocks to three night - time activity intervals of high, medium, and low; after the model training is completed, it is applied to the test dataset to verify the classification accuracy; the average value of night - time lighting data, the sub - level it belongs to, and the land use function attribute of the block to be classified are extracted and input into the XGBoost model, and the night - time activity interval to which the block belongs is output.

[0029] Further, in the fifth step, it is respectively determined whether the street - facing features of various blocks in the central area need to be optimized. Specifically, a case database for the central area is constructed, including image data before and after the optimization of no less than 500 street - facing feature optimization cases of central - area blocks; the image data of the space blocks to be optimized in the preliminary central area is pre - processed. Using Vision Transformer, the image is segmented into 10 * 10 pixel blocks, and the blocks are linearly embedded into a high - dimensional space to form a set of feature vectors, which are input into the Transformer encoder for information interaction and feature fusion to obtain the image feature encoding of the block; the similarity between the feature encodings is calculated to determine the case image data most similar to the block to be determined. If the case image data belongs to the pre - optimization data, it is determined that the block to be determined belongs to the space to be optimized; if the case image data belongs to the post - optimization data, it is determined that the block to be determined does not belong to the space to be optimized.

[0030] The beneficial effects of the present invention:

[0031] 1. Based on multi - output Gaussian process joint modeling, the present invention dynamically adjusts the class weight factors of building, green space, and road voxels, strengthening the three - dimensional combined feature differences in different zones. In the central area, the density of building and road voxels is relatively high, and their weights increase in the covariance calculation, ensuring that the high - density development characteristics are accurately captured; in the edge area, the density of green - space voxels is relatively high, and their weight ratio increases, highlighting the ecological characteristics. Through no less than 50 iterative trainings, the model adapts to the feature differences in different zones, significantly improving the division accuracy of the central area, transition area, and edge area, and avoiding the static deviation of traditional methods.

[0032] 2. The present invention extracts multi-dimensional indicators such as sky openness, plot ratio, and building density, and combines feature matching algorithms to quantify the similarity of adjacent block sets, ensuring the objectivity and accuracy of zoning correction. For block sets with an area of less than 1 square kilometer, by comparing the similarity scores of their three-dimensional combined features with those of adjacent block sets, they are classified into the most similar zones. This mechanism effectively solves the problem of ambiguous attribution of insufficient-area blocks and improves the scientificity and reliability of zoning results.

[0033] 3. The present invention separately counts the night light data of different periods for commercial, residential, and industrial land, uses the XGBoost model to divide high, medium, and low activity intervals, and scientifically locates low-activity blocks as spaces to be optimized. By separately dividing the night light data of different functional areas, the functional adaptability of identification is improved. This method not only improves the accuracy of identifying spaces to be optimized but also provides data support for urban planning.

[0034] 4. Based on the case databases of the central area, transition area, and edge area, the present invention extracts image feature encodings through Vision Transformer and convolutional neural networks, matches the case data before and after optimization, and realizes the automatic determination of the streetscape to be optimized. This method reduces the dependence on manual experience and ensures the objectivity and operability of the streetscape determination.

[0035] 5. The present invention interacts with users in real time through a holographic sand table display device, updates model parameters in combination with the corrected case database, and forms a "recognition - feedback - optimization" closed loop. This mechanism improves the dynamic iterative ability of space recognition results and provides strong support for urban planning decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is the flowchart of the method of the present invention;

[0037] Figure 2 is the night light recognition result of Shanghai;

[0038] Figure 3 is the sub-level recognition result of inefficient land in Shanghai. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention.

[0040] A method for identifying spaces to be optimized in the built environment based on a morphological sub-level model, as Figure 1 shown, includes the following steps:

[0041] 1. Built environment data collection and preprocessing. Using drones with more than 5TB of storage space, equipped with multispectral cameras and lidar systems with an accuracy of more than 100mm, collect and identify the three-dimensional morphological data of the target built environment, including building morphological data, green space morphological data, and road morphological data. Clean the data to remove outliers. Unify the data format into a voxel network with a voxel volume of 1 cubic decimeter. Each voxel stores the three-dimensional morphological information of the building, green space, or road at the corresponding spatial location, generating a voxel network dataset.

[0042] 2. Preliminary classification of built environment morphology. Collect the block boundary vector data of the target built environment and obtain the voxel network dataset within its coordinate range; Based on the voxel network dataset, use multi-task Bayesian federated learning (BFL) to divide the block boundary vector data of the target built environment into the central area, transition area and edge area; Specifically, first define three subtasks: identify the central area, transition area and edge area of the target built environment; On the local device, use the multi-output Gaussian process (MOGP) to jointly model the three subtasks, use the training set to optimize the model parameters, introduce the prior distribution, capture the three-dimensional combination characteristics of building voxels, green space voxels and road voxels in the voxel network data, and then upload the posterior distribution on different devices to the global processor for aggregation to form a global MOGP prior, which is distributed back to each local device for the next round of training. After no less than 50 iterations, the model converges and the block boundary vector data of the target built environment is classified into the central area, transition area and edge area;

[0043] The method captures the three-dimensional combination characteristics of building voxels, green space voxels and road voxels in voxel network data, specifically by: using a multi-output Gaussian process (MOGP) to jointly model three subtasks to capture the three-dimensional combination characteristics; first, based on the voxel network dataset, a prior distribution is defined, and the spatial distribution patterns of building, green space and road voxels are embedded in the covariance function, and the spatial correlation of different voxel types is quantified through the function, covering the average building height, building height staggeredness, green space coverage, green space height staggeredness, green space plant morphological richness, road network density, and road network connectivity (see the table below).

[0044]

[0045]

[0046] Based on this, a multi-dimensional covariance matrix is constructed to reflect the statistical laws of three-dimensional combined features. During the local device training stage, when optimizing the MOGP model parameters using the training set, a class weight factor is introduced to assign dynamic weights to building, green space, and road voxels respectively. By maximizing the marginal likelihood function, the weights are adaptively adjusted to strengthen the dominant features. The density of building voxels and road voxels in the central area is higher, and their weights account for a larger proportion in covariance calculation, decreasing in the transition area and the edge area in turn. The density of green space voxels in the edge area is higher, and its weight accounts for a larger proportion in covariance calculation, decreasing in the edge area and the central area in turn. During the training process, the posterior distribution of each subtask is updated through variational inference, and the weighted average method is used to fuse the parameters of each device to generate a global prior distribution. After no less than 50 iterations and the model converges, the three-dimensional combined feature differences of different partitions are characterized by the global prior.

[0047] III. Revise the classification of the built environment form; according to the classification result of the block boundary vector data of the target built environment, connect the blocks that belong to the same partition and are located in the spatial neighborhood to form several block sets. If the total area of the blocks in the block set is not less than 1 square kilometer, the verification passes; if it is less than 1 square kilometer, the partition to which it belongs is revised. During the revision process, if the adjacent block sets of this block set all belong to the same partition, then this block set also belongs to this partition; if the surrounding blocks of this block set belong to different partitions, compare and select which adjacent block set the three-dimensional combined features of the voxel network in the target block set are closest to, and classify this block set into the partition to which the most similar adjacent block set belongs.

[0048] The comparison and selection of which adjacent block set the three-dimensional combined features of the voxel network in the target block set are closest to is specifically achieved by: extracting the following indicators of the voxel network in this block set, three-dimensional morphological fractal dimension, sky openness, floor area ratio, building density, green space patch density, road intersection density (see the following table);

[0049]

[0050] Using a feature matching algorithm, compare the extracted three-dimensional combined features with the features of adjacent block sets one by one to obtain the similarity scores of each adjacent block set and the target block set. Select the classification type to which the block set with the highest score belongs as the classification type to which the target block set belongs.

[0052] IV. Preliminary identification of the space to be optimized; using a near-earth satellite with a storage space of more than 100T to carry a multispectral camera and a lidar system with an accuracy of more than 10m to collect the night light data of the target built environment within a natural month; according to the land use function attributes in the space to be optimized, respectively extract the commercial blocks, residential blocks and industrial blocks in the target built environment in its central area, transition area and edge area, and match the corresponding average night light data according to the coordinates; for the blocks in the same classification and with the same land use function attributes, according to the XGBoost model, divide each functional block in each classification into three intervals according to the average night light data, and the blocks in the lowest interval are the identified spaces to be optimized; according to the land use status information, verify and remove the blocks with the land use status of under construction and completed but not in operation, and obtain the preliminary dataset of the spaces to be optimized for commercial blocks, residential blocks and industrial blocks in the central area, transition area and edge area.

[0053] Dividing each functional block in each classification into three intervals according to the average night light data means independently carrying out the division work of different functional blocks in different classifications in the space to be optimized; for the average night light data of each functional block in each classification collected, remove the outliers and perform data normalization; set rules for model training, and the average night light of the blocks belonging to the same functional type in the central area, transition area and edge area decreases in turn; for commercial land, count the night light data from 18:00 to 22:00, for residential land, count the night light data from 18:00 to 24:00, and for industrial land, count the night light data from 18:00 to 6:00; use the preprocessed dataset to train the XGBoost model to learn how to map blocks to three intervals of high, medium and low night activity levels; after the model training is completed, apply it to the test dataset to verify the classification accuracy; extract the average night light data, the classification it belongs to and the land use function attributes of the block to be classified, input them into the XGBoost model, and output the night activity interval to which the block belongs.

[0054] V. Hierarchical Identification of Spaces to be Optimized: For the preliminary spaces to be optimized in the central area, use a drone equipped with a multispectral camera to collect image data of the corresponding coordinate blocks. Conduct a circumferential collection every 10 meters in height. Use VisionTransformer (ViT) to determine whether the street-facing landscapes of various types of blocks in the central area need to be optimized according to the central area case database; for the preliminary spaces to be optimized in the transition area, use a street view collection vehicle equipped with a 360° panoramic camera to collect the street-facing image data of the corresponding coordinate blocks. Use a convolutional neural network (CNN) to determine whether the street-facing landscapes of various types of blocks in the transition area need to be optimized according to the transition area case database; for the preliminary spaces to be optimized in the edge area, obtain satellite image data of the corresponding coordinate blocks through a low-earth satellite. Take pictures of the target every 60° through a stereo pair to obtain overlapping stereo images, perform matching and calculation, and extract the three-dimensional information of the graphic elements in the target area. Use CNN to determine whether the landscapes of various types of blocks in the edge area need to be optimized according to the edge area case database; obtain the hierarchical identification datasets of spaces to be optimized for commercial blocks, residential blocks, and industrial blocks in the central area, transition area, and edge area;

[0055] The determination of whether the street-facing landscapes of various types of blocks in the central area need to be optimized respectively refers to constructing a central area case database, including the pre- and post-optimization image data of no less than 500 street-facing landscape optimization cases of central area blocks; preprocess the image data of the blocks in the preliminary central area to be optimized. Use Vision Transformer to segment the image into 10*10 pixel blocks, linearly embed the blocks into a high-dimensional space to form a set of feature vectors, input them into the Transformer encoder for information interaction and feature fusion, and obtain the image feature encoding of the block; calculate the similarity between the feature encodings, determine the case image data most similar to the block to be determined. If the case image data belongs to the pre-optimization data, determine that the block to be determined belongs to the space to be optimized; if the case image data belongs to the post-optimization data, determine that the block to be determined does not belong to the space to be optimized;

[0056] VI. Interaction and Feedback of Identification Results: Import the output hierarchical identification datasets of spaces to be optimized for the target built environment into a holographic sand table display device equipped with a wearable three-dimensional motion capture system and a display screen with a resolution of more than 8K to display the distribution of commercial blocks, residential blocks, and industrial blocks to be optimized in the central area, transition area, and edge area of the target built environment, conduct human-computer virtual interaction, and record the feedback information; feed back the block information considered not to be a space to be optimized in the feedback to the corresponding hierarchical case database to update the identification results of the spaces to be optimized at the corresponding level.

[0057] Embodiment

[0058] The technical solution of the present invention will be described in detail below taking Shanghai as an example, asFigure 2 and Figure 3 as shown below:

[0059] (1) Built environment data collection and preprocessing, specifically including:

[0060] (1.1) Taking the Lujiazui Financial District in Shanghai as the core collection area, using an unmanned aerial vehicle equipped with a lidar system (accuracy up to 50 mm) and a multispectral camera to obtain 3D morphological data covering an area of 12.8 square kilometers; among them, the building morphological data includes building height, density, floor area, the green space morphological data includes vegetation species, quantity, green space area, green space height, and the road morphological data includes road length, floor area, and road grade;

[0061] (1.2) Conduct data cleaning through the CloudCompare platform to eliminate abnormal point cloud data caused by signal reflection (the daily cleaning volume is about 230 million points), and import the cleaned data into the CityEngine platform for voxelization processing; set the voxel volume to 1 cubic decimeter (1000×1000×1000 mm 3 ), use an octree data structure to construct a voxel network, and each voxel attribute field includes 3D morphological information such as building height (accuracy of 0.1 m), vegetation canopy thickness (accuracy of 0.05 m), and road level (encoded according to the municipal road classification standard), and finally generate a voxel network dataset of the Lujiazui Financial District containing 128 million voxels;

[0062] (2) Conduct preliminary classification of the built environment morphology, specifically including:

[0063] (2.1) Obtain the boundary vector data of 36 blocks in the Lujiazui Financial District from the official website of the Shanghai Municipal Planning and Natural Resources Bureau (coordinate system: Shanghai Local Coordinate System 2000), establish a spatial index in the Arcgis Pro platform, and extract the corresponding voxel network data of each block; define three federated learning subtasks: central area (building density ≥ 60%), transition area (30% ≤ building density < 60%), and edge area (building density < 30%), and configure NVIDIA A100 GPU nodes for local training for each subtask;

[0064] (2.2) In the multi-task Bayesian federated learning framework, the number of global iterations was set to 60 and the number of local training epochs was set to 50. The Matérn 3 / 2 covariance function was used to construct a multidimensional covariance matrix. The building height stagger index was calculated by calculating the internal standard deviation of the block (σ ≥ 15 m for high stagger), and the green plant morphological richness was evaluated using the Shannon index (H' ≥ 2.5 for high richness). The dynamic weight adjustment module set the initial weights to 0.6 for building voxels, 0.3 for road voxels, and 0.1 for green voxels. During the training process, the weights were automatically optimized according to the gradient descent direction. Finally, the building weight in the central area (such as the Yincheng Middle Road block) was increased to 0.82, and the green space weight in the peripheral area (such as the Binjiang Forest Park block) was increased to 0.68.

[0065] (3) Revise the built environment morphology classification, including:

[0066] (3.1) The spatial topology of the preliminary classification results was verified, and a Delaunay triangulation was established in the QGIS platform to analyze the spatial adjacency of the blocks. It was found that the area of the three transitional blocks on the north side of Pudong Avenue was insufficient (0.78 square kilometers), and a correction procedure was initiated. The three-dimensional morphological indicators of the target block set were extracted: sky openness (SVF = 0.62), building density (48%), and road intersection density (12 / km). 2 ), feature matching was performed with the adjacent central area (average SVF = 0.31, building density 68%) and edge area (SVF = 0.75, building density 22%);

[0067] (3.2) Using the cosine similarity algorithm to calculate the similarity of feature vectors, the target block set had a similarity score of 0.72 with the central area and 0.35 with the peripheral area. Based on the maximum similarity principle, the block set was modified into a transition area. After modification, the Lujiazui Financial District ultimately formed a three-level morphological hierarchy system with an 8.2 square kilometer central area (including the Xiaolujiazui core area), a 3.1 square kilometer transition area (including the Zhuyuan Commercial District), and a 1.5 square kilometer peripheral area (including the Binjiang Ecological Zone).

[0068] (4) Preliminary identification of the space to be optimized, including:

[0069] (4.1) Use a near-Earth satellite with a storage capacity of at least 100 terabytes, equipped with a multispectral camera and a lidar system with an accuracy of at least 10 meters, to collect nighttime light data for the target built environment in Shanghai over a calendar month;

[0070] (4.2) According to the land use function attributes in the Shanghai Urban Land Use Planning Map issued by the Shanghai Planning Department in accordance with the newly issued national "Urban Land Classification and Planning Construction Land Standard", extract the commercial blocks, residential blocks and industrial blocks in the target built environment in the central area, transition area and fringe area respectively, and match the corresponding average night light data according to the coordinates;

[0071] (4.3) For blocks in the same classification level and with the same land use function attribute, divide each functional block in each classification level into three intervals according to the average night light data. The division of different functional blocks in different classification levels is carried out independently; for the average night light data of each functional block in each classification level collected, remove the outliers and perform data normalization; set rules for model training, and the average night light of blocks belonging to the same functional type in the central area, transition area and fringe area decreases in turn; for the night light data of commercial land in Shanghai from 18:00 to 22:00, for the night light data of residential land in Shanghai from 18:00 to 24:00, and for the night light data of industrial land in Shanghai from 18:00 to 6:00;

[0072] (4.4) Use the preprocessed dataset to train the XGBoost model to learn how to map blocks to three night activity intervals of high, medium and low; after the model training is completed, apply it to the Shanghai test dataset to verify the classification accuracy; extract the average night light data, the classification level it belongs to and the land use function attribute of the blocks to be classified in Shanghai, input them into the XGBoost model, and output the night activity interval to which the block belongs;

[0073] (5) Spatial classification level identification to be optimized, specifically including:

[0074] (5.1) For the preliminary space to be optimized in the central area, use a drone equipped with a multispectral camera to collect image data of the corresponding coordinate blocks, and conduct a circumferential collection every 10 meters in height;

[0075] (5.2) Build a central area case database, including the pre- and post-optimization image data of no less than 500 streetscape optimization cases in the central area blocks, and perform feature marking on the pre- and post-optimization image data;

[0076] (5.3) Preprocess the image data of the spatial blocks to be optimized in the preliminary central area. Use Vision Transformer to segment the image into 10*10 pixel blocks, linearly embed the blocks into a high-dimensional space to form a set of feature vectors, input them into the Transformer encoder for information interaction and feature fusion, and obtain the image feature encoding of the blocks; calculate the similarity between the feature encodings, determine the case image data most similar to the block to be determined. If the case image data belongs to the data before optimization, determine that the block to be determined belongs to the space to be optimized; if the case image data belongs to the data after optimization, determine that the block to be determined does not belong to the space to be optimized.

[0077] (6) Recognition result interaction and feedback

[0078] (6.1) Import the hierarchical recognition dataset of the target built environment in Shanghai to be optimized, which is output, into the holographic sand table display device equipped with a wearable three-dimensional motion capture system and a display screen with a resolution of more than 8K, display the distribution of the commercial blocks, residential blocks and industrial blocks to be optimized in the central area, transition area and edge area of Shanghai respectively, conduct human-computer virtual interaction and record the feedback information.

[0079] (6.2) Feed back the block information considered not to be the space to be optimized in the feedback to the corresponding hierarchical case database to update the recognition result of the space to be optimized at the corresponding level.

[0080] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A method for identifying spaces to be optimized in the built environment based on a morphological hierarchical model, characterized in that, The steps include: Step 1: Built environment data collection and preprocessing Using drones with over 5TB of storage space, equipped with multispectral cameras and lidar systems with an accuracy of over 100mm, the system collects and identifies the three-dimensional morphological data of the target built environment, including building morphology data, green space morphology data, and road morphology data. The data is cleaned to remove outliers. The data is formatted as a voxel network with a voxel volume of 1 cubic decimeter. Each voxel stores the three-dimensional morphological information of the building, green space, or road at the corresponding spatial location, generating a voxel network dataset. Step 2: Preliminary stratification of built environment forms Collect block boundary vector data of the target built environment and obtain a voxel network dataset within its coordinate range. Based on the voxel network dataset, use multi-task Bayesian federated learning to divide the block boundary vector data of the target built environment into central area, transition area, and edge area. Specifically, three subtasks are first defined: identifying the central area, transition area, and edge area of the target built environment. On local devices, the three subtasks are jointly modeled using a multi-output Gaussian process. A training set is used to optimize model parameters, introducing a prior distribution to capture the three-dimensional combined features of building voxels, green space voxels, and road voxels in the voxel network data. The posterior distributions on different devices are then uploaded to a global processor for aggregation to form a global MOGP prior, which is then distributed back to each local device for the next round of training. After no fewer than 50 iterations, the model converges and the block boundary vector data of the target built environment is classified into the central area, transition area, and edge area. Step 3: Revise the built environment morphology Based on the hierarchical results of the block boundary vector data of the target built environment, blocks belonging to the same partition and located in spatial proximity are connected to form several block sets. The verification passes when the total area of the blocks in the block set is not less than 1 square kilometer. If the area is less than 1 square kilometer, the district to which it belongs will be revised; During the correction process, if the adjacent block sets of this block set all belong to the same partition, then this block set will also be assigned to this partition; if the blocks outside this block set belong to different partitions, the three-dimensional combination characteristics of the voxel network in the target block set are compared and selected to determine which adjacent block set is closest to it, and this block set will be assigned to the partition to which the most similar block set belongs; Step 4: Preliminary identification of the space to be optimized Use a near-earth satellite with a storage space of more than 100T to carry a multispectral camera and a lidar system with an accuracy of more than 10m to collect the night light data of the target built environment within a natural month; according to the land use function attributes, extract the commercial blocks, residential blocks and industrial blocks in the target built environment in the central area, transition area and edge area respectively, and match the corresponding average night light data according to the coordinates; for the blocks in the same classification and the same land use function attribute, according to the XGBoost model, divide each functional block in each classification into three intervals according to the average night light data, and the blocks in the lowest interval are the spaces to be optimized; according to the land use status information, check and remove the blocks with the land use status of under construction and completed but not in operation, and obtain the preliminary dataset of the spaces to be optimized for commercial blocks, residential blocks and industrial blocks in the central area, transition area and edge area; Step Five: Hierarchical Identification of Spaces to be Optimized For the preliminary spaces to be optimized in the central area, use a drone equipped with a multispectral camera to collect the image data of the corresponding coordinate blocks, conduct a circumferential collection every 10 meters in height, and use ViT to determine whether the street facades of various blocks in the central area need to be optimized according to the central area case database; for the preliminary spaces to be optimized in the transition area, use a street view collection vehicle equipped with a 360° panoramic camera to collect the street-facing image data of the corresponding coordinate blocks, and use CNN to determine whether the street facades of various blocks in the transition area need to be optimized according to the transition area case database; for the preliminary spaces to be optimized in the edge area, obtain the satellite image data of the corresponding coordinate blocks through a near-earth satellite, take pictures of the target every 60° through a stereo pair to obtain overlapping stereo images, perform matching and calculation and extract the three-dimensional information of the graphic elements of the target area, and use CNN to determine whether the styles of various blocks in the edge area need to be optimized according to the edge area case database; obtain the dataset of the hierarchically identified spaces to be optimized for commercial blocks, residential blocks and industrial blocks in the central area, transition area and edge area; Step Six: Interaction and Feedback of Identification Results Import the dataset of the hierarchically identified spaces to be optimized of the target built environment output in Step Five into a holographic sand table display device equipped with a wearable three-dimensional motion capture system and a display screen with a resolution of more than 8K to display the distribution of the commercial blocks, residential blocks and industrial blocks to be optimized in the central area, transition area and edge area of the target built environment, conduct human-computer virtual interaction and record the feedback information; feedback the block information considered not to be a space to be optimized in the feedback to the case database of the corresponding classification to update the identification results of the spaces to be optimized in the corresponding classification.

2. The method for identifying the space to be optimized in the built environment based on the morphological hierarchical model according to claim 1, wherein In the second step, the three-dimensional combined features of building voxels, green space voxels, and road voxels in the captured voxel network data are specifically obtained by: using multi-output Gaussian process to jointly model three sub-tasks to capture the three-dimensional combined features; First, based on the voxel network data set, a prior distribution is defined, and the spatial distribution rules of building, green space, and road voxels are embedded in the covariance function. The spatial correlation of different voxel types is quantified through the function, covering the average building height, building height irregularity, green space coverage rate, green space height irregularity, richness of green space plant forms, road network density, and road network connectivity, as shown in the following table; Based on this, a multi-dimensional covariance matrix is constructed to reflect the statistical laws of the three-dimensional combined features; In the local device training stage, when optimizing the MOGP model parameters using the training set, a class weight factor is introduced to assign dynamic weights to building, green space, and road voxels respectively. By maximizing the marginal likelihood function, the weights are adaptively adjusted to strengthen the dominant features; The densities of building voxels and road voxels in the central area are higher, and their weights account for a larger proportion in the covariance calculation, decreasing in the transition area and the edge area in turn; the density of green space voxels in the edge area is higher, and its weight accounts for a larger proportion in the covariance calculation, decreasing in the edge area and the central area in turn; During the training process, the posterior distribution of each sub-task is updated through variational inference, and the weighted average method is used to fuse the parameters of each device to generate a global prior distribution; After no less than 50 iterations and the model converges, the three-dimensional combined feature differences of different partitions are characterized by the global prior.

3. The method for identifying the space to be optimized in the built environment based on the morphological hierarchical model according to claim 2, wherein In the third step, to select which adjacent block set is closest to the three-dimensional combined features of the voxel network in the target block set, it is specifically done by: extracting the following indicators of the voxel network in this block set, three-dimensional morphological fractal dimension, sky openness, floor area ratio, building density, green space patch density, road intersection density, as shown in the following table; Using a feature matching algorithm, the extracted three-dimensional combined features are compared with the features of adjacent block sets one by one to obtain the similarity scores of each adjacent block set and the target block set; The hierarchical type to which the block set with the highest score belongs is selected as the hierarchical type to which the target block set belongs.

4. The method for identifying the space to be optimized in the built environment based on the morphological hierarchical model according to claim 3, characterized in that In the fourth step, each hierarchical and functional block is divided into three intervals according to the average value of night light data, specifically by independently carrying out the division work for different hierarchical and functional blocks; For each hierarchical and functional block, the average value of night light data is calculated, outliers are removed, and the data is normalized; Rules are set for model training, and the average night light values of blocks belonging to the same functional type in the central area, transition area, and edge area decrease in turn; For commercial land, the night light data from 18:00 to 22:00 is counted, for residential land, the night light data from 18:00 to 24:00 is counted, and for industrial land, the night light data from 18:00 to 6:00 is counted; Using the preprocessed data set to train the XGBoost model to learn how to map blocks to high, medium, and low night activity intervals; The model training is completed and applied to the test dataset to verify the classification accuracy; the average value of the night light data, the belonging classification level, and the land use function attributes of the block to be classified are extracted and input into the XGBoost model, and the night activity interval to which the block belongs is output.

5. A method for identifying the space to be optimized in the built environment based on the morphological hierarchical model according to claim 4, characterized in that, In step five, it is respectively determined whether the street-facing features of various blocks in the central area need to be optimized. Specifically, by constructing a case database for the central area, including the pre- and post-optimization image data of no less than 500 cases of street-facing feature optimization of blocks in the central area; the image data of the blocks to be optimized in the preliminary central area is preprocessed, and the images are segmented into 10*10 pixel blocks by using Vision Transformer, and the blocks are linearly embedded in a high-dimensional space to form a set of feature vectors, which are input into the Transformer encoder for information interaction and feature fusion to obtain the image feature encoding of the blocks. Calculate the similarity between the feature encodings to determine the case image data most similar to the block to be determined. If the case image data belongs to the pre-optimization data, it is determined that the block to be determined belongs to the space to be optimized; if the case image data belongs to the post-optimization data, it is determined that the block to be determined does not belong to the space to be optimized.

Images