Machine Learning-Based Approach to Adjusting the Structure of Green Spaces in Neighborhoods

Abstract

This invention discloses a machine learning-based method for adjusting the green space structure of urban blocks, relating to the fields of urban planning, landscape architecture, and artificial intelligence. The method includes the following processes: urban block spatial morphology identification, urban block green space structure identification, urban block “spatial morphology-green space structure” fit analysis, and urban block green space structure adjustment. This invention breaks through the scale limitations of traditional urban block green space structure identification, enabling efficient, large-scale, and cross-time period analysis of massive information samples. At the same time, it adjusts the urban block green space structure through urban block “spatial morphology-green space structure” fit analysis.

Description

Technical Field

[0001] This invention relates to the fields of urban planning, landscape architecture and artificial intelligence technology, specifically a method for adjusting the structure of green spaces in urban blocks based on machine learning. Background Technology

[0002] Urban development has entered a new stage of optimizing existing resources, moving from large-scale incremental construction to a model of intensive development and optimization based on existing assets. The spatial units for urban green space planning and regulation are shifting from the traditional large-scale "city" and "urban area" to more refined small-scale units such as "blocks" and "communities." Research on the structure of green spaces in urban blocks provides theoretical support and technical guidance for achieving refined regulation of green spaces, playing a crucial role in improving the service efficiency of urban green spaces and optimizing their spatial layout.

[0003] The existing methods have the following problems:

[0004] Thinking about green space planning in isolation, neglecting its compatibility with urban spatial form: Traditional urban green space planning often focuses too much on the size and area of ​​the green space itself, while ignoring its compatibility with the urban space. This poor matching often leads to problems such as low efficiency in the use of green spaces and waste of urban public resources.

[0005] Focusing on individual services while neglecting the spatial structure of clusters: Traditional green space planning methods at the street level mainly revolve around the addition of individual spaces, without paying enough attention to the collaborative effect of multiple green spaces forming a group. There is a lack of regulation on the overall form and structure of multiple green spaces, and green space planning is often too random and lacks coordination.

[0006] Data collection is difficult: Traditional planning and survey methods have significant spatial and temporal limitations and require a large amount of manpower and resources, making it difficult to collect and analyze large-scale, fine-grained green space information and community morphology. Summary of the Invention

[0007] To address the shortcomings mentioned in the background technology, the present invention aims to provide a machine learning-based method for adjusting the structure of green spaces in urban blocks. This method identifies the spatial morphology and green space cluster structure of urban blocks, analyzes the adaptability of the "spatial morphology-green space structure" of urban blocks, diagnoses and optimizes the green space structure of urban blocks, and provides guidance for optimizing the green space layout in different development areas of the city.

[0008] The objective of this invention can be achieved through the following technical solution: a machine learning-based method for adjusting the structure of green spaces in urban blocks, the method comprising the following steps:

[0009] By resampling the remote sensing images of the first city to the required accuracy of the Local Climate Zone (LCZ) model, and classifying the resampled remote sensing images according to the classification of urban morphology types in the LCZ theoretical model, we obtain samples of various types of blocks and construct a block spatial morphology dataset.

[0010] Based on the processed urban remote sensing images, a street spatial morphology training set is used as training samples to create street spatial morphology classification results. The street spatial morphology classification results are compared with the validation set, a confusion matrix is ​​established, and a street spatial morphology classification result with satisfactory classification accuracy is output.

[0011] Input the research unit according to the research needs, take the block spatial morphology type with the largest proportion within the research unit as the block spatial morphology of the unit, and output the block spatial morphology type of the research area unit.

[0012] The research units were classified and summarized according to the spatial morphology of the blocks, and a sufficient number of units with a high proportion of the main spatial morphology of the blocks were selected as representative samples.

[0013] Standard false-color remote sensing images are obtained by performing image correction processing on remote sensing images of the second city. Urban green spaces are identified using machine learning algorithms based on the standard false-color remote sensing images and cross-verified with remote sensing images of the second city. The identification results are divided into blocks of green space units using research units that are the same as the spatial morphology of the blocks.

[0014] Based on morphological theory, the characteristics of green space units in urban blocks are extracted and classified using cluster analysis to construct a system of green space structure types in urban blocks, and the cluster centers of each green space structure type are calculated. Based on the constructed system of green space structure types, representative green space units in urban blocks are selected by category to construct a dataset of green space structures in urban blocks.

[0015] A preliminary classification model for the structure of green spaces in urban areas is constructed based on an image recognition model. The urban green space structure dataset is used to train, adjust, and test the classification model, and finally outputs a classification model for the structure of green spaces in urban areas with satisfactory verification accuracy. The classification results of the urban green space structure are obtained by inputting the urban green space units into the classification model for the structure of green spaces in urban areas with satisfactory verification accuracy.

[0016] By combining the spatial form of the block with the green space structure type, a configuration model system of "spatial form - green space structure" of the block is obtained; the adaptability index of "spatial form - green space structure" of each block is calculated and classified.

[0017] Blocks with low compatibility between "spatial form and green space structure" are identified as the main adjustment areas; the green space structure of these blocks is gradually replaced with a green space structure that is highly compatible with the block, thus completing the adjustment of the green space structure of the block.

[0018] Preferably, the acquisition process of the first city remote sensing image and the second city remote sensing image includes:

[0019] Acquire high-precision multi-band remote sensing satellite images of the study area. The selection criteria include cloud cover of less than 10% and shooting time in summer (June-October). Crop the remote sensing images to the study area and resample the images to 100m-150m.

[0020] Preferably, the image correction process for the second city remote sensing image includes: geographic calibration, radiometric calibration, image registration, image mosaicking, image cropping, and band synthesis; the final output is a standard false-color remote sensing image with an accuracy of 10m.

[0021] Preferably, the process of constructing the street block spatial morphology dataset includes:

[0022] According to LCZ, the spatial morphology of the street is divided into α types. Based on the definition of each type in LCZ, training samples are selected, with no less than 50 samples in each type. All training samples are summarized to construct a street spatial morphology dataset. The training samples of each type are divided into ten parts, with nine parts as the training set and one part as the test set.

[0023] Preferably, the process of constructing a system of green space structure types in urban blocks includes:

[0024] A certain number of street blocks are randomly selected. Based on the morphological characteristics of green space volume, connectivity, volume uniformity, and distribution balance within each block, n corresponding indicators are selected to describe its spatial structure. The indicator description results of the green space structure of the sample blocks are calculated. Based on the sample results, the selected indicators are tested for collinearity, and collinear indicators are removed. Finally, m indicators are obtained to describe the characteristics of the green space structure of the blocks, and the green space structure indicators of each sample block are calculated. Random sampling is performed on each type of LCZ block block, with a units sampled each time. The sampled units are used to construct a sample analysis library of green space structure of the blocks. K-means clustering analysis is used to explore and classify the green space structure samples of a single type of LCZ block, obtaining the corresponding green space structure type b1 of the LCZ block, and recording the value range of each structural indicator of each type of green space structure. The previous clustering step is repeated for each type of LCZ block, and a total of m indicators are obtained. i(1≦i≦α) types of green space structures; compare the range of structural index values ​​of all obtained green space structure types, and merge green space structure types with high overlap of structural index value ranges; finally, obtain the block green space structure type system and the cluster center points of related structural indicators under each type.

[0025] Preferably, the process of constructing the neighborhood green space structure dataset includes:

[0026] Select no fewer than 500 street block unit samples of various green space structures, and summarize the selected unit samples to construct a street block green space structure dataset; randomly divide the street block green space structure dataset into a training set and a test set.

[0027] Preferably, the process of constructing the street block green space structure classification model with satisfactory verification accuracy includes:

[0028] A street green space structure classification model was built using deep learning image recognition tools. The training set was divided into 10 parts. One part was used for validation each time, and the rest were used for training. The green space structure classification model was trained using the training samples. Through multiple adjustments, the accuracy, precision, recall, and the harmonic mean of F1 precision and recall were optimized. The hyperparameters at this point were selected to obtain the final required model. Finally, the test set data was input into the model to check whether the output green space classification model met the requirements. If it met the requirements, the green space structure classification model was completed; otherwise, the training was repeated.

[0029] Preferably, the process of analyzing the fit between the "spatial morphology and green space structure" of the neighborhood is as follows:

[0030] The spatial morphology and green space structure types of overlapping blocks are analyzed, with the distance from the cluster center point of each type of LCZ block's green space structure type to the origin O of the coordinate axis. As an evaluation index for the adaptability of the "spatial form - green space structure" of the LCZ block, and based on The values ​​are used to rank and classify the suitability of the green space structure type for this type of LCZ block:

[0031]

[0032] Where, x Ai Let be the value of structural index i in green space structure type A, and n be the number of green space structural indicators.

[0033] Preferably, the process of adjusting the structure of the green space in the block is as follows:

[0034] Units with lower "spatial morphology-green space structure" compatibility levels in various LCZ blocks were selected as the main adjustment areas. By comparing the green space structure types and their compatibility levels covered by various LCZ blocks, the green space structure types with lower compatibility with the spatial morphology of the LCZ block were converted into green space structure types with higher compatibility, thus completing the green space structure adjustment of various LCZ blocks.

[0035] Preferably, an apparatus includes:

[0036] One or more processors;

[0037] Memory, used to store one or more programs;

[0038] When one or more of the programs are executed by one or more of the processors, the one or more of the processors implement the machine learning-based method for adapting the structure of green spaces in urban blocks as described above.

[0039] The beneficial effects of this invention are:

[0040] This invention breaks through the scale limitations of traditional street green space structure identification, enabling efficient, large-scale, and cross-time period analysis of massive information samples. It also obtains the street green space configuration pattern, analyzes the fit between "spatial form and green space structure", and provides guidance for optimizing the green space layout of cities at different stages of urbanization and different development areas of cities. Attached Figure Description

[0041] 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating the overall workflow of the present invention; Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] like Figure 1 As shown, the machine learning-based method for adjusting the structure of green spaces in urban blocks includes the following steps:

[0045] S1: Street spatial morphology recognition

[0046] The street spatial morphology identification includes S11 remote sensing image preprocessing, S12 street spatial morphology dataset construction, S13 street spatial morphology classification, S14 street spatial morphology identification verification and result output, S15 study unit street spatial morphology identification, and S16 representative street spatial morphology unit identification.

[0047] S11: Remote Sensing Image Preprocessing

[0048] Collect urban remote sensing images of the study area and resample them to the accuracy required by the theory of local climate zoning;

[0049] Specifically, it includes:

[0050] Acquire multi-band remote sensing images from common remote sensing satellites such as Landsat, Sentinel, and Spot for the study area. The selection criteria include cloud cover of less than 10% and the time of acquisition being summer (June-October). Crop the remote sensing images to the study area. Resample the remote sensing images to 100m-150m. In this example, the remote sensing images were resampled to 100m.

[0051] S12: Construction of Street Spatial Morphology Dataset

[0052] Based on the theory of local climate zone (LCZ), various types of samples are digitally processed to construct a street spatial morphology dataset;

[0053] Specifically, it includes:

[0054] Based on the actual conditions of the research site, the Local Climate Zone (LCZ) theory was improved to divide the street spatial morphology into n types. According to the definition of each type in the Local Climate Zone theory, training samples were selected using Google Maps. The number of samples for each type should not be less than 50. All training samples were summarized to construct a street spatial morphology dataset. The training samples of each type were divided into ten parts, of which nine parts were the training set and one part was the test set.

[0055] It should be further explained that, in this example, the spatial form of the block is subdivided into 10 types: compact high-rise (CHR), compact mid-rise (CMR), compact low-rise (CLR), open high-rise (OHR), open mid-rise (OMR), open low-rise (OLR), and large low-rise (LLR), as well as green space (GS), open space (OS) and water. Using Google Earth, approximately 70 training samples were identified for each type based on historical Google Earth imagery, and the training sample dataset was compiled and output. Approximately 55 samples from each type were randomly selected to form the training set, and the remainder were used as the validation set.

[0056] S13: Classification of Street Space Form

[0057] Based on the preprocessed remote sensing images of S11, and using the street spatial morphology training set constructed by S12 as training samples, the street spatial morphology classification results were created using local climate classification tools.

[0058] In this example, the local climate classification tool in SAGA GIS software is used to predict the spatial morphology classification of blocks based on the random forest algorithm; alternatively, deep learning network models such as ResNet and DenseNet can be used to predict the spatial morphology classification of blocks.

[0059] S14: Street spatial morphology recognition verification and result output

[0060] Compare the predicted street spatial morphology classification with the validation set, establish a confusion matrix, and output the street spatial morphology classification results with satisfactory classification accuracy; in this example, the output street spatial morphology classification results have an accuracy of over 80%.

[0061] S15: Research Unit Street Spatial Morphology Recognition

[0062] Input the research unit according to the research needs, and take the street space morphology type with the largest proportion within the research unit as the street space morphology of the unit.

[0063] In this example, 600m was selected to meet the research needs. A 600m grid was used as the research unit, and the block spatial morphology type with the largest proportion in the unit was selected as the block spatial morphology of the entire unit; if there were two or more block spatial morphology types with the largest proportion, the statistical range was expanded to 1200m. Within a 1200m range, calculate the block space morphology type with the largest proportion; repeat this operation until a unique block space morphology type with the largest proportion appears, and take this type as the block space morphology type of the research unit.

[0064] S16: Identification of Representative Block Spatial Morphology Units

[0065] The research units were categorized and summarized according to the type of street block spatial morphology, and the research units in which the main street block spatial morphology type accounted for more than 50% were selected as representative samples.

[0066] S2: Identification of Urban Green Space Structure

[0067] The identification of green space structure in urban blocks includes S21 remote sensing image preprocessing, S22 urban block green space unit identification, S23 construction of urban block green space structure type system, S24 construction of urban block green space structure dataset, S25 construction of urban block green space structure classification model, and S26 classification and output of urban block green space structure.

[0068] S21: Remote Sensing Image Preprocessing

[0069] Collect urban remote sensing images of the study area and perform image correction processing on them;

[0070] Specifically, it includes:

[0071] Acquire common remote sensing satellite data such as Sentinel and Spot; screening requirements include cloud cover of less than 10%, shooting time in summer (June-October), and panchromatic band accuracy of 10m and other band accuracy of at least 30m; remote sensing image data preprocessing includes geocalibration, radiometric calibration, image registration, image mosaicking, image cropping, and band synthesis; output standard false-color remote sensing images with an accuracy of 10m.

[0072] S22: Identification of Green Space Units in Neighborhoods

[0073] Urban green space identification is based on standard false-color remote sensing imagery preprocessed by S21. Support vector machine algorithm is used to cross-verify with high-precision urban remote sensing imagery to complete urban green space identification. The same research unit division and identification results as S15 are input to obtain street green space units.

[0074] Specifically, it includes:

[0075] Based on the preprocessing results in S21, samples of green spaces and other land use categories are manually selected. Machine learning is used to perform supervised classification on the preprocessed remote sensing images. The classification results are compared with the measured values ​​using a confusion matrix, and a raster image of urban green spaces with satisfactory classification accuracy is output. The recognition results are segmented into grids as needed for the research.

[0076] Further explanation is needed: In this example, ENVI software was used to select samples from categories such as green spaces, water bodies, and construction land. A support vector machine algorithm was used for supervised classification. The classification results were compared with the actual results of high-precision urban remote sensing imagery using a confusion matrix. The overall model accuracy was calculated, and the green space identification result was output when the model accuracy reached 80% or higher. The 600m input from S15 was used... The green space identification results of the 600m grid segmentation are used to obtain the green space units of the block.

[0077] S23: Construction of the Structural Type System of Urban Green Space

[0078] Select representative street block spatial morphology units corresponding to the street block green space units; select indicators to describe the spatial structure of the green space units within the units based on their morphological characteristics; construct a street block green space unit analysis library by randomly sampling according to street block spatial morphology types; and obtain the street block green space structure type system using cluster analysis methods.

[0079] Specifically, it includes:

[0080] A certain number of street blocks are randomly selected. Based on the morphological characteristics of green space within each block, such as green volume, connectivity, uniformity of volume, and balanced distribution, n corresponding indicators are selected to describe its spatial structure. The indicator description results of the green space structure of the sample units are calculated. Based on the sample results, the selected indicators are tested for collinearity, and collinear indicators are removed. Finally, m indicators are obtained to describe the characteristics of the green space structure of the blocks, and the green space structure indicators of each sample unit are calculated. Random sampling is performed on each type of LCZ block unit, and a unit samples are drawn each time. The sampled unit samples are used to construct a sample analysis library of green space structure of blocks. K-means clustering analysis is used to explore and classify the green space structure samples of single-class LCZ blocks to obtain the green space structure type b1 corresponding to this type of LCZ block. The value range of each structural indicator of each type of green space structure is recorded. The previous clustering step is repeated for each type of LCZ, and a total of m indicators are obtained. i (1≦i≦α) types of green space structures; compare the range of structural index values ​​of all obtained green space structure types, and merge green space structure types with high overlap of structural index value ranges; finally, obtain the block green space structure type system and the cluster center points of related structural indicators under each type.

[0081] It should be further explained that, in this example, among the street space morphology units generated by S16, the corresponding street green space units are selected as sample units; morphological characteristic indicators such as green volume, connectivity, volume uniformity, and distribution balance are selected to describe the spatial structure, and the indicator results of the sample units are calculated; the selected independent variables are subjected to collinearity test, one collinear indicator is deleted, and finally three indicators, green volume (P), connectivity (C), and volume uniformity (S), are selected to describe the morphological characteristics of the street green space units, and the three indicator results of the sample units are calculated;

[0082] Random sampling is performed on the green space unit samples corresponding to a single street spatial form type. 20 unit samples are drawn each time, and the sampling is performed 10 times. The sampled unit samples are used to construct a street green space unit sample analysis library.

[0083] In the unit sample analysis database, hierarchical clustering was first used to classify the green space unit samples corresponding to the spatial morphology type of compact high-rise (CHR) blocks, and the calculation was iteratively performed until the clustering coefficients tended to stabilize. The optimal number of classifications K was determined according to the curvature-based 'elbow' method. K-means clustering analysis was then used to classify the green space unit samples corresponding to a single block spatial morphology type, resulting in three types: A1, B1, and C1, with large inter-group differences and small intra-group differences. For compact mid-rise (CHR) blocks… The above operation was repeated for the classification of green space unit samples corresponding to the street spatial morphology types (CMR) to obtain two types, A2 and B2. Since the green space morphology types A1 and A2 have similar ranges of green quantity, connectivity, and evenness, they belong to the same category. The same comparison was performed on B1 and B2. Since the green quantity ranges differed significantly, B2 and B1 did not belong to the same category. Similarly, the connection ranges of C1 and B2 differed significantly, so B2 and C1 did not belong to the same category, and B2 was classified as category D, independent of A, B, and C. The above operation was repeated for the green space unit samples corresponding to the 10 street spatial morphology types, resulting in a total of 6 street green space types. The center point of each street green space type was calculated: Street green space type 1 (P1, C1, S1), Street green space type 2 (P2, C2, S2)... Street green space type 6 (P6, C...S1)...Street green space type 6 (P6, C...S1)...Street green space type 6 (P6, C...S1)...Street green space type 6 (P6, C...S1)...Street green space type 6 (P6, C...S1)...Street green space type 7 (P1, C1, S1)...Street green space type 8 (P1, C1, S1)...Street green space type 9 ... 6, S6).

[0084] S24: Construction of Block Green Space Structure Dataset

[0085] Select no fewer than 500 street block unit samples of various green space structures, and compile the selected unit samples to construct a street block green space structure dataset; randomly divide the street block green space structure dataset into a training set and a test set.

[0086] S25: Construction of a Classification Model for the Structure of Green Spaces in Urban Blocks

[0087] A preliminary classification model of the green space structure of the block was constructed based on the image recognition model. The block green space structure dataset constructed using S24 was used for training, adjustment and testing, and the output of the block green space structure classification model with verified accuracy was achieved.

[0088] Specifically, it includes:

[0089] A pre-defined algorithm with image classification capabilities is selected to build a street green space classification model. The training set is divided into 10 parts. Each time, one part is used as validation, and the rest are used as training samples. The green space classification model is trained using the training samples. Through multiple adjustments, the optimal values ​​of accuracy, precision, recall, and the harmonic mean of F1 precision and recall are reached. The hyperparameters at this point are selected to obtain the final required model. Finally, the test set data is input into the model to check whether the output green space classification model meets the requirements. If it meets the requirements, the green space classification model is completed; otherwise, the training is repeated.

[0090] Further explanation is needed. In this example, the VGG16 convolutional neural network model proposed by the Visual Geometry Group (VGG) at Oxford University was selected. The street green space structure dataset was randomly divided into training and validation sets. The training set contained approximately 500 samples of each street green space structure type unit. The training set was divided into 10 parts, and one of the 10 parts was used as the validation set each time, while the rest were used as the training set. The model convolution kernel size and stride parameters were adjusted until the optimal values ​​of accuracy, precision, recall, and the harmonic mean of F1 precision and recall were reached. The model convolution kernel size and stride parameters at this point were recorded. Using the model convolution kernel size and stride parameters at this point, the model was retrained using all 10 datasets as the training set to obtain the final model. The validation set and output results were compared using a confusion matrix to calculate the overall model accuracy. If the model accuracy reaches 80% or higher, the green space classification model is considered complete; otherwise, the training set samples need to be adjusted and retrained.

[0091] S26: Classification and Results Output of Urban Green Space Structure

[0092] Input all the green space units extracted in S22 into the green space structure classification model of the blocks constructed in S25, and output the classification results.

[0093] S3: Adaptability Analysis of "Spatial Morphology - Green Space Structure" in Neighborhoods

[0094] The spatial morphology and green space structure types of overlapping blocks are analyzed, with the distance from the cluster center point of each type of LCZ block's green space structure type to the origin O of the coordinate axis. As an evaluation index for the adaptability of the "spatial form - green space structure" of this type of LCZ block, and based on The values ​​are used to sort and classify the suitability of the green space structure type corresponding to this type of LCZ block.

[0095] Specifically, it includes:

[0096] By combining the spatial morphology of the block with the green space structure type, for each type of block spatial morphology, the cluster centers corresponding to different green space structure types are identified, such as type A (x A1 x A2 , ..., x An Type B (x) B1 x B2 , ..., x Bn ), ..., type M (x M1 y M2 , ..., z Mn Distance from the origin O of the coordinate axis As an evaluation index for the adaptability of "spatial form - green space structure" of this type of LCZ block, The higher the value, the better the fit.

[0097]

[0098] Where, x Ai Let be the value of structural index i in green space structure type A, and n be the number of green space structural indicators.

[0099] In this example, there are three types of green space structures, A, B, and C, within the compact high-rise block (LCZ) type, with cluster centers (x...). A1 ,x A2 ,x A3 ), (x B1 ,x B2 ,x B3 ), (x C1 ,x C2 ,x C3 ), calculate respectively , , ,like > > Therefore, the Class A green space structure is the most suitable for compact high-rise blocks, while the Class C green space structure is the least suitable for compact high-rise blocks.

[0100] Based on the actual situation and the type of street space, the adaptability of the green space structure types involved is classified from high to low.

[0101] In this example, for a certain type of street space form, the adaptability of the green space structure types involved is divided into three levels: high, medium, and low. The green space structure types with adaptability in the top 1 / 3 of the ranking are categorized as high adaptability for the street space form; those with adaptability in the top 1 / 3 to 2 / 3 of the ranking are categorized as medium adaptability; and those with adaptability in the bottom 1 / 3 of the ranking are categorized as low adaptability. Depending on actual needs, they can also be divided into two categories: high and low adaptability.

[0102] S4: Adjustment of Green Spaces in Neighborhoods

[0103] The adjustment of green spaces in the block includes the identification of areas for structural adjustment of green spaces in block S41 and the adjustment of specific units of green spaces in block S42.

[0104] S41: Identification of Areas for Adjusting the Structure of Green Spaces in Neighborhoods

[0105] Identify areas with low compatibility between "spatial morphology and green space structure" as the main areas for regulation;

[0106] Specifically, it includes:

[0107] Based on the "spatial form - green space structure" adaptation evaluation of the blocks obtained from S32, the units with low "spatial form - green space structure" adaptation are recorded as the main adaptation areas, and the blocks with medium adaptation are recorded as the secondary adaptation units.

[0108] S42: Specific Unit Adjustments for Block Green Spaces

[0109] Based on the adjustment areas obtained from S41, the structure type of green space in the blocks is changed by adding green spaces, altering the distribution of green spaces, and increasing the connectivity of green spaces, thereby completing the adjustment of the green space structure of various LCZ blocks.

[0110] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0111] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0112] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0113] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. A method for adjusting the structure of urban green spaces based on machine learning, characterized in that, The method includes the following steps: The remote sensing images of the first city were resampled to the required accuracy of the Local Climate Zone (LCZ) model. Based on the classification of urban morphology types in the LCZ theoretical model, the resampled remote sensing images were classified to obtain samples of various street block types, and a street block spatial morphology dataset was constructed. Based on the processed urban remote sensing images, a street spatial morphology training set is used as training samples to create street spatial morphology classification results. The street spatial morphology classification results are compared with the validation set, a confusion matrix is ​​established, and a street spatial morphology classification result with satisfactory classification accuracy is output. Input the research unit according to the research needs, take the block spatial morphology type with the largest proportion within the research unit as the block spatial morphology of the unit, and output the block spatial morphology type of the research area unit. The research units were classified and summarized according to the spatial morphology of the blocks, and a sufficient number of units with a high proportion of the main spatial morphology of the blocks were selected as representative samples. Standard false-color remote sensing images are obtained by performing image correction processing on remote sensing images of the second city. Urban green spaces are identified using machine learning algorithms based on the standard false-color remote sensing images and cross-verified with remote sensing images of the second city. The identification results are divided into blocks of green space units using research units that are the same as the spatial morphology of the blocks. Based on morphological theory, the characteristics of green space units in the block are extracted and classified by cluster analysis method to construct a system of green space structure types in the block, and the cluster center point of each green space structure type is calculated. Based on the constructed green space structure type system, representative block green space units were selected by category to construct a block green space structure dataset; The process of constructing a system of green space structure types in neighborhoods includes: A certain number of block units were randomly selected. Based on the morphological characteristics of green space within each unit, including green volume, connectivity, uniformity of volume, and balanced distribution, n corresponding indicators were selected to describe its spatial structure. Random sampling was performed on each type of LCZ block unit, with a units sampled each time. A block green space structure sample analysis database was constructed using these sampled units. K-means clustering analysis was used to explore and classify the green space structure samples of a single type of LCZ block, obtaining the corresponding green space structure type b1 for that type of LCZ block. The value range of each structural indicator for each type of green space structure was recorded. The previous clustering step was repeated for each type of LCZ block, resulting in a total of [number missing] indicators. The types of green space structures were identified; ultimately, the cluster centers of the block's green space structure type system and related structural indicators under each type were obtained. A preliminary classification model for the structure of green spaces in urban areas is constructed based on an image recognition model. The urban green space structure dataset is used to train, adjust, and test the classification model, and finally outputs a classification model for the structure of green spaces in urban areas with satisfactory verification accuracy. The classification results of the urban green space structure are obtained by inputting the urban green space units into the classification model for the structure of green spaces in urban areas with satisfactory verification accuracy. By combining the spatial morphology of the blocks with the green space structure type of the blocks, a configuration model system of "spatial morphology-green space structure" for the blocks is obtained; the adaptability index of "spatial morphology-green space structure" for each block is calculated and classified. Blocks with low compatibility between "spatial form and green space structure" are designated as the main adjustment areas; the green space structures of block-like areas are gradually replaced with green space structures that are highly compatible with the blocks, thus completing the adjustment of the green space structure of the blocks.

2. The method for adjusting the structure of urban green spaces based on machine learning according to claim 1, characterized in that, The acquisition process of the first city's remote sensing image and the second city's remote sensing image includes: Acquire high-precision multi-band remote sensing satellite images of the study area. The selection criteria include cloud cover of less than 10% and shooting time in summer (June-October). Crop the remote sensing images to the study area and resample the images to 100m-150m.

3. The method for adjusting the structure of urban green spaces based on machine learning according to claim 2, characterized in that, The image correction process for the second city remote sensing image includes: geographic calibration, radiometric calibration, image registration, image mosaicking, image cropping, and band synthesis; the final output is a standard false-color remote sensing image with an accuracy of 10m.

4. The method for adjusting the structure of urban green spaces based on machine learning according to claim 1, characterized in that, The process of constructing the street block spatial morphology dataset includes: According to LCZ, the spatial morphology of the street is divided into α types. Based on the definition of each type in LCZ, training samples are selected, with no less than 50 samples in each type. All training samples are summarized to construct a street spatial morphology dataset. The training samples of each type are divided into ten parts, with nine parts as the training set and one part as the test set.

5. The method for adjusting the structure of urban green spaces based on machine learning according to claim 1, characterized in that, A certain number of street blocks are randomly selected. Based on the morphological characteristics of green space volume, connectivity, volume uniformity, and distribution balance within each block, n corresponding indicators are selected to describe its spatial structure. The indicator description results of the green space structure of the sample blocks are calculated. Based on the sample results, the selected indicators are tested for collinearity, and collinear indicators are removed. Finally, m indicators are obtained to describe the characteristics of the green space structure of the blocks, and the green space structure indicators of each sample block are calculated. Random sampling is performed on each type of LCZ block block, with a units sampled each time. The sampled units are used to construct a sample analysis library of green space structure of the blocks. K-means clustering analysis is used to explore and classify the green space structure samples of a single type of LCZ block, obtaining the corresponding green space structure type b1 of the LCZ block, and recording the value range of each structural indicator of each type of green space structure. The previous clustering step is repeated for each type of LCZ block, and a total of m indicators are obtained. Classify green space structure types; compare the value ranges of structural indicators for all acquired green space structure types, and merge green space structure types with high overlap in the value ranges of various structural indicators; Finally, the cluster centers of the green space structure type system of the block and the relevant structural indicators under each type were obtained.

6. The method for adjusting the structure of urban green spaces based on machine learning according to claim 1, characterized in that, The process of constructing the neighborhood green space structure dataset includes: Select no fewer than 500 street block unit samples of various green space structures, and summarize the selected unit samples to construct a street block green space structure dataset; randomly divide the street block green space structure dataset into a training set and a test set.

7. The method for adjusting the structure of urban green spaces based on machine learning according to claim 1, characterized in that, The process of constructing the street green space structure classification model that meets the verification accuracy standard includes: A street green space structure classification model was built using deep learning image recognition tools. The training set was divided into 10 parts. One part was used for validation each time, and the rest were used for training. The green space structure classification model was trained using the training samples. Through multiple adjustments, the accuracy, precision, recall, and the harmonic mean of F1 precision and recall were optimized. The hyperparameters at this point were selected to obtain the final required model. Finally, the test set data was input into the model to check whether the output green space classification model met the requirements. If it met the requirements, the green space structure classification model was completed; otherwise, the training was repeated.

8. The method for adjusting the structure of urban green spaces based on machine learning according to claim 1, characterized in that, The process of analyzing the fit between the "spatial morphology and green space structure" of the neighborhood is as follows: The spatial morphology and green space structure types of overlapping blocks are analyzed, with the distance from the cluster center point of each type of LCZ block's green space structure type to the origin O of the coordinate axis. As an evaluation index for the adaptability of the "spatial form - green space structure" of the LCZ block, and based on The values ​​are used to rank and classify the suitability of the green space structure type for this type of LCZ block: Where, x Ai Let be the value of structural index i in green space structure type A, and n be the number of green space structural indicators.

9. The method for adjusting the structure of urban green spaces based on machine learning according to claim 8, characterized in that, The adjustment process for the green space structure of the block is as follows: Units with lower "spatial morphology-green space structure" compatibility levels in various LCZ blocks were selected as the main adjustment areas. By comparing the types of green space structures covered by various LCZ blocks with their compatibility levels, the green space structure types with lower compatibility with the spatial morphology of LCZ blocks were converted into green space structure types with higher compatibility, thus completing the green space structure adjustment of various LCZ blocks.

10. A device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more of the programs are executed by one or more of the processors, the one or more of the processors implement the machine learning-based method for adjusting the structure of green spaces in urban blocks as described in any one of claims 1-9.

Images