Regulation and control method and system adapting to block green space structure evolution mechanism

Abstract

The invention discloses a regulation and control method and system suitable for a block green space structure evolution mechanism, and relates to the technical field of urban planning, landscape gardens and artificial intelligence, and the method comprises the steps: receiving a high-resolution remote sensing image, carrying out the preprocessing of the high-resolution remote sensing image through ArcGIS software, obtaining a processed remote sensing image, and carrying out the processing of the processed remote sensing image; inputting the processed remote sensing image into a pre-established green space classification model, and outputting to obtain a trained green space classification model; receiving a research sample, inputting the research sample into the trained green space classification model, determining a land use type file in the research sample based on a classification result, and calculating a green space structure index based on the land use type file in the research sample; adopting a Spearman coefficient to screen influence factors significantly related to the structure indexes, constructing an initial data set of the shrunk city plots, and training a regression model to analyze a spatial evolution mechanism; and predicting a green space structure index in a natural evolution state of a target year through the trained regression model, comparing the green space structure index with an upper planning preset target value, and iteratively optimizing an influence factor parameter to reduce the difference so as to obtain a planning regulation and control scheme.

Description

Technical Field

[0001] The present invention relates to the technical fields of urban planning, landscape architecture, and artificial intelligence. Specifically, it is a regulation method and system adapted to the evolution mechanism of the green space structure in blocks. Background Art

[0002] The phenomenon of urban shrinkage is becoming increasingly significant globally, leading to a series of social and environmental problems. Green space is an effective means to mitigate the negative impacts of urban shrinkage. And the green space structure significantly affects the exertion of green space benefits. However, the existing planning and management methods have limitations in the research of green space structure, such as coarse research granularity (mostly at the urban scale) and insufficient consideration of shrinkage scenarios. Summary of the Invention

[0003] To solve the deficiencies mentioned in the above background art, the purpose of the present invention is to provide a regulation method and system adapted to the evolution mechanism of the green space structure in blocks.

[0004] In the first aspect, the purpose of the present invention can be achieved through the following technical solutions: A regulation method adapted to the evolution mechanism of the green space structure in blocks, the method comprising the following steps:

[0005] Receive high-resolution remote sensing images, preprocess the high-resolution remote sensing images using ArcGIS software to obtain processed remote sensing images, and input the processed remote sensing images into a pre-established green space classification model to output a trained green space classification model;

[0006] Receive research samples, input the research samples into the trained green space classification model, determine the green space type files in the research samples based on the classification results, and calculate the green space structure indicators based on the land use type files in the research samples;

[0007] Use the Spearman coefficient to screen the influencing factors significantly correlated with the structure indicators, construct an initial dataset of shrinking urban plots, and train a regression model to analyze the spatial evolution mechanism;

[0008] Predict the green space structure indicators in the natural evolution state of the target year through the trained regression model, compare them with the preset target values in the upper-level plan, and conduct multiple rounds of iteration and directional guidance on the green space structure plan by optimizing the influencing factor parameters, continuously narrowing the difference between the predicted structure plan indicator values and the preset target values to obtain the final planning regulation plan.

[0009] In combination with the first aspect, in some implementation manners of the first aspect, the method further comprises: The acquisition process of the high-resolution remote sensing images includes:

[0010] Obtain high-resolution remote sensing images of the target year for the research area. Among them, the high-resolution remote sensing images are from Google Maps, Esri World Image, Tianditu, GeoEye, Quickbird, and Ikonos.

[0011] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: the process of preprocessing the high-resolution remote sensing images by using ArcGIS software:

[0012] Use the ArcGISpro classification tool and the polygon selection tool. Based on the high-resolution remote sensing image, frame out the training samples, and stipulate that the number of samples of each type is not less than the preset number and is evenly distributed within the research area; in the attribute table, represent different classes according to the modified value of the field value; randomly divide the training set and the validation set at a preset ratio, input the training set raster image, and perform segmentation with label as the input feature class and value as the class value field, and finally obtain the processed remote sensing image.

[0013] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: the land use types of the pre-established green space classification model include multiple green spaces and non-green spaces. The multiple green spaces include: natural forests, natural shrublands, natural wetlands, artificial forests, and artificial grasslands.

[0014] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: the pre-established green space classification model uses ResNet-50 as the backbone model, and the formula is as follows:

[0015] y = F(x, W i ) + x

[0016] where F(x, W i ) is the output of the layer with weight W i , x is the input, and y is the output.

[0017] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: the green space structure index is measured from three aspects: scale, type, and space. The green space structure index part reflects the inherent characteristics of the green space in the block raster. In terms of scale, calculate the green quantity of the block raster, that is, the sum of the areas of all green spaces, denoted as P1; in terms of type, calculate the ratio of the area of natural green space to the area of artificial green space in the block raster, denoted as P2; in terms of spatial layout, calculate the average value of the minimum Euclidean distances between green spaces in the block raster, denoted as P3;

[0018] Among them, the natural green spaces include land use types: natural forests, natural shrublands, and natural wetlands; the artificial green spaces include land use types: artificial forests and artificial grasslands.

[0019] The calculation formula of P1 is as follows:

[0020]

[0021] In the formula, A i represents the area of the i-th type of green space, and n represents the total number of green space types;

[0022] The calculation formula of P2 is as follows:

[0023]

[0024] In the formula, A 自然,j represents the area of the j-th type of natural green space, A 人工,k represents the area of the k-th type of artificial green space, m represents the total number of natural green space types, and p represents the total number of artificial green space types;

[0025] The calculation formula of P3 is as follows:

[0026]

[0027] In the formula, N is the number of green spaces, (x l , y l ) and (x m , y m ) are the central coordinates of the l-th and m-th green spaces respectively.

[0028] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: the potential influencing factors of the green space structure are obtained by associating the green space structure with the block shrinkage situation, denoted as E{E1, E2, E3,..., En};

[0029] The construction of the initial dataset of the shrinking urban land parcels:

[0030] Each remote sensing image is divided into a matrix D0, Μ×Ν, according to the block grid, where each element V ij represents the block grid of the i-th row and the j-th column, where 1≤i≤M and 1≤j≤N;

[0031]

[0032] Each element V ij is represented in the form of an attribute vector. For an element V ij =V ij0 , V ij1, \, V ijn};

[0033] V ij0 is the evolution attribute vector of the green space structure, V ij0 = {P1, P2,..., P3}, and the remaining attributes V ij1 …V ijn then respectively correspond to the obtained potential correlation indicators E1, E2, E3,..., En.

[0034] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: the regression model constructs an artificial neural network using a time series regression prediction model based on the multi-layer perceptron (MLP) structure, which is composed of a multi-layer fully connected network, and uses the Leaky ReLU activation function after each layer;

[0035]

[0036] After each round of training, use the test set to evaluate the training results, and measure the quality of the regression model through the mean square error (MSE) and the coefficient of determination R 2 The calculation method is as follows:

[0037]

[0038] where, y i is the actual value of the test set, the predicted value of the neural network, is the average value of the test set, and n is the number of samples in the test set.

[0039] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: predicting the green space structure index in the natural evolution state of the target year through the trained regression model, comparing it with the preset target value of the upper-level plan, and carrying out multiple rounds of iteration and directional guidance on the green space structure plan by optimizing the influencing factor parameters, continuously narrowing the difference between the predicted structure plan index value and the preset target value to obtain the final planning and regulation plan.

[0040] In a second aspect, to achieve the above object, the present invention discloses a regulation system adapted to the evolution mechanism of the block green space structure, including:

[0041] An image processing module, configured to receive a high-resolution remote sensing image, preprocess the high-resolution remote sensing image using ArcGIS software to obtain a processed remote sensing image, and input the processed remote sensing image into a pre-established green space classification model to output a trained green space classification model;

[0042] A sample processing module for receiving research samples, inputting the research samples into the trained green space classification model, determining the land use type file in the research samples based on the classification results, and calculating the green space structure index based on the land use type file in the research samples;

[0043] A model training module for analyzing the potential influencing factors of the green space structure related to each index of the green space structure through Spearman correlation coefficient analysis, selecting them as regression independent variables to obtain the initial dataset of shrinking urban plots, inputting the initial dataset of shrinking urban plots into a pre-established regression model, outputting the trained regression model, and analyzing the evolution mechanism of the green space structure of the block based on the trained regression model;

[0044] A planning and regulation module for predicting the green space structure index under the natural evolution state in the target year through the trained regression model, comparing it with the preset target value of the upper-level plan, and obtaining the final planning and regulation plan by iteratively optimizing the influencing factor parameters.

[0045] Advantages of the present invention:

[0046] The present invention improves the accuracy and efficiency of green space classification: Through the application of the deep learning model, pixel-level classification of remote sensing images is achieved, significantly improving the accuracy and efficiency of green space classification and providing a reliable data basis for subsequent analysis. It realizes the analysis of land use type changes in multiple periods: The present invention can compare the land use type distribution maps of different years, deeply analyze the dynamic changes of urban land use types, especially the conversion of land use types during the process of urban shrinkage, providing important references for urban planning and environmental protection. It reveals the relationship between green space structure changes and urban shrinkage: By extracting variables related to green space structure changes and constructing a regression model for regression analysis, the present invention reveals the quantitative relationship between green space structure changes and urban shrinkage, providing a scientific basis for optimizing and regulating the green space structure. It provides a method for formulating an optimization and regulation plan for the green space structure: Based on the prediction results of the regression model, an optimization and regulation plan for the green space structure can be formulated through the present invention, and spatial simulation can be carried out using ArcGIS to evaluate the impact of the plan, providing a feasible solution for urban planning and environmental protection. It promotes the intelligent development of urban planning and environmental protection: The implementation of the present invention significantly improves the efficiency and accuracy of urban planning and environmental protection, promoting the intelligent development in this field. At the same time, significant improvements have been achieved in aspects such as analysis accuracy, cross-scale analysis ability, and data integration efficiency, providing strong support for the construction of smart cities and sustainable cities. It has a wide range of application scopes and social and economic values: The present invention can not only be applied to the field of urban planning, but also be extended to multiple fields such as ecological assessment and land resource management, with significant social and economic values, providing strong technical support for decision-making in multiple fields. Brief Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings;

[0048] Figure 1 is a schematic flowchart of the method of the present invention;

[0049] Figure 2 is a schematic diagram of the annotation of the classification model dataset of the present invention;

[0050] Figure 3 is a recognition result diagram of the green space recognition model of the present invention;

[0051] Figure 4 is a schematic diagram of the operation setting interface of the classification model training program of the present invention;

[0052] Figure 5 is a schematic diagram of the system structure of the present invention;

[0053] Figure 6 is the fitting result of the regression model of the present invention;

[0054] Figure 7 is the basic architecture of the green space recognition model of the present invention;

[0055] Figure 8 is the basic architecture of the regression model of the present invention. Detailed Embodiments

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0057] Embodiment 1:

[0058] As Figure 1 shown, a regulation method adapted to the evolution mechanism of the block green space structure, the method includes the following steps:

[0059] Receive high-resolution remote sensing images, preprocess the high-resolution remote sensing images using ArcGIS software to obtain processed remote sensing images, and input the processed remote sensing images into a pre-established green space classification model to output a trained green space classification model;

[0060] Specifically, the research unit of this application: Determine shrinking cities based on census data. Considering the road network spacing in the research area and the planning requirements for "15-minute community living circle residential areas" in the "Urban Residential Area Planning and Design Standard", the research area is delimited into block-scale research units with a side length of 1 km, hereinafter referred to as block units.

[0061] In this example, six shrinking cities are selected as the research area. According to the general road interval distance and block size in the planning area, the planning area is divided into block research units of 1000m * 1000m.

[0062] The acquisition process of the high-resolution remote sensing images includes:

[0063] Obtain high-resolution remote sensing images of the research area for the target year. The remote sensing images can come from free high-resolution data such as Google Maps, Esri World Image, and Tianditu, or paid high-resolution data such as the high-resolution series, GeoEye, Quickbird, and Ikonos. When screening the data, it is necessary to ensure that the cloud cover of the images is less than 10%, and the spatial resolution is not less than 2.5 m.

[0064] In this example, 17-level Google Maps remote sensing images for 2014, 2018, and 2022 are downloaded, and their spatial resolution is 2.15 m.

[0065] The process of preprocessing the high-resolution remote sensing images using ArcGIS software:

[0066] According to the definition of urban vacant land and the actual situation of the research site, urban land is divided into n types. According to the land type definition, use the ArcGISpro classification tool to select training samples, and use ResNet-50 as the backbone model for automatic classification based on the samples, and use accuracy indicators to evaluate the classification accuracy.

[0067] It should be further noted that the land type system is subdivided into 7 land types including 6 green spaces and non-green spaces. Among them, green space refers to the land covered by vegetation, specifically including: natural forest, natural shrubland and grassland, natural wetland, artificial forest, artificial grassland; non-green space is other land types except green space.

[0068] Among them, green space refers to "land that is partially or completely covered by herbs, trees, shrubs, or other vegetation", and non-green space is relative to green space. It refers to those lands or areas that are not covered by vegetation or do not have the functions of green space;

[0069] Using the ArcGIS pro classification tool and the polygon selection tool, based on the remote sensing images collected in S21, select the training samples by drawing a polygon box, ensuring that the number of samples per type is not less than 100 and they are evenly distributed within the study area; in the attribute table, represent different classes by modifying the value of the field "value": 1 to 7 represent 7 classes respectively, and 0 is the unknown class (default value); randomly divide the dataset and the validation set in an 8:2 ratio, input the training set raster image, and perform the split with "label" as the input feature class and "value" as the class value field (in pixels, the block size is 256*256, the stride is 128*128, and the metadata format selects classification slices). Finally, obtain the processed remote sensing image and save it in the "traindata" folder for subsequent deep learning model training.

[0070] Input multiple "traindata" folders for training. The number of training epochs is 200, the model type uses DeepLabV3 (pixel classification), the batch size is 8, the proportion of the validation set is 0.1, and the backbone model uses ResNet-50.

[0071] Use the accuracy evaluation tool to evaluate the recognition results. In this example, output the green space classification results with a model accuracy of over 80%. The recognition accuracy is consistent with the spatial resolution of the input image. When the recognition accuracy of the model for the validation set reaches over 75%, the training of the green space classification model is completed.

[0072] Use the trained land use classification model to classify the research samples, output the land use recognition results with qualified classification accuracy, and according to the research needs, resample the land use data of the input research unit to 20m using the mode resampling method to obtain all land use type files with a resolution of 20m*20m.

[0073] The pre-established green space classification model uses ResNet-50 as the backbone model, and the formula is as follows:

[0074] y = F(x, W i ) + x

[0075] where F(x, W i ) is the output of the layer with weight W i , x is the input, and y is the output.

[0076] Among the available backbone network models in ArcGIS pro, VGG, Resnet, etc. can be used. Among them, Resnet is the most widely used and most stable backbone model at present, and it is also the most common backbone model for land use classification based on ArcGIS pro. The recognition accuracy is shown in Table 1.

[0077] Table 1 Green Space Type Recognition Model Accuracy Data

[0078]

[0079] Receive research samples, input the research samples into the trained green space classification model, determine the land use type file in the research samples based on the classification results, and calculate the green space structure index based on the land use type file in the research samples;

[0080] Calculation of the green space structure index:

[0081] On the 1000m * 1000m block grid scale, analyze the evolution law of the green space structure, and extract and calculate the variables related to the change of the green space structure.

[0082] In this example, according to the existing research and the land use type summary file in S23, calculate from three aspects of the scale, type, and space of the green space structure index. In the green space structure index part, reflect the inherent characteristics of the green space in the block grid. In terms of scale, calculate the green quantity of the block grid, that is, the sum of the areas of all green spaces, denoted as P1; in terms of type, calculate the ratio of the area of natural green space to the area of artificial green space in the block grid, denoted as P2; in terms of spatial layout, calculate the average value of the minimum Euclidean distance between green spaces in the block grid, denoted as P3. Among them, natural green spaces include land use types: natural forests, natural shrublands, and natural wetlands; artificial green spaces include land use types: artificial forests, artificial grasslands. During the calculation process, to simplify the calculation, use the number of 20m * 20m land use type grids corresponding in the block grid to replace the area, and use the distance between the grid center points as the calculation basis.

[0083] Specifically, the reasons for describing the green space structure from the aspects of scale, type, and space: Combining the existing relevant literature, the scale describes the overall level of the green space, the type describes the composition of the green space, and the space describes the distribution characteristics of the green space. The three aspects comprehensively describe the green space structure from different perspectives. In previous studies, the green space was usually described only from one aspect, and a comprehensive display of the structure could not be achieved.

[0084] Reasons for selecting specific indicators:

[0085] It is a relatively common practice to use the green quantity and the proportion of the areas of different types of green spaces to reflect the scale and type; for the description of space, indicators such as uniformity and aggregation can also be used. In this study, the average value of the minimum Euclidean distance is selected because this indicator has no correlation with the scale and type indicators and can describe the characteristics of the green space structure from an independent dimension.

[0086] The calculation formula of the above-mentioned P1 is as follows:

[0087]

[0088] In the formula, A i represents the area of the i-th type of green space, and n represents the total number of green space types;

[0089] The calculation formula of P2 is as follows:

[0090]

[0091] In the formula, A 自然,j represents the area of the j-th type of natural green space, A 人工,k represents the area of the k-th type of artificial green space, m represents the total number of natural green space types, and p represents the total number of artificial green space types;

[0092] The calculation formula of P3 is as follows:

[0093]

[0094] In the formula, N is the number of green spaces, (x l , y l ) and (x m , y m ) are the central coordinates of the l-th and m-th green spaces respectively.

[0095] The Spearman coefficient is used to screen the influencing factors significantly correlated with the structural indicators, construct the initial dataset of shrinking urban plots, and train a regression model to analyze the spatial evolution mechanism.

[0096] The potential influencing factors of the green space structure are listed according to the aspects related to the green space structure and the context of block shrinkage, E{E1, E2, E3,..., En};

[0097] According to previous studies, the associations between the green space structure and the context of block shrinkage include but are not limited to the following categories: social economy, public services, development and construction, and natural environment. Further, a series of representative indicators are selected as potential influencing factors.

[0098] In this example, after confirming a strong potential correlation between the potential influencing factors and the dependent variables P1, P2, and P3 using the correlation analysis method of Spearman's correlation coefficient, the independent variables for regression are selected. In terms of social economy, the following are selected: block population (E1), lighting index (E2), population structure (E3), GDP (E4); in terms of public services, the following are selected: commercial facility POI (Point of Interest) (E5), office facility POI (E6), sports and leisure service facility POI (E7), cultural and educational facility POI (E8); in terms of development and construction, the following are selected: building density (E9), impervious surface area (E10), building height (E11), road network information (E12); in terms of natural environment, the following are selected: average elevation (E13), slope (E14), average annual temperature (E15), average annual precipitation (E16), PM2.5 index (E17). It should be noted that since the corresponding block grid in the method is at a scale of 1000m * 1000m, for data that is simultaneously located in multiple block grids, the average value is calculated based on the total number of block grids it is located in and then included in the data of each grid it is located in.

[0099] The specific data sources and calculation methods are shown in Table 2 below:

[0100] Table 2 Table of potential influencing factors

[0101]

[0102]

[0103] Construction of the initial dataset of shrinking city plots:

[0104] First, each remote sensing image is divided into a matrix D0(Μ×Ν) according to its 1000m * 1000m block grid, where each element V ij (1 ≤ i ≤ M, 1 ≤ j ≤ N) represents the block grid in the i-th row and j-th column.

[0105]

[0106] Each element V ij is represented in the form of an attribute vector. For an element V with n + 1 attributes ij = {V ij0 , V ij1 , …, V ijn}.

[0107] In this example, n = 17, which consists of the above 17 potential influencing factors of shrinking cities and the attribute vector of the green space structure evolution of this 1000m * 1000m block grid.

[0108] Specifically, Vij0 is the attribute vector of the evolution of the green space structure. In this example, combined with S32, V ij0 = {P1, P2,..., P3}. In other application cases, the attribute vector configuration should be adjusted according to the actual situation. The remaining attributes V ij1 …V ijn respectively correspond to the obtained potential correlation indicators E1, E2, E3,..., En.

[0109] Construction of the pre-established regression model:

[0110] Construct the change value ΔV of the attribute variables of the grid of blocks with the same number in adjacent research years of the same city in the initial database ijt (0 ≤ i ≤ n). In this example, n = 17. Subtract the V of 2022 from that of 2018, and the V of 2018 from that of 2014 ij1 to V ijn respectively in turn to get ΔV ij0 to ΔV ijn (denoted as ΔV).

[0111] Use a standardization or normalization algorithm to process ΔV to ensure the numerical stability during model training. In this example, the Z-Score standardization algorithm is adopted, which can be implemented by calling the StandardScaler function of the preprocessing tool in the scikitlearn library of Python, making the mean of the processed data 0 and the standard deviation 1 for the efficient convergence of gradient descent.

[0112] Take ΔV ij0 as the dependent variable, and ΔV ij1 to ΔV ijn as the independent variables to divide the dataset. Randomly divide the attribute vector V ij into a training set and a test set in a ratio of 8:2, and use the training set for training and the test set to evaluate the training results.

[0113] Regression models are constructed using regression methods such as the random forest regression algorithm, the logistic regression algorithm, or the artificial neural network algorithm. In this example, since there are many independent variables, an artificial neural network is constructed using a time series regression prediction model based on the structure of a multi-layer perceptron (MLP) to capture complex non-linear and time series logical relationships. The neural network consists of three fully connected layers. The first layer (self.fc1): input feature dimension -> 64 dimensions, the second layer (self.fc2): 64 dimensions -> 32 dimensions, and the output layer (self.fc3): 32 dimensions -> 1 dimension (the predicted value of the regression task). Due to the good effect of the piecewise linear function in dealing with non-linear relationships and its property of effectively preventing gradient disappearance, the Leaky ReLU (Leaky Rectified Linear Unit) activation function is used after each layer.

[0114]

[0115] During the training process, the training performance and training effect can be optimized by adjusting parameters such as the learning rate, the number of iterations, the regularization parameter, the number of neurons in the hidden layer, and the neural network structure.

[0116] After each round of training, the test set is used to evaluate the training results, and the mean squared error (MSE) and the coefficient of determination (R 2 ) are used to measure the quality of the model. The calculation method is as follows:

[0117]

[0118] where y i is the actual value of the test set, the predicted value of the neural network, is the average value of the test set, and n is the number of samples in the test set.

[0119] The Adam optimizer is used as the backpropagation optimizer to combine the advantages of momentum optimization and RMSprop, and to accelerate the convergence speed in the process of finding complex relationships among a large number of metrics.

[0120] In this example, when R 2 is greater than a certain threshold and the MSE is small, the model is considered to have a good explanatory ability, and the training of the regression model is completed.

[0121] During the addition of the regression model, the selectable regression algorithms include the random forest regression algorithm, the logistic regression algorithm, the artificial neural network algorithm, etc. In this example, due to the large number of independent variables, the multi-layer perceptron algorithm is adopted. The multi-layer sensor algorithm is based on the artificial neural network, which can handle highly complex non-linear relationships and patterns, has strong expressive ability, and has been proven to have stronger regression ability than polynomial regression, random forest regression, and decision tree regression in related research. In contrast, although polynomial regression can fit non-linear relationships, it has limited flexibility and is prone to overfitting; decision tree regression is easy to understand and interpret, but it performs poorly when dealing with complex relationships; although random forest regression improves accuracy by integrating multiple decision trees, parameter adjustment is complex and it may overfit. MLP regression performs excellently on large-scale datasets and can be trained end-to-end through deep learning methods. Therefore, this method uses MLP to build the regression model, as shown in Table 3.

[0122] Table 3 R-squared values of the MLP regression model

[0123] Dependent variable R-squared P1 0.44 P2 0.43 P3 0.37

[0124] Regression model weight analysis

[0125] In the trained MLP model, extract the weights between the input layer and the first hidden layer. These weights represent the influence of each input feature on the hidden layer neurons. The attributes can be qualitatively evaluated by analyzing the positive and negative values and the absolute values of the weights. Among them, the positive and negative values of the weights indicate the influence direction of the independent variable on the dependent variable. If the weight is positive, it means that the independent variable has a positive influence on the dependent variable; if the weight is negative, it means that the independent variable has a negative influence on the dependent variable. The absolute value of the weight represents the influence intensity of the independent variable on the dependent variable. The larger the absolute value of the weight, the stronger the influence of the independent variable on the dependent variable.

[0126] Regression model sensitivity analysis

[0127] From the trained MLP model, select input features for sensitivity analysis. While keeping other input features unchanged, gradually change the values of the selected input features (independent variables) and observe the changes in the output (dependent variable). Calculate the sensitivity of each input feature to the output through numerical differentiation or the finite difference method, that is, the ratio of the change in the output to the change in the input. The results of the sensitivity analysis can further verify the results of the weight analysis and provide more intuitive information on the influence direction and intensity. If a certain input feature has high sensitivity, it means that it has a greater influence on the output, which corresponds to the large weight value in the weight analysis. Both are helpful to find out which factors have significantly promoted the change of the green space structure during the process of urban shrinkage.

[0128] In this example, through the above two types of analysis and synthesis, it is concluded that the change of the green space structure is significantly correlated with a total of eight attributes, namely E1, E2, E3, E4, E9, E10, E15, and E16 in S41.

[0129] Formulation of the optimization and regulation plan for the green space structure:

[0130] First, based on the regression model, obtain the green space structure index in the natural evolution state. Subsequently, use the deviation between the predicted value and the preset target value of the upper-level plan as the optimization orientation. When the difference is positive (i.e., the target value is greater than the predicted value), adjust the influencing factor system by increasing the positively correlated parameters and decreasing the negatively correlated parameters. Conversely (i.e., the target value is less than the predicted value), decrease the positively correlated parameters and increase the negatively correlated parameters. Then, input the adjusted parameters into the model to verify the prediction results, and iterate repeatedly until the predicted value and the target value are fully matched (i.e., the absolute value of the difference is the smallest). Finally, formulate a planning regulation plan based on the difference between the optimal parameter combination and the current situation parameters.

[0131] Example 2: Second, as Figure 6 shown, to achieve the above objectives, the present invention discloses a regulation system adapted to the evolution mechanism of the green space structure in the block, including:

[0132] An image processing module 11, configured to receive high-resolution remote sensing images, preprocess the high-resolution remote sensing images using ArcGIS software to obtain processed remote sensing images, and input the processed remote sensing images into a pre-established green space classification model to output a trained green space classification model;

[0133] A sample processing module 12, configured to receive research samples, input the research samples into the trained green space classification model, determine the land use type file in the research samples based on the classification results, and calculate the green space structure index based on the land use type file in the research samples;

[0134] A model training module 13, configured to use the Spearman coefficient to screen the influencing factors significantly correlated with the structure index, construct an initial dataset of shrinking urban plots, and train a regression model to analyze the spatial evolution mechanism.

[0135] A planning regulation module 14, configured to predict the green space structure index in the natural evolution state of the target year through the trained regression model, compare it with the preset target value of the upper-level plan, and reduce the difference by iteratively optimizing the influencing factor parameters to obtain the final planning regulation plan.

[0136] Based on the same inventive concept, the present invention further provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or may also be 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 core and control core of the terminal, and is used to implement one or more instructions. Specifically, it is used to load and execute one or more instructions in the computer storage medium to implement the above method.

[0137] It should be further noted that, based on the same inventive concept, the present invention further provides a computer storage medium, on which a computer program is stored, and the computer program, when run by a processor, executes the above method. The storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but not be limited to, be an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a Random Access Memory (RAM), a Read Only Memory (ROM), an Erasable Programmable Read Only Memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or combined with an instruction execution system, apparatus, or device.

[0138] In the description of this specification, the description with reference to terms such as "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0139] The foregoing has shown and described the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and what is described in the above embodiments and the specification only illustrates the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure will have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure claimed.

Claims

1. A regulation method adapted to the evolution mechanism of the green space structure in the block, characterized in that The method includes the following steps: Receive high-resolution remote sensing images, preprocess the high-resolution remote sensing images using ArcGIS software to obtain processed remote sensing images, and input the processed remote sensing images into a pre-established green space classification model to output a trained green space classification model; Receive research samples, input the research samples into the trained green space classification model, determine the green space type file within the research samples based on the classification results, and calculate the green space structure index based on the green space type file within the research samples; Use the Spearman coefficient to screen the influencing factors significantly correlated with the structure index, construct an initial dataset of shrinking urban plots, and train a regression model to analyze the spatial evolution mechanism; Predict the green space structure index under the natural evolution state in the target year through the trained regression model, compare it with the preset target value of the upper-level plan, and obtain a planning control scheme by iteratively optimizing the influencing factor parameters to narrow the difference.

2. The regulation method for adapting to the evolution mechanism of the block green space structure according to claim 1, wherein, The acquisition process of the high-resolution remote sensing images includes: Obtain high-resolution remote sensing images of the research area in the target year, where the high-resolution remote sensing images come from one or more of Google Maps, Esri WorldImage, Tianditu, GeoEye, Quickbird, and Ikonos.

3. The regulation method for adapting to the evolution mechanism of the green space structure in the block according to claim 1, characterized in that, The process of preprocessing the high-resolution remote sensing images using ArcGIS software: Use the ArcGISpro classification tool and the polygon selection tool. Based on the high-resolution remote sensing image, select training samples, and stipulate that the number of samples of each type is not less than the preset number and is evenly distributed within the research area; in the attribute table, represent different classes by modifying the value of the field value; randomly divide the dataset and the validation set at a preset ratio, input the training set raster image, and perform segmentation with label as the input feature class and value as the class value field to finally obtain the processed remote sensing images.

4. A regulation method adapted to the evolution mechanism of the green space structure in the block according to claim 1, characterized in that, The land use types of the pre-established green space classification model include multiple green spaces and non-green spaces. The multiple green spaces include: natural forests, natural shrublands, natural wetlands, artificial forests, and artificial grasslands.

5. A regulation method adapted to the evolution mechanism of the green space structure in the block according to claim 4, characterized in that, The pre-established green space classification model uses ResNet-50 as the backbone model, and the formula is as follows: y = F(x, W i ) + x where F(x, W i ) is the output of the layer with weight W i , x is the input, and y is the output.

6. The regulation method for adapting to the evolution mechanism of the block green space structure according to claim 1, characterized in that, The green space structure index is measured from three aspects: scale, type, and space. In the part of the green space structure index, it reflects the inherent characteristics of the green space in the block raster. In terms of scale, calculate the green space volume in the block raster, that is, the sum of the areas of all green spaces, denoted as P1; in terms of type, calculate the ratio of the area of natural green space to the area of artificial green space in the block raster, denoted as P2; in terms of spatial layout, calculate the average value of the minimum Euclidean distances between green spaces in the block raster, denoted as P3; Among them, the natural green space includes land use types: natural forests, natural shrublands, and natural wetlands; the artificial green space includes land use types: artificial forests and artificial grasslands; The calculation formula of P1 is as follows: Where, A i represents the area of the i-th type of green space, and n represents the total number of green space types; The calculation formula of P2 is as follows: where A 自然,j represents the area of the j-th type of natural green space, and A 人工,k represents the area of the k-th type of artificial green space, m represents the total number of natural green space types, and p represents the total number of artificial green space types; The calculation formula of P3 is as follows: where N is the number of green spaces, (x l , y l ) and (x m , y m ) are the central coordinates of the l-th and m-th green spaces, respectively.

7. A regulation method adapted to the evolution mechanism of the green space structure in a block, as claimed in claim 1, wherein The potential influencing factors of the green space structure are obtained by correlating the green space structure with the context of block shrinkage, denoted as E{E1, E2, E3, ..., En}; Construction of the initial dataset of shrinking urban plots: Each remote sensing image is divided into a matrix D0, M×N, according to the block grid, where each element V ij represents the block grid of the i-th row and the j-th column, where 1 ≤ i ≤ M and 1 ≤ j ≤ N; Each element V ij is represented in the form of an attribute vector. For an element V with n + 1 attributes ij = {V ij0 , V ij1 , …, V ijn}; V ij0 is the evolution attribute vector of the green space structure, V ij0 = {P1, P2,..., P3}, and the remaining attributes V ij1 ...V ijn correspond to the obtained potential correlation indicators E1, E2, E3,..., En respectively.

8. A regulation method adapted to the evolution mechanism of the block green space structure according to claim 1, characterized in that, The regression model constructs an artificial neural network using a time series regression prediction model based on the multi-layer perceptron (MLP) structure, which consists of a multi-layer fully connected network, and uses the Leaky ReLU activation function after each layer; After each round of training is completed, the test set is used to evaluate the training results. The mean squared error (MSE) and the coefficient of determination R 2 are used to measure the quality of the regression model. The calculation methods are as follows: Among them, y i is the actual value of the test set, the predicted value of the neural network, is the average value of the test set, and n is the number of samples in the test set.

9. A regulation system adapted to the evolution mechanism of the green space structure in the block, characterized in that, Including: An image processing module for receiving high-resolution remote sensing images, preprocessing the high-resolution remote sensing images using ArcGIS software to obtain processed remote sensing images, and inputting the processed remote sensing images into a pre-established green space classification model to output a trained green space classification model; A sample processing module for receiving research samples, inputting the research samples into the trained green space classification model, determining the land use type files in the research samples based on the classification results, and calculating the green space structure indicators based on the land use type files in the research samples; A model training module that uses the Spearman coefficient to screen the influencing factors significantly correlated with the structure indicators, constructs an initial dataset of shrinking urban plots, and trains the regression model to analyze the spatial evolution mechanism; A planning and regulation module that predicts the green space structure indicators in the natural evolution state of the target year through the trained regression model, compares them with the preset target values of the upper-level plan, and obtains a planning and regulation plan by iteratively optimizing the influencing factor parameters to narrow the differences.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program that can run on the processor. When the processor loads and executes the computer program, it adopts a regulation method for analyzing the evolution of the green space structure of blocks according to any one of claims 1 to 9.

Images