A digital management method for towns based on knowledge graph

Through knowledge graph technology, the use of multivariate data and remote sensing image processing has solved the problem of low efficiency in traditional urban management, achieved efficient urban resource management and environmental protection, and promoted sustainable urban development.

CN118313560BActive Publication Date: 2024-12-06CHINA NAT INST OF STANDARDIZATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410577508.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-10
Publication Date
2024-12-06
Estimated Expiration
2044-05-10

AI Technical Summary

Technical Problem

The lack of information technology support in traditional urban management leads to low management efficiency and decision-making quality, inefficient resource utilization, resulting in waste of resources and environmental problems.

Method used

Using a knowledge graph-based urban digital management method, filter out low-efficiency land that needs to be optimized by obtaining multivariate data, preprocessing remote sensing image data, building land division models, and calculating the spatial distribution and spatial aggregation values ​​of low-efficiency land, and outputting results to improve Management accuracy and efficiency.

Benefits of technology

It improves the accuracy and efficiency of digital management of urban areas, saves resources and labor costs, promotes sustainable urban development, and is suitable for management of different towns.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118313560B_ABST
    Figure CN118313560B_ABST
Patent Text Reader

Abstract

The present invention discloses a digital management method for towns based on knowledge graphs, including obtaining multivariate data of towns, preprocessing the multivariate data; obtaining ideal values ​​of indicators, actual values ​​of indicators and weights of indicators according to the multivariate data; constructing a land use division model based on a convolutional neural network, wherein the land use division model divides the multivariate data into different types of land use patches; obtaining standard values ​​of indicators according to the actual values ​​of indicators and the ideal values ​​of indicators; screening out inefficient land use patches in the land use patches according to the standard values ​​of indicators and the weights of indicators; obtaining spatial distribution of inefficient land use according to the inefficient land use patches, calculating spatial aggregation values ​​according to the spatial distribution, screening out inefficient land use that needs to be optimized according to the spatial aggregation values, and outputting results. This method can not only improve the accuracy of the digital management method for towns based on knowledge graphs, but also has good interpretability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of geology and environmental science, and in particular to a digital urban management method based on a knowledge graph. Background Art

[0002] With the acceleration of urbanization and the rapid development of information technology, the concept of digital urban management has been proposed and quickly attracted global attention. Traditional urban management lacks the support of advanced information technology, which affects management efficiency and decision-making quality, and the use of land and energy resources is not efficient enough, resulting in resource waste and environmental problems. In contrast, digital urban management is based on knowledge graphs and introduces information technology, which can improve the efficiency and quality of urban management and promote the sustainable development of cities. Summary of the invention

[0003] The purpose of the present invention is to provide a method for digital management of towns based on knowledge graph.

[0004] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0005] The present invention comprises the following steps:

[0006] Acquire multivariate data of a town, wherein the multivariate data includes remote sensing image data, town data and expert opinions; the remote sensing image data includes sample remote sensing image data and regional remote sensing image data; the town data includes town planning data and town statistical data; pre-process the remote sensing image data; and acquire standard remote sensing image data according to the sample remote sensing image data;

[0007] A land use division model is constructed based on a convolutional neural network, the model is trained and tested according to the standard remote sensing image data, the regional remote sensing image data is input into the land use division model, and the land use division model divides the regional remote sensing image data into different types of land use patches according to commercial land, industrial land, public land, residential land and administrative land;

[0008] Obtaining the ideal value of the indicator according to the town planning data and the town statistical data, obtaining the indicator weight according to the expert opinion, obtaining the actual value of the indicator for each of the land use patches according to the town statistical data, and obtaining the standard value of the indicator according to the actual value of the indicator and the ideal value of the indicator; screening out the inefficient land use patches among the land use patches according to the standard value of the indicator and the weight of the indicator;

[0009] The spatial distribution of low-efficiency land is obtained according to the low-efficiency land map patches, the spatial aggregation value is calculated according to the spatial distribution, the low-efficiency land that needs to be optimized is screened out according to the spatial aggregation value, and the result is output.

[0010] Furthermore, the preprocessing method in step A includes radiation calibration, atmospheric correction, geometric correction, image enhancement and image registration.

[0011] Furthermore, the method for obtaining standard remote sensing image data according to the example remote sensing image data and the expert's first opinion includes:

[0012] The labeled Landsat remote sensing images obtained from Earth Explorer and various types of unlabeled remote sensing images obtained from Mechanical Turk are used as example remote sensing image data. The labeled Landsat remote sensing images are used as a labeled set, the labels corresponding to the labeled set are used as a labeled set, and the various types of unlabeled remote sensing images are used as an unlabeled set. The labels include commercial land, industrial land, public land, residential land, and administrative land. The one-to-one method of SVM is used to perform multi-classification on the example remote sensing image data, and the Gaussian radial basis function is used as the kernel function:

[0013]

[0014] Where K(x i ,x) is the kernel function, x i is the i-th labeled data, x is the unlabeled data, σ is the kernel coefficient, LXB(x i -x) is the mean of the difference between the labeled data and the unlabeled data, n is the total number of labeled data, and the classification function of the SVM model obtained according to the kernel function is:

[0015]

[0016] Where f(x) is the classification function, K(x i ,x) is the kernel function, x i is the i-th labeled data, x is the unlabeled data, y i is the i-th label data, b is the bias,

[0017]

[0018] α i is the Lagrange multiplier of the i-th training data, and the optimal solution α is solved by the SMO algorithm i , and α i satisfy:

[0019]

[0020] Wherein C is a penalty coefficient, and the penalty coefficient and the kernel coefficient are solved according to a particle swarm optimization algorithm; the marked set, the label set and the unmarked set are input into an SVM multi-classifier to obtain standard remote sensing image data, wherein the standard remote sensing image data includes commercial land standard remote sensing image data, industrial land standard remote sensing image data, public land standard remote sensing image data, residential land standard remote sensing image data and administrative land standard remote sensing image data;

[0021] Furthermore, a method for constructing a land use division model based on a convolutional neural network and training and testing the model according to the standard remote sensing image data includes:

[0022] A convolutional neural network is used to construct a land use division model. AlexNet is used as the convolutional neural network structure. The standard remote sensing image data is divided into a training data set and a test data set in a ratio of 7:3. The training data set is input into the convolutional neural network model to extract features:

[0023]

[0024] in is the jth feature map of the lth layer, is the i-th feature map of the l-1th layer, l is the number of convolutional layers, for and The convolution kernels connected between for The bias, M j is the input feature map, f is the activation function ReLU, Represents convolution; scales and maps the extracted features:

[0025]

[0026] in is the jth feature map of the lth layer, is the jth feature map of the l-1th layer, for The weight of for , f is the activation function ReLU, and down means downsampling; the extracted features are converted into vector form and the category labels corresponding to the features are calculated:

[0027] y i =f(W i x i +b i )

[0028] where y i For x i The corresponding category label, x iis the i-th feature map of the input, W i For x i The weight of b i For x i , f is the activation function ReLU; calculate the error between the category label corresponding to the feature and the actual category label:

[0029]

[0030] Where L is the error, n is the total amount of input feature maps, C is the number of categories, r is the scaling factor, the initialization value is r = 0.5, s is the hyperparameter cosine spacing, s = 0.2, θ j is the angle between the feature of category j and the i-th feature map, For x i The angle between the corresponding category label and the i-th feature map; the error is differentiated to calculate the gradient of the convolutional neural network model parameters, and the weights and bias of the convolutional neural network model are updated through the back propagation algorithm; the training data are repeatedly calculated to obtain an optimized convolutional neural network model; the test data set is input into the convolutional neural network model, and the accuracy, precision and recall rate of the test data set after classification are calculated to evaluate the performance of the convolutional neural network model and obtain the land use division model.

[0031] Furthermore, the method of obtaining the ideal value of the indicator according to the town planning data and the town statistical data, and obtaining the weight of the indicator according to the expert opinion includes:

[0032] The ideal values ​​of the indicators include consistency of planning use, building density, population density, comprehensive floor area ratio, completeness of infrastructure and ideal values ​​of per capita tax revenue; among them, the ideal values ​​of building density and comprehensive floor area ratio are determined according to the values ​​specified in the urban planning data, and the ideal values ​​of population density and per capita tax revenue are determined according to the urban statistical data using the average method; the indicator weights are determined based on the expert opinions using the Delphi method.

[0033] Furthermore, the method for obtaining the standard value of the indicator according to the actual value of the indicator and the ideal value of the indicator includes:

[0034] The indicators are divided into first-category indicators and second-category indicators. The standard values ​​of the first-category indicators are calculated as follows:

[0035]

[0036] where F i is the standard value of the i-th indicator, b i is the actual value of the i-th indicator, u i is the ideal value of the i-th indicator, F i0is the initial value of the i-th indicator standard, [e,f] is the boundary range value; calculate the indicator standard value of the second type of indicator:

[0037]

[0038] where F i is the standard value of the i-th indicator, b i is the actual value of the i-th indicator, u i is the ideal value of the i-th indicator, and [e,f] is the boundary range value.

[0039] Furthermore, the method of screening out the low-efficiency map spots in the map spots according to the indicator standard value and the indicator weight includes:

[0040] Calculate the land use value for each of the map patches:

[0041]

[0042] Among them U j is the land use value of the jth land use area, j = 1, 2, ..., J, J is the total number of land use areas, A m is the total number of the mth type of map patches, m = 1, 2, 3, 4, 5, is the weight of the i-th indicator of the j-th and m-th type of the map patch, F i is the standard value of the ith indicator; the land use threshold is set to 90, and the land use patches whose land use values ​​are less than the land use threshold are divided into low-efficiency land use patches to obtain low-efficiency land use patches.

[0043] Furthermore, the method for obtaining the spatial distribution of low-efficiency land according to the low-efficiency land use map patches includes:

[0044] The low-utility map patches are converted into low-utility location distributions, and according to the low-utility location distributions:

[0045] a. Obtain the standard deviation ellipse of the low-utility location of each land use type:

[0046] The center, azimuth and standard deviation are calculated to obtain the standard deviation ellipse of the low-efficiency land use point of each land use type. The calculation formula of the center is:

[0047]

[0048] in is the horizontal coordinate of the center of the circle, is the ordinate of the center of the circle, n is the number of low-utility locations, ω i is the weight of the ith low-utility location, which is assigned according to the local density of low-utility locations, x i is the horizontal coordinate of the i-th low-utility location, yi is the ordinate of the i-th low-utility location; the azimuth angle is calculated as:

[0049]

[0050]

[0051] Where θ is the azimuth angle, is the horizontal coordinate of the i-th low-utility location after translation transformation,

[0052] is the ordinate of the i-th low-utility location after translation transformation; the calculation formula of the standard deviation is:

[0053]

[0054] where σ x is the standard deviation along the x-axis, σ y is the standard deviation along the y-axis;

[0055] b. Obtain the kernel density distribution of low-utility locations for each land use type:

[0056] The kernel density distribution estimation formula is:

[0057]

[0058] Where p(μ) is the estimated value of the kernel density distribution, j is the number of low-utility locations within the window width, h is the window width, μ-μ i For low-utility location μ to the selected low-efficiency land sample point μ i Distance, set different width window values ​​h n , n = 1, 2, ..., N, calculate the kernel density estimation error under different width window values:

[0059]

[0060] in Indicates that when the width window is h n The kernel density estimation error when , W(μ) is the kernel function, the quadratic kernel is selected as the kernel function, p″(μ) is the quadratic derivative of the kernel density estimation value, and the wide window value that minimizes the error is selected as the wide window of the kernel density estimation; the spatial distribution of inefficient land is obtained according to the standard deviation ellipse and the kernel density distribution.

[0061] Furthermore, a method for calculating a spatial aggregation value according to the spatial distribution and screening out inefficient land that needs to be optimized according to the spatial aggregation value includes:

[0062] Calculate the spatial aggregation value:

[0063]

[0064] Where A(μ) is the spatial aggregation value of the low-utility location μ, γ is the angle between the low-utility location μ and the main axis of the standard deviation ellipse, θ is the azimuth of the standard deviation ellipse, p(μ) is the estimated value of the kernel density distribution of the low-utility location μ, d(μ,C) is the distance from the low-utility location μ to the center of the standard deviation ellipse, h is the wide window of the kernel density estimation, and j is the number of low-utility locations within the window width; the spatial aggregation values ​​are arranged in ascending order, and the low-utility locations with the last 10 spatial aggregation values ​​are selected as the low-utility lands that need to be optimized.

[0065] The beneficial effects of the present invention are:

[0066] The present invention is a digital town management method based on knowledge graph. Compared with the prior art, the present invention has the following technical effects:

[0067] The present invention can improve the accuracy of digital management of towns and cities through preprocessing, acquiring standard remote sensing image data, building a land division model, and calculating the spatial distribution and spatial aggregation value of inefficient land, thereby improving the precision of digital management of towns and cities. Digitizing town management can greatly save resources and manpower costs, improve the efficiency and quality of town management, and promote the sustainable development of cities. It is of great significance to town management, can adapt to the digital management of different towns, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a flow chart of the steps of a method for digital management of towns based on knowledge graphs in the present invention. DETAILED DESCRIPTION

[0069] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0070] The present invention provides a method for digital management of towns based on knowledge graph, comprising the following steps:

[0071] like Figure 1 As shown, in this embodiment, the following steps are included:

[0072] Acquire multivariate data of a town, wherein the multivariate data includes remote sensing image data, town data and expert opinions; the remote sensing image data includes sample remote sensing image data and regional remote sensing image data; the town data includes town planning data and town statistical data; pre-process the remote sensing image data; and acquire standard remote sensing image data according to the sample remote sensing image data;

[0073] A land use division model is constructed based on a convolutional neural network, the model is trained and tested according to the standard remote sensing image data, the regional remote sensing image data is input into the land use division model, and the land use division model divides the regional remote sensing image data into different types of land use patches according to commercial land, industrial land, public land, residential land and administrative land;

[0074] Obtaining the ideal value of the indicator according to the town planning data and the town statistical data, obtaining the indicator weight according to the expert opinion, obtaining the actual value of the indicator for each of the land use patches according to the town statistical data, and obtaining the standard value of the indicator according to the actual value of the indicator and the ideal value of the indicator; screening out the inefficient land use patches among the land use patches according to the standard value of the indicator and the weight of the indicator;

[0075] The spatial distribution of low-efficiency land is obtained according to the low-efficiency land map patches, the spatial aggregation value is calculated according to the spatial distribution, the low-efficiency land that needs to be optimized is screened out according to the spatial aggregation value, and the result is output.

[0076] In actual management, 80 labeled Landsat remote sensing images were obtained from Earth Explorer as the label set. The size of each remote sensing image is 1024×1024 pixels, and the spatial resolution is 0.3 meters. There are 20 images with commercial land labels, 15 images with industrial land labels, 20 images with public land labels, 15 images with residential land labels, and 10 images with administrative land labels. 120 unlabeled remote sensing images of various types obtained from Mechanical Turk were used as the label set. The label set was used to train the SVM classifier, and the labeled set and the unlabeled set were classified according to the one-to-one method of SVM. Finally, the standard remote sensing image data obtained are: 45 commercial land images, 40 industrial land images, 45 public land images, 40 residential land images, and 30 administrative land images.

[0077] In actual management, a total of 5987 high-resolution images covering a town in March 2024 with a total geographical area of ​​189.69 square kilometers are obtained as regional remote sensing image data;

[0078] In actual management, the statistical data of a town are as follows: a. Overall statistical data of the town include: total population of 270,000, urban population of 152,200, rural population of 117,800, permanent population of 270,400, area of ​​1,200 square kilometers, urbanization area of ​​480 square kilometers, and total commercial land area of ​​4.7km 2 The total industrial land area is 4.95km 2 The total area of ​​public land is 6.54km 2The total residential area is 40.12km 2 The total administrative land area is 2.87km 2 , the urbanization rate is 40%, the gross domestic product is 17.616 billion yuan, the output value of the primary industry is 503 million yuan, the output value of the secondary industry is 10.930 billion yuan, the output value of the tertiary industry is 6.183 billion yuan, the total industrial output value of the county has reached 9.76 billion yuan, the fixed asset investment has reached 18.62 billion yuan, the total fiscal revenue has reached 2.648 billion yuan, the total commercial tax revenue is 141 million yuan, the total industrial tax revenue is 0.99 billion yuan, the total public tax revenue is 0.262 billion yuan, the total residential tax revenue is 400 million yuan, the total administrative tax revenue is 0.23 billion yuan, the per capita disposable income of rural permanent residents is 14,273 yuan, and the per capita disposable income of urban residents is 31,624 yuan; b. Statistical data of urban districts include: statistical data on the consistency of planning use, building density, population density, comprehensive plot ratio, completeness of infrastructure and per capita tax revenue in various regions;

[0079] In actual management, the ideal values ​​of building density and comprehensive volume ratio are as follows: the planning requirements for commercial land in the town's "Overall Land Use Plan", "Detailed Control Plan for the Main Urban Area" and "Technical Regulations for Urban and Rural Planning Management" are that the building density is between 35 and 50, and the comprehensive volume ratio is between 2.5 and 3.5; the planning requirements for industrial land are that the building density is between 30 and 40, and the comprehensive volume ratio is between 0.5 and 1; the planning requirements for public land are that the building density is between 20 and 35, and the comprehensive volume ratio is between 0.8 and 1.2 ; The planning requirements for residential land are that the building density is 35 for ground floor, 28 for multi-story, 25 for mid-high-rise, and 20 for high-rise, and the comprehensive plot ratio is 1.2 for multi-story, 2 for mid-high-rise, and 3.5 for high-rise; the planning requirements for administrative land are that the building density is between 30 and 35, and the comprehensive plot ratio is between 2.0 and 2.5; the ideal value of per-unit tax revenue is obtained by dividing the total tax revenue of various types of land in urban statistical data by the total area of ​​various types of land; the ideal value of population density is obtained by dividing the total urban population in urban statistical data by the total area of ​​residential land;

[0080] In actual management, the scoring opinions of 15 experts in three rounds were obtained, and the Delphi method was used to obtain the mean, variance and coefficient of variation of the indicator weights of the three rounds of scoring. Since the variance and coefficient of variation of the third round of scoring were the smallest among the three rounds, the expert scoring weights of the third round were selected as the indicator weights, among which the consistency of planning use was a, the building density was b, the comprehensive volume ratio was c, the completeness of the basic setting was d, the average land tax was e, and the population density was f:

[0081]

[0082]

[0083] In actual management, the ideal values ​​and weights of the indicators are:

[0084]

[0085]

[0086]

[0087] In actual management, the land use division model divides the regional remote sensing image of the town with a total coverage area of ​​189.69 square kilometers into 40 commercial map patches with an area of ​​2.91km 2 ; 32 industrial plots, covering an area of ​​1.15km 2 ; 45 public land plots, covering an area of ​​3.46km 2 ; 60 residential plots, covering an area of ​​28.22km 2 ; 11 administrative map patches, covering an area of ​​1.95km 2 ;

[0088] In actual management, the actual values ​​of the indicators of a certain commercial land area are: planning use consistency of 80, building density of 40, comprehensive plot ratio of 2.6, infrastructure completeness of 80, and average land tax of 32,000 yuan / ha; the actual values ​​of the indicators of a certain industrial land area are: planning use consistency of 90, building density of 32, comprehensive plot ratio of 0.6, infrastructure completeness of 92, and average land tax of 18,000 yuan / ha; the actual values ​​of the indicators of a certain public land area are: planning use consistency of 95, building density of 26, comprehensive plot ratio of 2.6, infrastructure completeness of 88, and average land tax of 5,000 yuan / ha; the actual values ​​of the indicators of a certain residential land area are: planning use consistency of 82, building density of medium and high-rise: 19, comprehensive plot ratio of medium and high-rise: 1, infrastructure completeness of 90, average land tax of 2,000 yuan / ha, and population density of 230 people / m 2 The actual values ​​of the indicators for a certain residential plot are: planning consistency of 93, building density of 36, comprehensive plot ratio of 1.9, infrastructure completeness of 90, and average land tax of 10,000 yuan / ha;

[0089] In actual management, the standard values ​​of the indicators for a certain commercial land area are: 0.8 for planned use consistency, 1 for building density, 1 for comprehensive plot ratio, 0.8 for completeness of infrastructure, and 1 for average tax revenue per land area; the standard values ​​of the indicators for a certain industrial land area are: 0.9 for planned use consistency, 1 for building density, 1 for comprehensive plot ratio, 0.92 for completeness of infrastructure, and 0.9 for average tax revenue per land area; the standard values ​​of the indicators for a certain public land area are: 0.95 for planned use consistency, 1 for building density, and 1 for comprehensive plot ratio. The plot ratio is 1, the infrastructure completeness is 0.88, and the average tax revenue per land is 1; the actual values ​​of the indicators of a certain residential land area are: planning use consistency is 0.82, building density is 1.24, comprehensive plot ratio is 0.83, infrastructure completeness is 0.9, average tax revenue per land is 0.2, and population density is 0.84; the standard values ​​of the indicators of a certain administrative land area are: planning use consistency is 0.93, building density is 0.97, comprehensive plot ratio is 0.95, infrastructure completeness is 0.9, and average tax revenue per land is 1;

[0090] In actual management, the land use values ​​of 10 of the map plots are: 92, 94.74, 96.55, 94.23, 92.54, 91.78, 97.88, 92.83, 90.6, 96.79;

[0091] In actual management, the inefficient land use patches screened out are: 7 inefficient commercial land use patches, 8 inefficient industrial land use patches, 10 inefficient public land use patches, 18 inefficient residential land use patches, and 3 inefficient administrative land use patches, with a total area of ​​16.5km 2 ;

[0092] In actual management, the spatial distribution of inefficient land is obtained using ArcGIS10 platform to obtain the standard deviation ellipse of inefficient land and the kernel density distribution of inefficient land;

[0093] In actual management, the spatial aggregation values ​​of inefficient commercial land are 0.052, 0.036, 0.021, 0.033, 0.043, 0.034, and 0.0012 respectively; the spatial aggregation values ​​of inefficient industrial land are 0.25, 0.08, 0.055, 0.064, 0.052, 0.076, 0.053, and 0.064 respectively; the spatial aggregation values ​​of inefficient public land are 0.3, 0.05, 0.05, 0.061, 0.056, 0.073, and 0.0 43, 0.032, 0.044, 0.082; the spatial aggregation values ​​of inefficient residential land are 0.25, 0.07, 0.053, 0.087, 0.079, 0.084, 0.088, 0.054, 0.094, 0.067, 0.0753, 0.035, 0.21, 0.022, 0.047, 0.057, 0.076, 0.066; the spatial aggregation values ​​of inefficient administrative land are 0.0012, 0.01, 0.015;

[0094] In actual management, the inefficient land selected are: 2 inefficient industrial land, 2 inefficient public land, 6 inefficient residential land, with a total area of ​​6.2km 2 .

[0095] In this embodiment, the preprocessing method in step A includes radiation calibration, atmospheric correction, geometric correction, image enhancement and image registration.

[0096] In this embodiment, the method for obtaining standard remote sensing image data according to the example remote sensing image data and the expert's first opinion includes:

[0097] The labeled Landsat remote sensing images obtained from Earth Explorer and various types of unlabeled remote sensing images obtained from Mechanical Turk are used as example remote sensing image data. The labeled Landsat remote sensing images are used as a labeled set, the labels corresponding to the labeled set are used as a labeled set, and the various types of unlabeled remote sensing images are used as an unlabeled set. The labels include commercial land, industrial land, public land, residential land, and administrative land. The one-to-one method of SVM is used to perform multi-classification on the example remote sensing image data, and the Gaussian radial basis function is used as the kernel function:

[0098]

[0099] Where K(x i ,x) is the kernel function, x i is the i-th labeled data, x is the unlabeled data, σ is the kernel coefficient, LXB(x i-x) is the mean of the difference between the labeled data and the unlabeled data, n is the total number of labeled data, and the classification function of the SVM model obtained according to the kernel function is:

[0100]

[0101] Where f(x) is the classification function, K(x i ,x) is the kernel function, x i is the i-th labeled data, x is the unlabeled data, y i is the i-th label data, b is the bias,

[0102]

[0103] α i is the Lagrange multiplier of the i-th training data, and the optimal solution α is solved by the SMO algorithm i , and α i satisfy:

[0104]

[0105] Wherein C is a penalty coefficient, and the penalty coefficient and the kernel coefficient are solved according to a particle swarm optimization algorithm; the marked set, the label set and the unmarked set are input into an SVM multi-classifier to obtain standard remote sensing image data, and the standard remote sensing image data includes commercial land standard remote sensing image data, industrial land standard remote sensing image data, public land standard remote sensing image data, residential land standard remote sensing image data and administrative land standard remote sensing image data;

[0106] In this embodiment, a land use division model is constructed based on a convolutional neural network, and a method for training and testing the model according to the standard remote sensing image data includes:

[0107] A land use division model is constructed using a convolutional neural network. AlexNet is used as the convolutional neural network structure. The standard remote sensing image data is divided into a training data set and a test data set in a ratio of 7:3. The training data set is input into the convolutional neural network model to extract features:

[0108]

[0109] in is the jth feature map of the lth layer, is the i-th feature map of the l-1th layer, l is the number of convolutional layers, for and The convolution kernels connected between for The bias, M j is the input feature map, f is the activation function ReLU, Represents convolution; scales and maps the extracted features:

[0110]

[0111] in is the jth feature map of the lth layer, is the jth feature map of the l-1th layer, for The weight of for , f is the activation function ReLU, and down means downsampling; the extracted features are converted into vector form and the category labels corresponding to the features are calculated:

[0112] y i =f(W i x i +b i )

[0113] where y i For x i The corresponding category label, x i is the i-th feature map of the input, W i For x i The weight of b i For x i , f is the activation function ReLU; calculate the error between the category label corresponding to the feature and the actual category label:

[0114]

[0115] Where L is the error, n is the total amount of input feature maps, C is the number of categories, r is the scaling factor, the initialization value is r = 0.5, s is the hyperparameter cosine spacing, s = 0.2, θ j is the angle between the feature of category j and the i-th feature map, For x i The angle between the corresponding category label and the i-th feature map; the error is differentiated to calculate the gradient of the convolutional neural network model parameters, and the weights and bias of the convolutional neural network model are updated through the back propagation algorithm; the training data are repeatedly calculated to obtain an optimized convolutional neural network model; the test data set is input into the convolutional neural network model, and the accuracy, precision and recall rate of the test data set after classification are calculated to evaluate the performance of the convolutional neural network model and obtain the land use division model.

[0116] In this embodiment, the method of obtaining the ideal value of the indicator according to the town planning data and the town statistical data, and obtaining the weight of the indicator according to the expert opinion includes:

[0117] The ideal values ​​of the indicators include consistency of planning use, building density, population density, comprehensive floor area ratio, completeness of infrastructure and ideal values ​​of per capita tax revenue; among them, the ideal values ​​of building density and comprehensive floor area ratio are determined according to the values ​​specified in the urban planning data, and the ideal values ​​of population density and per capita tax revenue are determined according to the urban statistical data using the average method; the indicator weights are determined based on the expert opinions using the Delphi method.

[0118] In this embodiment, the method for obtaining the standard value of the indicator according to the actual value of the indicator and the ideal value of the indicator includes:

[0119] The indicators are divided into first-category indicators and second-category indicators. The standard values ​​of the first-category indicators are calculated as follows:

[0120]

[0121] where F i is the standard value of the i-th indicator, b i is the actual value of the i-th indicator, u i is the ideal value of the i-th indicator, F i0 is the initial value of the i-th indicator standard, [e,f] is the boundary range value; calculate the indicator standard value of the second type of indicator:

[0122]

[0123] where F i is the standard value of the i-th indicator, b i is the actual value of the i-th indicator, u i is the ideal value of the i-th indicator, and [e,f] is the boundary range value.

[0124] In this embodiment, the method of screening out the low-efficiency map blocks in the map blocks according to the indicator standard value and the indicator weight includes:

[0125] Calculate the land use value for each of the map patches:

[0126]

[0127] Among them U j is the land use value of the jth land use area, j = 1, 2, ..., J, J is the total number of land use areas, A m is the total number of the mth type of map patches, m = 1, 2, 3, 4, 5, is the weight of the i-th indicator of the j-th and m-th type of the map patch, F i is the standard value of the ith indicator; the land use threshold is set to 90, and the land use patches whose land use values ​​are less than the land use threshold are divided into low-efficiency land use patches to obtain low-efficiency land use patches.

[0128] In this embodiment, the method for obtaining the spatial distribution of low-efficiency land according to the low-efficiency land use map patches includes:

[0129] The low-utility map patches are converted into low-utility location distributions, and according to the low-utility location distributions:

[0130] a. Obtain the standard deviation ellipse of the low-utility location of each land use type:

[0131] The center, azimuth and standard deviation are calculated to obtain the standard deviation ellipse of the low-efficiency land use point of each land use type. The calculation formula of the center is:

[0132]

[0133] in is the horizontal coordinate of the center of the circle, is the ordinate of the center of the circle, n is the number of low-utility locations, ω i is the weight of the ith low-utility site, which is assigned according to the local density of low-utility sites, x i is the horizontal coordinate of the i-th low-utility location, y i is the ordinate of the i-th low-utility location; the azimuth angle is calculated as:

[0134]

[0135]

[0136] Where θ is the azimuth angle, is the horizontal coordinate of the i-th low-utility location after translation transformation,

[0137] is the ordinate of the i-th low-utility location after translation transformation; the calculation formula of the standard deviation is:

[0138]

[0139] where σ x is the standard deviation along the x-axis, σ y is the standard deviation along the y-axis;

[0140] b. Obtain the kernel density distribution of low-utility locations for each land use type:

[0141] The kernel density distribution estimation formula is:

[0142]

[0143] Where p(μ) is the estimated value of the kernel density distribution, j is the number of low-utility locations within the window width, h is the window width, μ-μ iFor low-utility location μ to the selected low-efficiency land sample point μ i distance, set different width window values ​​h n , n = 1, 2, ..., N, calculate the kernel density estimation error under different width window values:

[0144]

[0145] in Indicates that when the window width is h n The kernel density estimation error when , W(μ) is the kernel function, the quadratic kernel is selected as the kernel function, p″(μ) is the quadratic derivative of the kernel density estimation value, and the wide window value that minimizes the error is selected as the wide window of the kernel density estimation; the spatial distribution of inefficient land is obtained according to the standard deviation ellipse and the kernel density distribution.

[0146] In this embodiment, the method of calculating the spatial aggregation value according to the spatial distribution and screening out the inefficient land that needs to be optimized according to the spatial aggregation value includes:

[0147] Calculate the spatial aggregation value:

[0148]

[0149] Where A(μ) is the spatial aggregation value of the low-utility location μ, γ is the angle between the low-utility location μ and the main axis of the standard deviation ellipse, θ is the azimuth of the standard deviation ellipse, p(μ) is the estimated value of the kernel density distribution of the low-utility location μ, d(μ,C) is the distance from the low-utility location μ to the center of the standard deviation ellipse, h is the wide window of the kernel density estimation, and j is the number of low-utility locations within the window width; the spatial aggregation values ​​are arranged in ascending order, and the low-utility locations with the last 10 spatial aggregation values ​​are selected as the low-utility lands that need to be optimized.

[0150] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A digital town management method based on knowledge graph, It is characterized in that The following steps are involved: Acquire multivariate data of a town, wherein the multivariate data includes remote sensing image data, town data and expert opinions; the remote sensing image data includes sample remote sensing image data and regional remote sensing image data; the town data includes town planning data and town statistical data; pre-process the remote sensing image data; and acquire standard remote sensing image data according to the sample remote sensing image data; A land use division model is constructed based on a convolutional neural network, the model is trained and tested according to the standard remote sensing image data, the regional remote sensing image data is input into the land use division model, and the land use division model divides the regional remote sensing image data into different types of land use patches according to commercial land, industrial land, public land, residential land and administrative land; Obtaining the ideal value of the indicator according to the town planning data and the town statistical data, obtaining the indicator weight according to the expert opinion, obtaining the actual value of the indicator for each of the land use patches according to the town statistical data, and obtaining the standard value of the indicator according to the actual value of the indicator and the ideal value of the indicator; screening out the inefficient land use patches among the land use patches according to the standard value of the indicator and the weight of the indicator; Acquire the spatial distribution of low-efficiency land according to the low-efficiency land use map patches, calculate the spatial aggregation value according to the spatial distribution, screen out the low-efficiency land that needs to be optimized according to the spatial aggregation value, and output the result; The method for obtaining standard remote sensing image data according to the example remote sensing image data comprises: The labeled Landsat remote sensing images obtained from Earth Explorer and various types of unlabeled remote sensing images obtained from Mechanical Turk are used as example remote sensing image data. The labeled Landsat remote sensing images are used as a labeled set, the labels corresponding to the labeled set are used as a labeled set, and the various types of unlabeled remote sensing images are used as an unlabeled set. The labels include commercial land, industrial land, public land, residential land, and administrative land. The one-to-one method of SVM is used to perform multi-classification on the example remote sensing image data, and the Gaussian radial basis function is used as the kernel function: Where K(x i ,x) is the kernel function, x i is the i-th labeled data, x is the unlabeled data, σ is the kernel coefficient, LXB(x i -x) is the mean of the difference between the labeled data and the unlabeled data, n is the total number of labeled data, and the classification function of the SVM model obtained according to the kernel function is: Where f(x) is the classification function, K(x i ,x) is the kernel function, x i is the i-th labeled data, x is the unlabeled data, y i is the i-th label data, b is the bias, α i is the Lagrange multiplier of the i-th training data, and the optimal solution α is solved by the SMO algorithm i , and α i satisfy: Wherein C is a penalty coefficient, and the penalty coefficient and the kernel coefficient are solved according to a particle swarm optimization algorithm; the marked set, the label set and the unmarked set are input into an SVM multi-classifier to obtain standard remote sensing image data, wherein the standard remote sensing image data includes commercial land standard remote sensing image data, industrial land standard remote sensing image data, public land standard remote sensing image data, residential land standard remote sensing image data and administrative land standard remote sensing image data; The method for obtaining the standard value of an indicator according to the actual value of the indicator and the ideal value of the indicator includes: The indicators are divided into first-category indicators and second-category indicators. The standard values ​​of the first-category indicators are calculated as follows: Among them, F i is the standard value of the i-th indicator, b i is the actual value of the i-th indicator, u i is the ideal value of the i-th indicator, F i0 is the initial value of the i-th indicator standard, [e,f] is the boundary range value; calculate the indicator standard value of the second type of indicator: Among them, F i is the standard value of the i-th indicator, b i is the actual value of the i-th indicator, u i is the ideal value of the i-th indicator, and [e,f] is the boundary range value.

2. According to the method for digital management of towns based on knowledge graph according to claim 1, It is characterized in that The preprocessing method described in step A includes radiation calibration, atmospheric correction, geometric correction, image enhancement and image registration.

3. According to the method for digital management of towns based on knowledge graph as claimed in claim 1, It is characterized in that The method of constructing a land use division model based on a convolutional neural network and training and testing the model according to the standard remote sensing image data includes: A convolutional neural network is used to construct a land use division model. AlexNet is used as the convolutional neural network structure. The standard remote sensing image data is divided into a training data set and a test data set in a ratio of 7:

3. The training data set is input into the convolutional neural network model to extract features: in is the jth feature map of the lth layer, is the i-th feature map of the l-1th layer, l is the number of convolutional layers, for and The convolution kernels connected between for The bias, M j is the input feature map, f is the activation function ReLU, Represents convolution; scales and maps the extracted features: in is the jth feature map of the lth layer, is the jth feature map of the l-1th layer, for The weight of for , f is the activation function ReLU, and down means downsampling; the extracted features are converted into vector form and the category labels corresponding to the features are calculated: y i =f(W i x i +b i ) where y i For x i The corresponding category label, x i is the i-th feature map of the input, W i For x i The weight of b i For x i , f is the activation function ReLU; calculate the error between the category label corresponding to the feature and the actual category label: Where L is the error, n is the total amount of input feature maps, C is the number of categories, r is the scaling factor, the initialization value is r = 0.5, s is the hyperparameter cosine spacing, s = 0.2, θ j is the angle between the feature of category j and the i-th feature map, For x i The angle between the corresponding category label and the i-th feature map; the error is differentiated to calculate the gradient of the convolutional neural network model parameters, and the weights and bias of the convolutional neural network model are updated through the back propagation algorithm; the training data are repeatedly calculated to obtain an optimized convolutional neural network model; the test data set is input into the convolutional neural network model, and the accuracy, precision and recall rate of the test data set after classification are calculated to evaluate the performance of the convolutional neural network model and obtain the land use division model.

4. According to the method for digital management of towns based on knowledge graph according to claim 1, It is characterized in that The method of obtaining the ideal value of the indicator according to the town planning data and the town statistical data, and obtaining the weight of the indicator according to the expert opinion, comprises: The ideal values ​​of the indicators include consistency of planning use, building density, population density, comprehensive floor area ratio, completeness of infrastructure and ideal values ​​of per capita tax revenue; among them, the ideal values ​​of building density and comprehensive floor area ratio are determined according to the values ​​specified in the urban planning data, and the ideal values ​​of population density and per capita tax revenue are determined according to the urban statistical data using the average method; the indicator weights are determined based on the expert opinions using the Delphi method.

5. According to the method for digital management of towns based on knowledge graph according to claim 1, It is characterized in that The method for screening out the low-efficiency map spots in the map spots according to the indicator standard value and the indicator weight includes: Calculate the land use value for each of the map patches: Among them U j is the land use value of the jth land use area, j = 1, 2, ..., J, J is the total number of land use areas, A m is the total number of the mth type of map patches, m = 1, 2, 3, 4, 5, is the weight of the i-th indicator of the j-th and m-th type of the map patch, F i is the standard value of the ith indicator; the land use threshold is set to 200, and the land use patches whose land use values ​​are less than the land use threshold are divided into low-efficiency land use patches to obtain low-efficiency land use patches.

6. According to the method for digital management of towns based on knowledge graph according to claim 1, It is characterized in that The method for obtaining the spatial distribution of low-efficiency land according to the low-efficiency land use map patches includes: The low-utility map patches are converted into low-utility location distributions, and according to the low-utility location distributions: a. Obtain the standard deviation ellipse of the low-utility location of each land use type: The center, azimuth and standard deviation are calculated to obtain the standard deviation ellipse of the low-efficiency land use point of each land use type. The calculation formula of the center is: in is the horizontal coordinate of the center of the circle, is the ordinate of the center of the circle, n is the number of low-utility locations, ω i is the weight of the ith low-utility location, which is assigned according to the local density of low-utility locations, x i is the horizontal coordinate of the i-th low-utility location, y i is the ordinate of the i-th low-utility location; the azimuth angle is calculated as: Where θ is the azimuth angle, is the horizontal coordinate of the i-th low-utility location after translation transformation, is the ordinate of the i-th low-utility location after translation transformation; the calculation formula of the standard deviation is: where σ x is the standard deviation along the x-axis, σ y is the standard deviation along the y-axis; b. Obtain the kernel density distribution of low-utility locations for each land use type: The kernel density distribution estimation formula is: Where p(μ) is the estimated value of the kernel density distribution, j is the number of low-utility locations within the window width, h is the window width, μ-μ i For low-utility location μ to the selected low-efficiency land sample point μ i distance, set different width window values ​​h n , n = 1, 2, ..., N, calculate the kernel density estimation error under different width window values: in Indicates that when the width window is h n The kernel density estimation error when , W(μ) is the kernel function, the quadratic kernel is selected as the kernel function, p" (μ) is the quadratic derivative of the kernel density estimation value, and the wide window value that minimizes the error is selected as the wide window of the kernel density estimation; the spatial distribution of inefficient land is obtained according to the standard deviation ellipse and the kernel density distribution.

7. According to the method for digital management of towns based on knowledge graph as claimed in claim 1, It is characterized in that The method of calculating the spatial aggregation value according to the spatial distribution and screening out the low-efficiency land that needs to be optimized according to the spatial aggregation value includes: Calculate the spatial aggregation value: Where A(μ) is the spatial aggregation value of the low-utility location μ, γ is the angle between the low-utility location μ and the main axis of the standard deviation ellipse, θ is the azimuth of the standard deviation ellipse, p(μ) is the estimated value of the kernel density distribution of the low-utility location μ, d(μ,C) is the distance from the low-utility location μ to the center of the standard deviation ellipse, h is the wide window of the kernel density estimation, and j is the number of low-utility locations within the window width; the spatial aggregation values ​​are arranged in ascending order, and the low-utility locations with the last 10 spatial aggregation values ​​are selected as the low-utility lands that need to be optimized.