Land use change prediction method, device, medium and equipment

Abstract

This application discloses a method, apparatus, medium, and equipment for predicting land use change, relating to the field of image processing technology. The method includes: dividing initial land state images of the land to be predicted at multiple time points into regions to obtain segmented land state images at multiple time points; generating a dynamic land change map based on the segmented land state images at multiple time points; obtaining a transition probability matrix corresponding to the land to be predicted; inputting the dynamic land change map and the transition probability matrix into a pre-trained feature extraction model to obtain a land state feature prediction matrix; inputting the land state feature prediction matrix into a pre-trained land state prediction model to obtain the land prediction result of the land state prediction image. The dynamic land change map uses a graph structure to represent various elements and dynamic relationships in the land use system, better considering the historical dependence and future trends of land use change, and improving the accuracy of land use change prediction.

Description

Technical Field

[0001] This invention relates to the fields of image processing technology and digital healthcare, and particularly to a method, apparatus, medium, and device for predicting land use change. Background Technology

[0002] Land use forecasting technology predicts the use of a piece of land. Land use forecasting can make advance arrangements for land development, remediation and protection in time and space, which has important practical significance for the allocation and sustainable use of regional land resources.

[0003] A common method for land use forecasting is to build a forecasting model. Historical land use information is input into the model, which then predicts future land use patterns. For example, in a certain area of ​​a city, there are many good schools but only one hospital. Next to the hospital is a park. As the number of students increases, the population of this area increases year by year. The forecasting model predicts that at some point in the future, the hospital will need to be expanded, and the part of the park adjacent to the hospital will be converted into a hospital.

[0004] However, existing land use prediction models are static models that ignore the dynamics and nonlinearity of the land use system and cannot effectively reflect the historical dependence and future trends of land use change. Therefore, the accuracy of existing land use prediction methods is low. Summary of the Invention

[0005] In view of this, the present invention provides a method, apparatus, medium and equipment for predicting land use change, the main purpose of which is to solve the problem of low accuracy in current land use change prediction.

[0006] According to one aspect of this application, a land use change prediction method is provided, the method comprising:

[0007] The initial land state images of the land to be predicted at multiple time points are divided into regions to obtain land state segmentation images at multiple time points. Each land state segmentation image includes multiple sub-regions.

[0008] For any land state segmentation image, the land state segmentation image at the next time point corresponding to the land state segmentation image is used as a comparison image. Based on the comparison image, the number of pixels in each sub-region of the land state segmentation image that have undergone utilization state transition is identified. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, the utilization state transition probability corresponding to each sub-region is calculated.

[0009] Based on the utilization state transition probability of each sub-region in each land state segmentation image, a transition probability matrix corresponding to the land to be predicted is constructed. Based on land state segmentation images at multiple time points, a land change dynamic map is generated. The land change dynamic map and the transition probability matrix are input into a pre-trained feature extraction model to obtain a land state feature prediction matrix.

[0010] The land state feature prediction matrix is ​​input into a pre-trained land state prediction model to obtain the land prediction result of the land state prediction image.

[0011] Optionally, the step of dividing the initial land state images of the land to be predicted at multiple time points into regions to obtain segmented land state images at multiple time points includes:

[0012] Superpixel segmentation is performed on the initial land state images corresponding to multiple time points of the land to be predicted. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmentation images corresponding to multiple time points.

[0013] Optionally, generating a dynamic land change map based on land state segmentation images at multiple time points includes:

[0014] Each sub-region in the land state segmentation image is treated as a node. Based on the connection relationship between the sub-regions, edges are constructed between the nodes to generate a static land state map.

[0015] Determine the node category attribute and node feature vector corresponding to each node in each land state static map;

[0016] The node category attributes and node feature vectors are labeled in each node of the static land state map to obtain a land state data map;

[0017] Arrange the land status data maps in chronological order to obtain a dynamic map of land changes.

[0018] Optionally, before dividing the initial land state images of the land to be predicted at multiple time points into regions, the method further includes:

[0019] Acquire remote sensing images, land use data, and driving factor data of the land to be predicted at multiple time points;

[0020] Land use data, driving factor data, and remote sensing images from multiple time points are fused to generate an initial land status image for the specified time points.

[0021] Optionally, the pre-trained feature extraction model can be obtained using the following methods:

[0022] Superpixel segmentation is performed on the initial land state image samples of land samples at multiple time points. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmented image samples corresponding to multiple time points. Based on the land state segmented image samples, land state static map samples are obtained.

[0023] Determine the node category attribute and node feature vector corresponding to each node in each land state static map sample, and mark them in each node of the land state static map to obtain a land state data map sample. Arrange the land state data map samples in chronological order to obtain a land change dynamic map sample.

[0024] For any land state segmentation image sample, the land state segmentation image sample at the next time point corresponding to the land state segmentation image sample is used as a comparison image. Based on the comparison image, the number of pixels that have undergone utilization state transition in each sub-region of the land state segmentation image sample is identified. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, the utilization state transition probability corresponding to each sub-region is calculated.

[0025] Based on the utilization state transition probability of each sub-region in each land state segmentation image sample, a transition probability matrix corresponding to the land sample is constructed.

[0026] Calculate the feature matrix corresponding to each static land state map sample, and train the initial feature extraction model based on the feature matrix corresponding to each static land state map sample, the transition probability matrix corresponding to the land sample, and the dynamic land change map sample to obtain the pre-trained feature extraction model.

[0027] Optionally, a pre-trained land state prediction model can be obtained using the following methods:

[0028] Based on the land change dynamic map sample of the land sample before the target time point, obtain the land state feature prediction matrix sample corresponding to the land sample at the target time point;

[0029] The initial land state prediction model is trained based on the land state feature prediction matrix sample corresponding to the land sample at the target time point and the real land use data to obtain the pre-trained land state prediction model.

[0030] Optionally, the driving factor data includes at least one of the following: elevation, slope, precipitation, temperature, GDP, population density, distance from road, and distance from railway.

[0031] According to another aspect of this application, a land use change prediction device is provided, comprising:

[0032] The region segmentation module is used to segment the initial land state images of the land to be predicted at multiple time points to obtain land state segmentation images at multiple time points. Each land state segmentation image includes multiple sub-regions.

[0033] The utilization transition probability calculation module is used to take any land state segmentation image as a comparison image of the next time point corresponding to the land state segmentation image, identify the number of pixels that have undergone utilization state transition in each sub-region of the land state segmentation image based on the comparison image, and calculate the utilization transition probability corresponding to each sub-region based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition.

[0034] The land state feature prediction matrix acquisition module is used to construct the transition probability matrix corresponding to the land to be predicted based on the utilization state transition probability of each sub-region in each land state segmentation image, generate a land change dynamic map based on land state segmentation images at multiple time points, and input the land change dynamic map and the transition probability matrix into a pre-trained feature extraction model to obtain the land state feature prediction matrix.

[0035] The land state prediction module is used to input the land state feature prediction matrix into a pre-trained land state prediction model to obtain the land prediction result of the land state prediction image.

[0036] Optionally, the step of dividing the initial land state images of the land to be predicted at multiple time points into regions to obtain segmented land state images at multiple time points includes:

[0037] Superpixel segmentation is performed on the initial land state images corresponding to multiple time points of the land to be predicted. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmentation images corresponding to multiple time points.

[0038] Optionally, generating a dynamic land change map based on land state segmentation images at multiple time points includes:

[0039] Each sub-region in the land state segmentation image is treated as a node. Based on the connection relationship between the sub-regions, edges are constructed between the nodes to generate a static land state map.

[0040] Determine the node category attribute and node feature vector corresponding to each node in each land state static map;

[0041] The node category attributes and node feature vectors are labeled in each node of the static land state map to obtain a land state data map;

[0042] Arrange the land status data maps in chronological order to obtain a dynamic map of land changes.

[0043] Optionally, the land use change prediction device further includes:

[0044] The data acquisition module is used to acquire remote sensing images, land use data, and driving factor data of the land to be predicted at multiple time points;

[0045] The fusion module is used to fuse land use data, driving factor data and remote sensing images from multiple time points to generate an initial land status image for the time points.

[0046] Optionally, the pre-trained feature extraction model can be obtained using the following methods:

[0047] Superpixel segmentation is performed on the initial land state image samples of land samples at multiple time points. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmented image samples corresponding to multiple time points. Based on the land state segmented image samples, land state static map samples are obtained.

[0048] Determine the node category attribute and node feature vector corresponding to each node in each land state static map sample, and mark them in each node of the land state static map to obtain a land state data map sample. Arrange the land state data map samples in chronological order to obtain a land change dynamic map sample.

[0049] For any land state segmentation image sample, the land state segmentation image sample at the next time point corresponding to the land state segmentation image sample is used as a comparison image. Based on the comparison image, the number of pixels that have undergone utilization state transition in each sub-region of the land state segmentation image sample is identified. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, the utilization state transition probability corresponding to each sub-region is calculated.

[0050] Based on the utilization state transition probability of each sub-region in each land state segmentation image sample, a transition probability matrix corresponding to the land sample is constructed.

[0051] Calculate the feature matrix corresponding to each static land state map sample, and train the initial feature extraction model based on the feature matrix corresponding to each static land state map sample, the transition probability matrix corresponding to the land sample, and the dynamic land change map sample to obtain the pre-trained feature extraction model.

[0052] Optionally, a pre-trained land state prediction model can be obtained using the following methods:

[0053] Based on the land change dynamic map sample of the land sample before the target time point, obtain the land state feature prediction matrix sample corresponding to the land sample at the target time point;

[0054] The initial land state prediction model is trained based on the land state feature prediction matrix sample corresponding to the land sample at the target time point and the real land use data to obtain the pre-trained land state prediction model.

[0055] Optionally, the driving factor data includes at least one of the following: elevation, slope, precipitation, temperature, GDP, population density, distance from road, and distance from railway.

[0056] According to another aspect of this application, a storage medium is provided that stores at least one executable instruction that causes a processor to perform operations corresponding to the above-described land use change prediction method.

[0057] According to another aspect of this application, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0058] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described land use change prediction method.

[0059] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages:

[0060] This application provides a land use change prediction method, apparatus, equipment, and medium. The method involves dividing initial land state images of the land to be predicted at multiple time points into regions, obtaining segmented land state images at multiple time points, generating a dynamic land change map based on these images, and using the dynamic land change map and the transition probability matrix of the land to be predicted as input data for a pre-trained feature extraction model to obtain a land state feature prediction matrix. This prediction matrix is ​​then used as input data for a pre-trained land state prediction model to obtain the land prediction results from the land state prediction images. The dynamic land change map utilizes a graph structure to represent various elements and dynamic relationships within the land use system, better reflecting spatial heterogeneity and topological structure, better considering the historical dependence and future trends of land use change, and improving the accuracy of land use change prediction.

[0061] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0062] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0063] Figure 1 A flowchart of a land use change prediction method provided in an embodiment of this application is shown;

[0064] Figure 2 Another flowchart of a land use change prediction method provided in an embodiment of this application is shown;

[0065] Figure 3 This paper shows another flowchart of a land use change prediction method provided in an embodiment of this application;

[0066] Figure 4 This paper illustrates another flowchart of a land use change prediction method provided in an embodiment of this application;

[0067] Figure 5 Another flowchart of a land use change prediction method provided in an embodiment of this application is shown;

[0068] Figure 6 This paper shows a block diagram of a land use change prediction device provided in an embodiment of this application;

[0069] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention is shown.

[0070] in,

[0071] Figure 6 In Chinese: 602 - Regional division module; 604 - Calculation module using state transition probability; 606 - Land state feature prediction matrix acquisition module; 608 - Land state prediction module;

[0072] Figure 7 In Chinese: 702 - Processor; 704 - Communication interface; 706 - Memory; 708 - Communication bus; 710 - Program. Detailed Implementation

[0073] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present invention can be combined with each other.

[0074] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments, structures, features, and effects according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "an embodiment" or "an embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0075] To address the issue of low accuracy in current land use change prediction methods, this application provides a land use change prediction method, such as... Figure 1 As shown, the method includes:

[0076] 102: Divide the initial land state images of the land to be predicted at multiple time points into regions to obtain land state segmentation images at multiple time points. Each land state segmentation image includes multiple sub-regions.

[0077] In this embodiment, initial land state images of the land to be predicted are obtained at multiple time points. Each initial land state image at each time point is then divided into regions to obtain a segmented land state image for that time point. For example, when dividing the initial land state images into regions, the area where the hospital is located is divided into a sub-region.

[0078] In another embodiment of the invention, for further definition and explanation, the initial land state images of the land to be predicted at multiple time points are divided into regions to obtain segmented land state images corresponding to multiple time points, including:

[0079] Superpixel segmentation is performed on the initial land state images corresponding to multiple time points of the land to be predicted. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmentation images corresponding to multiple time points.

[0080] In this embodiment, image superpixel segmentation can construct an image scene composed of small target units. In the field of computer vision, image superpixel segmentation refers to the process of subdividing a digital image into multiple image sub-regions (sets of pixels) (also called superpixels). Superpixel image segmentation groups and aggregates image pixels based on the similarity of single or multiple features of the image, dividing the image matrix into several non-overlapping regions with consistent similarity from a two-dimensional plane perspective. The result of image superpixel segmentation is a set of sub-regions on the image (these sub-regions collectively cover the entire image). Each pixel in a sub-region is similar under a certain characteristic measure or calculated characteristic, such as color, brightness, and texture. Adjacent regions are significantly different under a certain characteristic measure. That is, each small region consists of a series of adjacent pixels with similar features such as color, brightness, and texture. These small regions mostly retain effective information for further image segmentation and generally do not destroy the boundary information of objects in the image. The overall principle is that the pixel features within the same region should be consistent and similar, while the features of pixels between different regions should remain heterogeneous or inconsistent. Based on the similarity within regions and the heterogeneity between regions in the definition of image segmentation, this application adopts a region growing strategy to achieve superpixel image segmentation, and gradually merges similar pixels to achieve multi-level superpixel construction.

[0081] In another embodiment of the invention, for further definition and explanation, such as Figure 2 As shown, before dividing the initial land state images of the land to be predicted at multiple time points into regions, the method also includes:

[0082] 202: Acquire remote sensing images, land use data, and driving factor data of the land to be predicted at multiple time points;

[0083] 204: Fuse land use data, driving factor data and remote sensing images from multiple time points to generate an initial land status image for each time point.

[0084] In this embodiment, multiple periods of land use data and remote sensing images of the land to be predicted are acquired, along with some driving factor data such as elevation, slope, precipitation, temperature, GDP, population density, distance from roads, and distance from railways. These data are fused together to obtain an initial land status image. The initial land status image is converted into TIFF format using RS / GIS software or tools, ensuring that the projection coordinate system and row and column numbers are consistent. The converted initial land status images are stored in a folder in chronological order.

[0085] In one embodiment, the driving factor data includes at least one of: elevation, slope, precipitation, temperature, GDP, population density, distance from road, and distance from railway.

[0086] 104: For any land state segmentation image, take the land state segmentation image of the next time point corresponding to the land state segmentation image as the comparison image, identify the number of pixels in each sub-region of the land state segmentation image that have undergone utilization state transition based on the comparison image, and calculate the utilization state transition probability corresponding to each sub-region based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition.

[0087] In this embodiment, the land state segmentation image at one time point is compared with the land state segmentation image at the next time point. Based on the comparison image, the number of pixels in each sub-region of the land state segmentation image that have undergone utilization state transition is identified. For each sub-region, the utilization state transition probability corresponding to that sub-region is obtained by dividing the number of pixels that have undergone utilization state transition by the total number of pixels.

[0088] 106: Based on the state transition probability of each sub-region in each land state segmentation image, construct the transition probability matrix corresponding to the land to be predicted. Based on the land state segmentation images at multiple time points, generate a dynamic map of land change. Input the dynamic map of land change and the transition probability matrix into a pre-trained feature extraction model to obtain the land state feature prediction matrix.

[0089] In this embodiment, based on the land change dynamic map and the transition probability matrix, a pre-trained feature extraction model is used to predict the land state features at the next time point, thereby obtaining the land state feature prediction matrix.

[0090] In another embodiment of the invention, for further definition and explanation, such as Figure 3 As shown, a dynamic land change map is generated based on land state segmentation images at multiple time points, including:

[0091] 302: Treat each sub-region in the land state segmentation image as a node, construct edges between nodes based on the connection relationships between sub-regions, and generate a static land state map;

[0092] 304: Determine the node category attribute and node feature vector corresponding to each node in each land state static map;

[0093] 306: Label the node category attributes and node feature vectors in each node of the static land state map to obtain the land state data map;

[0094] 308: Arrange the land status data maps in chronological order to obtain a dynamic map of land changes.

[0095] In this embodiment, the land state segmentation image includes multiple sub-regions, also called blocks. Each block is a superpixel unit. Each superpixel unit is regarded as a node of a graph. Adjacent units are connected by an edge. In this way, an image is transformed into graph data with superpixels as units, also called a static land state image.

[0096] Each period corresponds to a static graph. Arranging these static graphs chronologically creates a continuous dynamic graph from multiple static graphs. The dynamic graph and its representation learning are defined as follows: A dynamic graph is represented as a sequence of multiple static graphs, G = {G1, G2, ..., GT}, where Gt = (Vt, Et) represents the static graph at time t, t ∈ {1, 2, ..., T}. For the adjacency matrix At ∈ RN×N of graph Gt, if there is an edge connecting (vt)i and (vt)j, then (At)i,j = 1; otherwise, (At)i,j = 0. The node representation learning on the dynamic graph is the learning of a sequence F = {f1, f2, ..., fT} of a set of mappings. Each mapping transforms a node at time t into a low-dimensional vector (yt)v = ft(v), ensuring that the mapped vector retains the original information of the node. In other words, the more similar two points are in the original graph, the closer their mapped vectors will be.

[0097] Assume a dynamic graph consists of nodes and edges, where the edges appear in a specific order. Specifically, the appearance time of each edge is recorded as part of its information. Consider a dynamic graph with 8 nodes and 8 dynamic edges. Each directed edge in the dynamic graph can be represented as (v_s, v_g, t), where v_s represents the source node, v_g represents the target node, and t represents the time the edge occurs. For example, an edge (v_2, v_1, t_0) appears between nodes v_2 and v_1 at time t_0. When an edge appears between two nodes, they interact, and their node representations change. For convenience, these two nodes are called "interacting nodes." Furthermore, the newly appearing edge not only affects its two interacting nodes but also other nodes "near" these interacting nodes; these are called "affected nodes." The node representations of these "affected nodes" also change. Therefore, it is necessary to update the information of the edges that appear on this line to its two "interacting nodes" and propagate this information to the "affected nodes".

[0098] The Graph Neural Network (GNN) model is based on a message propagation mechanism. Each node updates its state by exchanging messages with others until a stable value is reached. The output of the GNN is calculated at each node based on its current state. Essentially, the main learning process of a graph neural network model involves iteratively aggregating and updating the neighbor information of nodes in the graph data. In one iteration, each node updates its information by aggregating the features of its neighbors and its own features from the previous layer, typically performing a non-linear transformation on the aggregated information. By stacking multiple layers of the network, each node can obtain information about its neighbors within a certain number of hops.

[0099] In graph network model learning, if we understand it through node message passing, it involves two processes: the message passing phase and the readout phase. The message passing phase is the forward propagation phase, which runs T steps in a loop and is completed through the function M. t Information is obtained through function U t Update the nodes; the equation for this stage is as follows.

[0100]

[0101]

[0102] Among them, e vw This represents the feature vector of the edge from node v to w.

[0103] The readout phase calculates a feature vector to represent the features of the entire graph. This is equivalent to combining the features of 8 nodes (superpixels) to read out or obtain a feature about the entire graph scene. The dimension is the same as the dimension of the nodes. The function R is used to read out or obtain the feature of the entire graph:

[0104]

[0105] Where T represents the total number of time steps, and M is a function... t U t R and R can use different model settings.

[0106] Representation of a dynamic graph: Let G t =(V t E t ,X t ,Y t ) represents a snapshot at time point t, where V t It is a set of nodes, E t It is a set of edges, X t ∈R n×dIt is the node feature matrix, also called the land state feature prediction matrix, Y t ∈R n ×c It is a node category matrix. Let... Let represent the transition probability matrix from time point t to time point t+1. Then the dynamic graph can be represented as follows: Where T is the total number of time points.

[0107] This application provides a land use change prediction method. Compared with the prior art, it constructs a multi-level representation of an image scene based on superpixel segmentation, and further constructs a static land state map of the scene. The nodes of the static land state map are represented by superpixel regions of the scene. By aggregating information from the time and spatial dimensions, more complete node features can be learned, thereby improving the accuracy of land use prediction.

[0108] In another embodiment of the invention, for further definition and explanation, such as Figure 4 As shown, the pre-trained feature extraction model is obtained using the following method, including:

[0109] 402: Perform superpixel segmentation on the initial land state image samples of land samples at multiple time points, divide adjacent pixels that meet the similarity conditions into the same sub-region, and obtain land state segmented image samples corresponding to multiple time points. Based on the land state segmented image samples, obtain land state static map samples.

[0110] 404: Determine the node category attribute and node feature vector corresponding to each node in each land status static map sample, and mark them in each node of the land status static map to obtain a land status data map sample. Arrange the land status data map samples in chronological order to obtain a land change dynamic map sample.

[0111] 406: For any land state segmentation image sample, take the land state segmentation image sample at the next time point corresponding to the land state segmentation image sample as the comparison image. Based on the comparison image, identify the number of pixels in each sub-region of the land state segmentation image sample that have undergone utilization state transition. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, calculate the utilization state transition probability corresponding to each sub-region.

[0112] 408: Based on the utilization state transition probability of each sub-region in each land state segmentation image sample, construct the transition probability matrix corresponding to the land sample;

[0113] 410: Calculate the feature matrix corresponding to each static land state map sample. Based on the feature matrix corresponding to each static land state map sample, the transition probability matrix corresponding to the land sample, and the dynamic land change map sample, train the initial feature extraction model to obtain the pre-trained feature extraction model.

[0114] In this embodiment, the transition probability matrix corresponding to the land sample and the dynamic land change map sample are used as training samples for the model, and the feature matrix corresponding to each static land state map sample is used as the data result of the training sample to train the initial feature extraction model and obtain the pre-trained feature extraction model.

[0115] As a preferred embodiment of this example, the feature extraction model is the FDGNN model.

[0116] The FDGNN model consists of three parts: attribute self-updating, association process, and message passing. Attribute self-updating refers to updating the attributes of nodes and edges based on external factors. The association process refers to updating the connection relationships between nodes and edges based on the generation, removal, and modification of edges. Message passing refers to updating the embedding vectors of nodes and edges based on information from neighboring nodes. Specifically, the FDGNN model can be represented as:

[0117] X t+1 =f(X) t ,P t,t+1 ,θ1)(AttributeSelf-Updating)

[0118] E t+1 =g(E t ,P t,t+1 ,θ2)(AssociationProcess)

[0119] H t+1 =h(H t ,X t+1 E t+1 ,θ3)(MessagePassing)

[0120] Among them, X t and X t+1 These are the node feature matrices at time point t and time point (t+1), respectively, E t and E t+1 These are the edge feature matrices at time point t and time point (t+1), respectively. t,t+1 It is the transition probability matrix, H t and H t+1θ1, θ2, and θ3 are the node embedding matrices at time point t and time point (t+1), respectively. θ1, θ2, and θ3 are the model parameters, and f, g, and h are the model functions. The specific function form can be designed and selected according to different land use types.

[0121] During model training, the weighted sum of the cross-union model loss (Dice loss) and cross-entropy loss (BCE loss) is used as the loss function. The Adam optimizer is employed with an initial learning rate of 0.001. After 200 training rounds, the model with the lowest loss is selected as the optimal model.

[0122] 108: Input the land state feature prediction matrix into the pre-trained land state prediction model to obtain the land prediction results of the land state prediction image.

[0123] In this embodiment, based on the land state feature prediction matrix, a pre-trained land state prediction model is used to predict the land state feature prediction matrix to obtain the prediction result of the land use status or attribute at the next time point.

[0124] For example, a city has a well-known hospital. As the city's population increases, a predictive model can be used to predict that at some point in the future, this hospital will establish a branch in another region. For example, an abandoned factory located at a predetermined distance from the current hospital could be converted into a branch.

[0125] In another embodiment of the invention, for further definition and explanation, such as Figure 5 As shown, the pre-trained land state prediction model is obtained using the following method, including:

[0126] 502: Based on the dynamic land change map sample of the land sample before the target time point, obtain the land state feature prediction matrix sample corresponding to the land sample at the target time point;

[0127] 504: The initial land state prediction model is trained based on the land state feature prediction matrix sample corresponding to the target time point and the real land use data to obtain the pre-trained land state prediction model.

[0128] Specifically, the land state feature prediction matrix sample corresponding to the land sample at the target time point is used as the training sample of the model, and the real land use data is used as the data result of the training sample to train the initial land state prediction model and obtain the pre-trained land state prediction model.

[0129] In a preferred embodiment of this invention, the land state prediction model is an LTM model. An LTM model is an artificial neural network model that can predict land use change using GIS data and various driving factors. Specifically, the LTM model can be represented as:

[0130]

[0131] in, Let Z be the predicted node category matrix at time point t+1, and Ht+1 be the node embedding matrix at time point t+1, where Z ∈ R. n×k This is the driving factor matrix, θ4 is the parameters of the prediction model, and p is the function of the prediction model. The specific form of the function can be designed and selected according to different tasks and data.

[0132] This application provides a land use change prediction method. Compared with existing technologies, it divides the initial land state images of the land to be predicted at multiple time points into regions to obtain segmented land state images at multiple time points. Based on the segmented land state images at multiple time points, a dynamic land change map is generated. The dynamic land change map and the transition probability matrix of the land to be predicted are used as input data for a pre-trained feature extraction model to obtain a land state feature prediction matrix. The land state feature prediction matrix is ​​then used as input data for a pre-trained land state prediction model to obtain the land prediction results of the land state prediction image. The dynamic land change map uses a graph structure to represent various elements and dynamic relationships in the land use system, which can better reflect spatial heterogeneity and topological structure, better consider the historical dependence and future trends of land use change, and improve the accuracy of land use change prediction.

[0133] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this invention provides a land use change prediction device, such as... Figure 6 As shown, the device includes:

[0134] The region division module 602 is used to divide the initial land state images of the land to be predicted at multiple time points into regions, so as to obtain land state segmentation images at multiple time points. Each land state segmentation image includes multiple sub-regions.

[0135] The utilization transition probability calculation module 604 is used to take any land state segmentation image as a comparison image of the land state segmentation image at the next time point corresponding to the land state segmentation image, identify the number of pixels that have undergone utilization state transition in each sub-region of the land state segmentation image based on the comparison image, and calculate the utilization transition probability corresponding to each sub-region based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition.

[0136] The land state feature prediction matrix acquisition module 606 is used to construct the transition probability matrix corresponding to the land to be predicted based on the utilization state transition probability of each sub-region in each land state segmentation image, generate a land change dynamic map based on land state segmentation images at multiple time points, and input the land change dynamic map and the transition probability matrix into a pre-trained feature extraction model to obtain the land state feature prediction matrix.

[0137] The land state prediction module 608 is used to input the land state feature prediction matrix into the pre-trained land state prediction model to obtain the land prediction results of the land state prediction image.

[0138] This application provides a land use change prediction device. Compared with the prior art, it divides the initial land state images of the land to be predicted at multiple time points into regions to obtain segmented land state images at multiple time points. Based on the segmented land state images at multiple time points, a dynamic land change map is generated. The dynamic land change map and the transition probability matrix of the land to be predicted are used as input data for a pre-trained feature extraction model to obtain a land state feature prediction matrix. The land state feature prediction matrix is ​​used as input data for a pre-trained land state prediction model to obtain the land prediction result of the land state prediction image. The dynamic land change map uses a graph structure to represent various elements and dynamic relationships in the land use system, which can better reflect spatial heterogeneity and topological structure, better consider the historical dependence and future trend of land use change, and improve the accuracy of land use change prediction.

[0139] In one embodiment, the initial land state images of the land to be predicted at multiple time points are divided into regions to obtain segmented land state images at multiple time points, including:

[0140] Superpixel segmentation is performed on the initial land state images corresponding to multiple time points of the land to be predicted. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmentation images corresponding to multiple time points.

[0141] In one embodiment, a dynamic land change map is generated based on land state segmentation images at multiple time points, including:

[0142] Each sub-region in the land state segmentation image is treated as a node. Based on the connection relationship between the sub-regions, edges are constructed between the nodes to generate a static land state map.

[0143] Determine the node category attribute and node feature vector corresponding to each node in each land state static map;

[0144] The node category attributes and node feature vectors are labeled in each node of the static land state map to obtain the land state data map;

[0145] Arrange the land status data maps in chronological order to obtain a dynamic map of land changes.

[0146] In one embodiment, the land use change prediction device further includes:

[0147] The data acquisition module is used to acquire remote sensing images, land use data, and driving factor data of the land to be predicted at multiple time points;

[0148] The fusion module is used to fuse land use data, driving factor data, and remote sensing images from multiple time points to generate an initial land status image for that time point.

[0149] In one embodiment, a pre-trained feature extraction model is obtained using the following method:

[0150] Superpixel segmentation is performed on the initial land state image samples of land samples at multiple time points. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmented image samples corresponding to multiple time points. Based on the land state segmented image samples, land state static map samples are obtained.

[0151] Determine the node category attribute and node feature vector corresponding to each node in each land status static map sample, and mark them in each node of the land status static map to obtain land status data map samples. Arrange the land status data map samples in chronological order to obtain land change dynamic map samples.

[0152] For any land state segmentation image sample, the land state segmentation image sample at the next time point corresponding to the land state segmentation image sample is used as a comparison image. Based on the comparison image, the number of pixels in each sub-region of the land state segmentation image sample that have undergone utilization state transition is identified. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, the utilization state transition probability corresponding to each sub-region is calculated.

[0153] Based on the utilization state transition probability of each sub-region in each land state segmentation image sample, a transition probability matrix corresponding to the land sample is constructed.

[0154] Calculate the feature matrix corresponding to each static land state map sample. Based on the feature matrix corresponding to each static land state map sample, the transition probability matrix corresponding to the land sample, and the dynamic land change map sample, train the initial feature extraction model to obtain the pre-trained feature extraction model.

[0155] In one embodiment, a pre-trained land state prediction model is obtained using the following method:

[0156] Based on the dynamic map of land change of land samples before the target time point, obtain the land state feature prediction matrix sample corresponding to the target time point.

[0157] The initial land state prediction model is trained based on the land state feature prediction matrix sample corresponding to the land sample at the target time point and the real land use data to obtain the pre-trained land state prediction model.

[0158] In one embodiment, the driving factor data includes at least one of: elevation, slope, precipitation, temperature, GDP, population density, distance from road, and distance from railway.

[0159] Figure 7 The diagram illustrates a structural schematic of a computer device according to an embodiment of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0160] like Figure 7 As shown, the computer device may include: a processor 702, a communications interface 704, a memory 706, and a communications bus 708.

[0161] The processor 702, communication interface 704, and memory 706 communicate with each other via communication bus 708.

[0162] The communication interface 704 is used to communicate with other network elements such as clients or other servers.

[0163] The processor 702 is used to execute program 710, specifically to perform the relevant steps in the above-described land use change prediction method embodiment.

[0164] Specifically, program 710 may include program code that includes computer operation instructions.

[0165] Processor 702 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0166] Memory 706 is used to store program 710. Memory 706 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0167] Specifically, program 710 can be used to cause processor 702 to perform the following operations:

[0168] The initial land state images of the land to be predicted at multiple time points are divided into regions to obtain land state segmentation images at multiple time points. Each land state segmentation image includes multiple sub-regions.

[0169] For any land state segmentation image, the land state segmentation image at the next time point corresponding to the land state segmentation image is used as a comparison image. Based on the comparison image, the number of pixels in each sub-region of the land state segmentation image that have undergone utilization state transition is identified. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, the utilization state transition probability corresponding to each sub-region is calculated.

[0170] Based on the state transition probability of each sub-region in each land state segmentation image, a transition probability matrix corresponding to the land to be predicted is constructed. Based on land state segmentation images at multiple time points, a dynamic map of land change is generated. The dynamic map of land change and the transition probability matrix are input into a pre-trained feature extraction model to obtain a land state feature prediction matrix.

[0171] The land state feature prediction matrix is ​​input into a pre-trained land state prediction model to obtain the land prediction results of the land state prediction image.

[0172] It will be apparent to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. In one embodiment, they can be implemented using device-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0173] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A method for predicting land use change, characterized in that, include: The initial land state images of the land to be predicted at multiple time points are divided into regions to obtain land state segmentation images at multiple time points. Each land state segmentation image includes multiple sub-regions. For any land state segmentation image, the land state segmentation image at the next time point corresponding to the land state segmentation image is used as a comparison image. Based on the comparison image, the number of pixels that have undergone utilization state transition in each sub-region of the land state segmentation image is identified. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, the utilization state transition probability corresponding to each sub-region is calculated. The utilization state transition probability is the ratio of the number of pixels that have undergone utilization state transition to the total number of pixels. Based on the utilization state transition probability of each sub-region in each land state segmentation image, a transition probability matrix corresponding to the land to be predicted is constructed. Based on land state segmentation images at multiple time points, a land change dynamic map is generated. The land change dynamic map and the transition probability matrix are input into a pre-trained feature extraction model to obtain a land state feature prediction matrix. The land state feature prediction matrix is ​​input into a pre-trained land state prediction model to obtain the land prediction result of the land state prediction image.

2. The land use change prediction method as described in claim 1, characterized in that, The step of dividing the initial land state images of the land to be predicted at multiple time points into regions to obtain segmented land state images at multiple time points includes: Superpixel segmentation is performed on the initial land state images corresponding to multiple time points of the land to be predicted. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmentation images corresponding to multiple time points.

3. The land use change prediction method as described in claim 2, characterized in that, The land state segmentation image based on multiple time points generates a dynamic land change map, including: Each sub-region in the land state segmentation image is treated as a node. Based on the connection relationship between the sub-regions, edges are constructed between the nodes to generate a static land state map. Determine the node category attribute and node feature vector corresponding to each node in each land state static map; The node category attributes and node feature vectors are labeled in each node of the static land state map to obtain a land state data map; Arrange the land status data maps in chronological order to obtain a dynamic map of land changes.

4. The land use change prediction method as described in claim 1, characterized in that, Before dividing the initial land state images of the land to be predicted at multiple time points into regions, the method further includes: Acquire remote sensing images, land use data, and driving factor data of the land to be predicted at multiple time points; Land use data, driving factor data, and remote sensing images from multiple time points are fused to generate an initial land status image for the specified time points.

5. The land use change prediction method as described in claim 1, characterized in that, The pre-trained feature extraction model is obtained using the following methods: Superpixel segmentation is performed on the initial land state image samples of land samples at multiple time points. Adjacent pixels that meet the similarity condition are divided into the same sub-region to obtain land state segmented image samples corresponding to multiple time points. Based on the land state segmented image samples, land state static map samples are obtained. Determine the node category attribute and node feature vector corresponding to each node in each land state static map sample, and mark them in each node of the land state static map to obtain a land state data map sample. Arrange the land state data map samples in chronological order to obtain a land change dynamic map sample. For any land state segmentation image sample, the land state segmentation image sample at the next time point corresponding to the land state segmentation image sample is used as a comparison image. Based on the comparison image, the number of pixels that have undergone utilization state transition in each sub-region of the land state segmentation image sample is identified. Based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition, the utilization state transition probability corresponding to each sub-region is calculated. Based on the utilization state transition probability of each sub-region in each land state segmentation image sample, a transition probability matrix corresponding to the land sample is constructed. Calculate the feature matrix corresponding to each static land state map sample, and train the initial feature extraction model based on the feature matrix corresponding to each static land state map sample, the transition probability matrix corresponding to the land sample, and the dynamic land change map sample to obtain the pre-trained feature extraction model.

6. The land use change prediction method as described in claim 1, characterized in that, The pre-trained land state prediction model is obtained using the following methods: Based on the land change dynamic map sample of the land sample before the target time point, obtain the land state feature prediction matrix sample corresponding to the land sample at the target time point; The initial land state prediction model is trained based on the land state feature prediction matrix sample corresponding to the land sample at the target time point and the real land use data to obtain the pre-trained land state prediction model.

7. The land use change prediction method as described in claim 4, characterized in that, The driving factor data includes at least one of the following: elevation, slope, precipitation, temperature, GDP, population density, distance from road, and distance from railway.

8. A land use change prediction device, characterized in that, include: The region segmentation module is used to segment the initial land state images of the land to be predicted at multiple time points to obtain land state segmentation images at multiple time points. Each land state segmentation image includes multiple sub-regions. The utilization transition probability calculation module is used to, for any land state segmentation image, take the land state segmentation image of the next time point corresponding to the land state segmentation image as a comparison image, identify the number of pixels that have undergone utilization state transition in each sub-region of the land state segmentation image based on the comparison image, and calculate the utilization transition probability corresponding to each sub-region based on the total number of pixels in each sub-region and the number of pixels that have undergone utilization state transition. The utilization transition probability is the ratio of the number of pixels that have undergone utilization state transition to the total number of pixels. The land state feature prediction matrix acquisition module is used to construct the transition probability matrix corresponding to the land to be predicted based on the utilization state transition probability of each sub-region in each land state segmentation image, generate a land change dynamic map based on land state segmentation images at multiple time points, and input the land change dynamic map and the transition probability matrix into a pre-trained feature extraction model to obtain the land state feature prediction matrix. The land state prediction module is used to input the land state feature prediction matrix into a pre-trained land state prediction model to obtain the land prediction result of the land state prediction image.

9. A storage medium storing at least one executable instruction that causes a processor to perform an operation corresponding to the land use change prediction method as described in any one of claims 1-7.

10. A computer device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation corresponding to the land use change prediction method as described in any one of claims 1-7.

Images