Artificial intelligence-based digital earth data analysis method and device

Abstract

The application discloses an artificial intelligence-based digital earth data analysis method and device, and relates to the technical field of data analysis. The method comprises the following steps: constructing a digital earth data analysis model based on an artificial intelligence deep learning algorithm; collecting multi-source digital earth data, and performing space-time registration and normalization processing on the multi-source digital earth data to obtain a space-time data cube; inputting the space-time data cube into the digital earth data analysis model to perform digital earth data analysis, obtaining a digital earth data analysis result, and performing visualization. The application solves the problems of large amount of calculation, weak space relationship modeling capability and lack of physical constraints in the prior art.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a digital earth data analysis method and apparatus based on artificial intelligence. Background Technology

[0002] With the rapid development of Earth observation technology, digital earth data has experienced explosive growth, encompassing multi-source information such as optical imagery, microwave remote sensing, meteorological data, and ground-based sensor network data. Efficiently and accurately extracting valuable information from this massive amount of multi-source digital earth data is of great significance for environmental monitoring, urban planning, and disaster early warning.

[0003] Traditional digital earth data analysis methods primarily rely on physical models or shallow machine learning algorithms. While physical models offer strong interpretability, they often require a large number of parameter inputs and have limitations in handling complex nonlinear relationships. Shallow machine learning algorithms, while capable of handling certain nonlinear problems, are insufficient in extracting high-dimensional abstract features and capturing spatiotemporal dependencies. In recent years, deep learning technology has achieved great success in the field of computer vision, providing new ideas for Earth big data analysis. However, most existing deep learning methods process data directly based on pixels, resulting in high computational costs, difficulty in considering the spatial topological relationships of ground objects, and a lack of physical constraints that leads to prediction results that do not conform to geophysical laws.

[0004] Therefore, there is an urgent need for an efficient digital earth data analysis method that can integrate multi-source data, take into account spatial topology, and have physical mechanism constraints. Summary of the Invention

[0005] This invention provides a digital earth data analysis method and apparatus based on artificial intelligence. This invention addresses the problems of high computational load, weak spatial relationship modeling capabilities, and lack of physical constraints in existing technologies.

[0006] In a first aspect, embodiments of the present invention provide a digital earth data analysis method based on artificial intelligence, the method comprising: A digital earth data analysis model is constructed based on deep learning algorithms for artificial intelligence. Multi-source digital earth data were collected, and spatiotemporal registration and normalization were performed on the multi-source digital earth data to obtain a spatiotemporal data cube. The spatiotemporal data cube is input into the digital earth data analysis model to perform digital earth data analysis, obtain the digital earth data analysis results, and then visualize them.

[0007] The technical solution provided in this application has at least the following beneficial effects: By employing superpixel segmentation technology, pixel-level analysis is transformed into object-level analysis, significantly reducing computational load while suppressing noise in individual pixels. A spatiotemporal topology-aware module is introduced, which effectively models the spatial adjacency and distance relationships between ground objects using graph convolutional networks, extracting spatially dependent spatiotemporal features. A dual-drive prediction module based on physical data is adopted, which incorporates physical constraints such as surface energy balance through a joint loss function, improving the accuracy and interpretability of the model in predicting physical quantities.

[0008] In one optional implementation, the digital earth data analysis model includes an adaptive multi-scale superpixel fusion network, a spatiotemporal topology sensing module, and a physical data dual-driven prediction module. The adaptive multi-scale superpixel fusion network is used to extract features from the spatiotemporal data cube to obtain an initial feature map containing multi-scale context information. The spatiotemporal topology sensing module is used to model the spatial relationships of the initial feature map based on the topology mask matrix and extract spatiotemporal feature vectors with spatial dependencies. The physical data dual-drive prediction module is used to make predictions based on the spatiotemporal feature vectors to obtain superpixel-level digital earth data analysis results.

[0009] In one optional implementation, the adaptive multi-scale superpixel fusion network includes a superpixel segmentation module, a pyramid feature extraction module, a dynamic superpixel pooling module, and a multi-scale feature fusion module. The superpixel segmentation module is used to perform superpixel segmentation on the spatiotemporal data cube to obtain a superpixel label map. The pyramid feature extraction module is used to extract feature maps of different resolutions from the spatiotemporal data cube; The dynamic superpixel pooling module is used to perform superpixel average pooling on the feature map at each resolution based on the superpixel label map to obtain object-level features at different resolutions. The multi-scale feature fusion module is used to perform multi-scale feature fusion on the feature maps and object-level features of different resolutions to obtain a fused feature map, i.e., an initial feature map.

[0010] In one optional implementation, the spatiotemporal topology sensing module includes a spatial topology graph construction module, a topology mask matrix generation module, and a spatiotemporal graph convolution propagation module. The spatial topology graph construction module is used to convert the superpixel region corresponding to the fused feature map into a spatial topology graph, and based on the spatial topology graph, convert the fused feature map into a node feature matrix. The topology mask matrix generation module is used to generate a corresponding topology mask matrix based on the spatial topology map. The spatiotemporal graph convolution propagation module is used to perform spatiotemporal graph convolution propagation on the jointly input node feature matrix and topological mask matrix to obtain spatiotemporal feature vectors.

[0011] In one optional implementation, the physical data dual-drive prediction module includes a data-driven prediction head, and the physical data dual-drive prediction module is further configured with a joint loss function of data fitting loss and physical constraint loss. The prediction head is used to make predictions based on the spatiotemporal feature vector to obtain superpixel-level digital earth data analysis results; The joint loss function is used to guide the optimization of network parameters in the CNN of the pyramid feature extraction module, the GCN of the spatiotemporal graph convolutional propagation module, and the MLP of the prediction head. During the training of the digital earth data analysis model, the Adam optimizer is used to update all network parameters of CNN, GCN and MLP simultaneously using the backpropagation algorithm. When the joint loss value generated by the joint loss function is less than the joint loss threshold, the optimal network parameters are obtained.

[0012] In one alternative implementation, the formula for the joint loss function is:

[0013] In the formula, This is the joint loss value; This represents the physical constraint loss value. For the first j The predicted value of the sample; For the first j The true value of the sample; Weights for physical constraint loss; j For sample indication; N The total number of samples;

[0014] In the formula, Net radiation; G Soil heat flux; It is the sensible heat physical flux; This represents latent heat physical flux.

[0015] In one optional implementation, multi-source digital earth data is collected, and spatiotemporal registration and normalization are performed on the multi-source digital earth data to obtain a spatiotemporal data cube, including: Collect multi-source digital earth data, which includes image data of the target area and corresponding auxiliary data; Based on auxiliary data, geometric correction, radiometric calibration and normalization are performed on the image data to obtain spatiotemporally registered image data. Define a unified grid space, convert the spatiotemporally registered image data into a binary mask, rasterize the binary mask into this grid space, and interpolate auxiliary data to the grid center point of the grid space to obtain the spatiotemporal data grid. The spatiotemporal data grid is stacked into cubes to obtain a spatiotemporal data cube.

[0016] In one alternative implementation, a spatiotemporal data cube is input into a digital earth data analysis model for digital earth data analysis, resulting in digital earth data analysis results, which are then visualized, including: Input the spatiotemporal data cube into the digital earth data analysis model; An adaptive multi-scale superpixel fusion network based on a digital earth data analysis model is used to extract features from the spatiotemporal data cube to obtain an initial feature map containing multi-scale contextual information. The spatiotemporal topology sensing module of the digital earth data analysis model performs spatial relationship modeling on the initial feature map based on the topology mask matrix, and extracts spatiotemporal feature vectors with spatial dependencies. The physical data dual-drive prediction module of the digital earth data analysis model makes predictions based on the spatiotemporal feature vectors to obtain superpixel-level digital earth data analysis results. Post-processing and visualization of digital earth data analysis results.

[0017] In one alternative implementation, the results of digital earth data analysis are post-processed and visualized, including: The results of digital earth data analysis are mapped back to the pixel grid through inverse SLIC to generate a smooth raster prediction map; Load raster prediction maps into the Digital Earth Engine, overlay them onto the 3D terrain, and use color gradients to visually display the data distribution of the raster prediction maps; By comparing the current raster forecast map with historical raster forecast maps, the difference map of the raster forecast maps is calculated, areas that have changed dramatically are automatically highlighted, and a briefing is generated.

[0018] Secondly, embodiments of the present invention provide an artificial intelligence-based digital earth data analysis device for implementing a digital earth data analysis method. The device includes: Model building unit, used to build digital earth data analysis models based on deep learning algorithms of artificial intelligence; The data acquisition unit is used to acquire multi-source digital earth data and perform spatiotemporal registration and normalization processing on the multi-source digital earth data to obtain a spatiotemporal data cube. The data analysis unit is used to input spatiotemporal data cubes into the digital earth data analysis model for digital earth data analysis, obtain digital earth data analysis results, and visualize them.

[0019] A third aspect of this invention provides an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the steps of a digital earth data analysis method based on artificial intelligence, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the functional units of a digital earth data analysis device based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0023] The present invention will be further described below with reference to the accompanying drawings.

[0024] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0025] like Figure 1As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0026] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0027] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program for an artificial intelligence-based digital earth data analysis device.

[0028] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the electronic program of the artificial intelligence-based digital earth data analysis device stored in the memory 1005 through the processor 1001 and executes the artificial intelligence-based digital earth data analysis method provided in the embodiment of the present invention.

[0029] Reference Figure 2 The present invention provides an artificial intelligence-based digital earth data analysis method, the method comprising: S201: Constructing a digital earth data analysis model based on deep learning algorithms using artificial intelligence; S202: Collect multi-source digital earth data and perform spatiotemporal registration and normalization on the multi-source digital earth data to obtain a spatiotemporal data cube; S203: Input the spatiotemporal data cube into the digital earth data analysis model to perform digital earth data analysis, obtain the digital earth data analysis results, and visualize them.

[0030] The technical solution provided in this application has at least the following beneficial effects: By employing superpixel segmentation technology, pixel-level analysis is transformed into object-level analysis, significantly reducing computational load while suppressing noise in individual pixels. A spatiotemporal topology-aware module is introduced, which effectively models the spatial adjacency and distance relationships between ground objects using graph convolutional networks, extracting spatially dependent spatiotemporal features. A dual-drive prediction module based on physical data is adopted, which incorporates physical constraints such as surface energy balance through a joint loss function, improving the accuracy and interpretability of the model in predicting physical quantities.

[0031] In one optional implementation, the digital earth data analysis model includes an adaptive multi-scale superpixel fusion network, a spatiotemporal topology sensing module, and a physical data dual-driven prediction module. The adaptive multi-scale superpixel fusion network is used to extract features from the spatiotemporal data cube to obtain an initial feature map containing multi-scale context information. The spatiotemporal topology sensing module is used to model the spatial relationships of the initial feature map based on the topology mask matrix and extract spatiotemporal feature vectors with spatial dependencies. The physical data dual-drive prediction module is used to make predictions based on the spatiotemporal feature vectors to obtain superpixel-level digital earth data analysis results.

[0032] In one optional implementation, the adaptive multi-scale superpixel fusion network includes a superpixel segmentation module, a pyramid feature extraction module, a dynamic superpixel pooling module, and a multi-scale feature fusion module. The superpixel segmentation module is used to perform superpixel segmentation on the spatiotemporal data cube to obtain a superpixel label map. The pyramid feature extraction module is used to extract feature maps of different resolutions from the spatiotemporal data cube; The dynamic superpixel pooling module is used to perform superpixel average pooling on the feature map at each resolution based on the superpixel label map to obtain object-level features at different resolutions. The multi-scale feature fusion module is used to perform multi-scale feature fusion on the feature maps and object-level features of different resolutions to obtain a fused feature map, i.e., an initial feature map.

[0033] In one optional implementation, the spatiotemporal topology sensing module includes a spatial topology graph construction module, a topology mask matrix generation module, and a spatiotemporal graph convolution propagation module. The spatial topology graph construction module is used to convert the superpixel region corresponding to the fused feature map into a spatial topology graph, and based on the spatial topology graph, convert the fused feature map into a node feature matrix. The topology mask matrix generation module is used to generate a corresponding topology mask matrix based on the spatial topology map. The spatiotemporal graph convolution propagation module is used to perform spatiotemporal graph convolution propagation on the jointly input node feature matrix and topological mask matrix to obtain spatiotemporal feature vectors.

[0034] In one optional implementation, the physical data dual-drive prediction module includes a data-driven prediction head, and the physical data dual-drive prediction module is further configured with a joint loss function of data fitting loss and physical constraint loss. The prediction head is used to make predictions based on the spatiotemporal feature vector to obtain superpixel-level digital earth data analysis results; The joint loss function is used to guide the optimization of network parameters in the Convolutional Neural Networks (CNN) of the pyramid feature extraction module, the Graph Convolutional Networks (GCN) of the spatiotemporal graph convolutional propagation module, and the Multi-Layer Perceptron (MLP) of the prediction head. During the training of the digital earth data analysis model, the Adam optimizer is used, and the backpropagation algorithm is used to update all network parameters of CNN, GCN, and MLP simultaneously. When the joint loss value generated by the joint loss function is less than the joint loss threshold, the optimal network parameters are obtained.

[0035] In one alternative implementation, the formula for the joint loss function is:

[0036] In the formula, This is the joint loss value; This represents the physical constraint loss value. For the first j The predicted value of the sample; For the first j The true value of the sample; Weights for physical constraint loss; j For sample indication; N The total number of samples;

[0037] In the formula, Net radiation; G Soil heat flux; It is the sensible heat physical flux; This represents latent heat physical flux.

[0038] In one optional implementation, multi-source digital earth data is collected, and spatiotemporal registration and normalization are performed on the multi-source digital earth data to obtain a spatiotemporal data cube, including: S2021: Collect multi-source digital earth data, which includes image data of the target area and corresponding auxiliary data; S2022: Based on auxiliary data, geometric correction, radiometric calibration and normalization are performed on the image data to obtain spatiotemporally registered image data; S2023: Define a unified grid space, convert the spatiotemporally registered image data into a binary mask, rasterize the binary mask into this grid space, and interpolate the auxiliary data to the grid center point of the grid space to obtain the spatiotemporal data grid; S2024: Stack the spatiotemporal data grid into cubes to obtain a spatiotemporal data cube.

[0039] In one alternative implementation, a spatiotemporal data cube is input into a digital earth data analysis model for digital earth data analysis, resulting in digital earth data analysis results, which are then visualized, including: S2031: Input the spatiotemporal data cube into the digital earth data analysis model; S2032: Using an adaptive multi-scale superpixel fusion network of the digital earth data analysis model, feature extraction is performed on the spatiotemporal data cube to obtain an initial feature map containing multi-scale contextual information. S2033: The spatiotemporal topology sensing module using the digital earth data analysis model performs spatial relationship modeling on the initial feature map based on the topology mask matrix, and extracts spatiotemporal feature vectors with spatial dependencies; S2034: The physical data dual-drive prediction module using the digital earth data analysis model makes predictions based on the spatiotemporal feature vector to obtain superpixel-level digital earth data analysis results. The digital earth data analysis results are superpixel-level physical quantities to be predicted, such as surface temperature, soil moisture, or evapotranspiration. S2035: Post-processing and visualization of digital earth data analysis results.

[0040] In one alternative implementation, an adaptive multi-scale superpixel fusion network of a digital earth data analysis model is used to extract features from the spatiotemporal data cube, resulting in an initial feature map containing multi-scale contextual information, including: S20321: Perform Canny edge detection on the spatiotemporal data cube to obtain an edge weight map; S20322: The superpixel segmentation module of the adaptive multi-scale superpixel fusion network of the digital earth data analysis model performs superpixel segmentation on the spatiotemporal data cube based on the Simple Linear Iterative Clustering (SLIC) algorithm to obtain a superpixel label map. In this embodiment, edge weights are introduced into the distance metric formula of SLIC, and the formula is as follows:

[0041] In the formula, The distance value is obtained from the SLIC algorithm; Spectral distance; Spatial distance; The maximum spatial distance normalization factor; This is the edge suppression coefficient; The edge weights of pixels in the edge weight map are used to make the dividing lines fit the actual boundaries of ground features more closely and reduce the phenomenon of "over-segmentation". S20323: The pyramid feature extraction module of the adaptive multi-scale superpixel fusion network of the digital earth data analysis model extracts feature maps of different resolutions of the spatiotemporal data cube. In this embodiment, the pyramid feature extraction module is equipped with a convolutional neural network backbone based on ResNet-50. The spatiotemporal data cube is passed through the four stages of the network, Conv2, Conv3, Conv4 and Conv5, to extract feature maps {F1, F2, F3, F4} of different resolutions, where F1 has the highest resolution (rich in detail) and F4 has the lowest resolution (rich in semantics). S20324: The dynamic superpixel pooling module of the adaptive multi-scale superpixel fusion network using the digital earth data analysis model performs superpixel average pooling on the feature map at each resolution based on the superpixel label map to obtain object-level features at different resolutions. The formula is:

[0042] In the formula, For the first The first superpixel region i Resolution-level object-level features; For the superpixel region The first pixel i Feature maps of resolution; For the first The set of pixels in a superpixel region; This is a superpixel region indicator; i This is a resolution indicator; This is a pixel indicator; this operation aggregates millions of pixel features into hundreds of object features, significantly reducing computation and suppressing noise in individual pixels. S20325: The multi-scale feature fusion module of the adaptive multi-scale superpixel fusion network of the digital earth data analysis model performs multi-scale feature fusion on the feature maps and object-level features of different resolutions to obtain a fused feature map, i.e., the initial feature map. In this embodiment, feature map F4 is upsampled so that its size is the same as that of F3; The upsampled features are concatenated with the F3 object features along the channel dimension; Repeat this process until the original input resolution is restored; the final result is a fused feature map containing "deep semantic guidance + shallow fine boundaries".

[0043] In one alternative implementation, a spatiotemporal topology-aware module of a digital earth data analysis model is used to model the spatial relationships of the initial feature map based on a topology mask matrix, extracting spatially dependent spatiotemporal feature vectors, including: S20331: Spatial topology map construction module of spatiotemporal topology perception module of digital earth data analysis model, which converts the superpixel region corresponding to the fused feature map into a spatial topology map, and converts the fused feature map into a node feature matrix based on the spatial topology map; In this embodiment, the superpixel regions corresponding to the fused feature map are converted into a spatial topology map, and adjacency relationships are determined: two adjacency rules are defined: Physical adjacency: If superpixel regions A and B share a common boundary, then an edge exists. e AB ; Distance adjacency: If the Euclidean distance between the centroids of two superpixels is less than a threshold, a connection is established; Since the nodes of a graph neural network correspond to superpixel blocks rather than individual pixels, it is necessary to aggregate the pixel features in the feature map into the corresponding superpixel regions. Index mapping: Represents the superpixel ID to which each pixel in the superpixel label map belongs by its value; Weighted pooling: For the fused feature map, iterate through each superpixel ID and extract the feature vector of all pixel positions corresponding to that ID; Averaging calculation: Calculate the average of these feature vectors to obtain the feature vector of the corresponding superpixel node; This step transforms the features of an image into a set of node features containing several vectors. Stack the feature vectors of several nodes to form a node feature matrix, where each row represents the feature of a superpixel block. S20332: Topology mask matrix generation module of spatiotemporal topology sensing module using digital earth data analysis model, which generates the corresponding topology mask matrix based on spatial topology map; In this embodiment, the adjacency matrix and geographical distance matrix of the spatial topology graph are calculated; Construct the Gaussian decay weight matrix using the following formula:

[0044] In the formula, For nodes a and b Gaussian decay weights; For nodes a and b The distance value; Hyperparameters for controlling the decay rate; It is a natural exponential function; The formula for calculating the topology mask matrix of the final input GCN is:

[0045] In the formula, It is a topological mask matrix; This is the adjacency matrix of the spatial topology graph; This is a Gaussian decay weight matrix; this matrix preserves the adjacency structure while assigning higher weights to nearby nodes. The symbol for the Hadamarda complex; S20333: The spatiotemporal graph convolutional propagation module of the spatiotemporal topology sensing module of the digital earth data analysis model, based on the spatiotemporal graph convolutional network (ST-GCN), performs spatiotemporal graph convolutional propagation on the joint input node feature matrix and topology mask matrix to obtain spatiotemporal feature vectors; In this embodiment, temporal extension: In the time dimension, the graph structure at several consecutive moments is input into the GRU (Gated Recurrent Unit) to capture the evolution characteristics of the superpixel region over time (such as the process of urban expansion).

[0046] In one alternative implementation, the results of digital earth data analysis are post-processed and visualized, including: S20351: The results of digital earth data analysis are mapped back to the pixel grid through inverse SLIC to generate a smooth raster prediction map; S20352: Load raster prediction maps into the Digital Earth Engine, overlay them onto the 3D terrain, and use color gradients to visually display the data distribution of the raster prediction maps; S20353: Compare the current raster prediction map with historical raster prediction maps, calculate the difference map of the raster prediction maps, automatically highlight areas where drastic changes have occurred, and generate a briefing.

[0047] This invention also provides an artificial intelligence-based digital earth data analysis device, referring to... Figure 3 The device may include the following units: Model building unit 301 is used to build a digital earth data analysis model based on deep learning algorithms of artificial intelligence; The data acquisition unit 302 is used to acquire multi-source digital earth data and perform spatiotemporal registration and normalization processing on the multi-source digital earth data to obtain a spatiotemporal data cube. The data analysis unit 303 is used to input the spatiotemporal data cube into the digital earth data analysis model to perform digital earth data analysis, obtain the digital earth data analysis results, and visualize them.

[0048] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. Memory, used to store computer programs; The processor, when executing a program stored in memory, implements the artificial intelligence-based digital earth data analysis method of the present invention.

[0049] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EI) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM), or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0050] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0051] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the artificial intelligence-based digital earth data analysis method of the present invention.

[0052] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0053] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0057] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A digital earth data analysis method based on artificial intelligence, characterized in that, The method includes: A digital earth data analysis model is constructed based on deep learning algorithms for artificial intelligence. Multi-source digital earth data were collected, and spatiotemporal registration and normalization were performed on the multi-source digital earth data to obtain a spatiotemporal data cube. The spatiotemporal data cube is input into the digital earth data analysis model to perform digital earth data analysis, obtain the digital earth data analysis results, and then visualize them.

2. The digital earth data analysis method based on artificial intelligence according to claim 1, characterized in that, The digital earth data analysis model includes an adaptive multi-scale superpixel fusion network, a spatiotemporal topology sensing module, and a physical data dual-drive prediction module. The adaptive multi-scale superpixel fusion network is used to extract features from the spatiotemporal data cube to obtain an initial feature map containing multi-scale context information. The spatiotemporal topology sensing module is used to model the spatial relationships of the initial feature map based on the topology mask matrix and extract spatiotemporal feature vectors with spatial dependencies. The physical data dual-drive prediction module is used to make predictions based on the spatiotemporal feature vectors to obtain superpixel-level digital earth data analysis results.

3. The digital earth data analysis method based on artificial intelligence according to claim 2, characterized in that, The adaptive multi-scale superpixel fusion network includes a superpixel segmentation module, a pyramid feature extraction module, a dynamic superpixel pooling module, and a multi-scale feature fusion module. The superpixel segmentation module is used to perform superpixel segmentation on the spatiotemporal data cube to obtain a superpixel label map. The pyramid feature extraction module is used to extract feature maps of different resolutions from the spatiotemporal data cube; The dynamic superpixel pooling module is used to perform superpixel average pooling on the feature map at each resolution based on the superpixel label map to obtain object-level features at different resolutions. The multi-scale feature fusion module is used to perform multi-scale feature fusion on the feature maps and object-level features of different resolutions to obtain a fused feature map, i.e., an initial feature map.

4. The digital earth data analysis method based on artificial intelligence according to claim 3, characterized in that, The spatiotemporal topology sensing module includes a spatial topology graph construction module, a topology mask matrix generation module, and a spatiotemporal graph convolutional propagation module. The spatial topology graph construction module is used to convert the superpixel region corresponding to the fused feature map into a spatial topology graph, and based on the spatial topology graph, convert the fused feature map into a node feature matrix. The topology mask matrix generation module is used to generate a corresponding topology mask matrix based on the spatial topology map. The spatiotemporal graph convolution propagation module is used to perform spatiotemporal graph convolution propagation on the jointly input node feature matrix and topological mask matrix to obtain spatiotemporal feature vectors.

5. The digital earth data analysis method based on artificial intelligence according to claim 4, characterized in that, The physical data dual-drive prediction module includes a data-driven prediction head, and the physical data dual-drive prediction module is also set with a joint loss function of data fitting loss and physical constraint loss. The prediction head is used to make predictions based on the spatiotemporal feature vector to obtain superpixel-level digital earth data analysis results; The joint loss function is used to guide the optimization of network parameters in the CNN of the pyramid feature extraction module, the GCN of the spatiotemporal graph convolutional propagation module, and the MLP of the prediction head. During the training of the digital earth data analysis model, the Adam optimizer is used to update all network parameters of CNN, GCN and MLP simultaneously using the backpropagation algorithm. When the joint loss value generated by the joint loss function is less than the joint loss threshold, the optimal network parameters are obtained.

6. The digital earth data analysis method based on artificial intelligence according to claim 5, characterized in that, The formula for the joint loss function is: In the formula, This is the joint loss value; This represents the physical constraint loss value. For the first j The predicted value of the sample; For the first j The true value of the sample; Weights for physical constraint loss; j For sample indication; N The total number of samples; In the formula, Net radiation; G Soil heat flux; It is the sensible heat physical flux; This represents latent heat physical flux.

7. The digital earth data analysis method based on artificial intelligence according to claim 6, characterized in that, Multi-source digital earth data was collected, and spatiotemporal registration and normalization were performed on the multi-source digital earth data to obtain a spatiotemporal data cube, including: Collect multi-source digital earth data, which includes image data of the target area and corresponding auxiliary data; Based on auxiliary data, geometric correction, radiometric calibration and normalization are performed on the image data to obtain spatiotemporally registered image data. Define a unified grid space, convert the spatiotemporally registered image data into a binary mask, rasterize the binary mask into this grid space, and interpolate auxiliary data to the grid center point of the grid space to obtain the spatiotemporal data grid. The spatiotemporal data grid is stacked into cubes to obtain a spatiotemporal data cube.

8. The digital earth data analysis method based on artificial intelligence according to claim 7, characterized in that, The spatiotemporal data cube is input into the digital earth data analysis model for digital earth data analysis, resulting in the visualization of the analysis results, including: Input the spatiotemporal data cube into the digital earth data analysis model; An adaptive multi-scale superpixel fusion network based on a digital earth data analysis model is used to extract features from the spatiotemporal data cube to obtain an initial feature map containing multi-scale contextual information. The spatiotemporal topology sensing module of the digital earth data analysis model performs spatial relationship modeling on the initial feature map based on the topology mask matrix, and extracts spatiotemporal feature vectors with spatial dependencies. The physical data dual-drive prediction module of the digital earth data analysis model makes predictions based on the spatiotemporal feature vectors to obtain superpixel-level digital earth data analysis results. Post-processing and visualization of digital earth data analysis results.

9. The digital earth data analysis method based on artificial intelligence according to claim 8, characterized in that, Post-processing and visualization of digital earth data analysis results, including: The results of digital earth data analysis are mapped back to the pixel grid through inverse SLIC to generate a smooth raster prediction map; Load raster prediction maps into the Digital Earth Engine, overlay them onto the 3D terrain, and use color gradients to visually display the data distribution of the raster prediction maps; By comparing the current raster forecast map with historical raster forecast maps, the difference map of the raster forecast maps is calculated, areas of drastic change are automatically highlighted, and a briefing is generated.

10. A digital earth data analysis device based on artificial intelligence, used to implement the digital earth data analysis method as described in any one of claims 1-9, characterized in that, The device includes: Model building unit, used to build digital earth data analysis models based on deep learning algorithms of artificial intelligence; The data acquisition unit is used to acquire multi-source digital earth data and perform spatiotemporal registration and normalization processing on the multi-source digital earth data to obtain a spatiotemporal data cube. The data analysis unit is used to input spatiotemporal data cubes into the digital earth data analysis model for digital earth data analysis, obtain digital earth data analysis results, and visualize them.

Images