Vegetation resource distribution high-resolution prediction method and system, terminal and medium

Abstract

The invention discloses a vegetation resource distribution high-resolution prediction method, a vegetation resource distribution high-resolution prediction system, a terminal and a medium, and relates to the technical field of vegetation resource prediction. The prediction precision and generalization ability of the model are significantly improved, the improvement ensures that the model can process growth laws of complex geographic features and different vegetation types, the output prediction result better conforms to ecological laws and geographic features, and reliable support is provided for vegetation resource management.

Description

Technical Field

[0001] The present invention relates to the technical field of vegetation resource prediction, and more specifically, to a method, system, terminal and medium for high-resolution prediction of vegetation resource distribution. Background Art

[0002] With the intensification of global climate change and the deterioration of the ecological environment, the protection and management of vegetation resources has become a core issue in the field of ecological protection. For biodiversity hotspots, due to their rich vegetation resources, they play a vital role in maintaining ecological balance, enhancing carbon sink function and promoting sustainable development. However, with the changes in climate and environment, traditional vegetation resource monitoring methods are difficult to meet the high-frequency and dynamic management needs, especially in terms of spatiotemporal resolution and data update efficiency.

[0003] Existing vegetation resource prediction technologies mainly include prediction based on statistical models, ecosystem simulation based on physical models, and vegetation change prediction based on machine learning algorithms. These technologies have improved the simulation accuracy of dynamic changes in vegetation resources at different time and spatial scales. However, existing technologies still face challenges such as data heterogeneity and insufficient spatial relationship processing.

[0004] Therefore, how to study and design a high-resolution prediction method, system, terminal and medium for vegetation resource distribution that can overcome the above-mentioned defects is a problem that we urgently need to solve. Summary of the invention

[0005] In order to address the deficiencies in the prior art, the purpose of the present invention is to provide a method, system, terminal and medium for high-resolution prediction of vegetation resource distribution. The present invention significantly improves the prediction accuracy and generalization ability of the model through adaptive spatial convolution, vegetation type-sensitive temporal convolution and a regularized objective function based on geographic information and ecological indicators.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] In a first aspect, a high-resolution prediction method for vegetation resource distribution is provided, comprising the following steps:

[0008] Obtain a sample data set consisting of multi-source information fusion;

[0009] The Geo-Eco-Ada ST-GCN network is trained using the sample data set to obtain the Geo-Eco-Ada ST-GCN model;

[0010] The Geo-Eco-Ada ST-GCN model is used to predict the vegetation coverage of the target area and obtain the prediction results;

[0011] Among them, the Geo-Eco-Ada ST-GCN network is built based on the ST-GCN network. The specific construction process is:

[0012] Adjust the adjacency matrix in the ST-GCN network according to the spatial shape and / or direction of the geographic features, and optimize and update the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix;

[0013] Dynamically adjust the temporal convolution kernel size in the ST-GCN network according to the growth rate of the vegetation type;

[0014] And, according to the geographic information and / or ecological laws, corresponding regularization terms are constructed, and all regularization terms are added to the objective function of the ST-GCN network to obtain the Geo-Eco-Ada ST-GCN network.

[0015] Furthermore, the sample data in the sample data set includes temperature and precipitation, slope, altitude, topography type and land use type.

[0016] Furthermore, the process of adjusting the adjacency matrix in the ST-GCN network according to the spatial shape of the geographical features is as follows:

[0017] Determine the geographic feature vector of node i and the geographic feature vector of node j;

[0018] Determine the Euclidean distance representing the similarity of geographical features according to the geographical feature vector of node i and the geographical feature vector of node j;

[0019] The weight of the corresponding adjacency matrix is ​​adjusted according to the Euclidean distance. The higher the similarity of geographical features, the higher the weight of the adjacency matrix.

[0020] Furthermore, the process of adjusting the adjacency matrix in the ST-GCN network according to the direction of the geographical features is as follows:

[0021] Determine the direction vector of a geographic feature;

[0022] The corresponding adjacency matrix is ​​weighted according to the directional consistency of the two direction vectors.

[0023] Furthermore, the process of optimizing and updating the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix is ​​specifically as follows:

[0024] The product of the adjusted adjacency matrix and the weight matrix of the spatial convolution kernel is used as the weight matrix of the optimized updated spatial convolution kernel.

[0025] Furthermore, the process of dynamically adjusting the temporal convolution kernel size in the ST-GCN network according to the growth rate of the vegetation type is specifically as follows:

[0026] Classify the vegetation type in each grid as crop, grassland, or forest using remote sensing data or ground observation data;

[0027] Automatically learn the optimal temporal convolution kernel size for different vegetation types by analyzing historical vegetation change data;

[0028] The corresponding temporal convolution kernel size is dynamically selected according to different vegetation types.

[0029] Furthermore, corresponding regularization terms are constructed according to geographic information, including regularization terms related to altitude and regularization terms related to slope;

[0030] And / or, the regularization term constructed according to the ecological law includes a regularization term based on ecological indicators and a regularization term related to the vegetation growth cycle, and the regularization term based on ecological indicators includes a regularization term related to climate and a regularization term related to soil type.

[0031] In a second aspect, a high-resolution prediction system for vegetation resource distribution is provided, the system being used to implement a high-resolution prediction method for vegetation resource distribution as described in any one of the first aspects, comprising:

[0032] A data acquisition module is used to acquire a sample data set formed by fusion of multi-source information;

[0033] The model training module is used to train the Geo-Eco-Ada ST-GCN network through the sample data set to obtain the Geo-Eco-Ada ST-GCN model;

[0034] The vegetation prediction module is used to predict the vegetation coverage of the target area using the Geo-Eco-Ada ST-GCN model to obtain the prediction results;

[0035] Among them, the Geo-Eco-Ada ST-GCN network is built based on the ST-GCN network. The specific construction process is:

[0036] Adjust the adjacency matrix in the ST-GCN network according to the spatial shape and / or direction of the geographic features, and optimize and update the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix;

[0037] Dynamically adjust the temporal convolution kernel size in the ST-GCN network according to the growth rate of the vegetation type;

[0038] And, according to the geographic information and / or ecological laws, corresponding regularization terms are constructed, and all regularization terms are added to the objective function of the ST-GCN network to obtain the Geo-Eco-Ada ST-GCN network.

[0039] In a third aspect, a computer terminal is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a high-resolution prediction method for vegetation resource distribution as described in any one of the first aspects is implemented.

[0040] In a fourth aspect, a computer-readable medium is provided, on which a computer program is stored, and the computer program is executed by a processor to implement a high-resolution prediction method for vegetation resource distribution as described in any one of the first aspects.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention significantly improves the prediction accuracy and generalization ability of the model through adaptive spatial convolution, vegetation type-sensitive temporal convolution and regularized objective function based on geographic information and ecological indicators. This improvement ensures that the model can handle complex geographical features and growth patterns of different vegetation types, and the output prediction results are more in line with ecological laws and geographical characteristics, providing reliable support for vegetation resource management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0044] Figure 1 is a flow chart of Embodiment 1 of the present invention;

[0045] Figure 2 is the model cross-validation result in Example 1 of the present invention;

[0046] Figure 3 It is the high-resolution prediction result of vegetation resources of 6 typical cities in Example 1 of the present invention;

[0047] Figure 4 It is a system block diagram in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0049] Embodiment 1: A high-resolution prediction method for vegetation resource distribution, such as Figure 1 As shown, the following steps are included:

[0050] S1: Obtain a sample data set consisting of multi-source information fusion;

[0051] S2: Train the Geo-Eco-Ada ST-GCN network using the sample data set to obtain the Geo-Eco-Ada ST-GCN model;

[0052] S3: Use the Geo-Eco-Ada ST-GCN model to predict the vegetation coverage of the target area and obtain the prediction results.

[0053] In step S1, multi-source information fusion technology can provide more comprehensive and accurate environmental information support by combining remote sensing data, geographic information system (GIS), meteorological data and ground observation data, breaking through the limitations of a single data source. Especially driven by artificial intelligence and deep learning technology, multi-source information fusion has significantly improved the dynamic monitoring and prediction capabilities of vegetation resources.

[0054] Multi-source data were strictly preprocessed before import, including value range distribution check, outlier removal and confidence interval screening, to eliminate noise and errors to the maximum extent and ensure data quality. The integration and cleaning of multi-source data provide a solid foundation for the high-resolution prediction of vegetation resources. Through these data, the study can deeply explore the spatiotemporal characteristics of different data sources. The basic data information in the multi-source data is shown in Table 1.

[0055] Table 1 Summary of basic data information

[0056]

[0057] In step S2, the Geo-Eco-Ada ST-GCN (geo-ecological adaptive spatiotemporal graph convolution) network is constructed based on the ST-GCN (spatiotemporal graph convolution) network. The specific construction process is: adjusting the adjacency matrix in the ST-GCN network according to the spatial shape and / or direction of the geographical features, and optimizing and updating the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix; dynamically adjusting the size of the temporal convolution kernel in the ST-GCN network according to the growth rate of the vegetation type; and constructing corresponding regularization terms according to geographical information and / or ecological laws, and adding all regularization terms to the objective function of the ST-GCN network to obtain the Geo-Eco-Ada ST-GCN network.

[0058] S21, Adaptive spatial convolution operation

[0059] In traditional ST-GCN, the size of the convolution kernel is usually fixed to 3×3 or 5×5, which is suitable for regular grid processing. However, some areas have complex geographical spaces and diverse terrain and natural features, such as rivers, mountains, cities, etc., where the geographical structures are directional or irregular. In order to adapt to specific spatial shapes and directions, the present invention designs the convolution kernel size to be adaptively adjusted according to different geographical features.

[0060] In order to handle the spatial differences of different geographical features, the kernel size can be dynamically adjusted by adjusting the adjacency matrix based on geographic information. For example, narrow terrain such as rivers should have kernels with larger long sides, while in complex areas such as mountains or cities, the kernels can be more evenly distributed.

[0061] (1) Adaptive Adjustment of the Adjacency Matrix

[0062] The adjustment formula is:

[0063]

[0064] in, is the adjacency matrix adjusted according to geographical features; d(G i , G j ) represents the geographic feature vector G of node i i and the geographic feature vector G of node j j The distance between them can be calculated based on the Euclidean distance or other metrics such as altitude, slope, vegetation type, etc. σ is a parameter that controls the sparsity of the matrix.

[0065] This adaptive adjustment allows the model to adjust the connection strength based on the similarity of geographical features between nodes. For example, if nodes i and j are both located in river areas and have the same direction, then The weight of will be higher, causing the convolution operation to be more concentrated along the river direction.

[0066] (2) Introduction of adjacency matrix directionality

[0067] For geographical features with obvious directionality (such as rivers or roads), the present invention introduces the direction vector viv_ivi to reflect the main direction of the grid. The direction vector can be calculated based on the direction of the river or the terrain features, and the adjacency matrix weight can be weighted according to the direction consistency:

[0068]

[0069] in, is the weighted adjacency matrix; θ ijis the angle between the direction vectors of nodes i and j; α is the weight coefficient for adjusting the directional influence. In this way, the convolution kernel will be strengthened on neighboring nodes with similar directions, thereby performing convolution operations along the direction of linear features such as rivers.

[0070] (3) Convolution operation formula

[0071] The convolution operation not only performs spatial convolution adaptively according to the weight of the adjacency matrix, but also according to the size of the convolution kernel. The formula is as follows:

[0072]

[0073] Among them, N(i) is the neighbor set of node i, which represents the grid involved in the calculation during the convolution process; is the weighted adjacency matrix, reflecting the spatial dependency between grids; W is the weight matrix of the convolution kernel; σ() is the activation function; Represents the feature matrix of node i at layer l+1; Represents the feature matrix of node j at layer l.

[0074] This adaptive convolution kernel design ensures that the model can flexibly cope with the spatial structure of different regions, especially when dealing with complex geographical features such as rivers and mountains, which improves the accuracy of convolution operations.

[0075] S22. Differentiated temporal convolution operation

[0076] Different types of vegetation have different growth rates and changing patterns in the time dimension. Therefore, the temporal convolution operation in the present invention is dynamically adjusted according to the growth rate of the vegetation type to better capture these differences.

[0077] (1) Vegetation type-sensitive temporal convolution

[0078] In order to adapt to the growth rate of different vegetation types, the temporal convolution kernel size K v Can be dynamically adjusted according to vegetation type:

[0079] Kv=f(v);

[0080] Among them, K v is the temporal convolution kernel size for vegetation type v; f(v) is the function that adjusts the convolution kernel size according to different vegetation growth rates.

[0081] For example, fast-growing crops can use a smaller convolution kernel K v = 2, while the slow-growing forest sets a larger convolution kernel K v = 5. At each time step t, the convolution kernel size is dynamically adjusted according to the vegetation type:

[0082]

[0083] in, represents the node feature matrix at time t in the l+1th layer; represents the node feature matrix at time tk in layer l; Represents the kth convolution kernel (or weight matrix) in layer l.

[0084] This approach ensures that the model can adapt to the growth rates and dynamic changes of different types of vegetation and capture their temporal dependencies. For example, for faster-growing crops, the model uses smaller convolution kernels to better capture their rapid changes, while for slower-growing forests, larger convolution kernels are used to capture long-term trends.

[0085] (2) Classification based on vegetation type

[0086] The vegetation type in each grid is classified into crops, grassland, forest, etc. through remote sensing data or ground observation data. According to the vegetation classification results, the corresponding temporal convolution kernel size K is dynamically selected v .

[0087] (3) Data-driven adjustments

[0088] By analyzing historical vegetation change data, the optimal temporal convolution kernel size for different vegetation types can be automatically learned, and the appropriate temporal convolution kernel size can be dynamically selected according to different vegetation types.

[0089] S23. Objective function of knowledge enhancement

[0090] To further enhance the performance of the model, the objective function of the model introduces geographic information and ecological indicators to help the model use domain knowledge to impose constraints on the prediction results so that they conform to the actual geographical and ecological laws. The addition of regularization terms can ensure that the model not only relies on data, but also combines prior knowledge of geography and ecology to improve the rationality and accuracy of the prediction results.

[0091] (1) Regularization term based on geographic information

[0092] Effect of altitude: The higher the altitude, the lower the vegetation coverage is generally, so an altitude-based regularization term can be added to ensure that the vegetation coverage in high-altitude areas is appropriately reduced:

[0093]

[0094] in, represents the regularization term related to altitude; G elev represents the altitude of the ith grid, ensuring that vegetation prediction in high-altitude areas is reasonably constrained; n represents the number of observed samples; yi is the actual value, is the predicted value.

[0095] The slope and aspect have an impact. There is less vegetation in places with larger slopes, so a slope-related regularization term can be introduced:

[0096]

[0097] in, represents the regularization term related to the slope; G slope，i is the slope of the ith grid. This regularization term ensures that the vegetation coverage is lower in areas with large slopes.

[0098] (2) Regularization term based on ecological indicators

[0099] Temperature and precipitation: Temperature and precipitation affect the growth rate of vegetation, so a climate-dependent regularization term can be added:

[0100]

[0101] in, represents the regularization term related to climate, which includes temperature and precipitation; G temp , G precip Representing temperature and precipitation respectively, this ensures that vegetation grows faster in areas with suitable climate.

[0102] Soil type: Soil type affects the suitability of vegetation growth. Different soil types have different supporting effects on vegetation, so a soil regularization term can be added:

[0103]

[0104] in, represents the regularization term related to soil type; G soil，i It is the soil suitability score, ensuring higher vegetation cover where soil conditions are suitable.

[0105] (3) Vegetation growth cycle regularization term

[0106] The growth cycle of vegetation affects the growth rate of vegetation in different seasons, so the regularization term can be defined based on the growth cycle:

[0107]

[0108] in, represents the regularization term related to the vegetation growth cycle; T season，i It is the growing season of vegetation, ensuring that the model can capture higher growth rates during the vegetation growing period.

[0109] (4) Integration of objective function:

[0110] Finally, the objective function can incorporate all these regularization terms and is given by:

[0111]

[0112] Among them, L represents the loss value, and the optimization goal is to minimize it; λ1, λ2, λ3, λ4, and λ5 are the weight parameters of each regularization term, and the optimal value can be determined by cross-validation or hyperparameter tuning. These regularization terms help the model incorporate geographic information and ecological indicators into the prediction, so that the model can provide a more realistic vegetation coverage prediction.

[0113] The prediction accuracy and generalization ability of the model are significantly improved through adaptive spatial convolution, vegetation type-sensitive temporal convolution, and regularized objective functions based on geographic information and ecological indicators. This improvement ensures that the model can handle complex geographical features and the growth patterns of different vegetation types, and the output prediction results are more in line with ecological laws and geographical characteristics, providing reliable support for vegetation resource management.

[0114] Model accuracy test

[0115] To ensure the generalization ability of the model, the study adopted the K-fold cross-validation method to comprehensively test the model and reduce the risk of overfitting. The training data was divided into K parts, and training and validation were performed separately, and the results of each were combined to ensure the robust performance of the model on different data sets.

[0116] The evaluation indicators include the coefficient of determination (R 2 ) and root mean square error (RMSE). RMSE measures the deviation between the predicted value and the actual value. The smaller the value, the lower the prediction error of the model. Through these indicators, the prediction accuracy and stability of the model are evaluated to ensure its reliability in real-world scenarios. The final model has performed well after multiple verifications and optimizations.

[0117] The present invention performs 10-fold cross validation on the training data, taking into account both computational efficiency and model accuracy. Figure 2 As shown in the figure, the scatter distribution of predicted vegetation values ​​and observed values ​​is shown, and color coding is used to reflect the density of different points. The color changes from blue (sparse points) to red (dense points), reflecting the distribution characteristics of different areas.

[0118] 1) Root mean square error (RMSE) = 8.7 ppm: The average deviation between the predicted vegetation value and the actual value is 8.7 ppm. A lower RMSE indicates that the model has a small error.

[0119] 2) Coefficient of determination (R 2 )=0.82:R2 The value indicates that the model explains 82% of the variation in vegetation values, showing its superior performance in reconstructing vegetation number distribution. 2 A value close to 1 indicates that the model fits well and the correlation between the predicted and actual values ​​is high.

[0120] 3) Slope = 0.78: The slope is 0.78, which means that the predicted value is consistent with the observed value, but slightly underestimates the higher vegetation values. The ideal slope should be 1, and the current model may underestimate the prediction of higher vegetation values.

[0121] 4) Number of samples (n=23085): The scatter plot contains 23,085 samples, ensuring the credibility of the verification results.

[0122] 5) Scatter distribution: The scatter points are mainly concentrated on the y=x diagonal line, indicating that the predicted values ​​are highly consistent with the observed values. However, the point density near low vegetation values ​​is higher and the color is concentrated in the blue-green area, indicating that the prediction error increases with the increase of vegetation values.

[0123] These results show that the model can efficiently process large-scale data and demonstrate superior performance in reconstructing the spatial and temporal distribution of vegetation values. 2 The indicators demonstrate the high prediction accuracy of the model and provide strong data support for ecological environment monitoring and management.

[0124] This paper uses multi-source information fusion technology to make a high-resolution prediction of the natural resources of six typical cities in Sichuan Province in 2022, and shows the spatial distribution of vegetation in each city, revealing the vegetation coverage status and density characteristics, such as Figure 3 These prediction results not only reflect the current status of natural resources, but also provide important reference for future ecological management and planning.

[0125] First, the vegetation values ​​in Ganzi Prefecture range from 0 to 120, showing significant regional differences. The vegetation values ​​in the northern region are close to 120, indicating that the vegetation in this area is very dense and may be composed of alpine forests or natural grasslands. Due to the complex geographical conditions in Ganzi Prefecture, especially the plateau and mountainous areas in the north, the vegetation resources are rich and the ecosystem is healthy. The geographical diversity of Ganzi Prefecture has caused the vegetation cover to fluctuate greatly between dense forests and sparse areas, showing the diversity of the ecological environment. The vegetation values ​​in the central and southern parts are relatively low, and the vegetation cover is reduced, which may be affected by climatic conditions or terrain.

[0126] The vegetation numerical forecast for Aba Prefecture shows that the vegetation coverage is relatively uniform, with a vegetation value range of 0 to 80. Although there are no particularly dense vegetation areas in the region, the vegetation distribution is generally stable. Due to the mild climate and complex terrain, the ecosystem in Aba Prefecture remains balanced, and the vegetation resources are relatively evenly distributed, without significant sparseness or density, which may be related to the stability of forest protection and agricultural activities.

[0127] The vegetation values ​​of Bazhong City range from 0 to 70, and the vegetation is mainly concentrated in the central region, reflecting the rich forest resources in the region. Bazhong City is located in hilly and mountainous areas, with favorable natural conditions, so the vegetation coverage density in the central region is higher. In contrast, the vegetation values ​​in the surrounding areas are lower, which may be related to agricultural land or urbanization development. This vegetation distribution pattern shows the impact of mountainous terrain and human activities on natural resources, especially the vegetation coverage in the central region has been well protected.

[0128] The vegetation values ​​in Chengdu range from 0 to 175, highlighting the significant impact of urbanization on natural resources. The vegetation values ​​in the city center are low, with sparse vegetation cover, especially in the urban area, where the expansion of construction land has led to a decrease in vegetation density. However, the vegetation values ​​in the suburbs have increased significantly, reaching 175, indicating that a large amount of green space or natural vegetation has been retained in the suburbs. The vegetation distribution in Chengdu shows a sharp contrast between the urban area and the suburbs. Urbanization has a significant impact on the ecosystem in the city center, while the natural resources in the suburbs are relatively abundant.

[0129] The vegetation value of Dazhou City ranges from 0 to 70, and the vegetation distribution is relatively scattered. The vegetation in the southwest and central regions is relatively dense, which may be due to the rich mountain and forest resources, while the vegetation values ​​in the north and east are lower, with less vegetation coverage, which may be related to the flat terrain and more human activities.

[0130] The vegetation values ​​in Deyang range from 0 to 60, with vegetation coverage mainly concentrated in the central region, but the overall coverage is low. The urbanization process in Deyang has had a significant impact on the distribution of vegetation, with less vegetation coverage in the city center, while some densely vegetated areas in the central region may be mountainous or ecologically protected areas. Overall, the vegetation distribution pattern in Deyang shows that urbanization and agricultural development have had a significant impact on natural resources. Although there are some densely vegetated areas in the central region, the overall vegetation coverage is low.

[0131] Overall, Garze Prefecture and Chengdu City have the most extensive vegetation coverage, indicating that these areas have significant ecosystem diversity. The northern mountainous areas of Garze Prefecture and the suburbs of Chengdu City have rich vegetation resources. The vegetation coverage in Aba Prefecture and Bazhong City is relatively uniform. Although there are no particularly dense vegetation areas, it is generally stable. The vegetation coverage in Dazhou City and Deyang City is relatively scattered, especially in Deyang City, where the overall vegetation coverage is low, which may be related to the urbanization process.

[0132] These prediction results obtained through multi-source information fusion technology provide a clear and high-resolution picture of the distribution of natural resources. This information provides important reference for the formulation of ecological protection policies, rational land use and response to climate change, and can help relevant departments better manage and protect natural resources in various regions.

[0133] Embodiment 2: A high-resolution prediction system for vegetation resource distribution, the system is used to implement a high-resolution prediction method for vegetation resource distribution as described in Embodiment 1, such as Figure 4 As shown, it includes a data acquisition module, a model training module and a vegetation prediction module.

[0134] Among them, the data acquisition module is used to obtain a sample data set composed of multi-source information fusion; the model training module is used to train the Geo-Eco-Ada ST-GCN network through the sample data set to obtain the Geo-Eco-Ada ST-GCN model; the vegetation prediction module is used to use the Geo-Eco-Ada ST-GCN model to predict the vegetation coverage of the target area and obtain the prediction result.

[0135] The Geo-Eco-Ada ST-GCN network is built based on the ST-GCN network. The specific construction process is: adjust the adjacency matrix in the ST-GCN network according to the spatial shape and / or direction of the geographical features, and optimize and update the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix; dynamically adjust the size of the temporal convolution kernel in the ST-GCN network according to the growth rate of the vegetation type; and construct corresponding regularization terms according to geographic information and / or ecological laws, and add all regularization terms to the objective function of the ST-GCN network to obtain the Geo-Eco-Ada ST-GCN network.

[0136] The present invention also provides a computer terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a high-resolution prediction method for vegetation resource distribution as described in Example 1 is implemented.

[0137] The present invention also provides a computer-readable medium having a computer program stored thereon, and the computer program can be executed by a processor to implement a high-resolution prediction method for vegetation resource distribution as described in Example 1.

[0138] Working principle: The present invention significantly improves the prediction accuracy and generalization ability of the model through adaptive spatial convolution, vegetation type-sensitive temporal convolution and regularized objective function based on geographic information and ecological indicators. This improvement ensures that the model can handle complex geographical features and the growth patterns of different vegetation types, and the output prediction results are more in line with ecological laws and geographical characteristics, providing reliable support for vegetation resource management.

[0139] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0140] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0143] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-resolution prediction method for vegetation resource distribution, characterized by: The following steps are involved: Obtain a sample data set consisting of multi-source information fusion; The Geo-Eco-Ada ST-GCN network is trained using the sample data set to obtain the Geo-Eco-Ada ST-GCN model; The Geo-Eco-Ada ST-GCN model is used to predict the vegetation coverage of the target area and obtain the prediction results; Among them, the Geo-Eco-Ada ST-GCN network is built based on the ST-GCN network. The specific construction process is: Adjust the adjacency matrix in the ST-GCN network according to the spatial shape and / or direction of the geographic features, and optimize and update the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix; Dynamically adjust the temporal convolution kernel size in the ST-GCN network according to the growth rate of the vegetation type; And, according to the geographic information and / or ecological laws, corresponding regularization terms are constructed, and all regularization terms are added to the objective function of the ST-GCN network to obtain the Geo-Eco-Ada ST-GCN network.

2. The high-resolution prediction method for vegetation resource distribution according to claim 1 is characterized in that: The sample data in the sample data set include temperature and precipitation, slope, altitude, topography type and land use type.

3. The high-resolution prediction method for vegetation resource distribution according to claim 1 is characterized in that: The process of adjusting the adjacency matrix in the ST-GCN network according to the spatial shape of geographic features is as follows: Determine the geographic feature vector of node i and the geographic feature vector of node j; Determine the Euclidean distance representing the similarity of geographical features according to the geographical feature vector of node i and the geographical feature vector of node j; The weight of the corresponding adjacency matrix is ​​adjusted according to the Euclidean distance. The higher the similarity of geographical features, the higher the weight of the adjacency matrix.

4. The high-resolution prediction method for vegetation resource distribution according to claim 1 is characterized in that: The process of adjusting the adjacency matrix in the ST-GCN network according to the direction of the geographical features is as follows: Determine the direction vector of a geographic feature; The corresponding adjacency matrix is ​​weighted according to the directional consistency of the two direction vectors.

5. The high-resolution prediction method for vegetation resource distribution according to claim 1 is characterized in that: The process of optimizing and updating the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix is ​​specifically as follows: The product of the adjusted adjacency matrix and the weight matrix of the spatial convolution kernel is used as the weight matrix of the optimized updated spatial convolution kernel.

6. The high-resolution prediction method for vegetation resource distribution according to claim 1 is characterized in that: The process of dynamically adjusting the temporal convolution kernel size in the ST-GCN network according to the growth rate of the vegetation type is specifically as follows: Classify the vegetation type in each grid as crop, grassland, or forest using remote sensing data or ground observation data; Automatically learn the optimal temporal convolution kernel size for different vegetation types by analyzing historical vegetation change data; The corresponding temporal convolution kernel size is dynamically selected according to different vegetation types.

7. The high-resolution prediction method for vegetation resource distribution according to claim 1 is characterized in that: The corresponding regularization terms constructed according to the geographic information include regularization terms related to altitude and regularization terms related to slope; And / or, the regularization term constructed according to the ecological law includes a regularization term based on ecological indicators and a regularization term related to the vegetation growth cycle, and the regularization term based on ecological indicators includes a regularization term related to climate and a regularization term related to soil type.

8. A high-resolution prediction system for vegetation resource distribution, characterized by: The system is used to implement a high-resolution prediction method for vegetation resource distribution according to any one of claims 1 to 7, comprising: A data acquisition module is used to acquire a sample data set formed by fusion of multi-source information; The model training module is used to train the Geo-Eco-Ada ST-GCN network through the sample data set to obtain the Geo-Eco-Ada ST-GCN model; The vegetation prediction module is used to predict the vegetation coverage of the target area using the Geo-Eco-Ada ST-GCN model to obtain the prediction results; Among them, the Geo-Eco-Ada ST-GCN network is built based on the ST-GCN network. The specific construction process is: Adjust the adjacency matrix in the ST-GCN network according to the spatial shape and / or direction of the geographic features, and optimize and update the weight matrix of the spatial convolution kernel in the ST-GCN network according to the adjusted adjacency matrix; Dynamically adjust the temporal convolution kernel size in the ST-GCN network according to the growth rate of the vegetation type; And, according to the geographic information and / or ecological laws, corresponding regularization terms are constructed, and all regularization terms are added to the objective function of the ST-GCN network to obtain the Geo-Eco-Ada ST-GCN network.

9. A computer terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, a high-resolution prediction method for vegetation resource distribution is implemented as described in any one of claims 1-7.

10. A computer readable medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement a high-resolution prediction method for vegetation resource distribution as described in any one of claims 1 to 7.

Images