Forest structure parameter inversion model obtaining method, use method and related equipment

Abstract

The invention provides a forest structure parameter inversion model obtaining method, a forest structure parameter inversion model using method and related equipment, and is applied to the technical field of artificial intelligence. According to the method, a training set containing space lag variables is constructed based on actual measurement data of current environment factors and forest structure parameters of target area samples; a neighborhood sample is determined through spatial neighborhood matching, nonlinear mapping is performed by using a multi-layer perceptron (MLP) network structure based on a spatial Durbin model, a gradient descent optimization model is adopted, a spatial autocorrelation effect is effectively considered, spatial dependence among pixels is eliminated, and the precision of forest structure parameter inversion is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method for obtaining forest structure parameter inversion models, a method for using such models, and related equipment. Background Technology

[0002] Forest structure parameters, such as average tree height, dominant tree height, and stand density, are crucial foundational data for forest resource management, ecological monitoring, and carbon sequestration assessment. Accurate and efficient extraction and inversion of large-scale forest structure parameters are of great significance for scientific decision-making and sustainable forestry management. With the development of remote sensing technology, big data, and artificial intelligence, deep learning and machine learning methods are widely used in forest structure parameter extraction.

[0003] However, existing technologies mostly use pixels as the unit to directly predict forest structure parameters using current environmental factors. Although they have strong data processing capabilities, they generally ignore spatial autocorrelation and fail to fully consider the spatial interaction between the target area and its neighboring areas, resulting in certain limitations in prediction accuracy and model robustness.

[0004] Therefore, how to improve the prediction accuracy of forest structure parameters has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method, method of use, and related equipment for obtaining a forest structure parameter inversion model that overcomes or at least partially solves the above problems. The technical solution is as follows:

[0006] A method for obtaining a forest structure parameter inversion model includes:

[0007] The original dataset was constructed using measured data of current environmental factors and forest structure parameters from multiple sample points in the target area.

[0008] The original dataset is subjected to spatial neighborhood sample matching processing to determine a preset number of neighborhood sample points for each target sample point;

[0009] Based on the measured data of the current environmental factors and forest structure parameters of each target sample point and its neighboring sample points, a training sample set is constructed, wherein the training sample set includes the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target region;

[0010] The current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters in the training sample set are input into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain the first forest structure parameter prediction data. The multilayer perceptron (MLP) network structure is constructed based on the spatial Durbin model.

[0011] The predicted data of the first forest structure parameters are fitted with the measured data of the forest structure parameters of the target sample points to obtain an initial model;

[0012] The initial model is iteratively trained using the gradient descent algorithm until the loss function converges, thereby obtaining a forest structure parameter inversion model based on the spatial Durbin model.

[0013] Optionally, the step of performing spatial neighborhood sample matching on the original dataset to determine a preset number of neighborhood sample points for each target sample point includes:

[0014] Calculate the spatial distance between any target sample point and all candidate sample points in the original dataset;

[0015] A preset number of neighboring sample points are selected according to the spatial distance from smallest to largest.

[0016] Optionally, the step of constructing a training sample set based on the measured data of the current environmental factors and forest structure parameters of each target sample point and its neighboring sample points includes:

[0017] Using the current environmental factors of each of the neighboring sample points, the spatial lag environmental factors are obtained;

[0018] Spatial lag forest structure parameters are obtained using measured data of forest structure parameters from each of the neighboring sample points.

[0019] Optionally, the multilayer perceptron (MLP) network structure includes a linear transformation layer, a layer normalization layer, and a nonlinear activation function layer, wherein the layer normalization layer is used to stabilize the distribution of input features and enhance the model's ability to fit different nonlinear relationships.

[0020] Optionally, the loss function is the mean squared error function, and the optimization objective is to minimize the error between the predicted value and the measured value.

[0021] A method for using a forest structure parameter inversion model includes:

[0022] Obtain current environmental factors and historical forest structure parameters for multiple sample points in the area to be inverted;

[0023] Using the current environmental factors and the historical forest structure parameter data, calculate the spatially lagged environmental factors and spatially lagged forest structure parameters;

[0024] The current environmental factors, the spatially lagged environmental factors, and the spatially lagged forest structure parameters are input into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. Nonlinear mapping and feature merging are performed through the multilayer perceptron (MLP) network structure in the forest structure parameter inversion model to output the second forest structure parameter prediction data of the region to be inverted. The forest structure parameter inversion model is obtained through the method of obtaining the forest structure parameter inversion model.

[0025] A device for obtaining a forest structure parameter inversion model includes: an original dataset construction unit, a neighborhood sample point determination unit, a training sample set construction unit, a nonlinear mapping unit, a regression fitting unit, and a parameter inversion model acquisition unit;

[0026] The original dataset construction unit is used to construct the original dataset using measured data of current environmental factors and forest structure parameters from multiple sample points in the target area.

[0027] The neighborhood sample point determination unit is used to perform spatial neighborhood sample matching processing on the original dataset to determine a preset number of neighborhood sample points for each target sample point.

[0028] The training sample set construction unit is used to construct a training sample set based on the measured data of the current environmental factors and forest structure parameters of each target sample point and its neighboring sample points, wherein the training sample set includes the current environmental factors, spatially lagged environmental factors and spatially lagged forest structure parameters of the target region;

[0029] The nonlinear mapping unit is used to input the current environmental factor, the spatial lag environmental factor, and the spatial lag forest structure parameters from the training sample set into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain the first forest structure parameter prediction data. The multilayer perceptron (MLP) network structure is constructed based on the spatial Durbin model.

[0030] The regression fitting unit is used to perform regression fitting between the predicted data of the first forest structure parameters and the measured data of the forest structure parameters of the target sample points to obtain an initial model.

[0031] The parameter inversion model acquisition unit is used to iteratively train the initial model using the gradient descent algorithm until the loss function converges, thereby obtaining a forest structure parameter inversion model based on the spatial Durbin model.

[0032] A device for using a forest structure parameter inversion model includes: a data acquisition unit for the area to be inverted, a spatial lag data acquisition unit, and a forest structure parameter prediction data acquisition unit.

[0033] The data acquisition unit for the area to be inverted is used to obtain current environmental factors and historical forest structure parameters of multiple sample points in the area to be inverted.

[0034] The spatial lag data acquisition unit is used to calculate spatial lag environmental factors and spatial lag forest structure parameters using the current environmental factors and the historical forest structure parameter data.

[0035] The forest structure parameter prediction data acquisition unit is used to input the current environmental factors, the spatially lagged environmental factors, and the spatially lagged forest structure parameters into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. The model performs nonlinear mapping and feature merging through the multilayer perceptron (MLP) network structure in the forest structure parameter inversion model, and outputs the second forest structure parameter prediction data of the region to be inverted. The forest structure parameter inversion model is obtained through the forest structure parameter inversion model acquisition device.

[0036] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the method for obtaining the forest structure parameter inversion model, and / or the method for using the forest structure parameter inversion model.

[0037] An electronic device includes at least one processor, at least one memory connected to the processor, and a bus; wherein the processor and the memory communicate with each other via the bus; the processor is used to call program instructions in the memory to execute the forest structure parameter inversion model acquisition method, and / or the forest structure parameter inversion model usage method.

[0038] By employing the above technical solution, the forest structure parameter inversion model acquisition method, usage method, and related equipment provided by this invention utilize the current environmental factors and measured forest structure parameters of multiple sample points in the target area to construct an original dataset; perform spatial neighborhood sample matching processing on the original dataset to determine a preset number of neighborhood sample points for each target sample point; construct a training sample set based on the current environmental factors and measured forest structure parameters of each target sample point and its neighborhood sample points, wherein the training sample set includes the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target area; input the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters in the training sample set into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain the first forest structure parameter prediction data, wherein the MLP network structure is constructed based on the spatial Durbin model; perform regression fitting between the first forest structure parameter prediction data and the measured forest structure parameter data of the target sample points to obtain an initial model; use the gradient descent algorithm to iteratively train the initial model until the loss function converges to obtain the forest structure parameter inversion model based on the spatial Durbin model. This invention combines a spatial Durbin model and considers spatial autocorrelation effects to eliminate the potential spatial autocorrelation influence of pixels. This enables forest structure parameter inversion to suppress the spatial autocorrelation influence of pixels, making it more suitable for forest area structure parameter inversion and thus effectively improving the prediction accuracy of forest structure parameters.

[0039] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0040] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0041] Figure 1 This diagram illustrates a flowchart of one implementation of the forest structure parameter inversion model acquisition method provided by an embodiment of the present invention.

[0042] Figure 2 A schematic diagram of the multilayer perceptron (MLP) network structure provided in an embodiment of the present invention is shown.

[0043] Figure 3This diagram illustrates a flowchart of one implementation of the forest structure parameter inversion model provided in this invention.

[0044] Figure 4 This diagram illustrates the structure of the forest structure parameter inversion model acquisition device provided in an embodiment of the present invention.

[0045] Figure 5 A schematic diagram of the device used for the forest structure parameter inversion model provided in an embodiment of the present invention is shown. Detailed Implementation

[0046] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0047] Forest structure parameters (such as average tree height, dominant tree height, and stand density) are crucial foundational data for forest resource management, ecological monitoring, and carbon sequestration assessment. Accurate and efficient extraction and inversion of large-scale forest structure parameters are essential for scientific decision-making and sustainable forestry management. Currently, with the development of remote sensing, big data, and artificial intelligence technologies, deep learning and machine learning methods have been widely applied to the extraction and inversion of forest structure parameters. Existing techniques commonly use pixel-based methods to directly predict forest structure parameters using current environmental factors in the target area. While these methods can efficiently model large amounts of data, they typically neglect spatial autocorrelation, meaning they do not fully consider the spatial connections and mutual influences between the target area and its surrounding areas.

[0048] In the field of spatial statistical analysis, the Spatial Durbin Model (SDM) is a multiple regression analysis method that can effectively characterize spatial neighborhood effects. The SDM not only considers the influence of dependent variables from neighboring regions (spatial lagged dependent variables) on the dependent variable of the local region, but also incorporates independent variables from neighboring regions (spatial lagged independent variables) into the analytical framework, thus reflecting spatial interactions more comprehensively. However, existing research mostly applies the SDM to construct linear regression models for the prediction and interpretation of spatial data. For the complex nonlinear relationship that commonly exists between forest structure parameters and current environmental factors, a purely linear SDM is insufficient to effectively capture this characteristic, thus limiting the model's applicability and predictive accuracy.

[0049] Furthermore, forest structure parameters are not only influenced by current external environmental factors, but also constrained by their own growth status and spatial distribution. Therefore, in the parameter inversion process, it is crucial to simultaneously consider the synergistic effects of environmental variables and intrinsic variables. Currently, most deep learning or traditional machine learning methods fail to effectively integrate spatial autocorrelation, relying solely on pixel-level data for prediction. They lack comprehensive modeling of spatial lag effects, making it difficult to overcome the bottleneck in their prediction accuracy.

[0050] In summary, existing methods for extracting and inverting large-scale forest structure parameters have not fully integrated the effects of spatial autocorrelation and nonlinear relationships. They lack a technical approach that can comprehensively consider the spatial lag effects of independent and dependent variables in the local and neighboring regions, and leverage the powerful nonlinear modeling capabilities of deep learning. Therefore, there is an urgent need to propose a spatial scale-up method that integrates the spatial Durbin model with deep learning to achieve efficient, accurate, and spatially sensitive inversion of forest structure parameters, meeting the practical needs of forest ecological monitoring and resource management.

[0051] Based on this, this invention provides a method for obtaining a forest structure parameter inversion model. First, an original dataset containing current environmental factors and forest structure parameters is constructed using measured data from the target region. Neighborhood matching is used to obtain the neighboring points of each sample point, and spatially lagged environmental factors and spatially lagged forest structure parameters are calculated accordingly, forming a training sample set. Then, this sample set is input into a multilayer perceptron based on a spatial Durbin model for nonlinear mapping, and its output is regressed and fitted with the measured values. Gradient descent is then used for iterative training until the loss function converges, obtaining the inversion model. It is evident that this invention, by combining a spatial Durbin model and introducing a spatial lag term, effectively considers spatial autocorrelation effects and significantly improves the inversion accuracy of forest region structure parameters.

[0052] like Figure 1 The diagram shows a flowchart of one embodiment of the forest structure parameter inversion model acquisition method provided by this invention. The method may include:

[0053] S100. Construct the original dataset using measured data of current environmental factors and forest structure parameters from multiple sample points in the target area.

[0054] The target area refers to the geographic spatial range or forest area studied by the forest structure parameter inversion, which is the spatial basis for data collection and model application.

[0055] Among them, sample points refer to selected measurement points within the target area, containing measured data of current environmental factors and forest structure parameters, which are used for model training and validation.

[0056] Among them, the current environmental factors refer to external natural or anthropogenic conditions that affect forest structure, such as climate, soil, hydrology and topography, which are input into the model as independent variables.

[0057] Among them, the measured data of forest structure parameters refers to forest structure index data obtained through field surveys or remote sensing, such as average forest height, dominant tree height and stand density, which are used as dependent variables in the model.

[0058] The original dataset refers to the basic dataset consisting of measured data of current environmental factors and forest structure parameters collected from multiple sample points within the target area.

[0059] Specifically, in this embodiment of the invention, multiple representative sample points can be selected within the target area, and measured data of current environmental factors and forest structure parameters can be collected for each sample point. These data are then organized into a unified data table according to the sample point number and corresponding attributes, forming an original dataset containing current environmental factors and forest parameters, providing basic data support for subsequent model training.

[0060] As examples, embodiments of the present invention can first identify the target geographic area (i.e., the target region) where forest structure parameters need to be retrieved; within this region, a series of representative sample points are systematically deployed or selected. Next, various current environmental factor data corresponding to the location of each sample point can be obtained through methods such as remote sensing interpretation, meteorological data interpolation, or digital elevation model analysis. Simultaneously, accurate measured forest structure parameter data at these identical sample points are obtained through field measurements (such as setting up quadrats or using lidar). The current environmental factor data of all sample points are then correlated with their corresponding measured forest structure parameter data to form a structured raw dataset.

[0061] S110. Perform spatial neighborhood sample matching on the original dataset to determine the preset number of neighborhood sample points for each target sample point.

[0062] The preset quantity refers to the number of neighboring sample points pre-determined for each target sample point during spatial neighborhood sample matching processing, ensuring full utilization of local spatial information. For example, the preset quantity can be 3.

[0063] Among them, neighborhood sample points refer to several sample points around a target sample point selected according to spatial distance or other rules, which are used to construct spatial lag variables to reflect spatial mutual influence.

[0064] Specifically, in this embodiment of the invention, spatial proximity can be calculated to select a preset number of nearest sample points as neighborhood sample points for each target sample point.

[0065] S120. Based on the measured data of current environmental factors and forest structure parameters of each target sample point and its neighboring sample points, construct a training sample set, which includes the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target area.

[0066] The training sample set refers to the current environmental factors and forest structure parameters of the target sample point and its neighboring sample points. It includes the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target area, and is used to train the model.

[0067] Among them, the spatial lag environmental factor refers to the weighted combination of the current environmental factor data of the neighboring sample points, which is used to reflect the impact of the environment of the neighboring sample points on the target sample point under spatial dependence.

[0068] Among them, the spatial lag forest structure parameter refers to the weighted combination of forest structure parameters of neighboring sample points, which reflects the spatial influence of the forest structure of neighboring sample points on the forest structure of the target sample point.

[0069] Specifically, in this embodiment of the invention, training samples can be constructed by aggregating the data of neighboring sample points corresponding to each target sample point to generate spatial lag environmental factors and spatial lag forest structure parameters that characterize spatial dependencies.

[0070] S130. Input the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters from the training sample set into the pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain the first forest structure parameter prediction data. The multilayer perceptron (MLP) network structure is constructed based on the spatial Durbin model.

[0071] Among them, the Multilayer Perceptron (MLP) network structure refers to a feedforward artificial neural network, which consists of multiple linear transformations, activation functions, and layer normalization, and is used to simulate the complex nonlinear relationship between input and output.

[0072] Optionally, the multilayer perceptron (MLP) network structure includes a linear transformation layer, a layer normalization layer, and a nonlinear activation function layer. The layer normalization layer is used to stabilize the distribution of input features and enhance the model's ability to fit different nonlinear relationships.

[0073] The first forest structure parameter prediction data refers to the predicted values ​​of the forest structure parameters of the target sample points obtained by inputting the features in the training sample set into the MLP network and performing nonlinear mapping.

[0074] The Spatial Durbin Model (SDM) is a spatial econometric model that incorporates spatial lag terms for dependent and independent variables to account for the spatial interactions between variables in the local and neighboring regions, and is used for linear analysis of spatial data.

[0075] The spatial Durbin model provided in this invention utilizes deep learning methods to express and model nonlinear features. In this model structure, a spatial autocorrelation term is introduced into the nonlinear regression model, simultaneously considering the influence of spatial features on the dependent variable through a spatial interaction mechanism. Its mathematical expression is as follows:

[0076]

[0077] in, Indicates forest structure parameters (dependent variables); X Indicates environmental factors (independent variables); Indicates spatially lagging environmental factors; Indicates the spatially lagging forest structure parameters; The error term represents the difference or residual between the model's predicted values ​​and the actual observed values.

[0078] refer to Figure 2 The multilayer perceptron (MLP) network structure shown is illustrated, where i represents the target sample point number. Based on sample points in the neighboring region and the spatial Durbin model, a multilayer perceptron (MLP) is used to simulate the functional relationships in the spatial Durbin model. Specifically, as... Figure 2 As shown, this deep learning network is built upon the spatial Durbin model. The MLP structure includes linear transformation layers, layer normalization layers, and nonlinear activation functions. The layer normalization layers maintain the numerical stability of each training feature by translating and scaling it, enabling the network to flexibly simulate various complex nonlinear functions. Furthermore, the model introduces a dependent variable... As a regression input, it embodies the autoregressive term in the spatial Durbin model. After feature fusion and linear mapping, the network can output the final regression result, achieving effective modeling and prediction of spatial dependencies.

[0079] Specifically, in this embodiment of the invention, the input variables of the training sample set (current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters) can be used as input layer data for a multilayer perceptron (MLP) network. The MLP network structure design incorporates the mechanism of the spatial Durbin model, including linear transformation, layer normalization, and nonlinear activation functions, which can capture complex nonlinear spatial dependencies. After forward propagation, the network outputs the prediction results of the first round of forest structure parameters.

[0080] As examples, embodiments of the present invention can construct a specific multilayer perceptron (MLP) network structure. The core design idea of ​​this network originates from the spatial Durbin model, and its input layer is designed to simultaneously receive current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters. These features are transformed and extracted layer by layer through multiple hidden layers containing linear transformations, layer normalization, and the nonlinear activation function ReLU. The network ultimately outputs a preliminary prediction value, namely the first forest structure parameter prediction data.

[0081] S140. Perform regression fitting between the predicted forest structure parameters and the measured forest structure parameters of the target sample points to obtain the initial model.

[0082] The initial model refers to the model state after the network structure has been constructed and the predicted data of the first forest structure parameters have been fitted with the measured data of the forest structure parameters for the first time, but before it has undergone sufficient iterative optimization.

[0083] Specifically, in this embodiment of the invention, regression analysis can be used to fit the forest structure parameters predicted by the MLP network with the measured values ​​of the corresponding target sample points to determine the initial estimates of the model parameters. This step ensures an initial match between the model prediction results and the actual observation data, providing a reasonable starting point for subsequent optimization training.

[0084] As examples, embodiments of the present invention can compare the predicted data of the first forest structure parameters with the actual measured data of the forest structure parameters of the target sample points. A regression layer (such as a linear transformation layer) maps the high-level features extracted by the network to the final predicted values, and calculates the error (such as mean squared error) between the predicted and true values. At this point, the entire network structure (feature extraction MLP + regression layer) and its current parameters constitute a complete, but unoptimized, initial model.

[0085] S150. The gradient descent algorithm is used to iteratively train the initial model until the loss function converges, thereby obtaining the forest structure parameter inversion model based on the spatial Durbin model.

[0086] Optionally, the loss function can be the mean squared error function, and the optimization objective is to minimize the error between the predicted value and the measured value.

[0087] The forest structure parameter inversion model refers to the final, usable model obtained after the initial model has been repeatedly trained using the gradient descent algorithm until the loss function converges. This model can accurately predict forest structure parameters in unknown regions using data from the target region and neighboring regions.

[0088] Specifically, in this embodiment of the invention, the MLP network can be iteratively trained using a gradient descent optimization algorithm based on the initial model parameters. The model parameters are adjusted to minimize the error loss function between the predicted and measured values. The model parameters are continuously updated during training until the loss function converges stably, ultimately yielding a high-precision forest structure parameter inversion model that integrates the spatial Durbin model.

[0089] As examples, embodiments of the present invention can use the Adam optimizer with gradient descent algorithm to iteratively train the initial model. In each iteration, the model processes a batch of training data, calculates the total error of the current prediction based on the mean squared error loss function, and then calculates the gradient of the error with respect to all model parameters using the backpropagation algorithm. Next, the optimizer updates the model parameters based on these gradients to reduce the error. This process is repeated until the value of the loss function stabilizes near a minimum point, i.e., it reaches a "convergence" state. The model with optimal parameters obtained at this point can then be used as an inversion model for actual forest structure parameters in prediction.

[0090] The present invention provides a method for obtaining a forest structure parameter inversion model, comprising: constructing an original dataset using measured data of current environmental factors and forest structure parameters of multiple sample points in a target region; performing spatial neighborhood sample matching on the original dataset to determine a preset number of neighborhood sample points for each target sample point; constructing a training sample set based on the measured data of current environmental factors and forest structure parameters of each target sample point and its neighborhood sample points, wherein the training sample set includes the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target region; inputting the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters in the training sample set into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain first forest structure parameter prediction data, wherein the MLP network structure is constructed based on a spatial Durbin model; performing regression fitting between the first forest structure parameter prediction data and the measured forest structure parameter data of the target sample points to obtain an initial model; and iteratively training the initial model using a gradient descent algorithm until the loss function converges to obtain a forest structure parameter inversion model based on a spatial Durbin model. This invention combines a spatial Durbin model and considers spatial autocorrelation effects to eliminate the potential spatial autocorrelation influence of pixels. This enables forest structure parameter inversion to suppress the spatial autocorrelation influence of pixels, making it more suitable for forest area structure parameter inversion and thus effectively improving the prediction accuracy of forest structure parameters.

[0091] Optionally, in the above Figure 1 Based on one or more corresponding embodiments, in another optional embodiment provided by the present invention, step S110 may specifically include:

[0092] Calculate the spatial distance between any target sample point in the original dataset and all candidate sample points; then select a preset number of neighboring sample points in ascending order of spatial distance.

[0093] Specifically, in this embodiment of the invention, neighborhood samples can be screened for each target sample point based on its spatial coordinates using a distance metric (e.g., Euclidean distance). A preset number of neighboring sample points are selected as neighborhood samples of the target point, sorted from closest to furthest by distance, to ensure that samples within the neighborhood reflect spatial correlation and facilitate the construction of spatially lagged variables.

[0094] As examples, embodiments of the present invention can calculate the Euclidean distance between each target sample point in the original dataset and all other sample points in the dataset. Based on a preset number (e.g., K=3), the K sample points closest to the current target sample point are selected from all candidate points and determined as the neighborhood sample points of the target sample point, thereby constructing a neighborhood set reflecting its local spatial background for each target sample point.

[0095] This invention calculates the spatial distance between any target sample point and all candidate sample points, and selects a preset number of neighboring sample points according to the distance from nearest to farthest. This effectively captures the spatial proximity relationships around the target sample point, ensuring that the neighboring samples truly reflect the similarity and correlation in geographic space. This constructs accurate spatial lag variables, enhances the model's ability to represent spatial autocorrelation, and improves the accuracy and reliability of forest structure parameter inversion. It also avoids interference caused by spatial heterogeneity and heteroscedasticity, promotes the rational use of spatial information, and improves the overall performance of the model.

[0096] Optionally, in the above Figure 1 Based on one or more corresponding embodiments, in another optional embodiment provided by the present invention, step S120 may specifically include:

[0097] Spatial lag environmental factors are obtained by using the current environmental factors of each neighboring sample point; spatial lag forest structure parameters are obtained by using the measured data of forest structure parameters of each neighboring sample point.

[0098] Specifically, in this embodiment of the invention, the current environmental factors and forest structure parameters of neighboring sample points can be weighted and averaged according to weights (such as a proximity weight matrix) to obtain spatially lagged environmental factors and spatially lagged forest structure parameters. The current environmental factors of the target sample points are combined with the corresponding spatial lags to construct a training sample set containing the original independent variables and their spatial lag terms, fully reflecting the spatial dependencies.

[0099] As examples, embodiments of the present invention can traverse each target sample point and its set of neighboring sample points. For each target sample point: its current environmental factor data is directly taken as the "local feature"; the current environmental factor data of all its neighboring sample points are weighted and averaged according to their distance from the target sample point (the closer the distance, the higher the weight), and the result is used as the spatial lag environmental factor; the measured forest structure parameter data of all its neighboring sample points are similarly weighted and averaged, and the result is used as the spatial lag forest structure parameter. These three features are combined to form the training sample set used to train the model.

[0100] This invention constructs a training sample set based on measured data of current environmental factors and forest structure parameters of the target sample point and its neighboring sample points. By using the current environmental factors of the neighboring sample points to calculate spatially lagged environmental factors and using the measured data of forest structure parameters of the neighboring sample points to calculate spatially lagged forest structure parameters, it can effectively characterize the spatial dependencies between the target area and neighboring areas, fully reflect the spatial autocorrelation characteristics of current environmental factors and forest structure parameters, thereby enhancing the model's ability to express complex spatial interactions, improving the accuracy and stability of forest structure parameter inversion, overcoming the local bias that may be caused by relying solely on single-point data, achieving reasonable modeling of spatial heterogeneity and spatial continuity, and facilitating the construction of a more generalizable and explanatory forest structure parameter inversion model.

[0101] like Figure 3 The diagram shows a flowchart illustrating one implementation of the forest structure parameter inversion model provided in this invention. The method may include:

[0102] S300: Obtain current environmental factors and historical forest structure parameters for multiple sample points in the area to be inverted.

[0103] Specifically, in this embodiment of the invention, a reasonable number of sample points can be deployed within the target area to be inverted. Current environmental factor data (such as temperature, precipitation, soil properties, and topography) and historical forest structure parameter data (such as historical measured values ​​of average tree height, dominant tree height, and stand density) are collected for each sample point to ensure data representativeness and spatiotemporal consistency. The collected data is then organized into a structured database or data tables to prepare for subsequent spatial lag variable calculations and model input.

[0104] As examples, embodiments of the present invention identify the areas to be inverted for forest structure parameter prediction. Within these areas, a series of sample points are systematically determined using methods such as regular grids or random point placement. Then, spatial information technologies such as remote sensing image interpretation, reanalysis meteorological data interpolation, or digital elevation model analysis are used to acquire environmental factor data corresponding to the location of each sample point, similar to those used during model training. Simultaneously, if available early forest resource survey data or historical remote sensing inversion products exist, they are used as the historical forest structure parameter data for that point; otherwise, this data item can be left blank, and the calculation of its spatial lag term will primarily rely on information from neighboring sample points. Finally, a dataset covering the area to be inverted and containing all necessary input variables is formed.

[0105] S310. Using current environmental factors and historical forest structure parameter data, calculate spatially lagged environmental factors and spatially lagged forest structure parameters.

[0106] Specifically, in this embodiment of the invention, the spatial distance between each sample point and its neighboring sample points within the area to be inverted can be calculated based on the geographical coordinates of each sample point, thus determining the neighborhood sample set (e.g., selecting the three closest neighboring points). Then, based on the current environmental factors and historical forest structure parameters of the neighboring sample points, spatially lagged environmental factors and spatially lagged forest structure parameters are calculated to reflect the spatial influence of the neighboring area.

[0107] As examples, embodiments of the present invention can traverse every target sample point in the region to be inverted. For each target sample point: calculate its Euclidean distance to all other sample points in the region. Based on a preset number of neighborhoods, select the K nearest sample points to the target sample point as its neighborhood sample points. Calculate a weighted average of the environmental factor data of these K neighborhood sample points according to their distances to the target sample point (the closer the distance, the higher the weight), generating a spatially lagged environmental factor vector that comprehensively reflects the surrounding environmental conditions. Similarly, calculate a distance-weighted average of the historical forest structure parameter data of these K neighborhood sample points to generate a spatially lagged forest structure parameter value that reflects the background of the surrounding forest structure.

[0108] S320. Input the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters into the pre-trained forest structure parameter inversion model based on the spatial Durbin model. Perform nonlinear mapping and feature merging through the multilayer perceptron (MLP) network structure in the forest structure parameter inversion model, and output the second forest structure parameter prediction data of the region to be inverted. The forest structure parameter inversion model is obtained by the above-mentioned method for obtaining the forest structure parameter inversion model.

[0109] Specifically, in this embodiment of the invention, the current environmental factors, corresponding spatially lagged environmental factors, and spatially lagged forest structure parameters of each sample point are used as input features and fed into a pre-trained multilayer perceptron (MLP) network based on a spatial Durbin model. This MLP network includes linear transformations, layer normalization, and nonlinear activation functions, enabling it to capture complex nonlinear spatial dependencies. The forest structure parameter inversion model utilizes these input features for nonlinear mapping and feature fusion to generate predicted forest structure parameters for the target sample points, i.e., second forest structure parameter prediction data, providing a scientific basis for forest resource management and ecological monitoring.

[0110] As examples, embodiments of the present invention simultaneously input the environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of each target sample point into a pre-trained forest structure parameter inversion model. This forest structure parameter inversion model is internally constructed based on a spatial Durbin model structure, with a multilayer perceptron (MLP) network at its core. After receiving these three types of inputs, the MLP network performs nonlinear mapping and feature merging layer by layer through its internal multilayer linear transformations, layer normalization, and nonlinear activation functions, thereby learning and expressing the complex nonlinear spatial relationship between environmental variables and forest structure parameters. Finally, the forest structure parameter inversion model outputs the predicted forest structure parameters for each sample point location; the set of all these predicted values ​​constitutes the second forest structure parameter prediction data for the region to be inverted.

[0111] The forest structure parameter inversion model and method provided in this invention obtains current environmental factors and historical forest structure parameter data of multiple sample points within the region to be inverted, and calculates spatially lagged environmental factors and spatially lagged forest structure parameters based on these data. This comprehensively captures the spatial dependencies between the target region and its neighborhood, fully reflecting the spatial mutual influence and continuity of environmental factors and forest structure parameters. Subsequently, these multidimensional spatial features are input into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. The multilayer perceptron (MLP) network structure accurately simulates the complex nonlinear spatial correlation between environmental variables and forest structure parameters through nonlinear mapping and feature fusion, thereby outputting high-precision second forest structure parameter prediction data. This effectively integrates the advantages of spatial autocorrelation and nonlinear feature expression, significantly improving the accuracy of forest structure parameter inversion.

[0112] Although the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous.

[0113] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0114] Corresponding to the above-mentioned method for obtaining forest structure parameter inversion models, this embodiment of the invention also provides a device for obtaining forest structure parameter inversion models, the structure of which is as follows: Figure 4 As shown, it may include: original dataset construction unit 10, neighborhood sample point determination unit 11, training sample set construction unit 12, nonlinear mapping unit 13, regression fitting unit 14, and parameter inversion model acquisition unit 15.

[0115] The original dataset construction unit 10 is used to construct the original dataset using measured data of current environmental factors and forest structure parameters from multiple sample points in the target area.

[0116] The neighborhood sample point determination unit 11 is used to perform spatial neighborhood sample matching processing on the original dataset and determine a preset number of neighborhood sample points for each target sample point.

[0117] The training sample set construction unit 12 is used to construct a training sample set based on the measured data of the current environmental factors and forest structure parameters of each target sample point and its neighboring sample points. The training sample set includes the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target area.

[0118] The nonlinear mapping unit 13 is used to input the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters from the training sample set into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain the first forest structure parameter prediction data. The multilayer perceptron (MLP) network structure is constructed based on the spatial Durbin model.

[0119] The regression fitting unit 14 is used to perform regression fitting between the predicted data of the first forest structure parameters and the measured data of the forest structure parameters of the target sample points to obtain the initial model.

[0120] Unit 15, which obtains the parameter inversion model, is used to iteratively train the initial model using the gradient descent algorithm until the loss function converges, thereby obtaining the forest structure parameter inversion model based on the spatial Durbin model.

[0121] Optionally, the neighborhood sample point determination unit 11 is specifically used to calculate the spatial distance between any target sample point in the original dataset and all candidate sample points; and to select a preset number of neighborhood sample points in order of increasing spatial distance.

[0122] Optionally, the training sample set construction unit 12 is specifically used to obtain spatially lagged environmental factors by utilizing the current environmental factors of each neighboring sample point; and to obtain spatially lagged forest structure parameters by utilizing the measured data of forest structure parameters of each neighboring sample point.

[0123] Optionally, the multilayer perceptron (MLP) network structure includes a linear transformation layer, a layer normalization layer, and a nonlinear activation function layer. The layer normalization layer is used to stabilize the distribution of input features and enhance the model's ability to fit different nonlinear relationships.

[0124] Optionally, the loss function is the mean squared error function, and the optimization objective is to minimize the error between the predicted value and the measured value.

[0125] The present invention provides a device for obtaining a forest structure parameter inversion model. This device is used to: construct an original dataset using measured data of current environmental factors and forest structure parameters from multiple sample points in a target region; perform spatial neighborhood sample matching on the original dataset to determine a preset number of neighborhood sample points for each target sample point; construct a training sample set based on the measured data of current environmental factors and forest structure parameters from each target sample point and its neighborhood sample points, wherein the training sample set includes current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target region; input the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters from the training sample set into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain first forest structure parameter prediction data, wherein the MLP network structure is constructed based on a spatial Durbin model; perform regression fitting between the first forest structure parameter prediction data and the measured forest structure parameter data of the target sample points to obtain an initial model; and iteratively train the initial model using a gradient descent algorithm until the loss function converges to obtain a forest structure parameter inversion model based on a spatial Durbin model. This invention combines a spatial Durbin model and considers spatial autocorrelation effects to eliminate the potential spatial autocorrelation influence of pixels. This enables forest structure parameter inversion to suppress the spatial autocorrelation influence of pixels, making it more suitable for forest area structure parameter inversion and thus effectively improving the prediction accuracy of forest structure parameters.

[0126] Corresponding to the above-described method for using the forest structure parameter inversion model, this embodiment of the invention also provides a device for using the forest structure parameter inversion model, the structure of which is as follows: Figure 5 As shown, it may include: a data acquisition unit 20 for the area to be inverted, a spatial lag data acquisition unit 21, and a forest structure parameter prediction data acquisition unit 22.

[0127] The data acquisition unit 20 for the area to be inverted is used to obtain current environmental factors and historical forest structure parameters of multiple sample points in the area to be inverted.

[0128] The spatial lag data acquisition unit 21 is used to calculate spatial lag environmental factors and spatial lag forest structure parameters using current environmental factor and historical forest structure parameter data.

[0129] The forest structure parameter prediction data acquisition unit 22 is used to input the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. The model performs nonlinear mapping and feature merging through the multilayer perceptron (MLP) network structure in the forest structure parameter inversion model, and outputs the second forest structure parameter prediction data of the region to be inverted. The forest structure parameter inversion model is obtained through the forest structure parameter inversion model acquisition device.

[0130] The forest structure parameter inversion model provided in this invention uses an apparatus to obtain current environmental factors and historical forest structure parameter data of multiple sample points within the area to be inverted. Based on this data, it calculates spatially lagged environmental factors and spatially lagged forest structure parameters, which can comprehensively capture the spatial dependencies between the target area and its neighborhood, fully reflecting the spatial mutual influence and continuity of environmental factors and forest structure parameters. Subsequently, these multidimensional spatial features are input into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. The multilayer perceptron (MLP) network structure accurately simulates the complex nonlinear spatial correlation between environmental variables and forest structure parameters through nonlinear mapping and feature fusion, thereby outputting high-precision second forest structure parameter prediction data. This effectively integrates the advantages of spatial autocorrelation and nonlinear feature expression, significantly improving the accuracy of forest structure parameter inversion.

[0131] Regarding the apparatus in the above embodiments, the specific manner in which each unit performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0132] The forest structure parameter inversion model acquisition device includes a processor and a memory. The original dataset construction unit 10, the neighborhood sample point determination unit 11, the training sample set construction unit 12, the nonlinear mapping unit 13, the regression fitting unit 14, and the parameter inversion model acquisition unit 15 are all stored in the memory as program units. The processor executes the above-mentioned program units stored in the memory to realize the corresponding functions.

[0133] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and by adjusting kernel parameters, a training set containing spatially lagged variables is constructed based on measured data of current environmental factors and forest structure parameters of the target region samples. Neighborhood samples are determined through spatial neighborhood matching, and a multilayer perceptron (MLP) network structure based on a spatial Durbin model is used for nonlinear mapping. Gradient descent is employed to optimize the model, effectively considering spatial autocorrelation effects, eliminating spatial dependencies between pixels, and significantly improving the accuracy of forest structure parameter inversion.

[0134] This invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements the method for obtaining the forest structure parameter inversion model.

[0135] This invention provides a processor for running a program, wherein the program executes the forest structure parameter inversion model acquisition method during runtime.

[0136] This invention provides an electronic device, which includes at least one processor, at least one memory connected to the processor, and a bus; wherein the processor and the memory communicate with each other via the bus; the processor is used to call program instructions in the memory to execute the above-described method for obtaining the forest structure parameter inversion model. The electronic device described herein may be a server, PC, PAD, mobile phone, etc.

[0137] The present invention also provides a computer program product, which, when executed on an electronic device, is suitable for executing the steps of initializing a method for obtaining a forest structure parameter inversion model.

[0138] The device used for the forest structure parameter inversion model includes a processor and a memory. The aforementioned data acquisition unit 20 for the area to be inverted, spatial lag data acquisition unit 21, and forest structure parameter prediction data acquisition unit 22 are all stored in the memory as program units. The processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0139] The processor contains a kernel that retrieves the corresponding program units from memory. One or more kernels can be configured. By adjusting kernel parameters, current environmental factors and historical forest structure parameters of multiple sample points within the region to be inverted are obtained. Based on this data, spatially lagged environmental factors and spatially lagged forest structure parameters are calculated. This comprehensively captures the spatial dependencies between the target region and its neighborhood, fully reflecting the spatial interaction and continuity of environmental factors and forest structure parameters. Subsequently, these multidimensional spatial features are input into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. The multilayer perceptron (MLP) network structure accurately simulates the complex nonlinear spatial correlation between environmental variables and forest structure parameters through nonlinear mapping and feature fusion, thereby outputting high-precision second forest structure parameter prediction data. This effectively integrates the advantages of spatial autocorrelation and nonlinear feature representation, significantly improving the accuracy of forest structure parameter inversion.

[0140] This invention provides another computer-readable storage medium storing a program that, when executed by a processor, implements the method for using the forest structure parameter inversion model.

[0141] This invention provides another processor for running a program, wherein the program executes the method for using the forest structure parameter inversion model during runtime.

[0142] This invention provides another electronic device, which includes at least one processor, at least one memory connected to the processor, and a bus; wherein the processor and the memory communicate with each other via the bus; the processor is used to call program instructions in the memory to execute the above-described method for using the forest structure parameter inversion model. The electronic device in this document can be a server, PC, PAD, mobile phone, etc.

[0143] The present invention also provides a computer program product that, when executed on an electronic device, is suitable for executing the steps of initializing a forest structure parameter inversion model.

[0144] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices (systems), and computer program products according to embodiments of the invention. It will 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 device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0145] In a typical configuration, an electronic device includes one or more processors (CPUs), memory, and a bus. The electronic device may also include input / output interfaces, network interfaces, etc.

[0146] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM, and memory includes at least one memory chip. Memory is an example of computer-readable media.

[0147] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0148] In the description of this invention, it should be understood that if the terms "upper", "lower", "front", "rear", "left" and "right" are used to indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the position or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0149] 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. It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0150] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0151] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the present invention.

Claims

1. A method for obtaining a forest structure parameter inversion model, characterized in that, include: The original dataset was constructed using measured data of current environmental factors and forest structure parameters from multiple sample points in the target area. The original dataset is subjected to spatial neighborhood sample matching processing to determine a preset number of neighborhood sample points for each target sample point; Based on the measured data of the current environmental factors and forest structure parameters of each target sample point and its neighboring sample points, a training sample set is constructed, wherein the training sample set includes the current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters of the target region; The current environmental factors, spatially lagged environmental factors, and spatially lagged forest structure parameters in the training sample set are input into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain the first forest structure parameter prediction data. The multilayer perceptron (MLP) network structure is constructed based on the spatial Durbin model. The predicted data of the first forest structure parameters are fitted with the measured data of the forest structure parameters of the target sample points to obtain an initial model; The initial model is iteratively trained using the gradient descent algorithm until the loss function converges, thereby obtaining a forest structure parameter inversion model based on the spatial Durbin model.

2. The method according to claim 1, characterized in that, The step of performing spatial neighborhood sample matching on the original dataset to determine a preset number of neighborhood sample points for each target sample point includes: Calculate the spatial distance between any target sample point and all candidate sample points in the original dataset; A preset number of neighboring sample points are selected according to the spatial distance from smallest to largest.

3. The method according to claim 1, characterized in that, The training sample set is constructed based on the measured data of the current environmental factors and forest structure parameters of each target sample point and its neighboring sample points, including: Using the current environmental factors of each of the neighboring sample points, the spatial lag environmental factors are obtained; Spatial lag forest structure parameters are obtained using measured data of forest structure parameters from each of the neighboring sample points.

4. The method according to claim 1, characterized in that, The multilayer perceptron (MLP) network structure includes a linear transformation layer, a layer normalization layer, and a nonlinear activation function layer. The layer normalization layer is used to stabilize the distribution of input features and enhance the model's ability to fit different nonlinear relationships.

5. The method according to any one of claims 1 to 4, characterized in that, The loss function is the mean squared error function, and the optimization objective is to minimize the error between the predicted value and the measured value.

6. A method for using a forest structure parameter inversion model, characterized in that, include: Obtain current environmental factors and historical forest structure parameters for multiple sample points in the area to be inverted; Using the current environmental factors and the historical forest structure parameter data, calculate the spatially lagged environmental factors and spatially lagged forest structure parameters; The current environmental factors, the spatially lagged environmental factors, and the spatially lagged forest structure parameters are input into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. Nonlinear mapping and feature merging are performed through the multilayer perceptron (MLP) network structure in the forest structure parameter inversion model to output the second forest structure parameter prediction data of the region to be inverted. The forest structure parameter inversion model is obtained by the method described in any one of claims 1 to 5.

7. A device for obtaining a forest structure parameter inversion model, characterized in that, include: The system comprises: a raw dataset construction unit, a neighborhood sample point determination unit, a training sample set construction unit, a nonlinear mapping unit, a regression fitting unit, and a parameter inversion model acquisition unit. The original dataset construction unit is used to construct the original dataset using measured data of current environmental factors and forest structure parameters from multiple sample points in the target area. The neighborhood sample point determination unit is used to perform spatial neighborhood sample matching processing on the original dataset to determine a preset number of neighborhood sample points for each target sample point. The training sample set construction unit is used to construct a training sample set based on the measured data of the current environmental factors and forest structure parameters of each target sample point and its neighboring sample points, wherein the training sample set includes the current environmental factors, spatially lagged environmental factors and spatially lagged forest structure parameters of the target region; The nonlinear mapping unit is used to input the current environmental factor, the spatial lag environmental factor, and the spatial lag forest structure parameters from the training sample set into a pre-constructed multilayer perceptron (MLP) network structure for nonlinear mapping to obtain the first forest structure parameter prediction data. The multilayer perceptron (MLP) network structure is constructed based on the spatial Durbin model. The regression fitting unit is used to perform regression fitting between the predicted data of the first forest structure parameters and the measured data of the forest structure parameters of the target sample points to obtain an initial model. The parameter inversion model acquisition unit is used to iteratively train the initial model using the gradient descent algorithm until the loss function converges, thereby obtaining a forest structure parameter inversion model based on the spatial Durbin model.

8. A device for using a forest structure parameter inversion model, characterized in that, include: Units for obtaining data of the area to be inverted, spatial lag data, and forest structure parameter prediction data; The data acquisition unit for the area to be inverted is used to obtain current environmental factors and historical forest structure parameters of multiple sample points in the area to be inverted. The spatial lag data acquisition unit is used to calculate spatial lag environmental factors and spatial lag forest structure parameters using the current environmental factors and the historical forest structure parameter data. The forest structure parameter prediction data acquisition unit is used to input the current environmental factors, the spatially lagged environmental factors, and the spatially lagged forest structure parameters into a pre-trained forest structure parameter inversion model based on a spatial Durbin model. The unit performs nonlinear mapping and feature merging through a multilayer perceptron (MLP) network structure in the forest structure parameter inversion model, and outputs the second forest structure parameter prediction data for the region to be inverted. The forest structure parameter inversion model is obtained using the forest structure parameter inversion model acquisition device as described in claim 7.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the method for obtaining the forest structure parameter inversion model as described in any one of claims 1 to 5, and / or the method for using the forest structure parameter inversion model as described in claim 6.

10. An electronic device, characterized in that, The electronic device includes at least one processor, at least one memory connected to the processor, and a bus; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the forest structure parameter inversion model acquisition method as described in any one of claims 1 to 5, and / or the forest structure parameter inversion model usage method as described in claim 6.

Images