A method for calculating the carbon sink value of ecological products

Abstract

The present invention discloses a method for calculating the value of carbon sink of an ecological product, which belongs to the technical field of resource assessment. The method solves the problem that the existing method has insufficient coverage and resolution of a single remote sensing data, is difficult to adapt to the spatial heterogeneity of the region to which the generated product belongs, and thus leads to deviations in the regional carbon sink estimation results. The method comprises: collecting multi-source heterogeneous carbon sink monitoring data associated with a sub-monitoring region in real time according to a preset collection period, preprocessing the multi-source heterogeneous carbon sink monitoring data, and performing data enhancement on a normalized data set based on a Mosaic‑Mixup hybrid method; executing a carbon sink value accounting model, and the carbon sink value accounting model outputs a carbon sink value accounting result; the present invention adopts a method for preprocessing the multi-source heterogeneous carbon sink monitoring data and performing data enhancement on the normalized data set to provide accurate and scientific data support for the carbon sink value accounting of ecological products, thereby ensuring that the carbon sink value accounting model accurately evaluates the carbon sink capacity of regional ecological products.

Description

Technical Field

[0001] The present invention belongs to the technical field of resource assessment, and in particular relates to a method for calculating the carbon sink value of an ecological product. Background Art

[0002] The carbon sink value of ecological products refers to the process of converting the carbon absorption and storage functions of ecosystems into economic value. The carbon sink value in ecological products is also an important way to achieve carbon neutrality and ecological value transformation. It not only helps to mitigate global climate change, but also promotes a virtuous cycle of ecological protection and economic development. The carbon sink value accounting of ecological products provides a scientific basis for carbon reduction targets. By systematically evaluating the ability to absorb and store carbon, the role and contribution of different ecosystems in the carbon cycle can be clarified.

[0003] Chinese patent CN118410304A discloses a method and system for dynamic monitoring and accounting of carbon sinks in a mulberry garden ecosystem, including: obtaining laser radar data and remote sensing optical images of a target area for preprocessing, and extracting multi-source characteristic variables in the accounting area; obtaining carbon sink baseline data in the accounting area, extracting the spatiotemporal characteristics of the carbon sink estimation sequence, using the spatiotemporal characteristics to screen high-response characteristic variables from the multi-source characteristic variables, and constructing a carbon sink dynamic monitoring model; obtaining corresponding characteristic variable parameters according to the current botanical characteristics and pruning characteristics of the target mulberry garden, and using the model output to obtain the current carbon sink estimation data of the target mulberry garden; however, when the existing method model uses a single remote sensing data to calculate the carbon sink estimation data, the single remote sensing data has insufficient coverage and resolution, and it is difficult to adapt to the spatial heterogeneity of the region to which the generated product belongs, which leads to deviations in the regional carbon sink estimation results, and it is also difficult to ensure that the model performs reliability analysis on regional carbon sink data at different spatial and temporal scales. In response to the above problems, we propose a method for calculating the carbon sink value of ecological products. Summary of the invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and provide a method for calculating the carbon sink value of ecological products. The method solves the problem that when the existing method model uses single remote sensing data to calculate carbon sink estimation data, the coverage and resolution of the single remote sensing data are insufficient, and it is difficult to adapt to the spatial heterogeneity of the area to which the generated product belongs, which in turn leads to deviations in the regional carbon sink estimation results.

[0005] When the existing method model uses a single remote sensing data to calculate the carbon sink estimation data, the coverage and resolution of the single remote sensing data are insufficient, and it is difficult to adapt to the spatial heterogeneity of the region to which the generated product belongs, which leads to deviations in the regional carbon sink estimation results. It is also difficult to ensure that the model can perform reliability analysis on regional carbon sink data at different spatial and temporal scales. To address the above problems, we propose an ecological product carbon sink value accounting method. In short, the method includes real-time collection of multi-source heterogeneous carbon sink monitoring data associated with sub-monitoring areas according to a preset collection period, preprocessing the multi-source heterogeneous carbon sink monitoring data to obtain a normalized data set, and enhancing the normalized data set based on the Mosaic-Mixup hybrid method to obtain a data enhancement set. A carbon sink value accounting model is constructed by combining support vector machines and self-attention mechanisms. The data enhancement set is used as input to execute the carbon sink value accounting model, and the carbon sink value accounting model outputs the carbon sink value accounting results. In an embodiment of the present invention, a method of preprocessing multi-source heterogeneous carbon sink monitoring data and enhancing normalized data sets can provide accurate and scientific data support for ecological product carbon sink value accounting, thereby ensuring that the carbon sink value accounting model accurately evaluates the carbon sink capacity of regional ecological products. At the same time, the carbon sink value accounting model combines basic geographic data, ecological product classification data, multispectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and other data, so that the carbon sink value accounting result evaluation efficiency and accuracy are higher than the existing technology, which overcomes the problem that when the existing method model uses a single remote sensing data to calculate carbon sink estimation data, the single remote sensing data has insufficient coverage and resolution, and is difficult to adapt to the spatial heterogeneity of the generated product area, which leads to deviations in the regional carbon sink estimation results, and it is also difficult to ensure the reliability of the model for regional carbon sink data at different spatial and temporal scales. Analysis.

[0006] The present invention is implemented as follows: a method for calculating the carbon sink value of an ecological product, the method comprising:

[0007] Obtain the coordinate data of the monitoring area, divide the monitoring area into N sub-monitoring areas based on the coordinate data of the monitoring area, and collect multi-source heterogeneous carbon sink monitoring data associated with the sub-monitoring areas in real time according to the preset collection cycle, wherein the carbon sink monitoring data includes basic geographic data, ecological product classification data, multi-spectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, and emission data;

[0008] Load multi-source heterogeneous carbon sink monitoring data, pre-process the multi-source heterogeneous carbon sink monitoring data to obtain a normalized data set, and perform data enhancement on the normalized data set based on the Mosaic-Mixup hybrid method to obtain a data enhancement set;

[0009] Based on the data enhancement set, a carbon sink dynamic database is constructed, and the ecological product classification data, multispectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and basic geographic data in the carbon sink dynamic database are associated, and the data enhancement set index code is automatically generated;

[0010] In response to the ecological product value accounting instruction, a database index instruction is generated based on the ecological product value accounting instruction. The carbon sink dynamic database responds to the database index instruction, outputs a modeling data set, divides the modeling data set into a test set and a training set, and constructs a carbon sink value accounting model in combination with a support vector machine and a self-attention mechanism. The carbon sink value accounting model is iteratively trained using the test set and the training set, and the trained carbon sink value accounting model is output;

[0011] The carbon sink dynamic database is traversed to obtain the preprocessed data enhancement set, and the carbon sink value accounting model is executed with the data enhancement set as input. The carbon sink value accounting model outputs the carbon sink value accounting result.

[0012] The carbon sink value accounting results are loaded and fed back to the carbon sink dynamic database in real time. The carbon sink dynamic database presents the carbon sink value accounting results of the sub-monitoring area in real time based on visualization tools.

[0013] Preferably, the method for obtaining the data enhancement set specifically includes:

[0014] Load multi-source heterogeneous carbon sink monitoring data, perform radiometric calibration and geometric correction preprocessing on multispectral remote sensing data, and perform missing value, difference and rasterization processing on multi-source heterogeneous carbon sink monitoring data based on the ArcGIS platform to obtain standardized data;

[0015] Retrieve historical carbon sink monitoring data, and build and train a multi-source heterogeneous fusion model based on the historical carbon sink monitoring data;

[0016] Obtain standardized data, and obtain a normalized data set by combining a multi-source heterogeneous fusion model with a federated weighted average fusion strategy and a decision fusion algorithm;

[0017] Based on the Mosaic-Mixup hybrid method, the normalized data set is enhanced to obtain a data enhancement set;

[0018] Among them, the principle formula for data enhancement of the normalized data set is:

[0019] (1)

[0020] in, The data augmentation output vector is for the normalized dataset, and is the dataset mixing enhancement coefficient, is the carbon sink factor label corresponding to a batch of normalized data, is the carbon sink factor label corresponding to another batch of normalized data;

[0021] (2)

[0022] in, is the dataset mixing enhancement coefficient, are the mixing pre-coefficient and post-coefficient respectively, and The sum is 1, is the number of data in the normalized dataset.

[0023] Preferably, the method for constructing and training a multi-source heterogeneous fusion model based on historical carbon sink monitoring data specifically includes:

[0024] Construct the initial model of the multi-source heterogeneous fusion model, define the multimodal fusion network and cross-modal aligner, and generate the initial model of the multi-source heterogeneous fusion model with the multimodal fusion network, cross-modal aligner, and graph neural fusion network;

[0025] Among them, the multimodal fusion network performs dimensionality reduction processing on the standardized data based on the federated weighted average fusion strategy, and the dimensionality reduction conversion data information is:

[0026] (3)

[0027] in, Output information for dimensionality reduction transformation of standardized data, is the type of standardized data, is the input standardized data, The conversion standard value is set. is the federal weighting coefficient corresponding to the standardized data;

[0028] (4)

[0029] in, is the federation weight coefficient corresponding to the standardized data, is the data dimension value, is the weight parameter corresponding to the standardized data;

[0030] Multiple sets of dimension reduction transformation data information are input into the cross-modal aligner. The cross-modal aligner introduces an adversarial mechanism and a transfer learning mechanism to achieve feature alignment and mapping learning between multiple sets of dimension reduction transformation data information to obtain a feature alignment set.

[0031] Preferably, the method for constructing and training a multi-source heterogeneous fusion model based on historical carbon sink monitoring data further includes:

[0032] Load the feature alignment set and input the feature alignment set into the graph neural fusion network for iterative training. The fusion convolution layer in the graph neural fusion network uses multiple sets of filters to perform convolution operations on the feature alignment set to extract local significant features of the feature alignment set.

[0033] Among them, the local salient features are expressed as:

[0034] (5)

[0035] is the local significant feature vector, is the activation function of the multi-source heterogeneous fusion model, is the bias term, The size is The filter, where is the dimension value of the data, is the number of filter channels, is the feature alignment set;

[0036] The maximum feature among the local significant features is extracted to obtain the maximum significant feature, and the maximum significant feature is fused and spliced ​​based on the decision fusion algorithm to obtain a normalized data set;

[0037] A historical verification set is randomly selected from the historical carbon sink monitoring data, and the initial model of the multi-source heterogeneous fusion model after training is verified using the historical verification set to determine whether the initial model of the multi-source heterogeneous fusion model meets the preset fusion threshold. If it meets the preset fusion threshold, the multi-source heterogeneous fusion model is output.

[0038] Preferably, the method of constructing a carbon sink value accounting model by combining a support vector machine and a self-attention mechanism specifically includes:

[0039] The PointNet model is used as the initial model of the carbon sink value accounting model. The support vector machine radial kernel function is used to improve the T-Net network in the PointNet model. The SimAM self-attention mechanism is introduced into the multi-layer perceptron in the feature learning network of the PointNet model. The lightweight detection structure Slim-Neck is used to replace the maximum pooling layer of the PointNet model, and the loss function of the carbon sink value accounting model is set.

[0040] Obtain the training set, define the model learning rate and hyperparameters, use the cross entropy loss function to put the training set into the initial model of the carbon sink value accounting model for iterative training, and obtain the improved initial model;

[0041] Based on the carbon sink calculation index, ablation experiments were carried out on the T-Net network, multi-layer perceptron, and lightweight detection structure Slim-Neck in the improved initial model to obtain the initial model after ablation.

[0042] Obtain the test set, input the test set into the initial model after ablation, obtain the predicted value accounting result, calculate whether the real value accounting result and the predicted value accounting result meet the accuracy threshold based on the cross entropy loss function, and output the trained carbon sink value accounting model if they meet the accuracy threshold;

[0043] If the accuracy threshold is not met, the learning rate is reinitialized, the optimization function of the Adam optimizer is used to tune the learning rate and hyperparameters of the initial model, and the cross entropy loss function is used to iteratively train the initial model.

[0044] Preferably, the method of executing the carbon sink value accounting model using the data enhancement set as input specifically includes:

[0045] Obtain a data enhancement set, and extract feature vectors associated with the sub-monitoring areas in the data enhancement set based on the T-Net network;

[0046] The radial kernel function of the vector machine based on the T-Net network maps the feature vector associated with the monitoring area to a high-dimensional space to obtain a high-dimensional vector set;

[0047] The high-dimensional vector set is input into a multi-layer perceptron, which performs convolution and pooling operations on the high-dimensional vector set, extracts the carbon sink calculation index of the high-dimensional vector set based on the index extraction function, and obtains a calculation index set;

[0048] The calculation indicator set is obtained, and the lightweight detection structure Slim-Neck evaluates the ecological product value accounting results in the sub-monitoring area based on the calculation indicator set.

[0049] Preferably, the indicator extraction function is expressed as:

[0050] (6)

[0051] in, The output representation of carbon sink calculation indicators is: is the type of standardized data, is the output vector after convolution pooling of the high-dimensional vector set, Contribution weight to ecological product carbon sink, is the coupling coefficient of the carbon sink value accounting model, The first and second association values ​​of the output vector and the carbon sink contribution are respectively;

[0052] (7)

[0053] in, Contribution weight to ecological product carbon sink, is the type of standardized data, is the output vector after convolution pooling of the high-dimensional vector set, Contribute to the ecological product carbon sink bias term, The initial coefficient of carbon sink contribution of ecological products.

[0054] Preferably, the method for the carbon sink dynamic database to present the carbon sink value accounting results of the sub-monitoring area in real time based on a visualization tool specifically includes:

[0055] Obtain basic geographic data of the sub-monitoring area, carbon sink value accounting results and format conversion processing;

[0056] Performing three-dimensional modeling on the sub-monitoring area using a three-dimensional visualization modeling tool to obtain an initial three-dimensional model;

[0057] Obtain the basic geographic data corresponding to the ecological products in the sub-monitoring area, and deploy carbon sink monitoring nodes based on the carbon sink contribution weights corresponding to the ecological products. The carbon sink monitoring nodes are divided into primary nodes, secondary nodes, tertiary nodes, and quaternary nodes.

[0058] The initial three-dimensional model is edited and optimized based on the deployment of carbon sink monitoring nodes to obtain an optimized visualized three-dimensional model.

[0059] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0060] In an embodiment of the present invention, the method of preprocessing multi-source heterogeneous carbon sink monitoring data and data enhancement of normalized data sets can provide accurate and scientific data support for the carbon sink value accounting of ecological products, thereby ensuring that the carbon sink value accounting model accurately evaluates the carbon sink capacity of regional ecological products. At the same time, the carbon sink value accounting model combines basic geographic data, ecological product classification data, multi-spectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and other data, so that the efficiency and accuracy of carbon sink value accounting result evaluation are higher than the existing technology.

[0061] In an embodiment of the present invention, when obtaining a data enhancement set, a multi-source heterogeneous fusion model combined with a federal weighted average fusion strategy and a decision fusion algorithm are used to effectively fuse multi-source heterogeneous carbon sink monitoring data, so that the data enhancement set can improve the robustness and data processing capabilities of the carbon sink value accounting model. At the same time, the multi-source heterogeneous carbon sink monitoring data can be mixed to generate carbon sink accounting training data with significant characteristics, so that the carbon sink value accounting model can perform robustly in more scenarios.

[0062] In an embodiment of the present invention, a multi-source heterogeneous fusion model and a training method thereof are provided. The multi-source heterogeneous fusion model is composed of a multimodal fusion network, a cross-modal aligner, and a graph neural fusion network. The multi-source heterogeneous fusion model is constructed and trained by considering the differences, uncertainties, and diversity of multi-source heterogeneous data, thereby ensuring the fusion performance of the multi-source heterogeneous fusion model for data, thereby improving the performance of the carbon sink value accounting model, and can effectively improve the efficiency and accuracy of data fusion, providing more accurate and comprehensive data analysis and decision support for the carbon sink value accounting model.

[0063] In an embodiment of the present invention, a carbon sink value accounting model is provided, and the carbon sink value accounting model uses the PointNet model as the initial model. At the same time, the initial model is improved by introducing the radial kernel function of the support vector machine, the SimAM self-attention mechanism, and the lightweight detection structure Slim-Neck, thereby ensuring that the carbon sink value accounting model can more effectively transition from low-level data enhancement set features to high-level data enhancement set features, and realize multi-level integration of carbon sink calculation indicators of high-dimensional vector sets, thereby enhancing the ecological product value accounting accuracy of the carbon sink value accounting model. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic diagram of the implementation process of the ecological product carbon sink value accounting method provided by the present invention.

[0065] Figure 2 It is a schematic diagram of the implementation flow of the method for obtaining a data enhancement set provided by the present invention.

[0066] Figure 3 It is a schematic diagram of the implementation process of the method for constructing and training a multi-source heterogeneous fusion model based on historical carbon sink monitoring data provided by the present invention.

[0067] Figure 4 It is a schematic diagram of the implementation process of the method for constructing a carbon sink value accounting model by combining a support vector machine and a self-attention mechanism provided by the present invention.

[0068] Figure 5 It is a schematic diagram of the implementation flow of the carbon sink value accounting model method provided by the present invention using the data enhancement set as input.

[0069] Figure 6 It is a schematic diagram of the implementation process of a method for presenting carbon sink value accounting results of a sub-monitoring area in real time in a carbon sink dynamic database based on a visualization tool provided by the present invention. DETAILED DESCRIPTION

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0071] When the existing method model uses a single remote sensing data to calculate the carbon sink estimation data, the coverage and resolution of the single remote sensing data are insufficient, and it is difficult to adapt to the spatial heterogeneity of the region to which the generated product belongs, which leads to deviations in the regional carbon sink estimation results. It is also difficult to ensure that the model can perform reliability analysis on regional carbon sink data at different spatial and temporal scales. To address the above problems, we propose an ecological product carbon sink value accounting method. In short, the method includes real-time collection of multi-source heterogeneous carbon sink monitoring data associated with sub-monitoring areas according to a preset collection period, preprocessing the multi-source heterogeneous carbon sink monitoring data to obtain a normalized data set, and enhancing the normalized data set based on the Mosaic-Mixup hybrid method to obtain a data enhancement set. A carbon sink value accounting model is constructed by combining support vector machines and self-attention mechanisms. The data enhancement set is used as input to execute the carbon sink value accounting model, and the carbon sink value accounting model outputs the carbon sink value accounting results. In an embodiment of the present invention, a method of preprocessing multi-source heterogeneous carbon sink monitoring data and enhancing normalized data sets can provide accurate and scientific data support for ecological product carbon sink value accounting, thereby ensuring that the carbon sink value accounting model accurately evaluates the carbon sink capacity of regional ecological products. At the same time, the carbon sink value accounting model combines basic geographic data, ecological product classification data, multispectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and other data, so that the carbon sink value accounting result evaluation efficiency and accuracy are higher than the existing technology, which overcomes the problem that when the existing method model uses a single remote sensing data to calculate carbon sink estimation data, the single remote sensing data has insufficient coverage and resolution, and is difficult to adapt to the spatial heterogeneity of the generated product area, which leads to deviations in the regional carbon sink estimation results, and it is also difficult to ensure the reliability of the model for regional carbon sink data at different spatial and temporal scales. Analysis.

[0072] The embodiment of the present invention provides a method for calculating the carbon sink value of ecological products. Figure 1 The figure shows a schematic diagram of the implementation process of the ecological product carbon sink value accounting method, which specifically includes:

[0073] Step S10, obtaining the coordinate data of the monitoring area, dividing the monitoring area into N sub-monitoring areas based on the coordinate data of the monitoring area, and collecting multi-source heterogeneous carbon sink monitoring data associated with the sub-monitoring areas in real time according to a preset collection period;

[0074] It should be noted that the carbon sink monitoring data include but are not limited to basic geographic data, ecological product classification data, multi-spectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, and emission data;

[0075] Basic geographic data can be obtained through satellite remote sensing, geographic databases and geographic information websites. Basic geographic data include but are not limited to regional types, regional areas, vegetation coverage, vegetation index, ecosystem biomass, digital soil maps, geological maps and soil organic matter content distribution maps. Ecological products include but are not limited to agricultural products (food crops, cotton, hemp, sugar, vegetables, fruits) and forestry products (timber, rubber, rosin, medicinal materials).

[0076] In an embodiment of the present invention, the monitoring area can be divided into N sub-monitoring areas based on the monitoring area coordinate data, and the sub-monitoring areas can be rasterized into N sub-monitoring areas in equal proportion, and the preset collection period can be set to 24 hours, 48 ​​hours, 7 days, 15 days or 30 days.

[0077] Step S20, loading multi-source heterogeneous carbon sink monitoring data, preprocessing the multi-source heterogeneous carbon sink monitoring data to obtain a normalized data set, and performing data enhancement on the normalized data set based on a Mosaic-Mixup hybrid method to obtain a data enhancement set;

[0078] Step S30, constructing a carbon sink dynamic database based on the data enhancement set, associating ecological product classification data, multispectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and basic geographic data in the carbon sink dynamic database, and automatically generating a data enhancement set index code;

[0079] It should be noted that when establishing an association, ArcGIS-Pro or Python scripts can be used to convert the data format, and then ArcGIS can be used to establish an association between the ecological product classification data, multispectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, and emission data in the carbon sink dynamic database and the basic geographic data.

[0080] Step S40, in response to the ecological product value accounting instruction, a database index instruction is generated based on the ecological product value accounting instruction, the carbon sink dynamic database responds to the database index instruction, outputs a modeling data set, divides the modeling data set into a test set and a training set, and constructs a carbon sink value accounting model in combination with a support vector machine and a self-attention mechanism, uses the test set and the training set to iteratively train the carbon sink value accounting model, and outputs the trained carbon sink value accounting model;

[0081] Step S50, traverse the carbon sink dynamic database, obtain the preprocessed data enhancement set, use the data enhancement set as input, execute the carbon sink value accounting model, and the carbon sink value accounting model outputs the carbon sink value accounting result.

[0082] Step S60, loading the carbon sink value accounting results, and feeding them back to the carbon sink dynamic database in real time, and the carbon sink dynamic database presents the carbon sink value accounting results of the sub-monitoring area in real time based on the visualization tool.

[0083] In an embodiment of the present invention, the method of preprocessing multi-source heterogeneous carbon sink monitoring data and data enhancement of normalized data sets can provide accurate and scientific data support for the carbon sink value accounting of ecological products, thereby ensuring that the carbon sink value accounting model accurately evaluates the carbon sink capacity of regional ecological products. At the same time, the carbon sink value accounting model combines basic geographic data, ecological product classification data, multi-spectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and other data, so that the efficiency and accuracy of carbon sink value accounting result evaluation are higher than the existing technology.

[0084] The embodiment of the present invention provides a method for obtaining a data enhancement set. Figure 2 A schematic diagram of the implementation flow of the method for obtaining a data enhancement set is shown. The method for obtaining a data enhancement set specifically includes:

[0085] Step S101, loading multi-source heterogeneous carbon sink monitoring data, performing radiation calibration and geometric correction preprocessing on the multispectral remote sensing data, and performing missing value, difference and rasterization processing on the multi-source heterogeneous carbon sink monitoring data based on the ArcGIS platform to obtain standardized data;

[0086] In this embodiment, standard multispectral remote sensing data can be obtained by performing radiation calibration and geometric correction preprocessing on the multispectral remote sensing data, thereby ensuring the data accuracy of the data enhancement set, reducing errors, and enhancing the reliability and comparability of the data.

[0087] Step S102, retrieving historical carbon sink monitoring data, and building and training a multi-source heterogeneous fusion model based on the historical carbon sink monitoring data;

[0088] Step S103, obtaining standardized data, combining the multi-source heterogeneous fusion model with the federated weighted average fusion strategy and the decision fusion algorithm to obtain a normalized data set;

[0089] Step S104, performing data enhancement on the normalized data set based on the Mosaic-Mixup hybrid method to obtain a data enhancement set.

[0090] In an embodiment of the present invention, when obtaining a data enhancement set, a multi-source heterogeneous fusion model combined with a federal weighted average fusion strategy and a decision fusion algorithm are used to effectively fuse multi-source heterogeneous carbon sink monitoring data, so that the data enhancement set can improve the robustness and data processing capabilities of the carbon sink value accounting model. At the same time, the multi-source heterogeneous carbon sink monitoring data can be mixed to generate carbon sink accounting training data with significant characteristics, so that the carbon sink value accounting model can perform robustly in more scenarios.

[0091] Among them, the principle formula for data enhancement of the normalized data set is:

[0092] (1)

[0093] in, The data augmentation output vector is for the normalized dataset, and is the dataset mixing enhancement coefficient, is a carbon sink factor label corresponding to a batch of normalized data. In this embodiment, the carbon sink factor label includes but is not limited to carbon footprint, carbon emission reduction, carbon neutrality, and visualized carbon factor labels. is the carbon sink factor label corresponding to another batch of normalized data;

[0094] (2)

[0095] in, is the dataset mixing enhancement coefficient, are the mixing pre-coefficient and post-coefficient respectively, and The sum is 1, is the number of data in the normalized dataset.

[0096] The embodiment of the present invention provides a method for constructing and training a multi-source heterogeneous fusion model based on historical carbon sink monitoring data. Figure 3 The schematic diagram of the implementation process of the method for constructing and training a multi-source heterogeneous fusion model based on historical carbon sink monitoring data is shown. The method for constructing and training a multi-source heterogeneous fusion model based on historical carbon sink monitoring data specifically includes:

[0097] Step S1021, constructing an initial model of a multi-source heterogeneous fusion model, defining a multimodal fusion network and a cross-modal aligner, and generating an initial model of the multi-source heterogeneous fusion model with the multimodal fusion network, the cross-modal aligner, and the graph neural fusion network;

[0098] It should be noted that the multimodal fusion network includes a data fusion layer, a data dimension reduction layer, a feature extraction layer, and a feature fusion layer. The data fusion layer is used to upsample and fuse multi-source heterogeneous data. The number of channels in the data fusion layer is 512, and the data dimension reduction layer is a principal component analysis layer. The data dimension reduction layer can be used to reduce the dimension and complexity of the data while retaining key information, thereby simplifying the model and improving computational efficiency. The feature extraction layer consists of an Upsample layer and a Concat layer.

[0099] Among them, the multimodal fusion network performs dimensionality reduction processing on the standardized data based on the federated weighted average fusion strategy, and the dimensionality reduction conversion data information is:

[0100] (3)

[0101] in, Output information for dimensionality reduction transformation of standardized data, is the type of standardized data. In this embodiment, the types of standardized data are 3-20. is the input standardized data, is the conversion standard value set, which can be the variance or mean of the standardized data. is the federal weighting coefficient corresponding to the standardized data;

[0102] (4)

[0103] in, is the federation weight coefficient corresponding to the standardized data, is the data dimension value. In this embodiment, the data dimension value can be 1-5. is the weight parameter corresponding to the standardized data. The weight parameter is a constant and can be 0.1-0.5;

[0104] Step S1022, inputting the multiple sets of dimension reduction transformation data information into the cross-modal aligner, the cross-modal aligner introduces an adversarial mechanism and a transfer learning mechanism to achieve feature alignment and mapping learning between the multiple sets of dimension reduction transformation data information, and obtain a feature alignment set.

[0105] It should be noted that, since features of different modalities may have different dimensions and distributions, feature alignment is required. In this embodiment, this can be achieved through linear transformation, kernel method or mapping layer in deep learning.

[0106] Step S1023, loading the feature alignment set, using the feature alignment set as the training set of the model, inputting the feature alignment set into the graph neural fusion network for iterative training, and the fusion convolution layer in the graph neural fusion network uses multiple sets of filters to perform convolution operations on the feature alignment set to extract local significant features of the feature alignment set;

[0107] Among them, the local salient features are expressed as:

[0108] (5)

[0109] is the local significant feature vector, is the activation function of the multi-source heterogeneous fusion model. In this embodiment, the activation function may be a ReLU activation function. is the bias term, The size is The filter, where is the dimension value of the data, The number of channels for the filter can be 3-16. is the feature alignment set;

[0110] Step S1024, extracting the maximum feature from the local significant features to obtain the maximum significant feature, fusing and splicing the maximum significant feature based on a decision fusion algorithm to obtain a normalized data set;

[0111] Step S1025, randomly selecting a historical verification set from the historical carbon sink monitoring data, and using the historical verification set to verify the initial model of the multi-source heterogeneous fusion model after training;

[0112] Step S1026, determining whether the initial model of the multi-source heterogeneous fusion model meets a preset fusion threshold;

[0113] Step S1027: if the preset fusion threshold is met, a multi-source heterogeneous fusion model is output.

[0114] If it does not meet the preset fusion threshold, return to step S1023.

[0115] In this embodiment, the preset fusion threshold may be 0.8-0.95.

[0116] In an embodiment of the present invention, a multi-source heterogeneous fusion model and a training method thereof are provided. The multi-source heterogeneous fusion model is composed of a multimodal fusion network, a cross-modal aligner, and a graph neural fusion network. The multi-source heterogeneous fusion model is constructed and trained by considering the differences, uncertainties, and diversity of multi-source heterogeneous data, thereby ensuring the fusion performance of the multi-source heterogeneous fusion model for data, thereby improving the performance of the carbon sink value accounting model, and can effectively improve the efficiency and accuracy of data fusion, providing more accurate and comprehensive data analysis and decision support for the carbon sink value accounting model.

[0117] The embodiment of the present invention provides a method for constructing a carbon sink value accounting model by combining a support vector machine and a self-attention mechanism. Figure 4 The schematic diagram of the implementation process of the method for constructing a carbon sink value accounting model by combining a support vector machine and a self-attention mechanism is shown. The method for constructing a carbon sink value accounting model by combining a support vector machine and a self-attention mechanism specifically includes:

[0118] Step S201, using the PointNet model as the initial model of the carbon sink value accounting model, using the support vector machine radial kernel function to improve the T-Net network in the PointNet model, and introducing the SimAM self-attention mechanism into the multi-layer perceptron in the PointNet model feature learning network, and using the lightweight detection structure Slim-Neck to replace the maximum pooling layer of the PointNet model, and setting the loss function of the carbon sink value accounting model;

[0119] It should be noted that when the SimAM self-attention mechanism is introduced into the multi-layer perceptron in the PointNet model feature learning network, the fully connected layer of the multi-layer perceptron can be replaced with a convolutional layer with a convolution kernel size of 3×3, which can reduce the number of model parameters while enhancing the domain classification accuracy.

[0120] Step S202, obtaining a training set, defining a model learning rate and hyperparameters, and using a cross entropy loss function to put the training set into the initial model of the carbon sink value accounting model for iterative training to obtain an improved initial model;

[0121] In this embodiment, the number of iterative training times may be 200-1000 times. During training, the model parameters that minimize the loss in the loss function may be found according to the stochastic gradient descent algorithm.

[0122] Step S203, based on the carbon sink calculation index, an ablation experiment is performed on the T-Net network, the multi-layer perceptron, and the lightweight detection structure Slim-Neck in the improved initial model to obtain the ablated initial model.

[0123] It should be noted that ablation experiments on the T-Net network, multi-layer perceptron, and lightweight detection structure Slim-Neck in the improved initial model based on carbon sink calculation indicators are an important means of verifying the model, so as to accurately evaluate their contribution to the model performance. This not only helps to optimize the model structure, but also improves the understanding of the internal mechanism of the model.

[0124] Step S204, obtaining a test set, inputting the test set into the initial model after ablation, and obtaining a prediction value calculation result;

[0125] Step S205, calculating whether the real value calculation result and the predicted value calculation result meet the accuracy threshold based on the cross entropy loss function;

[0126] Step S206: if the accuracy threshold is met, output the trained carbon sink value calculation model;

[0127] In this embodiment, the accuracy threshold may be set to 0.95-0.98.

[0128] Step S207, if the accuracy threshold is not met, reinitialize the learning rate, use the optimization function of the Adam optimizer to tune the learning rate and hyperparameters of the initial model, return to step S202, and use the cross entropy loss function to iteratively train the initial model.

[0129] In an embodiment of the present invention, a carbon sink value accounting model is provided, and the carbon sink value accounting model uses the PointNet model as the initial model. At the same time, the initial model is improved by introducing the radial kernel function of the support vector machine, the SimAM self-attention mechanism, and the lightweight detection structure Slim-Neck, thereby ensuring that the carbon sink value accounting model can more effectively transition from low-level data enhancement set features to high-level data enhancement set features, and realize multi-level integration of carbon sink calculation indicators of high-dimensional vector sets, thereby enhancing the ecological product value accounting accuracy of the carbon sink value accounting model.

[0130] The embodiment of the present invention provides a method for executing a carbon sink value accounting model using a data enhancement set as input. Figure 5 The schematic diagram of the implementation flow of the method for executing the carbon sink value accounting model with the data enhancement set as input is shown. The method for executing the carbon sink value accounting model with the data enhancement set as input specifically includes:

[0131] Step S301, obtaining a data enhancement set, and extracting feature vectors associated with sub-monitoring areas in the data enhancement set based on a T-Net network;

[0132] Step S302, mapping the feature vector associated with the monitoring area to a high-dimensional space based on the radial kernel function of the vector machine of the T-Net network to obtain a high-dimensional vector set;

[0133] Step S303, inputting the high-dimensional vector set into a multi-layer perceptron, the multi-layer perceptron performs a convolution pooling operation on the high-dimensional vector set, extracting the carbon sink calculation index of the high-dimensional vector set based on the index extraction function, and obtaining a calculation index set;

[0134] Step S304, obtaining a calculation index set, and the lightweight detection structure Slim-Neck evaluates the ecological product value accounting results in the sub-monitoring area based on the calculation index set.

[0135] In this embodiment, the indicator extraction function is expressed as:

[0136] (6)

[0137] in, The output representation of carbon sink calculation indicators is: is the type of standardized data, is the output vector after convolution pooling of the high-dimensional vector set, Contribution weight to ecological product carbon sink, is the coupling coefficient of the carbon sink value accounting model. In this embodiment, the coupling coefficient can be 0.1 or 0.2. are respectively a first correlation value and a second correlation value of the output vector and the carbon sink contribution. It should be noted that the first correlation value may be a carbon sink quantitative index value, and the second correlation value may be a carbon sink quality index value. In this embodiment, the first correlation value and the second correlation value may be constants;

[0138] (7)

[0139] in, Contribution weight to ecological product carbon sink, is the type of standardized data, is the output vector after convolution pooling of the high-dimensional vector set, Contribute to the ecological product carbon sink bias term, The initial coefficient for the carbon sink contribution of ecological products can be set based on the initial value of Xavier and the initial value of He of ecological products.

[0140] The embodiment of the present invention provides a method for a carbon sink dynamic database to present the carbon sink value accounting results of a sub-monitoring area in real time based on a visualization tool. Figure 6 The present invention shows a schematic diagram of the implementation process of the method for presenting the carbon sink value accounting results of the sub-monitoring area in real time based on the visualization tool in the carbon sink dynamic database. The method for presenting the carbon sink value accounting results of the sub-monitoring area in real time based on the visualization tool in the carbon sink dynamic database specifically includes:

[0141] Step S401, obtaining basic geographic data of the sub-monitoring area and carbon sink value calculation results and format conversion processing;

[0142] Step S402, using a three-dimensional visualization modeling tool to perform three-dimensional modeling on the sub-monitoring area to obtain an initial three-dimensional model;

[0143] In this embodiment, the 3D visual modeling tool may be VTK, Glumpy, Mayavi or other tools for 3D visual modeling.

[0144] Step S403, obtaining basic geographic data corresponding to ecological products in the sub-monitoring area, and deploying carbon sink monitoring nodes based on the carbon sink contribution weights corresponding to the ecological products, wherein the carbon sink monitoring nodes are divided into primary nodes, secondary nodes, tertiary nodes and quaternary nodes;

[0145] Step S404: Edit and optimize the initial three-dimensional model based on the deployed carbon sink monitoring nodes to obtain an optimized visualized three-dimensional model.

[0146] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer program instructions, and the computer program instructions can be executed by a processor. When the computer program instructions are executed, the method of any of the above embodiments is implemented.

[0147] Another aspect of the present invention provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0148] The computer-readable storage medium (e.g., memory) herein may be a volatile memory or a nonvolatile memory, or may include both a volatile memory and a nonvolatile memory. As an example and not by way of limitation, a nonvolatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. A volatile memory may include a random access memory (RAM), which may act as an external cache memory. As an example and not by way of limitation, RAM may be obtained in a variety of forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to include, but are not limited to, these and other suitable types of memory.

[0149] To sum up, the present invention provides a method for calculating the carbon sink value of ecological products. In an embodiment of the present invention, the method of preprocessing multi-source heterogeneous carbon sink monitoring data and enhancing the normalized data set can provide accurate and scientific data support for the carbon sink value accounting of ecological products, thereby ensuring that the carbon sink value accounting model accurately evaluates the carbon sink capacity of regional ecological products. At the same time, the carbon sink value accounting model combines basic geographic data, ecological product classification data, multi-spectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and other data, so that the efficiency and accuracy of carbon sink value accounting result evaluation are higher than the existing technology.

[0150] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0151] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A method for calculating the carbon sink value of ecological products, characterized in that: The ecological product carbon sink value accounting method includes: Obtain the coordinate data of the monitoring area, divide the monitoring area into N sub-monitoring areas based on the coordinate data of the monitoring area, and collect multi-source heterogeneous carbon sink monitoring data associated with the sub-monitoring areas in real time according to the preset collection cycle, wherein the carbon sink monitoring data includes basic geographic data, ecological product classification data, multi-spectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, and emission data; Load multi-source heterogeneous carbon sink monitoring data, pre-process the multi-source heterogeneous carbon sink monitoring data to obtain a normalized data set, and perform data enhancement on the normalized data set based on the Mosaic-Mixup hybrid method to obtain a data enhancement set; Based on the data enhancement set, a carbon sink dynamic database is constructed, and the ecological product classification data, multispectral remote sensing data, statistical data, rainfall data, water volume data, soil monitoring data, emission data and basic geographic data in the carbon sink dynamic database are associated, and the data enhancement set index code is automatically generated; In response to the ecological product value accounting instruction, a database index instruction is generated based on the ecological product value accounting instruction. The carbon sink dynamic database responds to the database index instruction, outputs a modeling data set, divides the modeling data set into a test set and a training set, and constructs a carbon sink value accounting model in combination with a support vector machine and a self-attention mechanism. The carbon sink value accounting model is iteratively trained using the test set and the training set, and the trained carbon sink value accounting model is output; Traverse the carbon sink dynamic database, obtain the preprocessed data enhancement set, use the data enhancement set as input, execute the carbon sink value accounting model, and the carbon sink value accounting model outputs the carbon sink value accounting result; The method for constructing and training a multi-source heterogeneous fusion model based on historical carbon sink monitoring data specifically includes: Construct the initial model of the multi-source heterogeneous fusion model, define the multimodal fusion network and cross-modal aligner, and generate the initial model of the multi-source heterogeneous fusion model with the multimodal fusion network, cross-modal aligner, and graph neural fusion network; Among them, the multimodal fusion network performs dimensionality reduction processing on the standardized data based on the federated weighted average fusion strategy, and the dimensionality reduction conversion data information is: （3） in, Output information for dimensionality reduction transformation of standardized data, is the type of standardized data, is the input standardized data, The conversion standard value is set. is the federal weighting coefficient corresponding to the standardized data; （4） in, is the federation weight coefficient corresponding to the standardized data, is the data dimension value, is the weight parameter corresponding to the standardized data; Input multiple sets of dimension reduction transformation data information into the cross-modal aligner. The cross-modal aligner introduces an adversarial mechanism and a transfer learning mechanism to achieve feature alignment and mapping learning between multiple sets of dimension reduction transformation data information, and obtain a feature alignment set. Load the feature alignment set and input the feature alignment set into the graph neural fusion network for iterative training. The fusion convolution layer in the graph neural fusion network uses multiple sets of filters to perform convolution operations on the feature alignment set to extract local significant features of the feature alignment set. Among them, the local salient features are expressed as: （5） is the local significant feature vector, is the activation function of the multi-source heterogeneous fusion model, is the bias term, The size is The filter, where is the dimension value of the data, is the number of filter channels, is the feature alignment set; The maximum feature among the local significant features is extracted to obtain the maximum significant feature, and the maximum significant feature is fused and spliced ​​based on the decision fusion algorithm to obtain a normalized data set; Randomly select a historical verification set from the historical carbon sink monitoring data, use the historical verification set to verify the initial model of the multi-source heterogeneous fusion model after training, and judge whether the initial model of the multi-source heterogeneous fusion model meets the preset fusion threshold. If it meets the preset fusion threshold, output the multi-source heterogeneous fusion model; The method of executing the carbon sink value accounting model using the data enhancement set as input specifically includes: Obtain a data enhancement set, and extract feature vectors associated with the sub-monitoring areas in the data enhancement set based on the T-Net network; The radial kernel function of the vector machine based on the T-Net network maps the feature vector associated with the monitoring area to a high-dimensional space to obtain a high-dimensional vector set; The high-dimensional vector set is input into a multi-layer perceptron, which performs convolution and pooling operations on the high-dimensional vector set, extracts the carbon sink calculation index of the high-dimensional vector set based on the index extraction function, and obtains a calculation index set; Obtain the calculation indicator set, and the lightweight detection structure Slim-Neck evaluates the ecological product value accounting results in the sub-monitoring area based on the calculation indicator set; The indicator extraction function is expressed as: （6） in, The output representation of carbon sink calculation indicators is: is the type of standardized data, is the output vector after convolution pooling of the high-dimensional vector set, Contribution weight to ecological product carbon sink, is the coupling coefficient of the carbon sink value accounting model, The first and second association values ​​of the output vector and the carbon sink contribution are respectively; （7） in, Contribution weight to ecological product carbon sink, is the type of standardized data, is the output vector after convolution pooling of the high-dimensional vector set, Contribute to the ecological product carbon sink bias term, The initial coefficient of carbon sink contribution of ecological products.

2. The method for calculating the carbon sink value of ecological products according to claim 1, characterized in that: The method further comprises: The carbon sink value accounting results are loaded and fed back to the carbon sink dynamic database in real time. The carbon sink dynamic database presents the carbon sink value accounting results of the sub-monitoring area in real time based on visualization tools.

3. The method for calculating the carbon sink value of ecological products according to claim 1, characterized in that: The method for obtaining the data enhancement set specifically includes: Load multi-source heterogeneous carbon sink monitoring data, perform radiometric calibration and geometric correction preprocessing on multispectral remote sensing data, and perform missing value, difference and rasterization processing on multi-source heterogeneous carbon sink monitoring data based on the ArcGIS platform to obtain standardized data; Retrieve historical carbon sink monitoring data, and build and train a multi-source heterogeneous fusion model based on the historical carbon sink monitoring data; Obtain standardized data, and obtain a normalized data set by combining a multi-source heterogeneous fusion model with a federated weighted average fusion strategy and a decision fusion algorithm; Based on the Mosaic-Mixup hybrid method, the normalized data set is enhanced to obtain a data enhancement set; Among them, the principle formula for data enhancement of the normalized data set is: （1） in, The data augmentation output vector is for the normalized dataset, and is the dataset mixing enhancement coefficient, is the carbon sink factor label corresponding to a batch of normalized data, is the carbon sink factor label corresponding to another batch of normalized data; （2） in, is the dataset mixing enhancement coefficient, are the mixing pre-coefficient and post-coefficient respectively, and The sum is 1, is the number of data in the normalized dataset.

4. The method for calculating the carbon sink value of ecological products according to claim 3, characterized in that: The method of constructing a carbon sink value accounting model by combining a support vector machine and a self-attention mechanism specifically includes: The PointNet model is used as the initial model of the carbon sink value accounting model. The support vector machine radial kernel function is used to improve the T-Net network in the PointNet model. The SimAM self-attention mechanism is introduced into the multi-layer perceptron in the feature learning network of the PointNet model. The lightweight detection structure Slim-Neck is used to replace the maximum pooling layer of the PointNet model, and the loss function of the carbon sink value accounting model is set. Obtain the training set, define the model learning rate and hyperparameters, use the cross entropy loss function to put the training set into the initial model of the carbon sink value accounting model for iterative training, and obtain the improved initial model; Based on the carbon sink calculation index, ablation experiments were carried out on the T-Net network, multi-layer perceptron, and lightweight detection structure Slim-Neck in the improved initial model to obtain the initial model after ablation.

5. The method for calculating the carbon sink value of ecological products according to claim 4, characterized in that: The method of building a carbon sink value accounting model by combining a support vector machine and a self-attention mechanism specifically includes: Obtain the test set, input the test set into the initial model after ablation, obtain the predicted value accounting result, calculate whether the real value accounting result and the predicted value accounting result meet the accuracy threshold based on the cross entropy loss function, and output the trained carbon sink value accounting model if they meet the accuracy threshold; If the accuracy threshold is not met, the learning rate is reinitialized, the optimization function of the Adam optimizer is used to tune the learning rate and hyperparameters of the initial model, and the cross entropy loss function is used to iteratively train the initial model.

6. The method for calculating the carbon sink value of ecological products according to claim 2, characterized in that: The method of the carbon sink dynamic database presenting the carbon sink value accounting results of the sub-monitoring area in real time based on visualization tools specifically includes: Obtain basic geographic data and carbon sink value accounting results of the sub-monitoring area and convert them into formats; Performing three-dimensional modeling on the sub-monitoring area using a three-dimensional visualization modeling tool to obtain an initial three-dimensional model; Obtain the basic geographic data corresponding to the ecological products in the sub-monitoring area, and deploy carbon sink monitoring nodes based on the carbon sink contribution weights corresponding to the ecological products. The carbon sink monitoring nodes are divided into primary nodes, secondary nodes, tertiary nodes, and quaternary nodes. The initial three-dimensional model is edited and optimized based on the deployment of carbon sink monitoring nodes to obtain an optimized visualized three-dimensional model.

Images