Vegetation carbon storage prediction method, device, and storage medium

Abstract

The application provides a vegetation carbon storage prediction method and device and a storage medium, wherein the method comprises the following steps: extracting a feature set of a to-be-measured region according to a remote sensing image, an elevation model and historical vegetation index information of the to-be-measured region; inputting the feature set into a random forest vegetation classification model for prediction to obtain at least one vegetation type of the to-be-measured region and regional information of each vegetation type in the remote sensing image; performing segmentation processing on the remote sensing image of the to-be-measured region based on a pre-trained vegetation segmentation model to obtain at least one vegetation proportion information of the to-be-measured region and corresponding regional information of each vegetation proportion information in the remote sensing image; and determining the carbon storage of the to-be-measured region according to the vegetation type of the to-be-measured region, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information and the corresponding regional information of each vegetation proportion information in the remote sensing image. The application effectively improves the accuracy of forest carbon storage estimation.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a method, device and storage medium for predicting vegetation carbon storage. Background Technology

[0002] Forest carbon storage estimation refers to estimating the carbon storage of a forest area based on its vegetation cover. Current methods for forest carbon storage estimation include three main approaches: plot inventory, flux observation, and remote sensing-based modeling.

[0003] However, when estimating carbon storage based on existing technologies, the accuracy of carbon storage estimation is reduced due to two main issues: firstly, the classification and identification of vegetation types in forest areas are not precise enough; secondly, the differences between different forest areas and the lack of complete basic data. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, device, and storage medium for predicting vegetation carbon storage, in order to address the shortcomings of the prior art and solve the problem of low accuracy in carbon storage estimation in the prior art.

[0005] To achieve the above objectives, the technical solution adopted in this application is as follows:

[0006] In a first aspect, this application provides a method for predicting vegetation carbon storage, the method comprising:

[0007] Based on the remote sensing image of the area to be measured, the elevation model of the area to be measured, and the historical vegetation index information of the area to be measured, a feature set of the area to be measured is extracted. The feature set includes: a vegetation feature set, a terrain feature set, and a texture feature set.

[0008] The feature set is input into a random forest vegetation classification model, and the random forest vegetation classification model predicts the feature set to obtain at least one vegetation type of the area to be tested and the regional information of each vegetation type in the remote sensing image.

[0009] The remote sensing image of the area to be tested is segmented based on a pre-trained vegetation segmentation model to obtain at least one vegetation proportion information of the area to be tested and the area information corresponding to each vegetation proportion information in the remote sensing image. The vegetation proportion information is used to characterize the area proportion of a type of vegetation in the remote sensing image.

[0010] The carbon storage of the area to be tested is determined based on the vegetation type of the area to be tested, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information, and the regional information corresponding to each vegetation proportion information in the remote sensing image.

[0011] Optionally, the step of extracting a feature set of the area to be measured based on the remote sensing image of the area to be measured, the elevation model of the area to be measured, and the historical vegetation index information of the area to be measured includes:

[0012] Feature extraction is performed on the historical vegetation index information of the area to be tested to obtain the vegetation feature set, which includes at least one vegetation index feature, and the vegetation index feature is used to point to a vegetation type.

[0013] Feature extraction is performed on the elevation model of the area to be measured to obtain the terrain feature set, which includes at least one terrain feature, and the terrain feature includes at least: slope feature, aspect feature, and elevation feature.

[0014] Feature extraction is performed on the remote sensing image of the area to be tested to obtain the texture feature set, which includes at least one texture feature. The texture feature is used to characterize pixels in the remote sensing image that are homogeneous and correlated.

[0015] Optionally, the random forest vegetation classification model includes multiple decision tree models;

[0016] The process by which the random forest vegetation classification model predicts the feature set to obtain at least one vegetation type in the area to be tested and the regional information of each vegetation type in the remote sensing image includes:

[0017] The features in the feature set are combined to obtain at least one feature combination, and the feature combination is used to indicate a vegetation type.

[0018] The aforementioned feature combinations are input into each decision tree model of the random forest vegetation classification model, and each decision tree model performs decision calculations to obtain at least one vegetation type of the area to be tested and the regional information of each vegetation type in the remote sensing image.

[0019] Optionally, the step of inputting the combinations of the aforementioned features into the decision tree models of the random forest vegetation classification model, and having the decision tree models perform decision calculations to obtain at least one vegetation type of the area to be tested and the regional information of each vegetation type in the remote sensing image, includes:

[0020] The feature combinations are input into each decision tree model of the random forest vegetation classification model, and each decision tree model performs decision calculations based on the preset baseline feature combinations to obtain the predicted vegetation type of each decision tree model.

[0021] The predicted vegetation types of each decision tree model are weighted and calculated to obtain at least one vegetation type for the area to be tested, and the regional information of each vegetation type in the remote sensing image is recorded.

[0022] Optionally, the process of determining the combination of reference features includes:

[0023] Features are extracted from remote sensing images, elevation models, and historical vegetation indices of each sample plot area to obtain a baseline feature set. The sample plot area is a pre-defined sample plot in the area to be measured. The baseline feature set includes: a baseline vegetation feature set, a baseline topographic feature set, and a baseline texture feature set.

[0024] The features in the baseline feature set are combined to obtain at least one initial feature combination, and the initial feature combination is used to indicate a vegetation type.

[0025] The initial feature combinations are input into the random forest vegetation classification model for decision calculation to obtain the baseline feature combinations corresponding to each sample area.

[0026] Optionally, the vegetation segmentation model includes an encoder and a decoder. The encoder includes a dilated convolution module, a channel attention module, a multi-scale feature extraction module, an encoding fusion module, a spatial attention module, and a first convolution module connected in sequence.

[0027] The pre-trained vegetation segmentation model segments the remote sensing image of the area to be tested, obtaining at least one vegetation proportion information of the area to be tested and the corresponding area information of each vegetation proportion information in the remote sensing image, including:

[0028] The remote sensing image is input into the encoder, and the dilated convolution module in the encoder performs feature extraction to obtain the first encoded feature. The channel attention module, multi-scale feature extraction module, encoding fusion module, spatial attention module and the first convolution module in the encoder sequentially extract the first encoded feature to obtain the second encoded feature. The first encoded feature and the second encoded feature are then input into the decoder.

[0029] The decoder performs decoding processing on the first encoded feature and the second encoded feature to obtain at least one vegetation proportion information of the area to be tested and the area information corresponding to each vegetation proportion information in the remote sensing image.

[0030] Optionally, determining the carbon storage of the area to be measured based on the vegetation type of the area to be measured, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information, and the regional information corresponding to each vegetation proportion in the remote sensing image includes:

[0031] Based on the vegetation type of the area to be measured, the area information of each vegetation type in the remote sensing image, the vegetation proportion information, the area information corresponding to each vegetation proportion in the remote sensing image, and the total area of ​​the area to be measured, the area occupied by each vegetation in the area to be measured is determined.

[0032] The baseline carbon density of each quadrat area was determined based on the vegetation type of the quadrat area.

[0033] The carbon storage of the test area is obtained by inputting the area occupied by each vegetation in the test area and the baseline carbon density of each sample plot into a pre-trained carbon storage estimation model.

[0034] Optionally, determining the baseline carbon density of the quadrat area based on its vegetation type includes:

[0035] The biomass corresponding to the vegetation type is determined based on the unit area volume, biomass conversion factor, and root-to-stem ratio.

[0036] The carbon density corresponding to the vegetation type is determined based on the biomass and carbon content corresponding to the vegetation type.

[0037] The baseline carbon density of the quadrat area is determined based on the carbon density corresponding to each vegetation type in the quadrat area.

[0038] Secondly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of a vegetation carbon storage prediction method as described in any one of the first aspects.

[0039] Thirdly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of a vegetation carbon storage prediction method as described in any one of the first aspects.

[0040] The beneficial effects of this application are as follows: By extracting features from remote sensing images, elevation models, and historical vegetation index information of the area under test, a richer feature representation of the area can be obtained from multiple dimensions, thereby improving the classification accuracy of the random forest vegetation classification model. By classifying and identifying the vegetation types of the area under test based on the random forest vegetation classification model, and then segmenting the remote sensing images of the area under test using a vegetation segmentation model, the accuracy of vegetation segmentation and classification of the area under test can be improved, thereby enhancing the accuracy of subsequent carbon storage estimation.

[0041] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 A flowchart of a vegetation carbon storage prediction method provided in an embodiment of this application is shown;

[0044] Figure 2 This document illustrates a flowchart of an embodiment of the present application for extracting a feature set.

[0045] Figure 3 This document illustrates a flowchart of a vegetation classification method provided in an embodiment of this application.

[0046] Figure 4 This document illustrates another flowchart of vegetation classification provided in an embodiment of this application.

[0047] Figure 5 A flowchart illustrating a method for determining a combination of reference features according to an embodiment of this application is shown;

[0048] Figure 6 A schematic diagram of the architecture of a vegetation segmentation model provided in an embodiment of this application is shown;

[0049] Figure 7 This document illustrates a flowchart of a vegetation segmentation method provided in an embodiment of this application.

[0050] Figure 8 This application provides a flowchart for carbon storage estimation according to an embodiment of the present application.

[0051] Figure 9A flowchart illustrating a method for determining a baseline carbon density, as provided in an embodiment of this application, is shown.

[0052] Figure 10 This paper shows a schematic diagram of the structure of a vegetation carbon storage prediction device provided in an embodiment of this application;

[0053] Figure 11 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0055] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0056] In existing technologies, methods for estimating carbon storage in forest areas include plot inventory, flux observation, and remote sensing-based modeling. For large-scale forest areas, carbon storage can be estimated by combining forest resource inventory data with specific vegetation planting conditions.

[0057] However, due to the differences and complexity of forest regions, the basic data of different forest regions are different. Even if the forest regions are all temperate forests, due to differences in vegetation coverage and vegetation type, it is necessary to collect basic data separately for each forest region. Otherwise, the accuracy of carbon storage estimation will still be reduced.

[0058] Furthermore, existing technologies for estimating carbon storage in forest areas suffer from inaccurate classification and identification of vegetation types, which also affects the accuracy of carbon storage estimation.

[0059] Based on this, this application proposes a vegetation carbon storage prediction method. By establishing quadrat areas in forest regions, basic data on each vegetation type in the forest regions can be collected in a targeted manner, thereby improving the accuracy of subsequent carbon storage estimation. By performing image segmentation and vegetation type identification on remote sensing images of the target area based on random forest vegetation classification models and vegetation segmentation models, the accuracy of vegetation identification and segmentation can be improved, thereby improving the accuracy of subsequent carbon storage estimation.

[0060] Next, combine Figure 1 This application describes the vegetation carbon storage prediction method, which can be implemented by electronic devices, such as... Figure 1 As shown, the method includes:

[0061] S101. Based on the remote sensing image of the area to be measured, the elevation model of the area to be measured, and the historical vegetation index information of the area to be measured, the feature set of the area to be measured is extracted. The feature set includes: vegetation feature set, terrain feature set, and texture feature set.

[0062] The area to be tested can be a forest area where vegetation carbon storage prediction is required. The area to be tested includes at least one type of vegetation. The size of the area to be tested is not limited in this application.

[0063] Before feature extraction, the acquired remote sensing images and elevation models can be preprocessed, such as by performing image correction and registration, to ensure that the spatial positions of the remote sensing images and elevation models correspond accurately. Then, features can be extracted and integrated using algorithms to form a feature set.

[0064] In this embodiment, quadrat areas can be pre-defined within the area to be tested. These quadrat areas are pre-defined within the area to be tested, corresponding to different vegetation types. For example, in a forest area to be tested, multiple representative quadrat areas can be selected. Each quadrat area can include one dominant vegetation type, and the shape of each quadrat may be square, rectangular, or other regular shapes. Its size needs to be reasonably determined based on the research objectives and vegetation distribution characteristics, and is used to subsequently obtain various information for vegetation analysis.

[0065] Optionally, remote sensing images can be images reflecting surface information of forest areas acquired by remote sensing sensors carried by satellites, aircraft, or other means. Remote sensing images contain data in different bands, and different vegetation types have different characteristics such as reflectivity in each band. Based on these characteristics, it is helpful to distinguish vegetation types, etc.

[0066] Optionally, the elevation model can be a digital elevation model of the area to be measured, which is a three-dimensional representation of the topographic surface of the forest area. The elevation model can provide topographic information such as altitude, slope, and aspect of the area to be measured. Topographic factors can affect the growth environment of vegetation. For example, the vegetation type and growth status of sunny and shady slopes of mountains may be different. Elevation differences can also lead to changes in climate conditions, which in turn affect vegetation distribution.

[0067] Optionally, vegetation indices can be indicators constructed based on the reflection and absorption characteristics of vegetation in different wavelength bands, such as normalized vegetation index, ratio vegetation index, difference vegetation index, soil-optimized vegetation index, modified anthocyanin emission index, carotenoid response index, enhanced vegetation index, photochemical response index, normalized water index, and green light normalized vegetation index.

[0068] Optionally, the feature set includes a vegetation feature set, a terrain feature set, and a texture feature set. By extracting features from remote sensing images, at least one texture feature can be obtained to form a texture feature set. By extracting features from elevation models, at least one terrain feature can be obtained to form a terrain feature set. By extracting features from historical vegetation index information, at least one vegetation feature can be obtained to form a vegetation feature set.

[0069] Optionally, the vegetation features included in the vegetation feature set can reflect the time-series characteristics of vegetation, characterizing vegetation from aspects such as growth state and phenology. Topographic features can characterize the topographic features of the terrain where various types of vegetation grow, such as slope, altitude, and aspect. Texture features can characterize homogeneous vegetation. It should be understood that the same vegetation type usually has similar texture features, while different vegetation types have different textures. In one possible implementation, texture features can be quantified by calculating relevant indicators such as the pixel gray-level co-occurrence matrix.

[0070] S102. Input the feature set into the random forest vegetation classification model, and use the random forest vegetation classification model to predict the feature set to obtain at least one vegetation type in the area to be tested and the regional information of each vegetation type in the remote sensing image.

[0071] The random forest vegetation classification model is a classification algorithm based on ensemble learning. It consists of multiple decision trees, each trained by randomly sampling and selecting a subset of features from the training dataset. When predicting a new set of features, each decision tree outputs a classification result, which can be vegetation type. The final vegetation type prediction is then obtained by combining the judgments of all decision trees through methods such as majority voting and weighted voting.

[0072] In one possible implementation, features can be extracted in advance for the sample plot area to obtain a feature set of the sample plot area. The feature set of the sample plot area is then filtered based on the random forest vegetation classification model to obtain the optimal feature combination that can represent the sample plot area. Next, in step S102, the random forest vegetation classification model can make a decision on the input feature set based on the feature combination of the vegetation types contained in each sample plot area to obtain the vegetation type of the area to be tested.

[0073] Optionally, the regional information of vegetation type in the remote sensing image can indicate the approximate area of ​​the vegetation type in the remote sensing image. The shape of the area can be a rectangular area formed by the smallest bounding box of the area covered by the vegetation type or other regular-shaped areas. This application does not impose any restrictions here.

[0074] In another possible implementation, the regional information of vegetation type in the remote sensing image can also be the location coordinates of the center of the area where the vegetation type is located, which is used to indicate the location of different vegetation types in the remote sensing image.

[0075] Vegetation type is used to indicate the types of vegetation contained in the area to be tested. The area to be tested may include multiple types of vegetation, that is, there can be multiple types of vegetation.

[0076] Optionally, sample data of a vegetation type and its corresponding features can be collected as a training dataset to train a random forest vegetation classification model. After training, the feature set of the area to be tested is input into the random forest vegetation classification model to obtain the vegetation type of the area to be tested and the location information of each vegetation type in the remote sensing image.

[0077] S103. Based on the pre-trained vegetation segmentation model, the remote sensing image of the area to be tested is segmented to obtain at least one vegetation proportion information of the area to be tested and the corresponding area information of each vegetation proportion information in the remote sensing image. The vegetation proportion information is used to characterize the area proportion of a type of vegetation in the remote sensing image.

[0078] Optionally, a vegetation segmentation model is a pre-trained deep learning model used for semantic segmentation or instance segmentation of remote sensing images. As an example, a vegetation segmentation model can be a segmentation model built on a convolutional neural network (CNN), such as the U-Net architecture, or a segmentation model built on the DeeplabV3+ algorithm. The vegetation segmentation model can learn the spatial and shape features of vegetation in remote sensing images, thereby accurately separating different vegetation individuals or vegetation regions from the entire image.

[0079] Vegetation segmentation models can analyze input remote sensing images of the target area, identify parts of the image that belong to different vegetation, and refine the details to the boundaries of each individual vegetation or vegetation community, so as to obtain the specific location and distribution information of vegetation in the image.

[0080] Optionally, vegetation proportion information can characterize the area proportion of a type of vegetation in a remote sensing image. As one possible implementation, the number of pixels of vegetation in the remote sensing image can be determined, and the proportion of the number of pixels in the total number of pixels can be used as the area proportion of vegetation in the remote sensing image.

[0081] Optionally, the area information corresponding to the vegetation proportion information in the remote sensing image can indicate the location of the segmented vegetation area in the remote sensing image. The area information can be the position coordinates of the edge points of the segmented vegetation area in the remote sensing image, or the position coordinates of the center of the segmented vegetation area in the remote sensing image.

[0082] S104. Determine the carbon storage of the area to be measured based on the vegetation type of the area to be measured, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information, and the regional information corresponding to each vegetation proportion information in the remote sensing image.

[0083] Specifically, by using the regional information of each vegetation type in the remote sensing image and the regional information corresponding to each vegetation proportion in the remote sensing image, the vegetation type corresponding to each vegetation region segmented in step S103 can be determined.

[0084] After determining the vegetation type of each segmented region in the remote sensing image, the biomass of each vegetation type can be determined based on its area proportion in the image. This biomass can then be used as the basis for carbon storage estimation to obtain the carbon storage of the area to be measured. For example, the InVEST model can be used to estimate the carbon storage of the area to be measured.

[0085] In this embodiment, feature extraction is performed on the remote sensing images, elevation models, and historical vegetation index information of the area under test. This allows for richer feature representations of the area from multiple dimensions, thereby improving the classification accuracy of the random forest vegetation classification model. By classifying and identifying the vegetation types in the area under test based on the random forest vegetation classification model, and then segmenting the remote sensing images of the area under test using a vegetation segmentation model, the accuracy of vegetation segmentation and classification in the area under test can be improved, thus enhancing the accuracy of subsequent carbon storage estimation.

[0086] The following is a further explanation of step S101, such as Figure 2 As shown, the steps for extracting the feature set of the area to be measured based on the remote sensing image, elevation model, and historical vegetation index information of the area to be measured include:

[0087] S201. Extract features from the historical vegetation index information of the area to be tested to obtain a vegetation feature set. The vegetation feature set includes at least one vegetation index feature, which is used to point to a vegetation type.

[0088] Optionally, historical vegetation index information includes vegetation indices measured in the past for the area to be measured, including normalized vegetation index, ratio vegetation index, difference vegetation index, soil-optimized vegetation index, modified anthocyanin emission index, carotenoid response index, enhanced vegetation index, photochemical response index, normalized water index, and green light normalized vegetation index.

[0089] By extracting features from historical vegetation index information, at least one vegetation index feature can be obtained. For example, this can be achieved through statistical analysis or time series analysis of the vegetation index. Specifically, this involves calculating statistical quantities such as the average, maximum, and minimum values ​​of the vegetation index, or analyzing the trend of the vegetation index over time, thereby extracting the vegetation index feature.

[0090] Each vegetation index feature is associated with a specific vegetation type. For example, some vegetation types may have higher values ​​for a particular vegetation index, while others may have lower values. By analyzing the vegetation indices, features that indicate different vegetation types can be extracted.

[0091] S202. Extract features from the elevation model of the area to be measured to obtain a set of terrain features. The set of terrain features includes at least one terrain feature, which includes at least: slope feature, aspect feature, and elevation feature.

[0092] Optionally, the elevation model includes the slope, aspect, and elevation value of each location point. By extracting features from the elevation model, slope features, aspect features, and elevation features can be obtained.

[0093] Slope features can help identify steep areas of terrain, aspect features can indicate the orientation of the slope, and elevation features provide information on the elevation of the terrain.

[0094] In one possible implementation, the elevation model can be calculated using terrain analysis algorithms to obtain the terrain features of the remote sensing image.

[0095] S203. Extract features from the remote sensing image of the area to be measured to obtain a texture feature set. The texture feature set includes at least one texture feature, which is used to characterize pixels in the remote sensing image that are homogeneous and correlated.

[0096] Feature extraction from the remote sensing image of the area to be measured can involve calculating the gray-level co-occurrence matrix of each pixel in the remote sensing image and analyzing the texture characteristics of the image. The texture information of the image can reflect the visual characteristics of homogeneous phenomena in the image, and is represented by the gray-level distribution of pixels and their neighbors.

[0097] For example, the gray-level co-occurrence matrix and gray-level difference vector of each pixel can be calculated first from the remote sensing image of the area to be measured. Then, by calculating the homogeneity, correlation, two moments of the gray-level co-occurrence matrix of texture features, and the mean and entropy of the gray-level difference vector, a variety of texture features can be obtained.

[0098] Optionally, texture features are used to describe pixels in a remotely sensed image that exhibit homogeneity and correlation. Homogeneity refers to the relatively uniform pixel values ​​in certain areas of the image, while correlation reflects the spatial relationships between pixels. Texture features help distinguish different types of vegetation.

[0099] like Figure 3 As shown, the process by which the above-mentioned random forest vegetation classification model predicts the feature set to obtain at least one vegetation type in the area to be tested and the regional information of each vegetation type in the remote sensing image includes:

[0100] S301. Combine the features in the feature set to obtain at least one feature combination, and the feature combination is used to indicate a vegetation type.

[0101] Optionally, multiple features can be arbitrarily selected from the feature set and combined to form a feature combination. For example, one feature can be selected from each of the terrain feature set, vegetation feature set, and texture feature set to obtain a feature combination.

[0102] In another possible implementation, each type of vegetation can also be pre-set with at least one corresponding feature combination template. When performing step S301, features can be extracted from the feature set according to the pre-set feature combination template and a feature combination can be formed.

[0103] For example, the Normalized Difference Vegetation Index (NDVI) can be extracted from historical vegetation indices as a vegetation feature, slope can be extracted from an elevation model as a terrain feature, and the contrast of the gray-level co-occurrence matrix can be extracted from a remote sensing image as a texture feature. Then, the three features—NDVI, slope, and contrast—are combined to form a feature set that indicates a specific vegetation type.

[0104] S302. Input the combination of features into each decision tree model of the random forest vegetation classification model, and have each decision tree model perform decision calculations to obtain at least one vegetation type of the area to be tested and the regional information of each vegetation type in the remote sensing image.

[0105] Optionally, the random forest vegetation classification model includes multiple decision tree models. Each decision tree is an independent classifier that can make classification decisions on the input features. Each feature combination formed above is input into a separate decision tree of the random forest model. Each decision tree performs decision calculations based on its own training rules and feature combinations to determine which vegetation type the feature combination belongs to.

[0106] Specifically, in a decision tree, each internal node represents a test for a feature, each branch represents a possible value of the test result, and each leaf node represents a category label (vegetation type). The decision tree tests the input feature combinations starting from the root node until it reaches a leaf node, thus obtaining a prediction result for the vegetation type.

[0107] Suppose we have three feature combinations A, B, and C, which are input into three decision trees of a random forest model. Each decision tree determines the vegetation type based on its own rules. For example, decision tree 1 determines vegetation type X based on feature combination A, decision tree 2 determines vegetation type Y based on feature combination B, and decision tree 3 determines vegetation type Z based on feature combination C. Finally, by combining the results of all decision trees, we can obtain the vegetation type for each feature combination and record the regional information of each vegetation type in the remote sensing image.

[0108] Random Forest is a multi-tree ensemble algorithm based on decision trees. Its implementation process is as follows: First, a new training subset is constructed by randomly selecting several samples from the original training set. At each node split, a subset of features is randomly selected from all input features to construct multiple structurally differentiated decision trees. Second, multiple decision tree models are trained iteratively, allowing each tree to learn on different data and feature subsets, thereby improving the model's generalization ability. Finally, all decision trees are used to independently predict the test samples, and the final classification result is determined through a majority voting mechanism; the category with the most votes is determined as the predicted category of the test sample.

[0109] Based on the above principles, this application first extracts a multidimensional feature set from the original feature data, including vegetation index features, terrain features, and texture features, and uses these feature sets as input variables for the random forest vegetation classification model. Subsequently, fully split decision tree models are trained on multiple feature subsets, and each decision tree learns and classifies on its independently constructed sample subset. After forming multiple decision tree models, the classification results of all trees are integrated through a voting mechanism, and finally the optimal classification category is output to construct the random forest vegetation classification model.

[0110] The following is a further explanation of how the decision tree models of the random forest vegetation classification model, which input the various feature combinations into the model, perform decision calculations to obtain at least one vegetation type in the area under test and the regional information of each vegetation type in the remote sensing image, are further explained. Figure 4 As shown, step S302 above includes:

[0111] S401. Input each feature combination into each decision tree model of the random forest vegetation classification model. Each decision tree model performs decision calculations based on the preset baseline feature combination to obtain the predicted vegetation type of each decision tree model.

[0112] During the model training phase, some baseline feature combinations are pre-defined. These baseline feature combinations can be determined based on known vegetation types and their corresponding features, serving as a reference standard for classification decisions in the decision tree. The baseline feature combinations represent the feature patterns of typical vegetation types.

[0113] Optionally, each decision tree model can compare and analyze the input feature combination with a preset baseline feature combination. Based on the similarity or matching degree between the feature combination and the baseline feature combination, the decision tree model will make a series of judgments and decisions to arrive at a predicted vegetation type.

[0114] S402. Perform weighted calculations based on the predicted vegetation types of each decision tree model to obtain at least one vegetation type for the area to be tested, and record the regional information of each vegetation type in the remote sensing image.

[0115] Each decision tree model calculates its own prediction and outputs a predicted vegetation type. Since a random forest model consists of multiple decision trees, these trees may produce different predictions. To synthesize the predictions from all decision trees, a weighted calculation can be performed. The weighting method may determine the weights based on factors such as the performance of the decision trees (e.g., accuracy during training) and the similarity between the feature combinations and the baseline feature combinations. For example, some decision trees may perform better during training, so their predictions may be given higher weights.

[0116] By integrating the predictions from all decision trees, the vegetation type of the area to be tested can be obtained. Simultaneously, the specific regional information of these vegetation types in the remote sensing image can be recorded. For example, the classification results can be mapped back to the geographical location or pixel location of the remote sensing image to visually display the distribution range of each vegetation type on the image.

[0117] In this embodiment, feature extraction transforms sample data from high-dimensional to low-dimensional, thereby reducing processing complexity while retaining more useful information. Combining features reduces the computational load of the random forest vegetation classification model and improves its classification accuracy, thus enhancing the accuracy of vegetation type identification in the tested area.

[0118] The following is a further explanation of the process for determining the above-mentioned combination of benchmark features, such as... Figure 5 As shown, the process includes:

[0119] S501. Extract features from remote sensing images, elevation models, and historical vegetation indices of each sample plot area to obtain a baseline feature set. The sample plot area is a pre-defined sample plot in the area to be measured. The baseline feature set includes: baseline vegetation feature set, baseline topographic feature set, and baseline texture feature set.

[0120] It is worth noting that when selecting quadrat areas, it is advisable to conduct on-site investigations of the area to be tested or obtain prior knowledge in advance, so as to ensure that the quadrat areas can cover the main vegetation types of the area to be tested.

[0121] Optionally, the baseline vegetation feature set includes at least one baseline vegetation feature, which may be a feature extracted from historical vegetation index information of the quadrat area, such as the average or maximum value of the vegetation index of the quadrat area.

[0122] The set of benchmark topographic features includes at least one benchmark topographic feature, which can be a feature extracted from the elevation model of the sample area, such as slope, aspect, and elevation value.

[0123] The baseline texture feature set includes at least one baseline texture feature, which may be a feature extracted from a remote sensing image of the sample area.

[0124] S502. Combine the features in the baseline feature set to obtain at least one initial feature combination, and the initial feature combination is used to indicate a vegetation type.

[0125] Optionally, features from the baseline feature set can be combined to form initial feature combinations. Each initial feature combination indicates a specific vegetation type. Each vegetation type can correspond to at least one initial feature combination.

[0126] Feature combination can be performed in various ways, such as combining vegetation features, topographic features, and texture features according to certain rules. The combined features can more comprehensively describe the feature patterns of vegetation types.

[0127] S503. Input each initial feature combination into the random forest vegetation classification model for decision calculation to obtain the baseline feature combination corresponding to each sample area.

[0128] Optionally, the initial feature combinations can be input into a random forest vegetation classification model for decision-making. The random forest vegetation classification model can classify the initial feature combinations, obtain the predicted vegetation types, and record the confidence level of each vegetation type.

[0129] After predicting all initial feature combinations, they can be sorted according to their prediction confidence levels, and the top N initial feature combinations are used as the baseline feature combinations. Higher confidence levels indicate that the initial feature combination more accurately identifies the vegetation type of the sample plot area.

[0130] In this embodiment, by extracting features from the sample plot area and filtering feature combinations to obtain a baseline feature combination, a feature combination with better classification effect can be selected as the baseline for subsequent vegetation classification, thereby improving the accuracy of subsequent vegetation classification.

[0131] The vegetation segmentation model in step S103 above will be explained next, such as... Figure 6 The image shown is a schematic diagram of a vegetation segmentation model. Figure 6 The vegetation segmentation model includes an encoder and a decoder. The encoder comprises a dilated convolution module, a channel attention module, a multi-scale feature extraction module, an encoding fusion module, a spatial attention module, and a first convolution module, connected in sequence. The dilated convolution module is... Figure 6 The CAM (Channel Attention Module) and spatial attention modules in the text are... Figure 6 The SAM (Spatial Attention Module) module in the text.

[0132] The decoder includes an upsampling module and a second convolutional module, a fusion module, a third convolutional module, and another upsampling module connected in sequence.

[0133] The output of the dilated convolution can be input into the second convolution module of the decoder, and the final output of the encoder can be input into the fusion module of the decoder.

[0134] like Figure 7 As shown, the process of segmenting the remote sensing image of the test area based on the pre-trained vegetation segmentation model to obtain at least one vegetation proportion information of the test area and the corresponding area information of each vegetation proportion information in the remote sensing image includes:

[0135] S701. Input the remote sensing image into the encoder. The dilated convolution module in the encoder performs feature extraction to obtain the first encoded feature. The channel attention module, multi-scale feature extraction module, encoding fusion module, spatial attention module and the first convolution module in the encoder sequentially extract the first encoded feature to obtain the second encoded feature. The first encoded feature and the second encoded feature are then input into the decoder.

[0136] The dilated convolution module is used for feature extraction. Dilated convolution expands the receptive field by inserting holes into the convolution kernel, enabling the capture of more contextual information without increasing computational cost or parameter count. This helps in better understanding the distribution of vegetation in images.

[0137] Optionally, the first encoded feature can be a feature map obtained after processing by a dilated convolution module, containing preliminarily extracted vegetation-related information.

[0138] Optionally, the channel attention module emphasizes important feature channels and suppresses unimportant channels, improving the model's focus on key information. The multi-scale feature extraction module captures features at different scales, and the encoding fusion module fuses features from different scales. The spatial attention module focuses on spatial location information in the image, strengthening attention to the target region and suppressing interference from irrelevant background regions, thereby helping the model better locate vegetation areas.

[0139] It is worth noting that the above Figure 6 The dilated spatial pyramid pooling module comprises a dilated convolution module, a channel attention module, and a multi-scale feature extraction module. By introducing the dilated spatial pyramid pooling module, multi-scale semantic information is extracted using dilated convolutions at different scales. Without increasing additional computational complexity, it can capture pixel relationships within different receptive field ranges at each scale, significantly improving the model's ability to capture target boundaries and spatial information.

[0140] In one possible implementation, Figure 6 Each convolutional module in the system can be implemented using dilated convolution. By combining dilated convolution to further optimize the receptive field, the receptive field can be effectively expanded under dilated convolution with different hole rates. Especially for the segmentation needs of small areas and insignificant vegetation, it can achieve accurate identification of detailed features in remote sensing images.

[0141] S702, The decoder performs decoding processing on the first and second encoded features to obtain at least one vegetation proportion information of the area to be tested and the corresponding area information of each vegetation proportion information in the remote sensing image.

[0142] Optionally, the decoder can perform operations such as fusing and upsampling the first and second encoded features to recover the spatial details of the image. This process progressively transforms the low-resolution, high-semantic feature map into a high-resolution output that includes vegetation coverage and location information.

[0143] In this embodiment, the vegetation segmentation model can capture rich information from remote sensing images through multiple feature extraction modules in the encoder, and the decoder is used to convert this information into vegetation proportion and regional information that can be used for analysis and decision-making.

[0144] The following is a further explanation of how the carbon storage of the area under test was determined based on the vegetation type, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information, and the corresponding regional information of each vegetation proportion in the remote sensing image. Figure 8 As shown, step S104 above includes:

[0145] S801. Based on the vegetation type of the area to be measured, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information, the regional information corresponding to each vegetation proportion in the remote sensing image, and the total area of ​​the area to be measured, determine the area occupied by each vegetation in the area to be measured.

[0146] Optionally, based on the regional information of each vegetation type in the remote sensing image and the corresponding regional information of each vegetation proportion in the remote sensing image, the vegetation proportion information and the vegetation type can be associated to determine the vegetation type to which the vegetation area indicated by the vegetation proportion information belongs.

[0147] The total area of ​​the area to be measured can be determined based on remote sensing images and elevation models. Vegetation proportion information can describe the area proportion of various types of vegetation in the remote sensing images. Therefore, by multiplying the area proportion of each type of vegetation in the remote sensing images by the total area, the area occupied by each type of vegetation in the area to be measured can be obtained.

[0148] S802. Determine the baseline carbon density of each quadrat area based on the vegetation type of the quadrat area and the carbon density information corresponding to each vegetation type.

[0149] It should be noted that after determining the vegetation type of the quadrat area, each vegetation type has corresponding carbon density information, which includes: volume per unit area, biomass conversion factor, root-to-stem ratio, and carbon content.

[0150] Among these, volume per unit area indicates the biomass volume per hectare or square meter of vegetation, reflecting the growth of that type of vegetation within a specific area. Biomass conversion factor is used to convert vegetation volume into biomass. Root-to-stem ratio refers to the ratio of underground biomass to aboveground biomass. Carbon content refers to the proportion of carbon in vegetation biomass.

[0151] Optionally, the baseline carbon density can be the carbon storage per unit area of ​​the vegetation type contained in the quadrat.

[0152] S803. Input the area occupied by each vegetation in the area to be tested and the baseline carbon density of each quadrat into the pre-trained carbon storage estimation model to obtain the carbon storage of the area to be tested.

[0153] Among them, the carbon storage estimation model can be the InVEST carbon storage estimation model, and the mathematical expression of the InVEST model is as follows (1):

[0154] (1)

[0155] In the above formula, This represents the total carbon storage in the ecosystem. These represent aboveground total biomass carbon storage, belowground total biomass carbon storage, and soil total carbon storage, respectively. Belowground total biomass carbon storage mainly refers to the carbon storage of vegetation with low height beneath the forest vegetation (which is obscured by the forest vegetation and will not be displayed in remote sensing images). Belowground total biomass carbon storage can be estimated from aboveground total biomass carbon storage.

[0156] The formulas for calculating the total carbon storage of aboveground organisms are as follows: (2) (3):

[0157] (2)

[0158] (3)

[0159] in, Carbon density of a region divided by a certain type of vegetation. This represents the total area of ​​a segmented region for a specific type of vegetation within the test area. After obtaining the total area, the area of ​​the segmented region for a specific type of vegetation within the test area can be obtained through remote sensing image classification. The carbon storage in the test area represents the carbon storage of a specific type of vegetation segment. This refers to the total aboveground biomass carbon storage of all vegetation in all specific regions.

[0160] The formula for calculating the total underground biological carbon storage is as follows (4):

[0161] (4)

[0162] in, This indicates the carbon storage in the forest understory. The carbon conversion coefficient of forest understory vegetation can be 0.195.

[0163] The formula for calculating total soil carbon storage is as follows (5):

[0164] (5)

[0165] in, This indicates the total carbon storage in the soil. Indicates soil carbon density, This represents the area covered by vegetation in the area to be measured.

[0166] The following is a further explanation of the determination of the baseline carbon density of the quadrat area based on the vegetation type of the quadrat area, such as... Figure 9 As shown, the above S802 step includes:

[0167] S901. Determine the biomass corresponding to the vegetation type based on the unit area volume, biomass conversion factor, and root-to-stem ratio.

[0168] Alternatively, biomass can be calculated using the following formula (6):

[0169] (6)

[0170] Wherein, B is the total biomass (Mg) corresponding to the vegetation type; V is the volume per unit area corresponding to the vegetation type; BEF is the biomass transformation factor corresponding to the vegetation type; R is the root-to-stem ratio; and D is the timber density corresponding to the vegetation type.

[0171] S902. Determine the carbon density corresponding to the vegetation type based on the biomass and carbon content corresponding to the vegetation type.

[0172] Alternatively, the carbon density can be calculated using the following formula (7):

[0173] (7)

[0174] in, Carbon density of a certain tree species in a forest ( ); Carbon content of a particular tree species; Given the biomass of a particular tree species, the carbon density of each tree species in the tested area can be calculated using the above formula.

[0175] S903. Determine the baseline carbon density of the quadrat area based on the carbon density corresponding to each vegetation type in the quadrat area.

[0176] Optionally, the baseline carbon density includes the average carbon density of the sample plot area and the soil carbon density.

[0177] Optionally, if the area to be tested contains multiple types of vegetation, the vegetation types with a proportion greater than a preset threshold can be selected, and the average carbon density of these vegetation types can be calculated. The average carbon density can be used as the carbon density of the sample plot area. The average carbon density can be calculated as shown in the following formula (8):

[0178] (8)

[0179] in, These are the vegetation types whose proportion in each vegetation segmentation area exceeds a preset threshold. denoted as the average carbon density of the quadrat area, and n as the number of vegetation types.

[0180] The soil carbon density of the sample plot area can be calculated using the following formula (9):

[0181] (9)

[0182] in, G represents soil carbon density, H represents soil bulk density, and FC represents the percentage of gravel larger than 2 mm in diameter in the top 30 cm soil layer. This indicates the soil organic carbon content.

[0183] Based on the same inventive concept, this application also provides a vegetation carbon storage prediction device corresponding to the vegetation carbon storage prediction method. Since the principle of the device in this application is similar to the vegetation carbon storage prediction method described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0184] Figure 10 A schematic diagram of the structure of a vegetation carbon storage prediction device provided in an embodiment of this application is shown.

[0185] The extraction module 1001 is used to extract the feature set of the area to be measured based on the remote sensing image of the area to be measured, the elevation model of the area to be measured, and the historical vegetation index information of the area to be measured. The feature set includes: vegetation feature set, terrain feature set, and texture feature set.

[0186] Prediction module 1002 is used to input the feature set into the random forest vegetation classification model, and the random forest vegetation classification model predicts the feature set to obtain at least one vegetation type of the area to be tested and the regional information of each vegetation type in the remote sensing image.

[0187] The segmentation module 1003 is used to segment the remote sensing image of the area to be tested based on a pre-trained vegetation segmentation model, and to obtain at least one vegetation proportion information of the area to be tested and the corresponding area information of each vegetation proportion information in the remote sensing image. The vegetation proportion information is used to characterize the area proportion of a type of vegetation in the remote sensing image.

[0188] The determination module 1004 is used to determine the carbon storage of the area to be measured based on the vegetation type of the area to be measured, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information, and the regional information corresponding to each vegetation proportion in the remote sensing image.

[0189] Optionally, the extraction module 1001 is also used for:

[0190] Feature extraction is performed on the historical vegetation index information of the area to be tested to obtain a vegetation feature set. The vegetation feature set includes at least one vegetation index feature, which is used to point to a vegetation type.

[0191] Feature extraction is performed on the elevation model of the area to be measured to obtain a set of terrain features. The set of terrain features includes at least one terrain feature, which includes at least: slope feature, aspect feature, and elevation feature.

[0192] Feature extraction is performed on the remote sensing image of the area to be measured to obtain a texture feature set. The texture feature set includes at least one texture feature, which is used to characterize pixels in the remote sensing image that are homogeneous and correlated.

[0193] Optionally, the random forest vegetation classification model includes multiple decision tree models;

[0194] Optionally, the prediction module 1002 is also used for:

[0195] The features in the feature set are combined to obtain at least one feature combination, and each feature combination is used to indicate a vegetation type;

[0196] The various feature combinations are input into the decision tree models of the random forest vegetation classification model. Each decision tree model performs decision calculations to obtain at least one vegetation type in the area to be tested and the regional information of each vegetation type in the remote sensing image.

[0197] Optionally, the prediction module 1002 is also used for:

[0198] Each feature combination is input into each decision tree model of the random forest vegetation classification model. Each decision tree model performs decision calculations based on the preset baseline feature combination to obtain the predicted vegetation type of each decision tree model.

[0199] The predicted vegetation types of each decision tree model are weighted and calculated to obtain at least one vegetation type for the area to be tested, and the regional information of each vegetation type in the remote sensing image is recorded.

[0200] Optionally, the prediction module 1002 is also used for:

[0201] Features were extracted from remote sensing images, elevation models, and historical vegetation indices of each sample plot area to obtain a baseline feature set. The sample plot area is a pre-defined sample plot in the area to be measured. The baseline feature set includes: baseline vegetation feature set, baseline topographic feature set, and baseline texture feature set.

[0202] The features in the baseline feature set are combined to obtain at least one initial feature combination, and the initial feature combination is used to indicate a vegetation type.

[0203] The initial feature combinations are input into the random forest vegetation classification model for decision calculation, and the baseline feature combinations corresponding to each sample area are obtained.

[0204] Optionally, the vegetation segmentation model includes an encoder and a decoder. The encoder includes a dilated convolution module, a channel attention module, a multi-scale feature extraction module, an encoding fusion module, a spatial attention module, and a first convolution module connected in sequence.

[0205] Optionally, the segmentation module 1003 is also used for:

[0206] The remote sensing image is input into the encoder, where the dilated convolution module in the encoder performs feature extraction to obtain the first encoded feature. The channel attention module, multi-scale feature extraction module, encoding fusion module, spatial attention module and the first convolution module in the encoder sequentially extract the first encoded feature to obtain the second encoded feature. The first and second encoded features are then input into the decoder.

[0207] The decoder decodes the first and second encoded features to obtain at least one vegetation proportion information of the area to be tested and the corresponding area information of each vegetation proportion information in the remote sensing image.

[0208] Optionally, the determining module 1004 is also used for:

[0209] Based on the vegetation type of the area to be measured, the regional information of each vegetation type in the remote sensing image, the vegetation proportion information, the regional information corresponding to each vegetation proportion in the remote sensing image, and the total area of ​​the area to be measured, determine the area occupied by each vegetation in the area to be measured.

[0210] The baseline carbon density of each quadrat area was determined based on the vegetation type of the quadrat area.

[0211] The carbon storage of the area under test is obtained by inputting the area occupied by each vegetation in the test area and the baseline carbon density of each quadrat into a pre-trained carbon storage estimation model.

[0212] Optionally, the determining module 1004 is also used for:

[0213] The biomass corresponding to a vegetation type is determined based on the unit area volume, biomass conversion factor, and root-to-stem ratio.

[0214] The carbon density corresponding to a vegetation type is determined based on the biomass and carbon content of that vegetation type.

[0215] The baseline carbon density of the quadrat area is determined based on the carbon density corresponding to each vegetation type in the quadrat area.

[0216] In this embodiment, feature extraction is performed on remote sensing images, elevation models, and historical vegetation index information of the area under test. This allows for richer feature representations of the area from multiple dimensions, thereby improving the classification accuracy of the random forest vegetation classification model. By classifying and identifying the vegetation types in the area under test based on the random forest vegetation classification model, and then segmenting the remote sensing images of the area under test using a vegetation segmentation model, the accuracy of vegetation segmentation and classification in the area under test can be improved, thereby enhancing the accuracy of subsequent carbon storage estimation.

[0217] Figure 11 A schematic diagram of an electronic device provided in an embodiment of this application is shown, including: a processor 1101, a storage medium 1102, and a bus 1103. The storage medium 1102 stores machine-readable instructions executable by the processor 1101. When the electronic device runs a vegetation carbon storage prediction method as described in the embodiment, the processor 1101 communicates with the storage medium 1102 via the bus 1103. The processor 1101 executes the machine-readable instructions. The preamble of the method item of the processor 1101 executes the steps in the above-described vegetation carbon storage prediction method.

[0218] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor, which performs the steps in the above-described vegetation carbon storage prediction method.

[0219] In this embodiment, the computer program, when run by the processor, can also execute other machine-readable instructions to perform other methods as described in the embodiments. For details on the specific execution steps and principles, please refer to the description of the embodiments, which will not be repeated here.

[0220] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0221] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0222] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0223] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0224] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0225] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method of vegetation carbon stock prediction, characterized by, The method comprises the following steps: According to the remote sensing image of the to-be-measured region, the elevation model of the to-be-measured region, and the historical vegetation index information of the to-be-measured region, a feature set of the to-be-measured region is extracted, which comprises a vegetation feature set, a terrain feature set, and a texture feature set; The feature set is input into a random forest vegetation classification model, and the feature set is predicted by the random forest vegetation classification model to obtain at least one vegetation type of the to-be-measured region and the region information of each vegetation type in the remote sensing image; Based on a pre-trained vegetation segmentation model, the remote sensing image of the to-be-measured region is segmented to obtain at least one vegetation proportion information of the to-be-measured region and the corresponding region information of each vegetation proportion information in the remote sensing image, wherein the vegetation proportion information respectively represents the area proportion of one vegetation in the remote sensing image; According to the vegetation type of the to-be-measured region, the region information of each vegetation type in the remote sensing image, the vegetation proportion information, and the corresponding region information of each vegetation proportion information in the remote sensing image, the carbon storage of the to-be-measured region is determined; The determination of the carbon storage of the to-be-measured region according to the vegetation type of the to-be-measured region, the region information of each vegetation type in the remote sensing image, the vegetation proportion information, and the corresponding region information of each vegetation proportion information in the remote sensing image comprises: According to the vegetation type of the to-be-measured region, the region information of each vegetation type in the remote sensing image, the vegetation proportion information, the corresponding region information of each vegetation proportion information in the remote sensing image, and the total area of the to-be-measured region, the land area of each vegetation in the to-be-measured region is determined; According to the vegetation type of the sample region, the reference carbon density of each sample region is determined; The land area of each vegetation in the to-be-measured region and the reference carbon density of each sample region are input into a pre-trained carbon storage estimation model to obtain the carbon storage of the to-be-measured region.

2. The method of claim 1, wherein, The extraction of the feature set of the to-be-measured region according to the remote sensing image of the to-be-measured region, the elevation model of the to-be-measured region, and the historical vegetation index information of the to-be-measured region comprises: Feature extraction is performed on the historical vegetation index information of the to-be-measured region to obtain the vegetation feature set, wherein at least one vegetation index feature is included in the vegetation feature set, and the vegetation index feature is used to point to one vegetation type; Feature extraction is performed on the elevation model of the to-be-measured region to obtain the terrain feature set, wherein at least one terrain feature is included in the terrain feature set, and the terrain feature at least comprises a slope feature, a slope direction feature, and an elevation feature; Feature extraction is performed on the remote sensing image of the to-be-measured region to obtain the texture feature set, wherein at least one texture feature is included in the texture feature set, and the texture feature is used to represent the pixel points with homogeneity and correlation in the remote sensing image.

3. The method of claim 1, wherein, The random forest vegetation classification model comprises a plurality of decision tree models; The process in which the random forest vegetation classification model predicts the feature set to obtain at least one vegetation type of the to-be-detected region and region information of each vegetation type in the remote sensing image comprises: combining features in the feature set to obtain at least one feature combination, the feature combination being used to indicate one vegetation type respectively; inputting each feature combination into each decision tree model of the random forest vegetation classification model, and performing decision calculation by each decision tree model to obtain at least one vegetation type of the to-be-detected region and region information of each vegetation type in the remote sensing image.

4. The method of claim 3, wherein, The process in which the random forest vegetation classification model predicts the feature set to obtain at least one vegetation type of the to-be-detected region and region information of each vegetation type in the remote sensing image comprises: inputting each feature combination into each decision tree model of the random forest vegetation classification model, and performing decision calculation by each decision tree model based on a preset reference feature combination to obtain a predicted vegetation type of each decision tree model; performing weighted calculation according to the predicted vegetation type of each decision tree model to obtain at least one vegetation type of the to-be-detected region, and recording region information of each vegetation type in the remote sensing image.

5. The method of claim 4, wherein, The process of determining the reference feature combination comprises: extracting features from remote sensing images, elevation models and historical vegetation indexes of each sample region to obtain a reference feature set, the sample region being a sample plot pre-set in the to-be-detected region, the reference feature set comprising a reference vegetation feature set, a reference terrain feature set and a reference texture feature set; combining features in the reference feature set to obtain at least one initial feature combination, the initial feature combination being used to indicate one vegetation type respectively; inputting each initial feature combination into the random forest vegetation classification model to perform decision calculation to obtain a reference feature combination corresponding to each sample region.

6. The method of claim 1, wherein, The vegetation segmentation model comprises an encoder and a decoder, the encoder comprising, in sequence, a cavity convolution module, a channel attention module, a multi-scale feature extraction module, an encoding fusion module, a spatial attention module and a first convolution module; The process in which the pre-trained vegetation segmentation model performs segmentation processing on the remote sensing image of the to-be-detected region to obtain at least one vegetation proportion information of the to-be-detected region and region information corresponding to each vegetation proportion information in the remote sensing image comprises: inputting the remote sensing image into the encoder, extracting features by the cavity convolution module in the encoder to obtain first encoding features, sequentially extracting features from the first encoding features by the channel attention module, the multi-scale feature extraction module, the encoding fusion module, the spatial attention module and the first convolution module in the encoder to obtain second encoding features, and inputting the first encoding features and the second encoding features into the decoder; The decoder decodes the first encoded feature and the second encoded feature to obtain at least one vegetation coverage information of the to-be-measured region and region information corresponding to each vegetation coverage information in the remote sensing image.

7. The method of claim 1, wherein, The determining the reference carbon density of the sample plot region according to the vegetation type of the sample plot region comprises: determining the biomass corresponding to the vegetation type according to the unit area stock volume corresponding to the vegetation type, the biomass conversion factor and the root stem ratio; determining the carbon density corresponding to the vegetation type according to the biomass corresponding to the vegetation type and the carbon content rate corresponding to the vegetation type; determining the reference carbon density of the sample plot region according to the carbon density corresponding to each vegetation type in the sample plot region.

8. An electronic device, comprising: comprise: a processor, a storage medium and a bus, the storage medium storing machine readable instructions executable by the processor, when the electronic device is running, the processor and the storage medium communicate through the bus, the processor executes the machine readable instructions to perform the steps of the vegetation carbon storage prediction method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program performs the steps of the vegetation carbon storage prediction method according to any one of claims 1 to 7 when the processor runs.

Images