Forest aboveground biomass estimation method based on transfer learning
Through transfer learning technology, a forest AGB migration regression estimation model was constructed, which solved the poor universality and insufficient data of forest biomass estimation model when applied in different regions, and achieved efficient and accurate forest biomass prediction.
Patent Information
- Application Number
- CN202411376071.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-01-14
AI Technical Summary
The prior art has poor universality in forest biomass estimation, making it difficult to effectively apply in different regions, especially in the case of insufficient sample data and different data distribution, resulting in low prediction accuracy and difficult data acquisition.
Using a transfer learning method, through data preprocessing, vegetation index calculation and machine learning model training, combined with Two-stage TrAdaBoost.R2 transfer learning algorithm, a forest AGB migration regression estimation model is constructed to realize the regional migration and generalization capabilities of the model. promote.
It improves the universality and accuracy of forest biomass estimation models, can accurately predict in small samples, alleviate the problem of insufficient data, and enhances the applicability and efficiency of the model.
Smart Images
Figure CN119314047A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of machine learning, and in particular to a method for estimating forest aboveground biomass based on transfer learning. Background Art
[0002] At present, there are many models for estimating forest biomass, and the established forest AGB estimation models mainly include traditional statistical regression models and machine learning models. However, existing research mainly focuses on modeling and estimating forest biomass in a selected area. Due to the complex structure of forests, the estimation models established based on different regions contain corresponding experience and knowledge, and there are also differences in model structure and parameters, and the model has poor universality. At the same time, in remote or harsh environmental areas, it is difficult to obtain biomass in forest plots, and there are often problems of insufficient sample data and low quality, resulting in generally low estimation accuracy of models built based on small sample data.
[0003] Most of the machine learning methods currently used to estimate forest biomass are based on statistical learning, which requires that the training data and test data have the same distribution. Once the distribution changes, the training data must be collected again to build the model. The heterogeneity of forest structure means that the data collected in different regions have their own characteristics, which makes it difficult to meet the data distribution requirements. Therefore, when the model established based on a certain region using traditional machine learning methods is applied to other regions, it is prone to problems such as low prediction accuracy and model failure. In the new research area, a large number of tree samples need to be collected again for modeling and prediction, which is not only time-consuming and labor-intensive, but also reduces the utilization rate of existing samples. How to apply the biomass estimation model of one region to other regions and achieve low-cost and high-efficiency construction of forest biomass estimation models has become a problem that remains to be solved.
[0004] Therefore, it is an urgent problem for technicians in this field to propose a forest aboveground biomass estimation method based on transfer learning to solve the difficulties existing in the existing technology. Summary of the invention
[0005] In view of this, the present invention provides a forest aboveground biomass estimation method based on transfer learning, which is conducive to solving the problem of different data distribution, thereby realizing the regional migration of the forest biomass estimation model within a certain accuracy range, enhancing the generalization ability of the existing biomass model, and further realizing the accurate prediction of forest biomass with few samples, thereby improving the universality of the forest AGB estimation model.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] A method for estimating forest aboveground biomass based on transfer learning comprises the following steps:
[0008] S1. Data collection: Collect the measured data and auxiliary factor data of the sample site;
[0009] S2. Data preprocessing: preprocess the collected measured data and auxiliary factor data to obtain preprocessed measured data and auxiliary factor data, and use the multi-value extraction to point method in ArcGIS to extract the preprocessed measured data and auxiliary factor data to the sample point according to the central coordinates of the sample site to form a characteristic factor set;
[0010] S3. Calculation of vegetation index: Based on the auxiliary factor data obtained by preprocessing, extract the characteristic variables of remote sensing factors, extract the reflectance of the red light band, the reflectance of the blue light band, and the reflectance of the near-infrared band to calculate various vegetation index data, and add the various vegetation index data to the characteristic factor set;
[0011] S4. Base learner training: The aboveground biomass per unit area of the plot is used as the dependent variable, and the data in the characteristic factor set is used as the independent variable. The dependent variable and the independent variable are used to train the base learner composed of four machine learning methods: K nearest neighbor, support vector regression, random forest and extreme random tree;
[0012] S5. Model construction: Four forest AGB migration regression estimation models were constructed using the base learner combined with the Two-stage TrAdaBoost.R2 migration learning algorithm;
[0013] S6. Forest AGB estimation: Based on the four constructed forest AGB migration regression estimation models, the estimated value of forest AGB in the entire study area of the target domain was obtained.
[0014] Optionally, the auxiliary factor data in S1 include remote sensing data, meteorological data and terrain data; wherein the remote sensing image data is obtained based on Landsat 8 images; the meteorological data include the temperature, precipitation, wind speed and surface runoff of the study area; and the terrain data include slope, aspect and altitude.
[0015] Optionally, based on the preprocessed Landsat8 multispectral remote sensing image, S3 extracts the reflectance of the red band, the reflectance of the blue band, and the reflectance of the near-infrared band to calculate the normalized vegetation index, difference vegetation index, ratio vegetation index, and enhanced vegetation index. The calculation formula is:
[0016]
[0017] DVI=NIR-Red
[0018]
[0019]
[0020] Among them, NDVI is the normalized vegetation index, DVI is the difference vegetation index, RVI is the ratio vegetation index, EVI is the enhanced vegetation index, NIR is the reflectance of the near-infrared band, Red is the reflectance of the red light band, and BLUE is the reflectance of the blue light band.
[0021] Optionally, in S4, the corresponding biomass allometric growth equation is selected according to the tree species, and the aboveground biomass of each tree is calculated. The ratio of the sum of the AGB of all trees in the sample plot to the sample plot area is the aboveground biomass per unit area of the sample plot, and the aboveground biomass per unit area of the sample plot is used as the dependent variable;
[0022] In Python, the random forest feature importance ranking is used to sort the data in the feature factor set, and the model is built by adding the variables after positive sorting one by one, and the R value of each time is calculated. 2 Determine the best feature set, and use the best feature set as the independent variable.
[0023] Optionally, the specific contents of using dependent variables and independent variables to train base learners composed of four machine learning methods, namely, K nearest neighbor, support vector regression, random forest and extreme random tree, are as follows:
[0024] Firstly, the independent variables and dependent variables were randomly divided into training set and test set in a ratio of 7:3. The training set was used to construct and optimize the parameters of four forest AGB regression estimation models, and the test set was used to verify the accuracy of the four forest AGB regression estimation models.
[0025] Then, four machine learning algorithms, namely K nearest neighbor, support vector regression, random forest and extreme random tree, were used to train four forest AGB regression estimation models, and the prediction accuracy of the four forest AGB regression estimation models was continuously improved by adjusting parameters.
[0026] Optionally, in S5, the specific steps of using the base learner combined with the Two-stage TrAdaBoost.R2 transfer learning algorithm to construct four forest AGB migration regression estimation models are as follows:
[0027] S51. The target domain data is randomly divided into a test set and a training set in a ratio of 7:3, and all the source domain data is used as the training set. The test set of the target domain and all the source domain data are combined into the training set of the migration model, and the test set of the target domain is used as the test set of the migration model;
[0028] S52. Perform simulation training on the training set of the migration model through the base learner, and use the Two-stageTrAdaBoost.R2 algorithm to adjust the sample weights, reduce the weights of the source domain data with large errors, and increase the weights of the target domain data;
[0029] S53. Freeze the weights of all source domain samples, continue to use the base learner to train the target domain data, further update the weights of the target domain samples, and obtain the final four forest AGB migration regression estimation models.
[0030] Optionally, it also includes using the test set to evaluate the final four forest AGB migration regression estimation models. Specifically, the root mean square error and determination coefficient of the four forest AGB migration regression estimation models are calculated to evaluate the accuracy and fitting effect of the model, thereby obtaining the evaluation results of the four forest AGB migration regression estimation models.
[0031] Optionally, the root mean square error and determination coefficient of the four forest AGB migration regression estimation models are calculated, wherein the larger the determination coefficient value and the smaller the root mean square error value, the better the model fitting effect, specifically:
[0032]
[0033]
[0034] Among them, RMSE is the root mean square error, R 2 is the coefficient of determination, y i is the measured value of biomass in the sample plot, is the model-estimated value of the biomass of the sample plot, is the mean of the measured biomass of the sample plots, and n is the number of samples.
[0035] It can be seen from the above technical solutions that, compared with the prior art, the present invention provides a forest aboveground biomass estimation method based on transfer learning, which has the following beneficial effects:
[0036] (1) The transfer learning adopted by the present invention can transfer the knowledge in the source domain task to the target domain task, allowing the target domain to learn the knowledge in the source domain and solve the problem of different data distribution, thereby realizing the regional migration of the forest biomass estimation model within a certain accuracy range, enhancing the generalization ability of the existing biomass model, and improving the universality of the forest AGB estimation model;
[0037] (2) The Two-stage TrAdaBoost.R2 transfer learning algorithm can be used to migrate existing, high-quality source domain data to the target domain, thereby alleviating the problem of insufficient data and improving the accuracy of model estimation under small sample conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0039] Figure 1 A flow chart of a forest aboveground biomass estimation method based on transfer learning provided by the present invention. DETAILED DESCRIPTION
[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] Reference Figure 1 As shown, the present invention discloses a method for estimating forest aboveground biomass based on transfer learning, comprising the following steps:
[0042] S1. Data collection: Collect the measured data and auxiliary factor data of the sample site;
[0043] S2. Data preprocessing: preprocess the collected measured data and auxiliary factor data to obtain preprocessed measured data and auxiliary factor data, and use the multi-value extraction to point method in ArcGIS to extract the preprocessed measured data and auxiliary factor data to the sample point according to the central coordinates of the sample site to form a characteristic factor set;
[0044] S3. Calculation of vegetation index: Based on the auxiliary factor data obtained by preprocessing, extract the characteristic variables of remote sensing factors, extract the reflectance of the red light band, the reflectance of the blue light band, and the reflectance of the near-infrared band to calculate various vegetation index data, and add the various vegetation index data to the characteristic factor set;
[0045] S4. Base learner training: The aboveground biomass per unit area of the plot is used as the dependent variable, and the data in the characteristic factor set is used as the independent variable. The dependent variable and the independent variable are used to train the base learner composed of four machine learning methods: K nearest neighbor, support vector regression, random forest and extreme random tree;
[0046] S5. Model construction: Four forest AGB migration regression estimation models were constructed using the base learner combined with the Two-stage TrAdaBoost.R2 migration learning algorithm;
[0047] S6. Forest AGB estimation: Based on the four constructed forest AGB migration regression estimation models, the estimated value of forest AGB in the entire study area of the target domain was obtained.
[0048] Furthermore, the auxiliary factor data in S1 include remote sensing data, meteorological data and terrain data; among them, the remote sensing image data is obtained based on Landsat 8 images; the meteorological data include temperature, precipitation, wind speed and surface runoff in the study area; the terrain data include slope, aspect and altitude.
[0049] Specifically, sample plot measured data: select multiple sample plots in the forest areas of the source domain and the target domain respectively, measure the tree height and breast diameter data of each tree in the sample plot, and record the GPS coordinates of the center point of the sample plot.
[0050] Furthermore, based on the preprocessed Landsat8 multispectral remote sensing image, S3 extracts the reflectance of the red band, the reflectance of the blue band, and the reflectance of the near-infrared band to calculate the normalized vegetation index, difference vegetation index, ratio vegetation index, and enhanced vegetation index. The calculation formula is:
[0051]
[0052] DVI=NIR-Red
[0053]
[0054]
[0055] Among them, NDVI is the normalized vegetation index, DVI is the difference vegetation index, RVI is the ratio vegetation index, EVI is the enhanced vegetation index, NIR is the reflectance of the near-infrared band, Red is the reflectance of the red light band, and BLUE is the reflectance of the blue light band.
[0056] Furthermore, in S4, the corresponding biomass allometric growth equation is selected according to the tree species to calculate the aboveground biomass of each tree. The ratio of the sum of the AGB of all trees in the plot to the area of the plot is the aboveground biomass per unit area of the plot, and the aboveground biomass per unit area of the plot is taken as the dependent variable;
[0057] In Python, the random forest feature importance ranking is used to sort the data in the feature factor set, and the model is built by adding the variables after positive sorting one by one, and the R value of each time is calculated. 2 Determine the best feature set, and use the best feature set as the independent variable.
[0058] Furthermore, the specific contents of using dependent variables and independent variables to train the base learners composed of four machine learning methods, namely, K nearest neighbor, support vector regression, random forest and extreme random tree, are as follows:
[0059] Firstly, the independent variables and dependent variables were randomly divided into training set and test set in a ratio of 7:3. The training set was used to construct and optimize the parameters of four forest AGB regression estimation models, and the test set was used to verify the accuracy of the four forest AGB regression estimation models.
[0060] Then, four machine learning algorithms, namely K nearest neighbor, support vector regression, random forest and extreme random tree, were used to train four forest AGB regression estimation models, and the prediction accuracy of the four forest AGB regression estimation models was continuously improved by adjusting parameters.
[0061] Furthermore, in S5, the specific steps of constructing four forest AGB migration regression estimation models by using the base learner combined with the Two-stage TrAdaBoost.R2 migration learning algorithm are as follows:
[0062] S51. The target domain data is randomly divided into a test set and a training set in a ratio of 7:3, and all the source domain data is used as the training set. The test set of the target domain and all the source domain data are combined into the training set of the migration model, and the test set of the target domain is used as the test set of the migration model;
[0063] S52. Perform simulation training on the training set of the migration model through the base learner, and use the Two-stageTrAdaBoost.R2 algorithm to adjust the sample weights, reduce the weights of the source domain data with large errors, and increase the weights of the target domain data;
[0064] S53. Freeze the weights of all source domain samples, continue to use the base learner to train the target domain data, further update the weights of the target domain samples, and obtain the final four forest AGB migration regression estimation models.
[0065] Furthermore, it also includes using the test set to evaluate the final four forest AGB migration regression estimation models. Specifically, the root mean square error and determination coefficient of the four forest AGB migration regression estimation models are calculated to evaluate the accuracy and fitting effect of the model, thereby obtaining the evaluation results of the four forest AGB migration regression estimation models.
[0066] Furthermore, the root mean square error and determination coefficient of the four forest AGB migration regression estimation models were calculated. The larger the determination coefficient value and the smaller the root mean square error value, the better the model fitting effect is. Specifically:
[0067]
[0068]
[0069] Among them, RMSE is the root mean square error, R 2 is the coefficient of determination, y i is the measured value of biomass in the sample plot, is the model-estimated value of the biomass of the sample plot, is the mean of the measured biomass of the sample plots, and n is the number of samples.
[0070] In a specific embodiment, the study area is divided into a source domain and a target domain, wherein the area with a larger number of samples is set as the source domain, and the area with a smaller number of samples is set as the target domain. The forest tree species in the two areas are the same and there is a geographical span.
[0071] Step 1: Data collection, (1) Sample plot measured data: Select multiple sample plots in the forest area of the source domain and the target domain respectively, measure the tree height and breast diameter of each tree in the sample plot, and record the GPS coordinates of the center point of the sample plot. Select the corresponding biomass allometric growth equation according to the tree species, calculate the aboveground biomass of each tree, and the ratio of the sum of the AGB of all trees in the sample plot to the area of the sample plot is the aboveground biomass per unit area of the sample plot. (2) Auxiliary factor data: Collect remote sensing, meteorological and topographic data of the study area (including the source domain and the target domain). Among them, the remote sensing image data is obtained based on Landsat 8 images, and the images are pre-processed by atmospheric correction, radiation correction, splicing and cropping; the meteorological data obtained include temperature, precipitation, wind speed and surface runoff in the study area; the topographic data include slope, aspect and altitude.
[0072] Step 2: Preprocess the collected measured data and auxiliary factor data. Based on the extracted remote sensing characteristic factors and the acquired meteorological and terrain factor data, use the "multi-value extraction to point" method in ArcGIS to extract the remote sensing, meteorological and terrain factor data values to the sample point according to the central coordinates of the sample point. Together with the measured data of the sample point, they form a characteristic factor set to achieve the fusion of multi-source data.
[0073] Step 3: Based on the preprocessed Landsat 8 multispectral remote sensing image, extract the red, blue, and near-infrared band reflectance, and calculate various vegetation index data such as the Normalized Difference Vegetation Index (NDVI), Difference Vegetation Index (DVI), Ratio Vegetation Index (RVI), and Enhanced Vegetation Index (EVI), and add various vegetation index data to the characteristic factor set. The calculation formula of the vegetation index is as follows:
[0074]
[0075] DVI=NIR-Red
[0076]
[0077]
[0078] Among them, NDVI is the normalized vegetation index, DVI is the difference vegetation index, RVI is the ratio vegetation index, EVI is the enhanced vegetation index, NIR is the reflectance of the near-infrared band, Red is the reflectance of the red light band, and BLUE is the reflectance of the blue light band.
[0079] Step 4: Take the aboveground biomass per unit area of the plot as the dependent variable and the other factors in the characteristic factor set as independent variables, and build a model with the machine learning algorithm. Before model training, in Python, use the random forest feature importance ranking to sort the dependent variables, and add the variables after positive sorting one by one to build the model, and calculate the R 2 To determine the best feature set, and select the best feature set. When training the base learner, the data is first randomly divided into a training set and a test set in a ratio of 7:3. The training set is used for model construction and parameter optimization, and the test set is used to verify the accuracy of the model. Then, machine learning algorithms such as K nearest neighbor (KNN), support vector regression (SVR), random forest (RF) and extreme random tree (ET) are used for model training, and the prediction accuracy of the model is continuously improved by adjusting the parameters.
[0080] Step 5: Using KNN, SVR, RF and ET as base learners, combined with the Two-stage TrAdaBoost.R2 transfer learning algorithm, four forest AGB migration regression estimation models are constructed. The construction process mainly includes:
[0081] (1) Data reorganization. The target domain data was randomly divided into a test set and a training set in a ratio of 7:3, and all the source domain data was used as the training set. The target domain test set and all the source domain data were combined into the training set of the four forest AGB migration regression estimation models, and the target domain test set was used as the test set of the four forest AGB migration regression estimation models.
[0082] (2) Knowledge transfer. The training set data of the four forest AGB migration regression estimation models are simulated and trained by the base learner. The sample weights are adjusted using the Two-stage TrAdaBoost.R2 algorithm to reduce the weight of the source domain data with large errors and increase the weight of the target domain data.
[0083] First, perform initial weight allocation. Suppose the number of source domain samples is n s , the number of samples in the target domain is n t , the total number of samples in the training set is N = n s +n t . Assign equal initial weights to each sample:
[0084]
[0085] Set the initialization error weight adjustment parameter β:
[0086]
[0087] Where T is the number of iterations.
[0088] Secondly, perform model training and weight adjustment. For each iteration t=1,2,…,T, use the training set of the migration model and adjust the weights based on the current weights. Train the base learner and calculate the model error on source and target domain samples:
[0089]
[0090] In the formula, y i is the true value, f (t) (X i ) is the predicted value of the model.
[0091] The weight of the source domain sample is adjusted according to the error calculated above. If the sample comes from the target domain, the weight remains unchanged; if the sample comes from the source domain, the updated weight is:
[0092]
[0093] Renormalize the weights:
[0094]
[0095] (3) Model training and testing. Freeze the weights of all source domain samples, that is:
[0096]
[0097] Continue to use the base learner to train the target domain data and further update the weights of the target domain samples so that the model can better adapt to the environment and data characteristics of the target domain.
[0098] (4) Model accuracy evaluation. The final migration regression model was evaluated using the test set of the migration model, and the root mean square error (RMSE) and coefficient of determination (R 2 ) to evaluate the accuracy and fit of the model. 2 The larger the value, the smaller the RMSE value, and the better the model fitting effect. The calculation formulas for each evaluation index are as follows:
[0099]
[0100]
[0101] Among them, RMSE is the root mean square error, R 2 is the coefficient of determination, y i is the measured value of biomass in the sample plot, is the model-estimated value of the biomass of the sample plot, is the mean of the measured biomass of the sample plots, and n is the number of samples.
[0102] Step 6: Based on the evaluation results of the four forest AGB migration regression estimation models, the forest AGB migration regression estimation model with the highest fitting accuracy is selected and used for the remote sensing inversion of forest aboveground biomass in the target domain, thereby obtaining the forest AGB estimation value of the entire study area in the target domain.
[0103] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0104] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for estimating forest aboveground biomass based on transfer learning, characterized in that: The following steps are involved: S1. Data collection: Collect the measured data and auxiliary factor data of the sample site; S2. Data preprocessing: Preprocess the collected measured data and auxiliary factor data to obtain the preprocessed measured data and auxiliary factor data. Use the multi-value extraction to point method in ArcGIS to extract the preprocessed measured data and auxiliary factor data to the sample point according to the center coordinates of the sample site. Composition of characteristic factor set; S3. Calculation of vegetation index: Based on the auxiliary factor data obtained by preprocessing, extract the characteristic variables of remote sensing factors, extract the reflectance of the red light band, the reflectance of the blue light band, and the reflectance of the near-infrared band to calculate various vegetation index data, and add the various vegetation index data to the characteristic factor set; S4. Base learner training: The aboveground biomass per unit area of the plot is used as the dependent variable, and the data in the characteristic factor set is used as the independent variable. The dependent variable and the independent variable are used to train the base learner composed of four machine learning methods: K nearest neighbor, support vector regression, random forest and extreme random tree; S5. Model construction: Four forest AGB migration regression estimation models were constructed using the base learner combined with the Two-stage TrAdaBoost.R2 migration learning algorithm; S6. Forest AGB estimation: Based on the four constructed forest AGB migration regression estimation models, the estimated value of forest AGB in the entire study area of the target domain was obtained.
2. The method for estimating forest aboveground biomass based on transfer learning according to claim 1, characterized in that: The auxiliary factor data in S1 include remote sensing data, meteorological data and terrain data; among them, the remote sensing image data is obtained based on Landsat 8 images; the meteorological data include temperature, precipitation, wind speed and surface runoff in the study area; the terrain data include slope, aspect and altitude.
3. The method for estimating forest aboveground biomass based on transfer learning according to claim 2, characterized in that: In S3, based on the preprocessed Landsat 8 multispectral remote sensing image, the reflectance of the red light band, the reflectance of the blue light band, and the reflectance of the near-infrared band are extracted to calculate the normalized vegetation index, difference vegetation index, ratio vegetation index, and enhanced vegetation index. The calculation formula is: DVI=NIR-Red; Among them, NDVI is the normalized vegetation index, DVI is the difference vegetation index, RVI is the ratio vegetation index, EVI is the enhanced vegetation index, NIR is the reflectance of the near-infrared band, Red is the reflectance of the red light band, and BLUE is the reflectance of the blue light band.
4. The method for estimating forest aboveground biomass based on transfer learning according to claim 1, characterized in that: In S4, the corresponding biomass allometric growth equation is selected according to the tree species to calculate the aboveground biomass of each tree. The ratio of the sum of the AGB of all trees in the plot to the plot area is the aboveground biomass per unit area of the plot, and the aboveground biomass per unit area of the plot is taken as the dependent variable; In Python, the random forest feature importance ranking is used to sort the data in the feature factor set, and the model is built by adding the variables after positive sorting one by one, and the R value of each time is calculated. 2 Determine the best feature set, and use the best feature set as the independent variable.
5. The method for estimating forest aboveground biomass based on transfer learning according to claim 1, characterized in that: The specific contents of using dependent variables and independent variables to train the base learners composed of four machine learning methods, namely K nearest neighbor, support vector regression, random forest and extreme random tree, are as follows: Firstly, the independent variables and dependent variables were randomly divided into training set and test set in a ratio of 7:
3. The training set was used to construct and optimize the parameters of four forest AGB regression estimation models, and the test set was used to verify the accuracy of the four forest AGB regression estimation models. Then, four machine learning algorithms, namely K nearest neighbor, support vector regression, random forest and extreme random tree, were used to train four forest AGB regression estimation models, and the prediction accuracy of the four forest AGB regression estimation models was continuously improved by adjusting parameters.
6. The method for estimating forest aboveground biomass based on transfer learning according to claim 1, characterized in that: In S5, the specific steps of constructing four forest AGB migration regression estimation models by using the base learner combined with the Two-stage TrAdaBoost.R2 migration learning algorithm are as follows: S51. The target domain data is randomly divided into a test set and a training set in a ratio of 7:3, and all the source domain data is used as the training set. The test set of the target domain and all the source domain data are combined into the training set of the migration model, and the test set of the target domain is used as the test set of the migration model; S52. Perform simulation training on the training set of the migration model through the base learner, and use the Two-stageTrAdaBoost.R2 algorithm to adjust the sample weights, reduce the weights of the source domain data with large errors, and increase the weights of the target domain data; S53. Freeze the weights of all source domain samples, continue to use the base learner to train the target domain data, further update the weights of the target domain samples, and obtain the final four forest AGB migration regression estimation models.
7. The method for estimating forest aboveground biomass based on transfer learning according to claim 6, characterized in that: It also includes using the test set to evaluate the final four forest AGB migration regression estimation models. Specifically, the root mean square error and determination coefficient of the four forest AGB migration regression estimation models are calculated to evaluate the accuracy and fitting effect of the models, thereby obtaining the evaluation results of the four forest AGB migration regression estimation models.
8. The method for estimating forest aboveground biomass based on transfer learning according to claim 7, characterized in that: The root mean square error and determination coefficient of the four forest AGB migration regression estimation models were calculated. The larger the determination coefficient value and the smaller the root mean square error value, the better the model fitting effect is. Specifically: Among them, RMSE is the root mean square error, R 2 is the coefficient of determination, y i is the measured value of biomass in the sample plot, is the model-estimated value of the biomass of the sample plot, is the mean of the measured biomass of the sample plots, and n is the number of samples.