Artificial forest aboveground biomass monitoring method based on model fusion

Abstract

The present invention discloses a method for monitoring aboveground biomass of artificial forests based on model fusion, which relates to the technical field of biomass monitoring. It collects airborne hyperspectral data, airborne LiDAR data, and the true value of the biomass in any part of the observation area within the observation area; preprocesses the airborne hyperspectral data and airborne LiDAR data respectively, constructs an aboveground biomass model based on the airborne LiDAR data and an aboveground biomass model based on the airborne hyperspectral data, calculates the prediction error of the aboveground biomass model based on the airborne LiDAR data for the monitored area and the prediction error of the aboveground biomass model based on the airborne hyperspectral data for the monitored area respectively, thereby setting the weight of the airborne LiDAR data and the weight of the airborne hyperspectral data, and generating an aboveground biomass monitoring model for artificial forests. The present invention fuses airborne hyperspectral and LiDAR data, provides accurate forest resource data for forestry departments, and helps with more scientific resource assessment and ecological planning.

Description

Technical Field

[0001] The present invention relates to the technical field of biomass monitoring, and in particular to a method for monitoring aboveground biomass of artificial forests based on model fusion. Background Art

[0002] In terms of artificial forest monitoring, traditional single data monitoring of artificial forest biomass only relies on a certain type of data, and the information acquisition is one-sided. For example, only using satellite remote sensing data, although it can cover a large area, has low resolution and is difficult to accurately identify the biochemical components of vegetation, resulting in large errors in biomass estimation. Forestry resource assessment has extremely high requirements for data accuracy. Inaccurate data will mislead ecological planning. For example, when formulating forest carbon sink accounting, erroneous biomass data will cause carbon sink estimation deviations and affect the correct judgment of the ecological value of forests; manual field measurement of biomass was the main method in the early days. Workers needed to measure each tree in a large area of artificial forests. The process was cumbersome and time-consuming. In some vast forest farms, it may take months or even longer to complete a comprehensive measurement. With the development of forestry, the timeliness of monitoring dynamic changes in forests is becoming increasingly demanding. Sudden situations such as forest pests and diseases and fire hazards require rapid response. Traditional manual monitoring methods cannot provide dynamic data on forest resources in a timely manner, and it is difficult to meet the needs of forestry departments for real-time supervision of forests. Therefore, it is urgent to use advanced technology to improve the efficiency of biomass monitoring; traditional manual monitoring of biomass not only requires a lot of manpower, but also requires professional measurement tools, which is costly. Moreover, in order to ensure the accuracy of the measurement, it is often necessary to repeat the measurement many times, which further increases the cost. For some forestry projects with limited funds, it is difficult to bear such high monitoring costs. In addition, the limited measurement range also limits the comprehensive monitoring of large areas of forests. In the context of pursuing sustainable development, there is an urgent need for a new method that can reduce costs and achieve high-precision, large-area biomass monitoring to promote the sustainable development of forestry resource monitoring.

[0003] Therefore, people need a method for monitoring aboveground biomass of artificial forests based on model fusion to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a method for monitoring aboveground biomass of artificial forests based on model fusion, so as to solve the problems raised in the above-mentioned background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solution: a method for monitoring aboveground biomass of artificial forests based on model fusion, comprising the following steps:

[0006] S1. Collect airborne hyperspectral data, airborne LiDAR data and the real value of biomass in any part of the observation area; pre-process the airborne hyperspectral data and airborne LiDAR data respectively;

[0007] S2. Based on airborne LiDAR data and airborne hyperspectral data respectively, construct an aboveground biomass model based on airborne LiDAR data and an aboveground biomass model based on airborne hyperspectral data;

[0008] S3. Calculate the prediction error of the aboveground biomass model of airborne LiDAR data for the monitoring area and the prediction error of the aboveground biomass model of airborne hyperspectral data for the monitoring area respectively, so as to set the weight of airborne LiDAR data and the weight of airborne hyperspectral data;

[0009] S4. Integrate the aboveground biomass model based on airborne LiDAR data and the aboveground biomass model based on airborne hyperspectral data to generate an artificial forest land biomass monitoring model, and use the artificial forest land biomass monitoring model to monitor the biomass.

[0010] Further, in step S1, N characteristic data {x 1L , x 2L , …, x nL , …, x NL} of airborne LiDAR data are obtained through preprocessing, and I characteristic parameters {x 1H , x 2H , …, x iL , …, x IH} of hyperspectral data are obtained through preprocessing. The preprocessing includes removing data noise and correcting radiation error. The preprocessing operation is the key foundation of the entire artificial forest land biomass monitoring method based on model fusion. Removing data noise eliminates the messy signals that interfere with data accuracy, ensuring that the airborne LiDAR data and hyperspectral data can clearly and accurately reflect the actual characteristics of the forest. This greatly improves the stability and reliability of the model when constructing the biomass model using these data later, reducing the error risk caused by data fluctuations; correcting radiation error is an important measure to ensure data authenticity. Through correction, the data obtained under different conditions can be analyzed under the same standard, making the comparison and fusion of data more scientific, so that the extracted characteristic data can truthfully display key information such as the growth status and spatial structure of forest vegetation, laying a solid foundation for accurately monitoring the aboveground biomass of artificial forests and strongly supporting the subsequent model construction and biomass estimation work.

[0011] Further, in step S2, analyze the nth characteristic data of the airborne LiDAR data, and use the random forest algorithm to calculate the importance score P(x nL ) of the nth characteristic data x nL :

[0012] ;

[0013] where T is the number of decision trees in the random forest, err(OL ) is the overall error without using the feature data x nL , and err(t L ) is the error of the t-th decision tree when using the feature data x nL . Furthermore, the importance scores of N feature data of the airborne LiDAR data are obtained. K feature data are selected as the important feature data for the above-ground biomass model based on the airborne LiDAR data in descending order of the importance scores. Similarly, the hyperspectral data is analyzed, and B feature data are obtained as the important feature data for the above-ground biomass model based on the airborne hyperspectral data. The above-ground biomass model based on the airborne LiDAR data is constructed as follows:

[0014] ;

[0015] where β kL is the k-th regression coefficient of the above-ground biomass model based on the airborne LiDAR data, k = 1, 2, …, K, and x kL represents the k-th of the K important feature data of the above-ground biomass model based on the airborne LiDAR data, represents the error term of the above-ground biomass model based on the airborne LiDAR data, and represents the predicted biomass of the above-ground biomass based on the airborne LiDAR data;

[0016] The method for obtaining the overall error without using the feature data x nL is as follows: When constructing any decision tree in the random forest, a part of the observation area is randomly selected from the entire observation area to construct the training set of the tree. The set of the part of the observation area that is not selected to enter the training set becomes the out-of-bag sample of any decision tree;

[0017] After all decision trees are constructed, for any decision tree, calculate the error MSE of any decision tree:

[0018] ;

[0019] where y i represents the true value of the j-th observation area in the out-of-bag sample, Y i represents the predicted value of the j-th observation area in the out-of-bag sample, and α represents the number of observation areas in the out-of-bag sample. The predicted value is obtained by not considering the feature data x nL when splitting the nodes of any decision tree;

[0020] Furthermore, calculate the average value of the errors of all decision trees as the overall error err(O nL ) without using the feature data x L ;

[0021] The method for obtaining the error of the t-th decision tree when using the feature data x nL is as follows: Randomly and with replacement, select a part of the observation regions from the observation regions to form a training subset for constructing the t-th decision tree. During the node splitting process of constructing the t-th decision tree, all feature data including the feature data x nL will be used. According to the Gini impurity criterion, determine the splitting method of the node, thereby completing the construction of the t-th decision tree;

[0022] The set of the part of the observation regions in the observation regions that are not selected for constructing the t-th decision tree constitutes the out-of-bag data of the t-th decision tree. For the J-th observation region in the out-of-bag data, traverse downward according to the t-th decision tree, and finally reach a leaf node. The predicted value z J represented by this leaf node is the predicted value of the J-th observation region in the out-of-bag data;

[0023] Furthermore, calculate the error err(t nL ) of the t-th decision tree when using the feature data x L ):

[0024] ;

[0025] where Z J represents the true value of the J-th observation region in the out-of-bag data, and μ represents the number of observation regions in the out-of-bag data;

[0026] The k-th regression coefficient of the above-ground biomass model based on airborne LiDAR data is obtained by the least squares method;

[0027] Adopt the method of constructing the above-ground biomass model based on airborne LiDAR data to construct the above-ground biomass model based on airborne hyperspectral data:

[0028] ;

[0029] where β bH is the b-th regression coefficient of the above-ground biomass model based on airborne hyperspectral data, b = 1, 2, …, B, x bH represents the b-th of the B important feature data of the above-ground biomass model based on airborne hyperspectral data, represents the error term of the above-ground biomass model based on airborne hyperspectral data, Indicates the predicted biomass of aboveground biomass based on airborne hyperspectral data; from the perspective of data processing and analysis, calculating the importance scores of feature data using the random forest algorithm is a major highlight. By comparing the errors of decision trees when using and not using specific feature data, the value of each feature can be accurately evaluated. This process can deeply excavate the effective information in airborne LiDAR data and hyperspectral data, avoid model redundancy caused by blindly using all data, and make the selected K (for LiDAR data) and B (for hyperspectral data) important feature data more representative. For example, in a complex forest environment, key features such as vegetation height and spectral reflectance that contribute the most to biomass estimation can be quickly identified, laying a solid foundation for subsequent model construction.

[0030] From the perspective of model construction, aboveground biomass models are constructed respectively based on the selected important feature data, which improves the pertinence and accuracy of the models. The regression coefficients are obtained using the least squares method, enabling the models to optimally fit the data and minimizing errors to the greatest extent. Taking the airborne LiDAR data model as an example, by assigning appropriate weights to different feature data, the relationship between the three-dimensional structure of the forest and biomass can be more accurately reflected. Moreover, this method of constructing models is universal and also applicable to airborne hyperspectral data, ensuring that the model construction logic for the two data sources is consistent, facilitating subsequent model fusion and result analysis.

[0031] From the aspect of error evaluation and optimization, by calculating the decision tree errors using out-of-bag samples and then finding the average of the overall errors, the reliability of the models is effectively verified. When not using the feature data x nL , calculating the overall error using out-of-bag samples can clearly understand the impact of the absence of this feature on the model; when using the feature data, calculating the errors of each decision tree for further refined evaluation helps to adjust the model parameters in a timely manner and optimize the model performance, thereby significantly improving the prediction accuracy of aboveground biomass in plantations and providing more reliable data support for forestry resource monitoring.

[0032] Furthermore, in step S3, for any monitored area, the initial biomass M is obtained by an unmanned aerial vehicle, the predicted biomass of aboveground biomass based on airborne LiDAR data is calculated as M L using the aboveground biomass model based on airborne LiDAR data, and the predicted biomass of aboveground biomass based on airborne hyperspectral data is calculated as M H using the aboveground biomass model based on airborne hyperspectral data, so as to set the weight of airborne LiDAR data (|M H - M|) / (|M L - M| + |M H - M|) and the weight of airborne hyperspectral data (|M L - M|) / (|M L-M|+|M H -M|), obtain the initial biomass through UAV, combine the two models to set the weights, and dynamically adjust the role of airborne LiDAR and hyperspectral data in biomass calculation according to the actual situation of different monitoring areas, so that data fusion is more in line with reality and the monitoring accuracy is improved.

[0033] Furthermore, in step S4, a plantation biomass monitoring model is generated:

[0034] ;

[0035] in It represents the predicted biomass of artificial forests, effectively integrating the advantages of the two models, giving full play to the ability of LiDAR data to capture forest structure information and hyperspectral data to capture vegetation biochemical information, achieving more accurate and comprehensive prediction of artificial forest biomass, and providing strong data support for the scientific management of forestry resources.

[0036] Compared with the prior art, the beneficial effects achieved by the present invention are: on the one hand, the traditional single data monitoring method has large errors in the estimation of biomass. The present invention integrates airborne hyperspectral and LiDAR data. The former can accurately identify the biochemical components of vegetation and judge the growth status of plants by virtue of continuous spectral characteristics; the latter uses laser detection to obtain the three-dimensional structure of the forest and clarify parameters such as tree height and density. The combination of the two fully covers the information required for biomass estimation. In actual monitoring, the prediction results of the fusion model are closer to the actual biomass, reducing prediction errors, providing the forestry department with accurate forest resource data, and facilitating more scientific resource assessment and ecological planning;

[0037] On the one hand, in the past, when measuring biomass manually on the spot, workers had to measure each tree one by one. In large areas of artificial forests, this process is time-consuming and difficult. The monitoring method based on model fusion can quickly collect data on large areas of forests with the help of remote sensing technology. By using equipment mounted on aircraft, large-area scans can be completed in a short period of time to obtain massive amounts of data. The data is quickly analyzed and processed by the model, and the biomass monitoring results can be output. Compared with traditional methods, the monitoring cycle is shortened, allowing the forestry department to keep abreast of forest dynamics and changes, and gain valuable time to respond to emergencies such as forest pests and diseases and fire hazards in a timely manner;

[0038] On the other hand, traditional manual monitoring requires a lot of manpower and professional measurement tools, and due to the limited measurement range, repeated measurements are often required to ensure accuracy. The new method reduces field measurement work and manpower requirements. One remote sensing measurement can obtain multi-dimensional data to meet a variety of analyses. The efficient collection capability and accurate results reduce the overall monitoring cost, enabling forestry projects to achieve high-precision, large-area biomass monitoring, thereby promoting the sustainable development of forestry resource monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0040] Figure 1 is a flowchart of the method for monitoring aboveground biomass in artificial forests based on model fusion of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Please refer to Figure 1 , the present invention provides a technical solution: a method for monitoring aboveground biomass in artificial forests based on model fusion, including the following steps:

[0043] S1. Collect airborne hyperspectral data, airborne LiDAR data, and the true value of the biomass in any part of the observation area within the observation area; preprocess the airborne hyperspectral data and airborne LiDAR data respectively;

[0044] S2. Based on the airborne LiDAR data and airborne hyperspectral data respectively, construct an aboveground biomass model based on the airborne LiDAR data and an aboveground biomass model based on the airborne hyperspectral data;

[0045] S3. Calculate the prediction error of the aboveground biomass model based on the airborne LiDAR data for the monitoring area and the prediction error of the aboveground biomass model based on the airborne hyperspectral data for the monitoring area respectively, so as to set the weight of the airborne LiDAR data and the weight of the airborne hyperspectral data;

[0046] S4. Fusion the aboveground biomass model based on the airborne LiDAR data and the aboveground biomass model based on the airborne hyperspectral data to generate an artificial forest land biomass monitoring model, and use the artificial forest land biomass monitoring model to monitor the biomass.

[0047] In step S1, N characteristic data {x 1L , x 2L , …, x nL , …, x NL} of the airborne LiDAR data are obtained through preprocessing, and I characteristic parameters {x 1H , x 2H , …, xiL ,…,x IH}, and the preprocessing includes removing data noise and correcting radiation error. The preprocessing operation is the key foundation of the entire method for monitoring aboveground biomass in artificial forests based on model fusion. Removing data noise eliminates the messy signals that interfere with data accuracy, ensuring that the airborne LiDAR data and hyperspectral data can clearly and accurately reflect the actual characteristics of the forest. This significantly improves the stability and reliability of the model when constructing the biomass model using these data later, reducing the error risk caused by data fluctuations; correcting radiation error is an important measure to ensure data authenticity. Through correction, the data obtained under different conditions can be analyzed under the same standard, making the comparison and fusion of data more scientific, so that the extracted characteristic data can truthfully show the key information such as the growth status and spatial structure of forest vegetation, laying a solid foundation for accurately monitoring the aboveground biomass in artificial forests and strongly supporting the subsequent model construction and biomass estimation work.

[0048] In step S2, analyze the nth characteristic data of the airborne LiDAR data, and use the random forest algorithm to calculate the importance score P(x nL ) of the nth characteristic data x nL :

[0049] ;

[0050] where T is the number of decision trees in the random forest, err(O L ) is the overall error without using the characteristic data x nL , and err(t L ) is the error of the tth decision tree when using the characteristic data x nL . Then, obtain the importance scores of the N characteristic data of the airborne LiDAR data, and select K characteristic data with the highest to lowest importance scores as the important characteristic data of the aboveground biomass model based on the airborne LiDAR data; similarly, analyze the hyperspectral data to obtain B characteristic data as the important characteristic data of the aboveground biomass model based on the airborne hyperspectral data, and construct the aboveground biomass model based on the airborne LiDAR data:

[0051] ;

[0052] where β kL is the kth regression coefficient of the aboveground biomass model based on the airborne LiDAR data, k = 1, 2,…, K, x kL represents the kth of the K important characteristic data of the aboveground biomass model based on the airborne LiDAR data, represents the error term of the aboveground biomass model based on the airborne LiDAR data, Indicates the predicted biomass of aboveground biomass based on airborne LiDAR data;

[0053] The method for obtaining the overall error when not using the feature data x nL When constructing any decision tree in the random forest, a part of the observation area is randomly selected from the entire observation area to construct the training set of this tree. The set of the part of the observation area that is not selected into the training set becomes the out-of-bag sample of any decision tree;

[0054] When all decision trees are constructed, for any decision tree, calculate the error MSE of any decision tree:

[0055] ;

[0056] where y i represents the true value of the j-th observation area in the out-of-bag sample, Y i represents the predicted value of the j-th observation area in the out-of-bag sample, α represents the number of observation areas in the out-of-bag sample, and the predicted value is obtained by not considering the feature data x nL when splitting nodes in any decision tree;

[0057] Furthermore, calculate the average value of the errors of all decision trees as the overall error err(O nL ) when not using the feature data x L ;

[0058] The method for obtaining the error of the t-th decision tree when using the feature data x nL Randomly select a part of the observation area with replacement from the observation area to form a training subset for constructing the t-th decision tree. During the node splitting process of constructing the t-th decision tree, all feature data including the feature data x nL will be used, and the splitting method of the node is determined according to the Gini impurity criterion, thus completing the construction of the t-th decision tree;

[0059] The set of the part of the observation area that is not selected for constructing the t-th decision tree in the observation area constitutes the out-of-bag data of the t-th decision tree. For the J-th observation area in the out-of-bag data, traverse downward according to the t-th decision tree until reaching a leaf node, and the predicted value z J represented by this leaf node is the predicted value of the J-th observation area in the out-of-bag data;

[0060] Furthermore, calculate the error err(t nL ) of the t-th decision tree when using the feature data x L :

[0061] ;

[0062] where Z J represents the true value of the Jth observation area in the out-of-bag data, and μ represents the number of observation areas in the out-of-bag data;

[0063] The kth regression coefficient of the aboveground biomass model based on airborne LiDAR data is obtained by the least squares method;

[0064] Adopt the method of constructing an aboveground biomass model based on airborne LiDAR data to construct an aboveground biomass model based on airborne hyperspectral data:

[0065] ;

[0066] where β bH is the bth regression coefficient of the aboveground biomass model based on airborne hyperspectral data, b = 1, 2, …, B, and x bH represents the bth of the B important feature data of the aboveground biomass model based on airborne hyperspectral data, represents the error term of the aboveground biomass model based on airborne hyperspectral data, represents the predicted biomass of the aboveground biomass based on airborne hyperspectral data; From the perspective of data processing and analysis, calculating the importance score of feature data using the random forest algorithm is a major highlight. By comparing the errors of decision trees when using and not using specific feature data, the value of each feature can be accurately evaluated. This process can deeply excavate the effective information in airborne LiDAR data and hyperspectral data, avoid model redundancy caused by blindly using all data, and make the selected K (for LiDAR data) and B (for hyperspectral data) important feature data more representative. For example, in a complex forest environment, key features such as vegetation height and spectral reflectance that contribute the most to biomass estimation can be quickly identified, laying a solid foundation for subsequent model construction.

[0067] From the perspective of model construction, aboveground biomass models are respectively constructed based on the selected important feature data, which improves the pertinence and accuracy of the models. The regression coefficients are obtained using the least squares method, enabling the models to optimally fit the data and minimizing the error to the greatest extent. Taking the airborne LiDAR data model as an example, by assigning appropriate weights to different feature data, the relationship between the three-dimensional structure of the forest and biomass can be more accurately reflected. Moreover, this method of constructing models is universal and also applicable to airborne hyperspectral data, ensuring the consistency of the model construction logic for the two data sources and facilitating subsequent model fusion and result analysis.

[0068] From the perspective of error evaluation and optimization, by calculating the decision tree error using out-of-bag samples and then averaging the overall error, the reliability of the models is effectively verified. When not using the feature data x nLWhen using out-of-bag samples to calculate the overall error, the impact of the missing feature on the model can be clearly understood; while when using feature data, calculating the error of each decision tree for further refined evaluation helps to adjust the model parameters in a timely manner, optimize the model performance, and thus significantly improve the prediction accuracy of above-ground biomass in plantations, providing more reliable data support for forestry resource monitoring.

[0069] In step S3, for any monitoring area, the initial biomass M is obtained by an unmanned aerial vehicle, and the predicted biomass M of above-ground biomass based on airborne LiDAR data is calculated through the above-ground biomass model based on airborne LiDAR data. L The predicted biomass M of above-ground biomass based on airborne hyperspectral data is calculated through the above-ground biomass model of airborne hyperspectral data. H Thus, the weight of airborne LiDAR data (|M H - M|) / (|M L - M| + |M H - M|) and the weight of airborne hyperspectral data (|M L - M|) / (|M L - M| + |M H - M|) are set. By obtaining the initial biomass through an unmanned aerial vehicle and setting weights by combining the predicted values of the two models, the roles of airborne LiDAR and hyperspectral data in biomass calculation can be dynamically adjusted according to the actual situation of different monitoring areas, making the data fusion more in line with the actual situation and improving the monitoring accuracy.

[0070] Furthermore, in step S4, an above-ground biomass monitoring model for plantations is generated:

[0071] ;

[0072] where represents the predicted value of above-ground biomass in plantations, effectively integrating the advantages of the two models, giving full play to the ability of LiDAR data to capture forest structure information and hyperspectral data to capture vegetation biochemical information, and realizing more accurate and comprehensive prediction of above-ground biomass in plantations, providing strong data support for the scientific management of forestry resources.

[0073] Example 1: In stage S1, first use professional airborne remote sensing equipment to collect airborne hyperspectral data and airborne LiDAR data within the observation area of this eucalyptus forest. At the same time, select representative small areas in the eucalyptus forest and measure the tree diameter at breast height and tree height on the spot. Then, preprocess the obtained airborne hyperspectral data and airborne LiDAR data to remove data noise generated by factors such as equipment noise and atmospheric interference, and correct the radiation error. Finally, N feature data of airborne LiDAR data, such as canopy height and crown width, and I feature parameters of hyperspectral data, such as spectral reflectance in different bands, are obtained.

[0074] Enter stage S2. For each feature data of the airborne LiDAR data, use the random forest algorithm to calculate its importance score. For example, for the feature data "canopy height", calculate the overall error err(O L ) of the random forest model when not using it, and the error err(t L ) of each decision tree when using it, so as to obtain its importance score P(x nL ). Select K important feature data according to the score, determine the regression coefficients by the least squares method, and construct a aboveground biomass model based on the airborne LiDAR data. Similarly, analyze the hyperspectral data and construct a aboveground biomass model based on the airborne hyperspectral data.

[0075] In stage S3, use the drone to obtain the initial biomass M in the forest. Then calculate the first prediction M L and the second prediction M H respectively with the aboveground biomass models based on the airborne LiDAR data and the airborne hyperspectral data, and set the weights of the two kinds of data.

[0076] Finally, in stage S4, fuse the two models according to the determined weights to generate an artificial forestland biomass monitoring model, accurately predict the aboveground biomass of the entire eucalyptus forest, and provide data support for subsequent forest resource management and ecological assessment.

[0077] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any regard, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes within the meaning and scope of the equivalent elements of the claims in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed claims.

Claims

1. An above-ground biomass monitoring method for plantation forests based on model fusion, characterized in that: The method includes the following steps: S1. Collect airborne hyperspectral data, airborne LiDAR data, and the true value of the biomass of any part of the observation area within the observation area; preprocess the airborne hyperspectral data and the airborne LiDAR data respectively. S2. Based on the airborne LiDAR data and the airborne hyperspectral data respectively, construct an above-ground biomass model based on the airborne LiDAR data and an above-ground biomass model based on the airborne hyperspectral data. S3. Calculate the prediction error of the above-ground biomass model based on the airborne LiDAR data for the monitoring area and the prediction error of the above-ground biomass model based on the airborne hyperspectral data for the monitoring area respectively, so as to set the weight of the airborne LiDAR data and the weight of the airborne hyperspectral data. S4. Integrate the above-ground biomass model based on the airborne LiDAR data and the above-ground biomass model based on the airborne hyperspectral data to generate an artificial forestland biomass monitoring model, and use the artificial forestland biomass monitoring model to monitor the biomass. In step S2, analyze the nth feature data of the airborne LiDAR data, and use the random forest algorithm to calculate the importance score P(x nL ) of the nth feature data x nL : ； where T is the number of decision trees in the random forest, and err(O L ) is the overall error without using the feature data x nL , and err(t L ) is the error of the t-th decision tree when using the feature data x nL . Furthermore, the importance scores of N feature data of the airborne LiDAR data are obtained, and K feature data are selected from high to low according to the importance scores as the important feature data of the above-ground biomass model based on the airborne LiDAR data; similarly, the hyperspectral data is analyzed, and B feature data are obtained as the important feature data of the above-ground biomass model based on the airborne hyperspectral data, and the above-ground biomass model based on the airborne LiDAR data is constructed; Adopt the method of constructing the above-ground biomass model based on the airborne LiDAR data, consider the error term of the above-ground biomass model based on the airborne hyperspectral data and the predicted biomass of the above-ground biomass based on the airborne hyperspectral data, and construct the above-ground biomass model based on the airborne hyperspectral data.

2. The method for monitoring aboveground biomass of plantation based on model fusion according to claim 1, characterized in that: In step S1, N feature data {x 1L , x 2L , …, x nL , …, x NL} of airborne LiDAR data are obtained through preprocessing, and I feature parameters {x 1H , x 2H , …, x iL , …, x IH} of hyperspectral data are obtained through preprocessing. The preprocessing includes removing data noise and correcting radiation errors.

3. The method for monitoring above-ground biomass of planted forests based on model fusion according to claim 2, characterized in that: In step S2, the method for obtaining the overall error when the feature data x is not used nL : When constructing any decision tree in the random forest, a part of the observation area is randomly selected from the entire observation area to construct the training set of the tree, and the set of the part of the observation area that is not selected to enter the training set becomes the out-of-bag sample of any decision tree; After all decision trees are constructed, for any decision tree, calculate the mean squared error (MSE) of any decision tree, and then calculate the average of the errors of all decision trees as the overall error err(O nL when the feature data x L ) is not used.

4. The method for monitoring the above-ground biomass of a plantation based on model fusion according to claim 3, wherein: The use of the feature data x nL The method for obtaining the error of the t-th decision tree when using the feature data x: Randomly draw a part of the observation area with replacement from the observation area to form a training subset for constructing the t-th decision tree. During the node splitting process of constructing the t-th decision tree, all feature data including the feature data x nL will be used. According to the Gini impurity criterion, the splitting method of the node is determined, thereby completing the construction of the t-th decision tree; The set of partial observation regions in the observation region that are not drawn to construct the t-th decision tree constitutes the out-of-bag data of the t-th decision tree. For the J-th observation region in the out-of-bag data, traversing down according to the t-th decision tree, a leaf node is finally reached, and the predicted value z represented by this leaf node J is the predicted value of the J-th observation region in the out-of-bag data; Furthermore, based on the true value of the J-th observation region in the out-of-bag data and the number of observation regions in the out-of-bag data, calculate the error err(t nL of the t-th decision tree when using the feature data x L ).

5. The method for monitoring aboveground biomass of plantation based on model fusion according to claim 4, wherein: The k-th regression coefficient of the above-ground biomass model based on the airborne LiDAR data is obtained by the least squares method.

6. The method for monitoring the above-ground biomass of a planted forest based on model fusion according to claim 5, wherein: In step S3, for any monitoring area, the initial biomass M is obtained by an unmanned aerial vehicle, and the predicted biomass M of the above-ground biomass based on the airborne LiDAR data is calculated through the above-ground biomass model based on the airborne LiDAR data L , and the predicted biomass M of the above-ground biomass based on the airborne hyperspectral data is calculated through the above-ground biomass model of the airborne hyperspectral data H , so as to set the weight of the airborne LiDAR data (|M H -M|) / (|M L -M|+|M H -M|) and the weight of the airborne hyperspectral data (|M L -M|) / (|M L -M|+|M H -M|).

7. The method for monitoring above-ground biomass of artificial forests based on model fusion according to claim 6, wherein: In step S4, an artificial forest land biomass monitoring model is generated to obtain the artificial forest land biomass prediction amount .

Images