CN-122017870-A - Subtropical eucalyptus artificial forest parameter extraction method based on unmanned aerial vehicle laser radar

CN122017870ACN 122017870 ACN122017870 ACN 122017870ACN-122017870-A

Abstract

The invention discloses a method for extracting parameters of a (subtropical) eucalyptus artificial forest based on an unmanned aerial vehicle laser radar, which comprises the steps of carrying out sample plot investigation and measurement on actual forest parameters such as average diameter, average height, cross-sectional area, forest stand accumulation and the like, obtaining and preprocessing point cloud data of the high-density unmanned aerial vehicle laser radar, extracting three characteristic variables of height, density and vertical structure, deriving a model structural formula based on an accumulation abnormal speed growth equation, screening the variables according to rules, constructing an exponentiation model by using a regular exhaustion method, and screening an optimal model by estimating model parameters, checking, significance analysis and the like. The invention digs the variable synergistic effect, builds a multidimensional checking system and customizes the exclusive model through an exhaustion method, solves the problems of insufficient reliability of the model and poor suitability of a specific forest stand in the prior art, realizes high-precision estimation of the core forest parameters, and provides scientific and effective technical support for fine management and sustainable operation of eucalyptus artificial forest resources.

Inventors

HUANG XIAOQIAN
LIAO ZHIHONG
ZHANG YIMING
WEI CHANGWEN
WANG LUYING
SU KAI

Assignees

广西大学
黄晓倩

Dates

Publication Date: 20260512
Application Date: 20250522

Claims (7)

1. Claim 1: the subtropical eucalyptus artificial forest parameter extraction method based on the unmanned aerial vehicle laser radar is characterized by comprising the following steps of: (1) Sample plot investigation and data calculation: a. measuring the breast diameter, dominant tree height, average age, canopy density and the like of the forest in each sample area; b. Calculating actual forest parameters such as average Diameter (DBH), average height (H), cross-sectional area (BA), forest stand accumulation (VOL) and the like of the forest trees in various places based on the measured data; (2) Acquiring and preprocessing laser radar point cloud data of a high-density unmanned aerial vehicle: a. after the steps of denoising filtering, point cloud classification, DEM generation, elevation normalization processing and the like, normalized point cloud data of various patterns are obtained; b. extracting corresponding point cloud characteristic variables from the laser radar point cloud data normalized by each pattern; (3) Deducing a forest parameter estimation model structural formula based on an abnormal growth equation model of the accumulation amount, and screening point cloud characteristic variables; (4) According to the actual parameters of the sample plot and the characteristic variables of the laser radar point cloud of the sample plot, constructing an exponentiation model by using a regular exhaustion method; (5) And carrying out correlation analysis and model significance analysis on the model variables, carrying out parameter estimation and precision evaluation on the model, screening out an optimal model for estimating forest parameters of the eucalyptus artificial forest of the unmanned aerial vehicle laser radar according to the selected evaluation indexes, and finally carrying out adaptability test on the optimal model of each forest parameter of the eucalyptus artificial forest.
2. Claim 2: the method of claim 1, wherein the stand accumulation (VOL) in step (1) is calculated as: Wherein: Is the forest stand accumulation (m 3 /ha); Dividing the area (m 2 /ha) for the forest; Is the average height (m) of the stand; 、、 Is a model parameter; Is an error term.
3. Claim 3: The method of claim 1, wherein the corresponding point cloud feature variables in step (2) are lidar point cloud variables extracted using full echoes, and the correlation calculation formulas of the extracted point cloud feature variables with respect to canopy leaf area density correlation variables and vertical branch and leaf profile correlation variables are: ; ; ; ; ; ; ; ; Wherein: is the first Leaf area index of the individual height layers, In order to achieve a gap rate of the material, For the extinction coefficient to be a function of the extinction coefficient, Are highly spaced. In the form of a height interval, the height interval, Is the first Gap rate of the individual height layers.
4. Claim 4: the method of claim 1, wherein the deriving the forest parameter estimation model structure in step (3) is performed by: (1) Based on the stand accumulation abnormal speed growth equation of claim 2, the cross-sectional area (BA) is replaced by the density index (P) of the canopy density, and the equation can be expressed as: Wherein: is a stand density variable. Is derived to be available Substitution of . (2) The other 3 forest parameters have a certain relation with the accumulation amount, and the above formula can be changed into: wherein: the accumulation amount may be a cross-sectional area, an average diameter, or an average height, Is a stand height variable. (3) And introducing a related variable reflecting the vertical structural information of the forest stand, namely a vertical structural variable, so as to deduce a structural formula of the forest parameter estimation model as follows: wherein: forest parameters such as DBH, H, BA, VOL and the like can be adopted, As a variable of the density of the stand, As a stand height variable, the height of the stand, Is a vertical structural variable of a forest stand, 、、 Is a model parameter; Is an error term.
5. Claim 5: The method of claim 1, wherein the screened point cloud variables in step (3) comprise (1) a set of height variables including point cloud average height (hmean), 95% quantile height (hp 95), standard deviation of point cloud height distribution (hstdv) and coefficient of variation (hcv), (2) a set of density variables including canopy coverage (cc), dp50, dp75, and (3) a set of vertical structural variables including leaf area density mean (LADmean) and coefficient of variation (LADcv), leaf vertical profile mean (VFPmean), standard deviation (VFPstdv), and coefficient of variation (VFPcv).
6. Claim 6: The method of claim 1, wherein the variable selection and combination rules used in step (4) are that the forest parameter estimation model is constructed from 3-5 variables, each model should contain any 1 of the primary height variables hmean and hp95, 1 to 2 density variables (if two density variables are selected, they should be constructed from the primary density variable cc and one of the secondary density variables dp50 or dp 75) and 1 vertical structural variable.
7. Claim 7: The method of claim 1, wherein the model variable analysis in the step (5) uses SPSS to perform corresponding pearson correlation analysis, quantitatively evaluates correlation degrees between 12 laser radar point cloud variables and parameters, analyzes model significance mainly to perform F-test, verifies validity of the selected laser radar variables in an estimated model, determines which variables have the largest prediction contribution to the model, evaluates model prediction capacity, adopts a gaussian-newton iterative method to evaluate model parameter estimation, and finds an optimal solution of the parameters through iteration, adopts a ten-fold cross test method to accurately evaluate model precision in model precision evaluation, and mainly calculates three indexes of a determination coefficient (R2), a relative root mean square error (rRMSE) and an average prediction error (MPE) of the model, determines the fitting capacity of model interpretation observation data by calculating the coefficient (R2), and the higher value range [0,1] indicates stronger model interpretation capacity. The relative root mean square error (rRMSE) is a relative measure of the difference between the observed and model predictions, which analyzes the percentage of estimation accuracy, facilitating comparison between different models and studies. Average predictive error (MPE) measures the degree of deviation of the predicted values, and MPE approaching zero indicates that the model has no systematic deviation. The adaptability test of the model is to verify the generalization capability of the model, a ten-fold cross test method is adopted, and the model is evaluated by adding Mean Absolute Error (MAE) and Mean Square Error (MSE) on the basis of three indexes adopted by the model fitting evaluation of the structure test, wherein the generalization capability of the model can be evaluated by testing the performance of the model on an independent data set. If the model remains high accuracy and low error on the new data set, it can be concluded that the model has good adaptability and can be generalized for wider forestry practice. The corresponding calculation formula of the overall related evaluation index is as follows: ; ; ; *100 Wherein: in order to observe the value of the value, In order to be able to predict the value, As an average value of the observed values, In order to obtain the number of samples, For the t value when the confidence level is α, SEE is the standard error of the predicted value.

Description

Subtropical eucalyptus artificial forest parameter extraction method based on unmanned aerial vehicle laser radar Technical Field The invention belongs to the technical field of intersection of forestry remote sensing and forest resource investigation, and particularly relates to a subtropical eucalyptus artificial forest parameter extraction method based on an unmanned aerial vehicle laser radar, which is suitable for the fields of forest resource accurate investigation, growth situation monitoring, structural parameter analysis, efficient management and sustainable operation of forestry resources and the like of a subtropical area eucalyptus artificial forest. Background Accurate forest parameter estimation can provide key data support for sustainable management and ecological protection of forest resources. The unmanned aerial vehicle laser radar technology has important application value in forest parameter quantitative inversion by virtue of an efficient and flexible operation mode, high-density laser radar point cloud data acquisition capability and high-precision and large-scale acquisition advantages of forest three-dimensional structure data. Although the existing researches have constructed various estimation models aiming at different forest types and parameters and obtain staged results, the following problems still exist: 1. The variable screening method is limited in that the variable selection and the precision performance of the forest parameter estimation model are obviously influenced by specific factors such as the forest stand type, the environmental condition and the like of a research area, and often neglect the multidimensional variable synergistic effect, so that the model has insufficient capture of the three-dimensional structural characteristics of the forest, and the estimation precision is limited; 2. the model test is incomplete, the existing model test depends on simple fitting indexes, lacks multidimensional quantitative evaluation on systematic deviation, generalization capability and statistical significance, is difficult to verify the reliability under complex forest conditions, and is uncontrollable in cross-regional application errors; 3. The special forest stand is not suitable for the shortage of special models of unique structures of subtropical eucalyptus artificial forests, the characteristic variables of the special models are easily ignored when a general model is directly applied, and the estimation precision of core parameters cannot meet the requirement of fine management. Disclosure of Invention 1. Object of the invention The invention provides a method for extracting parameters of a (subtropical) eucalyptus artificial forest based on an unmanned aerial vehicle laser radar, which is characterized in that characteristic variables are extracted from high-density unmanned aerial vehicle laser radar point cloud data, then an exponentiation model is constructed by selecting the variables through an exhaustion method, and the parameters of each model are fitted, evaluated and inspected to select an optimal estimated model structure suitable for the subtropical eucalyptus artificial forest, so that accurate estimation of core forest parameters such as tree height, breast diameter, cross-sectional area, accumulation and the like is realized, the problems of insufficient mining of variable screening synergistic effect, construction of a multi-dimensional model analysis system and lack of dedicated model for parameter estimation of the subtropical eucalyptus artificial Lin Linfen in the prior art are solved, and scientific and effective technical support is provided for fine management and sustainable operation of eucalyptus artificial forest resources. 2. Technical proposal A method for extracting parameters of (subtropical) eucalyptus artificial forests based on unmanned aerial vehicle laser radar comprises the following steps: Step 1, sample plot investigation and data calculation (1) Measuring the breast diameter, dominant tree height, average age, canopy density and the like of the forest in each sample area; (2) Actual forest parameters such as average Diameter (DBH), average height (H), cross-sectional area (BA), stand accumulation (VOL) and the like of the forest in each plot are calculated based on the measured data, wherein the stand accumulation is calculated according to a corresponding different-speed growth equation, and the formula is as follows: wherein: Is the forest stand accumulation (m 3/ha); Dividing the area (m 2/ha) for the forest; Is the average height (m) of the stand; 、、 Is a model parameter; Is an error term. Step 2, obtaining and preprocessing laser radar point cloud data of high-density unmanned aerial vehicle After the steps of denoising filtering, point cloud classification, DEM generation, elevation normalization processing and the like, normalized point cloud data of various patterns are obtained; and using all echoes to respectively