Forest crown lightweight three-dimensional modeling method based on unmanned aerial vehicle oblique photography

Abstract

The invention relates to the technical field of remote sensing three-dimensional modeling, in particular to a forest crown lightweight three-dimensional modeling method based on unmanned aerial vehicle oblique photography, and the method comprises the following steps: controlling an unmanned aerial vehicle to execute fixed-point multi-angle flight shooting; according to the method, the comprehensive point cloud data is generated through pixel matching calculation, and the accuracy and practicability of the data are improved. The point cloud classification is realized based on the point cloud three-dimensional coordinates and the pixel reflectivity, the crown point cloud and the ground point cloud are distinguished, the generation of the classification point cloud further accurately distinguishes the crown point cloud and the non-crown point cloud, and the accuracy and the efficiency of data processing are improved. In addition, the boundary of the individual tree is determined through the spatial aggregation of the point cloud, the point cloud segmentation of the crown of the individual tree is realized, the envelope construction of the crown of the individual tree is realized, and a lightweight individual tree model is generated. Through the lightweight tree crown three-dimensional modeling method, the forest refined monitoring efficiency is improved under the condition that certain forest monitoring precision is guaranteed, and therefore forest management and ecological research are better served.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing three-dimensional modeling, and particularly to a lightweight three-dimensional modeling method for forest tree crowns based on unmanned aerial vehicle (UAV) oblique photography. Background Art

[0002] A UAV equipped with a photographic device is used to obtain image data of a forest from multiple angles. The main purpose of this method is to generate a three-dimensional model of the forest area, which is convenient for canopy structure analysis, ecological research, and environmental monitoring. Through the oblique photography technology, detailed information of trees can be captured from different perspectives.

[0003] Although the existing technology can obtain forest image data from multiple angles to generate a three-dimensional model, in the face of the complexity of the forest environment, the lack of continuous data acquisition in the existing methods easily leads to the loss or incompleteness of information, which forms a shortcoming in the detail and accuracy of the model. In terms of data processing, the lack of a classification mechanism and error information screening makes the model contain more noise and errors, affecting the reliability and practicality of the model, thereby reducing the application value of the technology and the accuracy of decision support. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings existing in the prior art, and a lightweight three-dimensional modeling method for forest tree crowns based on UAV oblique photography is proposed.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions. A lightweight three-dimensional modeling method for forest tree crowns based on UAV oblique photography includes the following steps: Control the UAV to perform fixed-point multi-angle flight shooting, collect continuous image data of a specified area, match the pixels at the same location between each image data one by one, and generate comprehensive point cloud data through pixel matching calculation; Load the comprehensive point cloud data, analyze the coordinates and reflectivity of each data point one by one, distinguish the tree crown and non-tree crown point clouds according to a preset threshold to generate classified point clouds; screen the abnormal reflectivity values in the classified point clouds, exclude non-tree crown data points, and retain the tree crown point cloud data; Analyze the tree crown point cloud data, determine the boundary of individual trees through the spatial aggregation of the point cloud, separate the point cloud of individual trees, and generate the point cloud of individual trees; perform boundary closing processing on the point cloud of individual trees, and adjust the closing parameters according to the point cloud density to form the contour data of individual trees; Simplify the contour data of individual trees, remove redundant point clouds, adjust the precision and quantity of data points, construct a three-dimensional envelope of the individual tree crown, and generate a lightweight individual tree model.

[0006] The acquisition step of the comprehensive point cloud data is as follows: The drone performs fixed-point multi-angle flight shooting to collect continuous image data within the area and generates an original image dataset; Extract the pixels of each image from the original image dataset, and calculate the gray-scale difference of the pixels at the same location in adjacent images. The calculation formula is: ; Among them, is the pixel gray-scale difference, and are the pixel gray-scale values at the same position in two images respectively; Based on the pixel gray-scale difference, aggregate similar pixel points to form comprehensive point cloud data.

[0007] The steps for obtaining the classified point cloud are as follows: Load the comprehensive point cloud data, extract the coordinates and reflectivity of each data point to obtain the original point cloud feature data; Based on the original point cloud feature data, calculate the reflectivity deviation of each point. The calculation formula is: ; Among them, is the reflectivity deviation, is the reflectivity of the current point, is the average value of the reflectivities of all points, is the standard deviation of the reflectivities of all points; According to the reflectivity deviation and the preset threshold , classify each point into canopy and non-canopy. If , it is canopy point cloud, otherwise it is non-canopy point cloud, forming classified point cloud data.

[0008] The steps for obtaining the canopy point cloud data are as follows: Load the classified point cloud, perform statistical analysis on the classified point cloud, determine the reflectivity value of each point cloud, and compare it with the preset abnormal reflectivity threshold to identify the points with abnormally high or low reflectivity, obtaining abnormal reflectivity point cloud data; Based on the abnormal reflectivity point cloud data, remove the data points of non-canopy, and at the same time remove the point cloud data marked as canopy but with abnormal reflectivity, retaining the canopy point cloud data that conforms to the canopy characteristics and has normal reflectivity, forming canopy point cloud data.

[0009] The steps for obtaining the individual tree point cloud are as follows: Load the canopy point cloud data; Based on the canopy point cloud data, calculate the average distance from each point in the canopy point cloud data to the nearest neighbor point , and the calculation formula is: ; Among them, represents the total number of points in the point cloud, is the coordinate of the th point, is the coordinate of the nearest neighbor point of the th point; Based on the average distance and a preset distance threshold, the points close to each other in the canopy point cloud dataset are clustered into groups, and each group represents an independent tree, forming a single-tree point cloud.

[0010] The steps for obtaining the single-tree contour data are as follows: Obtain the single-tree point cloud, calculate the spatial range and volume for the point cloud data of each tree, and obtain the spatial parameters of each tree; Based on the spatial parameters of each tree, calculate the point cloud density , and the calculation formula is: ; Among them, represents the number of points in the single-tree point cloud, represents the total volume within the spatial range of the point cloud; Use the point cloud density to analyze and adjust the boundary closing parameters of each tree, define the three-dimensional contour of each tree, and form the single-tree contour data.

[0011] The steps for obtaining the lightweight single-tree model are as follows: Based on the single-tree contour data, adjust the point cloud accuracy and quantity point by point, and the formula is: ; Calculate the new accuracy of each point, where is the accuracy of the current point, is the average value of the accuracies of all points, is the total number of points.

[0012] Preferably, the steps for obtaining the lightweight single-tree model further include: according to the new accuracy of each point, reorganize the point cloud data to generate a lightweight single-tree model.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The present invention generates comprehensive point cloud data through pixel matching calculation, improving the accuracy and practicality of the data. Based on the three-dimensional coordinates of the point cloud and the pixel reflectivity, point cloud classification is realized to distinguish between tree crown point clouds and ground point clouds. The generation of classified point clouds further accurately distinguishes between tree crown and non-tree crown point clouds, improving the accuracy and efficiency of data processing. In addition, the single-tree boundary is determined through the spatial aggregation of the point cloud, the single-tree crown point cloud is segmented, and the single-tree crown envelope is constructed to generate a lightweight single-tree model. Through the lightweight three-dimensional modeling method of the tree crown, the efficiency of fine forest monitoring is improved while ensuring a certain forest monitoring accuracy, so as to better serve forest management and ecological research. Brief Description of the Drawings

[0014] Figure 1 It is a schematic diagram of the steps of the present invention. Detailed Embodiment

[0015] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0016] Please refer to Figure 1 , the present invention provides a technical solution, a lightweight three-dimensional modeling method for forest tree crowns based on UAV oblique photography, including the following steps: Control the UAV to perform fixed-point multi-angle flight shooting, collect continuous image data of a specified area, match the pixels at the same location between each image data one by one, and generate comprehensive point cloud data through pixel matching calculation; Load the comprehensive point cloud data, analyze the coordinates and reflectivity of each data point one by one, distinguish between tree crown and non-tree crown point clouds according to a preset threshold, and generate classified point clouds; screen the abnormal reflectivity values in the classified point clouds, exclude non-tree crown data points, and retain the tree crown point cloud data; Analyze the tree crown point cloud data, determine the single-tree boundary through the spatial aggregation of the point cloud, separate the point cloud of independent trees, and generate single-tree point clouds; perform boundary closing processing on the single-tree point clouds, and adjust the closing parameters according to the point cloud density to form single-tree contour data; Simplify the single-tree contour data, remove redundant point clouds, adjust the accuracy and quantity of data points, construct a three-dimensional envelope of the single-tree crown, and generate a lightweight single-tree model.

[0017] Specifically, the comprehensive point cloud data is generated through pixel matching calculation, which improves the accuracy and practicality of the data. Based on the three-dimensional coordinates of the point cloud and the pixel reflectivity, the point cloud classification is realized to distinguish the crown point cloud and the ground point cloud. The generation of the classified point cloud further accurately distinguishes the crown and non-crown point clouds, improving the accuracy and efficiency of data processing. In addition, the single-tree boundary is determined through the spatial aggregation of the point cloud, the single-tree crown point cloud segmentation is realized, and the single-tree crown envelope is constructed to generate a lightweight single-tree model. Through the lightweight three-dimensional modeling method of the crown, the efficiency of forest fine monitoring is improved while ensuring a certain forest monitoring accuracy, so as to better serve forest management and ecological research.

[0018] The steps for obtaining the comprehensive point cloud data are as follows: The drone performs fixed-point multi-angle flight shooting to collect continuous image data within the area and generate an original image dataset; Extract the pixels of each image from the original image dataset, and calculate the gray difference of the pixels at the same location in adjacent images. The calculation formula is: ; where is the pixel gray difference, and are the pixel gray values at the same position in the two images respectively; Based on the pixel gray difference, aggregate the same type of pixel points to form comprehensive point cloud data.

[0019] Specifically, when performing the drone flight shooting task, the operator first sets the flight path of the drone according to the task schedule to ensure that the designated forest area can be fully covered. The camera carried by the drone automatically adjusts shooting parameters such as aperture, shutter speed, and ISO according to the flight altitude and area characteristics to adapt to different lighting and weather conditions. The flight position and attitude are corrected in real time through GPS and gyroscopes to ensure the continuity and overlap of the images. The quality of these original image data directly affects the subsequent pixel matching and point cloud generation processes. After the image collection is completed, the data is sent to the ground station through wireless transmission for preliminary quality inspection and data backup to prepare for the next processing.

[0020] The beneficial effect of the formula is that by calculating the gray difference of the pixels at the same location in adjacent images, the subtle changes between images can be accurately measured, which is crucial for pixel matching and generating high-precision point cloud data; the steps for obtaining the gray difference are as follows: first, extract the gray values and of the corresponding pixels in adjacent images, and then calculate the difference between them; Calculation process: The gray values of two adjacent pixel points are respectively and , then: ; This result indicates that the pixel gray - level difference between the two images at the same position is 10, which means there is a certain visual change and can be used for further point - cloud generation and 3D model construction.

[0021] After obtaining the gray - level difference data, the aggregation process of similar pixel points begins. First, a gray - level difference threshold is set, usually based on previous test data or empirical values. For example, if the gray - level difference between two pixel points is less than 10, they are considered part of the same object. The aggregation algorithm traverses the entire gray - level difference data set, evaluates each pixel point, and marks the points that meet the conditions as the same object. In this way, the algorithm can gradually construct a 3D point - cloud representation of the object. These point - cloud data very detailedly reflect the geometric features and structural details of the object surface, providing high - precision basic data for the final 3D modeling. These refined point - cloud data will then be used for further analysis and model optimization to ensure the accuracy and practicality of the final model.

[0022] The steps to obtain classified point clouds are as follows: Load the comprehensive point - cloud data, extract the coordinates and reflectivity of each data point to obtain the original point - cloud feature data; Based on the original point - cloud feature data, calculate the reflectivity deviation of each point. The calculation formula is: ; where, is the reflectivity deviation, is the reflectivity of the current point, is the average value of the reflectivities of all points, is the standard deviation of the reflectivities of all points; According to the reflectivity deviation and the preset threshold , classify each point into canopy and non - canopy. If , it is a canopy point cloud, otherwise it is a non - canopy point cloud, forming classified point - cloud data.

[0023] Specifically, during the process of loading the comprehensive point - cloud data, first, it is necessary to obtain the point - cloud files stored on the cloud server through the data interface. These files contain the 3D coordinates and reflectivity of each data point. According to the coordinates and reflectivity of each data point.

[0024] The advantage of formula is that by introducing the standard deviation of the reflectivity and the constant 0.1, it increases the stability of the calculation, while improving the sensitivity to reflectivity variation and making the classification more accurate; The parameter represents the reflectivity of the current point, which is obtained through the reflectivity field in the point cloud data. The parameter represents the average value of the reflectivities of all points, which is calculated by summing the reflectivity values of all point cloud data and then dividing by the number of points. The parameter is the standard deviation of the reflectivities of all points, which is obtained by calculating the sum of the squares of the differences between the reflectivity of each point and the average value, and then taking the square root after dividing by the number of points.

[0025] Calculation process: The reflectivities of three points are 0.15, 0.20, and 0.25 respectively. Calculate as follows: ; The calculation is as follows: ; For the point with a reflectivity of 0.15, . The result shows that the reflectivity deviation value is less than 1. If the threshold is set to 1, this point is classified as non-tree crown point cloud; if is set lower, it may be classified as tree crown point cloud, indicating that this formula can effectively distinguish point clouds with different reflectivity levels.

[0026] According to the reflectivity deviation data and the preset threshold , the classification process of each point into tree crown and non-tree crown involves comparing the reflectivity deviation value of each point with the threshold . If is greater than , then this point is marked as a tree crown point, otherwise it is marked as a non-tree crown point. This process should also include looping through each point cloud data point to ensure that each point is accurately classified. The classified data will be used to generate a new point cloud data set containing the marked tree crown and non-tree crown points, providing basic data for subsequent analysis and processing; the threshold T can be determined using an iterative method; for example, it can start with a preliminary threshold and, based on the classification results (such as the number of tree crown point clouds and non-tree crown point clouds or other features), adjust the threshold until the number of misclassifications is minimized or the accuracy of a certain classification is maximized.

[0027] The steps to obtain the tree crown point cloud data are as follows: Load the classified point cloud, perform statistical analysis on the classified point cloud, determine the reflectivity value of each point cloud, and compare it with the preset abnormal reflectivity threshold to identify points with abnormally high or low reflectivity, obtaining the abnormal reflectivity point cloud data; Based on the abnormal reflectance point cloud data, remove the data points that are not tree crowns, and at the same time remove the point cloud data that is marked as tree crown but has abnormal reflectance. Retain the tree crown point cloud data that conforms to the tree crown characteristics and has normal reflectance to form the tree crown point cloud data.

[0028] Specifically, after loading the classified point cloud data, first perform a comprehensive statistical analysis of these data. This analysis involves the extraction and recording of the reflectance values of each point cloud, which is completed through data traversal operations. The reflectance value of each data point is compared with a preset abnormal reflectance threshold, which is set based on historical data and expert knowledge and is designed to capture possible data anomalies, such as too high or too low reflectance. All the point cloud data identified as abnormal will be aggregated together to form a separate abnormal reflectance point cloud data set.

[0029] Based on the abnormal reflectance point cloud data obtained previously, screening and cleaning operations are carried out. The goal is to remove all non-tree crown data points marked with abnormal reflectance, and at the same time, also remove those point cloud data that are marked as tree crowns but have abnormal reflectance values. Filter out the point clouds that do not meet the tree crown characteristics through the set logical checks, and retain the data points that conform to the tree crown characteristics and have reflectance values within the normal range to form the optimized tree crown point cloud data.

[0030] The steps to obtain the single-tree point cloud are as follows: Load the optimized tree crown point cloud data; Based on the tree crown point cloud data, calculate the average distance from each point in the tree crown point cloud data to its nearest neighbor point , and the calculation formula is: ; Where represents the total number of points in the point cloud, is the coordinate of the th point, is the coordinate of the nearest neighbor point of the th point; Based on the average distance and a preset distance threshold, cluster the points that are close to each other in the tree crown point cloud data set. Each cluster represents an independent tree to form the single-tree point cloud.

[0031] Specifically, the advantage of the formula is that by calculating the average distance from each point to its nearest neighbor point, it effectively measures the spatial aggregation of the point cloud data, which helps to more accurately identify and define the boundaries of each independent tree; Calculation process: Assume that there are 500 points in the point cloud data, evenly distributed, and the coordinates of each point are randomly generated. Calculate the average neighbor distance of these points. For example, the coordinate of point 1 is (10, 10, 10), and its nearest neighbor point is (11, 10, 12). Substitute into the formula for calculation: ; And so on, calculate the average neighbor distance of all points and find the average value; this result indicates that if the average neighbor distance is less than a preset threshold, such as 3 meters, it indicates that these point clouds are likely to belong to a single tree crown; Further cluster the point clouds in the tree crown point cloud dataset based on their average neighbor distances. By calculating the spatial positions of each point cloud, find the set of point clouds that are close to each other and regard them as an independent tree entity. For those point clouds that are far away, it is determined that they do not belong to the current tree's point clouds and are excluded.

[0032] The steps for obtaining the single-tree contour data are as follows: Obtain the single-tree point cloud, calculate the spatial range and volume for the point cloud data of each tree to obtain the spatial parameters of each tree; Calculate the point cloud density based on the spatial parameters of each tree , and the calculation formula is: ; where, represents the number of points in the single-tree point cloud, represents the total volume within the spatial range of the point cloud; Use the point cloud density to analyze and adjust the boundary closure parameters of each tree, define the three-dimensional contour of each tree, and form the single-tree contour data.

[0033] Specifically, obtain the single-tree point cloud dataset from the previous steps. After extracting the three-dimensional coordinate information of the point cloud data point by point, perform spatial range analysis. Determine the minimum bounding box of the entire point cloud in three-dimensional space according to the coordinates of each point. Further use the three-dimensional convex hull method to calculate the minimum geometric volume of the point cloud. Construct the convex hull surface through point-by-point iteration, and at the same time calculate the formed closed geometric volume. Count the total number of points in the point cloud data, accumulate the number of all points in an iterative traversal manner and generate the point number data. Combine the two data of the convex hull volume and the number of points to finally form a single-tree point cloud dataset suitable for further density analysis.

[0034] In the formula , the acquisition step of is to count the number of points in the point cloud dataset of each tree, and use the iterative accumulation method to count the points, the acquisition step of is to use three-dimensional convex hull analysis to calculate the minimum geometric volume of the point cloud. This method accumulates the internal space volume after generating the convex hull surface through point-by-point nested analysis; ; The result shows that the density of the point cloud is 50 points per cubic unit. The high-density point cloud indicates that the point distribution of the tree is relatively concentrated, which is conducive to accurately defining the boundary closing parameters.

[0035] Using the calculated point cloud density, density threshold analysis is performed on the point cloud of each tree. According to the point cloud density value, the point regions with more concentrated distribution are screened out, and the spatial range of the point cloud is redefined. Each point in the point cloud data is checked to see if it meets the density threshold requirement. The density threshold is adjusted empirically. Points that do not meet the density threshold are removed one by one. The boundary range of the point cloud is adjusted recursively, and the remaining point data is limited to the region with higher density. According to the finally limited point cloud range, the boundary closing parameters are gradually shrunk. At the same time, during the adjustment of the closed boundary, it is checked whether the remaining point cloud forms a continuous and closed spatial range until the closing parameters are completely adjusted and a complete three-dimensional single-tree point cloud contour is generated, forming single-tree contour data.

[0036] The steps to obtain the lightweight single-tree model are as follows: Based on the single-tree contour data, the accuracy and quantity of the point cloud are adjusted point by point. The formula is: ; Calculate the new accuracy of each point , where is the accuracy of the current point, is the average value of the accuracies of all points, is the total number of points;

[0037] According to the new accuracy of each point, the point cloud data is reorganized to generate a lightweight single-tree model.

[0038] Specifically, when simplifying the single-tree contour data, through on-site measurement and high-precision laser scanning technology, the position and attributes of each data point are obtained to ensure that the integrity and accuracy of the data are not damaged during the optimization process. Then, a spatial data structure such as KD-tree is used to organize the point cloud for quick access and processing of the data in the point cloud. By setting a tolerance threshold to determine which point cloud can be considered redundant, this threshold is set based on the statistical data of the point cloud density and the distance between points. By comparing each point with its neighboring points one by one, points with low contribution to the main structure are deleted. This not only reduces the data volume but also improves the processing speed and data manageability, obtaining optimized single-tree contour data.

[0039] The beneficial effect of the formula is that by calculating the new accuracy of each point cloud data point, the accuracy and quantity of data processing are optimized, making the model more in line with the streamlined degree of actual requirements; in the formula, is the adjusted accuracy of the data point, is the accuracy of the current point, is the average value of the precision of all points, is the total number of points, and the values of these parameters are obtained through statistical analysis of the actual point cloud data. For example, if Z = 500, , then each will be adjusted according to the actual measurement of the point cloud data; Formula calculation process: There is a point cloud containing 500 data points, and the original precision of each point is obtained through measurement, and the average precision is 0.5mm. By summing the squares of the differences between the precision of each point and the average value and then dividing by the total number of points Z, a new precision is obtained, such as: ; By substituting specific numerical values, can be calculated; this result indicates that through precise adjustment and calculation, the data precision and size of the model can be effectively controlled, making the model more suitable for subsequent processing and applications. For the step result, it means efficient data processing and storage optimization.

[0040] Based on the adjusted point cloud precision and quantity, the process of reorganizing the point cloud data involves reassigning the attributes of each point, which includes but is not limited to updating the three-dimensional coordinates, reflection intensity, and color information of the points. By comparing the original data and the adjusted data, the effect of the precision adjustment is evaluated to ensure that the physical and visual attributes are maintained after the data is adjusted. Using octree decomposition to reduce the collective volume of the data while maintaining the integrity of the structure, effectively simplifying the data processing flow and improving the data processing efficiency. The finally generated lightweight single-tree model retains the necessary geometric and visual features and is suitable for rapid rendering and analysis.

[0041] The above is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A lightweight 3D modeling method for forest canopy based on drone oblique photography, characterized in that: The following steps are involved: Control the drone to perform fixed-point multi-angle flight shooting, collect continuous image data of the specified area, match the pixels of the same location between each image data one by one, and generate comprehensive point cloud data through pixel matching calculation; The comprehensive point cloud data is loaded, the coordinates and reflectivity of each data point are analyzed point by point, the crown and non-crown point clouds are distinguished according to a preset threshold value, and a classified point cloud is generated; abnormal reflectivity values ​​in the classified point cloud are screened, non-crown data points are excluded, and the crown point cloud data is retained; Analyze the tree crown point cloud data, determine the single tree boundary through the spatial aggregation of the point cloud, separate the point cloud of the independent tree, and generate the single tree point cloud; perform boundary closing processing on the single tree point cloud, adjust the closing parameters according to the point cloud density, and form the single tree outline data; The single tree outline data is simplified, redundant point clouds are removed, the accuracy and quantity of data points are adjusted, a three-dimensional envelope of the single tree crown is constructed, and a lightweight single tree model is generated.

2. The method for lightweight 3D modeling of forest canopy based on oblique photography by unmanned aerial vehicle according to claim 1, characterized in that: The steps for obtaining the comprehensive point cloud data are as follows: The drone performs fixed-point multi-angle flight photography, collects continuous image data in the area, and generates a raw image data set; The pixels of each image are extracted from the original image data set, and the grayscale difference of the pixels at the same location in adjacent images is calculated. The calculation formula is: ； in, is the pixel grayscale difference, and are the grayscale values ​​of pixels at the same position in the two images; Based on the pixel grayscale difference, similar pixel points are aggregated to form comprehensive point cloud data.

3. The method for lightweight 3D modeling of forest canopy based on oblique photography by unmanned aerial vehicle according to claim 1, characterized in that: The steps for obtaining the classified point cloud are: Loading the comprehensive point cloud data, extracting the coordinates and reflectivity of each data point, and obtaining original point cloud feature data; Based on the original point cloud feature data, the reflectivity deviation of each point is calculated using the following formula: ； in, is the reflectivity deviation, is the reflectivity of the current point, is the average reflectivity of all points, is the standard deviation of reflectivity at all points; According to the reflectivity deviation and the preset threshold , classify each point into crown and non-crown. If , then it is a crown point cloud, otherwise it is a non-crown point cloud, forming classified point cloud data.

4. The method for lightweight 3D modeling of forest canopy based on oblique photography by unmanned aerial vehicle according to claim 1, characterized in that: The steps for obtaining the tree crown point cloud data are as follows: Loading the classified point cloud, performing statistical analysis on the classified point cloud, determining the reflectivity value of each point cloud, and comparing it with a preset abnormal reflectivity threshold, thereby identifying abnormally high or abnormally low reflectivity points, and obtaining abnormal reflectivity point cloud data; Based on the abnormal reflectivity point cloud data, non-crown data points are removed, and point cloud data identified as crowns but with abnormal reflectivity are removed, and crown point cloud data that meets the crown characteristics and has normal reflectivity is retained to form crown point cloud data.

5. The method for lightweight 3D modeling of forest canopy based on oblique photography by unmanned aerial vehicle according to claim 1, characterized in that: The steps for obtaining the single tree point cloud are as follows: Loading the tree crown point cloud data; Based on the tree crown point cloud data, calculate the average distance from each point in the tree crown point cloud data to the nearest neighbor point , the calculation formula is: ； in, Represents the total number of points in the point cloud, It is The coordinates of the points, It is The coordinates of the nearest neighbor of a point; Based on the average distance and a preset distance threshold, points close to each other in the tree crown point cloud dataset are clustered into groups, each group represents an independent tree, and a single tree point cloud is formed.

6. The method for lightweight 3D modeling of forest canopy based on oblique photography by unmanned aerial vehicle according to claim 1, characterized in that: The steps for obtaining the single tree outline data are: Obtaining the single tree point cloud, calculating the spatial range and volume for the point cloud data of each tree, and obtaining the spatial parameters of each tree; Based on the spatial parameters of each tree, calculate the point cloud density , the calculation formula is: ； in, represents the number of points in a single tree point cloud, Represents the total volume within the point cloud space; The point cloud density is used to analyze and adjust the boundary closure parameters of each tree, define the three-dimensional outline of each tree, and form single tree outline data.

7. The method for lightweight 3D modeling of forest canopy based on oblique photography by unmanned aerial vehicle according to claim 1, characterized in that: The steps for obtaining the lightweight single wood model are as follows: Based on the single tree outline data, the point cloud accuracy and quantity are adjusted point by point. The calculation formula is: ； Calculate the new precision for each point ,in is the accuracy of the current point, is the average accuracy of all points, is the total number of points.

8. The method for lightweight three-dimensional modeling of forest canopy based on oblique photography by unmanned aerial vehicle according to claim 7, characterized in that: The step of acquiring the lightweight single wood model also includes: reorganizing the point cloud data according to the new precision of each point to generate the lightweight single wood model.

Images