CN110569805A

CN110569805A - Unmanned aerial vehicle image point cloud-based method for extracting individual forest trees and evaluating quality of forest trees

Info

Publication number: CN110569805A
Application number: CN201910855294.9A
Authority: CN
Inventors: 曹林; 李平昊; 徐仲
Original assignee: Nanjing Forestry University
Current assignee: Nanjing Forestry University
Priority date: 2019-09-10
Filing date: 2019-09-10
Publication date: 2019-12-13
Anticipated expiration: 2039-09-10
Also published as: CN110569805B

Abstract

the invention discloses an unmanned aerial vehicle image point cloud-based method for extracting individual trees of an artificial forest and evaluating quality of the individual trees, and belongs to the technical field of forest resource management. Firstly, an unmanned aerial vehicle is used for carrying a multispectral camera to collect multispectral photos, the unmanned aerial vehicle photos with coordinate information are matched by using an SFM technology to generate photogrammetric point clouds, and the photogrammetric point clouds are processed to obtain normalized point cloud data; then, single-tree extraction and precision verification are carried out on the artificial forest by adopting a PCS method based on normalized point cloud; and finally, compiling a place level index table by combining the average height of the dominant trees and the standard age of the sample plot, and evaluating the quality of the artificial forest. The invention can be applied to the fields of forest resource monitoring, ecological factor investigation, biodiversity research and the like, can master the distribution and dynamic change of forest tree species resources, and has important significance on the aspects of forest management and management, ecological environment protection and climate change.

Description

Unmanned aerial vehicle image point cloud-based method for extracting individual forest trees and evaluating quality of forest trees

Technical Field

The invention belongs to the technical field of forest resource management, and particularly relates to an unmanned aerial vehicle image point cloud-based method for extracting artificial forest trees and evaluating quality of forest trees.

Background

The evaluation of the quality of the field is an important application technology basis for realizing scientific and accurate cultivation of the artificial forest. The evaluation of the quality of the forest land provides a basis for selecting better afforestation tree species, formulating proper artificial forest cultivation measures and estimating the wood yield, and meanwhile, efficient and accurate forest land quality evaluation is needed when operation strategies are formulated in a classified mode in forest operation and estimation is made according to forest operation benefit estimation forest cultivation investment. The traditional method for evaluating the quality of the forest often needs a large amount of manual investigation, the forest information is difficult to measure timely and efficiently, a certain limitation exists in the sample size, and the rapidly developed remote sensing technology can meet the requirements of efficient and scientific cultivation of the forest. The application unmanned aerial vehicle platform carries on the sensor is a high accuracy remote sensing technology of rapid development in recent years, and the unmanned aerial vehicle photogrammetry system that the unmanned aerial vehicle platform carried on high resolution camera and constitutes compares in airborne data and laser radar data relatively lower on the cost of acquireing, can carry out the low-speed flight at low altitude, gathers high resolution's image to collect the ground data of forest. Meanwhile, the unmanned aerial vehicle photogrammetry system using the unmanned aerial vehicle carrying camera can efficiently and accurately acquire abundant forest stand and single-tree-scale forest space structure and type information, and provides favorable conditions for realizing stand quality evaluation of the forest stand, thereby providing effective guarantee for realizing accurate cultivation and management of the forest.

In recent years, the study of extracting information of single trees of artificial forests and evaluating quality is as follows: james et al published "Site Index Models for Tree specifices in the Northeaster United States" in reel 63 of Forest Science, 2017, compiled a Site Index table of 22 Tree Species for evaluation of the Site quality of the entire Forest, based on the Site survey data from 1999 to 2013, in conjunction with the Tree height information and age. The research adopts standard field sample actual measurement, manually obtains single-tree data, and uses a relative advantage height method to compile the mountain east China pine tree standing place index table, which can be used for evaluating the quality of mountain east China Pinus nigra and Pinus densiflora artificial forests. However, the above studies have not seen a comprehensive and deep comparison of the geodetic indices of different tree species. The data sources compiled by the ground level index table comprise fixed standard ground data, temporary standard ground data and dominant wood data, wherein the best data source is fixed standard ground data which are continuously observed for many years, but the traditional fixed standard ground data have the problems of huge acquisition workload, time consumption and labor consumption, and accurate data are difficult to obtain in actual work; in addition, ground level index tables based on fixed standard ground data are typical but lack randomness. Generally, crown structure parameters such as breast diameter, tree height, breast height cross-sectional area and the like can be used as indexes for evaluating the quality of the site, but the tree height is less influenced by forest stand density and thinning measures, and data is easy to measure, so that the site level index table is mostly compiled by the tree height in actual work for evaluating the quality of the site. The conventional methods are based on results of manual measurement on fixed standard places, the subjectivity is high, and the data acquisition is time-consuming and labor-consuming.

Disclosure of Invention

The method aims to solve the problems that the method for acquiring the single-tree information and evaluating the quality in the prior art is time-consuming and labor-consuming in acquiring data, too high in cost, strong in subjectivity and not beneficial to large-area popularization and use. In order to solve the technical problems, the invention provides a method for extracting artificial forest trees and evaluating quality of forest trees based on unmanned aerial vehicle image point cloud.

In order to solve the technical problems, the technical scheme adopted by the invention is as follows:

a method for extracting artificial forest trees and evaluating quality of forest trees based on unmanned aerial vehicle image point cloud comprises the steps of firstly, carrying a multispectral camera by an unmanned aerial vehicle to collect multispectral photos, matching the unmanned aerial vehicle photos with coordinate information by applying a Structure From Motion (SFM) technology to generate photogrammetric point cloud data, and processing the photogrammetric point cloud data to obtain normalized point cloud data; then, performing single-tree extraction and precision verification on the artificial forest by adopting a Point Cloud Segmentation (PCS) method based on normalized point cloud data; finally, compiling a place level index table by the average height of the dominant trees and combining the reference age information of the class where the sample plot is located, and performing quality evaluation on the artificial forest; the class here refers to a forest of almost tree species and forest age. The method specifically comprises the following steps:

(1) Carrying out high-resolution camera acquisition of high-resolution image data by means of an unmanned aerial vehicle, and carrying out data preprocessing: matching the unmanned aerial vehicle photo with coordinate information by using an SFM technology to generate photogrammetric point cloud, and processing photogrammetric point cloud data to obtain normalized point cloud data;

(3) extracting the single-wood canopy: extracting the single tree canopy breadth by adopting a distance judgment and segmentation method based on point clouds, wherein the method is used for sequentially segmenting the point clouds from top to bottom, the elevation value of each point in the normalized point clouds represents the height of the point from the ground surface, the highest point in the canopy breadth is regarded as a tree top point, and as the distance between the tree tops is larger than that between the tree crown bottoms, the point clouds of one tree are gathered together and are distinguished from the point clouds of other adjacent trees by utilizing the relative horizontal distance between the point in the point clouds and the tree tops according to the sequence from top to bottom, so that the single tree extraction is completed;

(4) And (5) compiling a status index table: the method comprises the steps of firstly fitting a ground level index guide preselection model to determine a ground level index guide curve, then bringing a tree species reference age into the guide curve to obtain a superior tree height value on the guide curve, and unfolding according to ground level index distances by taking the superior tree height value as a reference to obtain ground level index curves of all levels, so as to compile a ground level index table and perform vertical quality evaluation.

the method for extracting the single tree of the artificial forest and evaluating the quality of the forest based on the unmanned aerial vehicle image point cloud comprises the steps of extracting the single tree of the artificial forest and the quality of the forest.

according to the method for extracting the single trees of the artificial forest and evaluating the quality of the single trees of the artificial forest based on the unmanned aerial vehicle image point cloud, when a high-resolution image is collected, sample plots are selected to be arranged in the orthographic image strips of the artificial forest at different ages, each wood scale is carried out on the sample plots, the average height of the sample plots is calculated, 5 dominant trees are selected in each sample plot, and the average value of the heights of the dominant trees is used as the average height of the dominant trees; the patterns are circular patterns, and the radius of the circular patterns is 15 m.

The method for extracting the artificial forest single trees and evaluating the quality of the forest trees based on the unmanned aerial vehicle image point cloud comprises the following steps of processing image point cloud data to obtain normalized point cloud data: after photogrammetric point cloud data are obtained, space coordinates are given to the image point cloud through a Ground Control Point (GCP) by using an automatic matching technology, then the point cloud is converted into an actual geographic coordinate system, and secondary coordinate correction is carried out on the unmanned aerial vehicle image point cloud by using LiDAR data to obtain a corrected unmanned aerial vehicle photogrammetric point cloud; and then, combining the corrected unmanned aerial vehicle photogrammetry point cloud, and utilizing the ground points of LiDAR data to realize the normalization of the unmanned aerial vehicle photogrammetry point cloud.

According to the method for extracting the individual trees of the artificial forest and evaluating the quality of the stand-up area based on the unmanned aerial vehicle image point cloud, the reference age is that the growth speed of the trees tends to be gentle at a certain age in the process of growing the trees, the quality difference of the stand-up area can be reflected sensitively, and the reference age (A) is taken as the reference age of the tree species₀). The tree species with fast growth rate have a smaller standard age, and the tree species with slow growth rate have a larger standard age.

according to the method for extracting the single trees of the artificial forest and evaluating the quality of the forest based on the unmanned aerial vehicle image point cloud, the elevation value of each point in the point cloud after normalization represents the height of the point from the ground surface, and the highest point in the crown width is regarded as the top point of the tree.

According to the method for extracting the single trees of the artificial forest and evaluating the quality of the forest based on the unmanned aerial vehicle image point cloud, the position index step is determined according to the absolute variation range (delta H) of the tree height of the tree species at the reference age and the forest farm operation level, and the calculation formula is as follows:

C＝ΔH/k

In the formula, C is a position exponential step distance, Delta H is the absolute variation range of the tree height at the reference age, k is the number of exponential steps, the position exponential step distance is 1-4m, and the number of exponential steps is 10.

According to the method for extracting the artificial forest single trees and evaluating the quality of the forest based on the unmanned aerial vehicle image point cloud, the position-level index guide curve is as follows:

ln(H)＝ln(a)+bA

In the formula, A is the average forest age of the forest stand, and a and b are model coefficients.

Has the advantages that: compared with the prior art, the invention has the advantages that:

(1) According to the method, high-precision ground level index tables are compiled by adopting artificial forest tree height information obtained by dividing single trees by using high-resolution images of an unmanned aerial vehicle, combining forest stand age information obtained by a forest farm management file and adopting a standard deviation adjustment method according to the relationship between the tree height and the age; the process has the advantages of easy data acquisition, strong randomness and applicability, and can accurately reflect the growth conditions of the artificial forest stand at all ages and the relationship between the high growth of the dominant trees and the productivity of the forest stand.

(2) according to the method, each tree canopy is accurately segmented from the point cloud of the aerial photogrammetry of the unmanned aerial vehicle through a point cloud segmentation algorithm, so that single tree information is obtained. Unmanned aerial vehicle data acquisition is with low costs, rapid and resolution ratio is high, and the point cloud data possesses three-dimensional structure information simultaneously, can carry out the list wood information acquisition with high accuracy. Because the single-tree information directly contains the accurate information of the height of the single-tree for compiling the ground level index table, the method enhances the precision of the compiled ground level index table and is better applied to quality evaluation. The verification result shows that the geographical index table compiled by the method for the main tree species (the metasequoia and the poplar) in the forest area has better precision, compared with other methods for compiling the geographical index table, the overall precision is improved by more than 5 percent, and the method has good application prospect for evaluating the quality of the standing place.

(3) the method can be applied to the fields of forest resource monitoring, ecological factor investigation, biodiversity research and the like, can master the distribution and dynamic change of forest tree species resources, and has important significance on the aspects of forest management and management, ecological environment protection and climate change relief. The method is not only beneficial to the quality evaluation of a single tree species, but also easy to transplant, namely, the method can be applied to forest stands of different tree species.

Drawings

FIG. 1 is a schematic diagram of the segmentation result and the segmentation accuracy of the single tree in 9 plots of metasequoia forest, and FIGS. 1a, 1b, 1c, 1d, 1e, 1f, 1g, 1h, and 1i are schematic diagrams of the segmentation effect of the single tree in the image point cloud in 1-9 plots of metasequoia respectively;

Fig. 2 is a schematic diagram of the segmentation result and the segmentation accuracy of 9 plots of the poplar forest, and fig. 2a, 2b, 2c, 2d, 2e, 2f, 2g, 2h, and 2i are schematic diagrams of the segmentation effect of the single tree in the image point cloud of the poplar plots 1-9, respectively.

Detailed Description

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with examples are described in detail below.

Example 1

The research area is located in a national Dongtai forest farm in salt cities in Jiangsu province, the annual average air temperature is 14.6 ℃, the relative humidity is 88.3%, the annual rainfall is 1050mm, and the frost-free period is 220 d. The ground elevation of the forest farm in the research area is about 11-14m, and the elevation difference is small. The research area belongs to a typical subtropical monsoon climate area and has the floor area of about 2239hm²The forest coverage rate reaches 85%, and the main tree species of the artificial forest are Metasequoia japonica (Metasequoia glucopyranosides), poplar (Populus deltoids) and the like. From the 80 s of the 20 th century, the China-Yingdongtai forest farm begins to carry out large-scale forest planting, the forest farm carries out tending management on middle and young forests, and in the period, a certain degree of intermediate cutting is carried out to remove inferior trees and leave excellent trees.

And carrying out multispectral camera by means of an unmanned aerial vehicle to collect multispectral photos. According to the east forest farm management archive information, circular sample areas (r is 15m) are set in orthophoto image strips of Chinese fir and poplar of different ages, each sample area is checked to measure the average height of the sample areas, 5 dominant trees are selected from each sample area, and the average value of the tree heights is used as the average tree height of the dominant trees.

During data preprocessing, firstly, matching unmanned aerial vehicle photos with coordinate information through an SFM technology to generate aerial photogrammetry point clouds, then endowing space coordinates to the image point clouds through a Ground Control Point (GCP) by utilizing an automatic matching technology, then converting the point clouds into an actual geographic coordinate system, additionally acquiring LiDAR data of Chinese fir and poplar, verifying LiDAR data of single-tree segmentation precision, and utilizing the LiDAR data to perform secondary coordinate correction on the unmanned aerial vehicle image point clouds. And then combining the corrected unmanned aerial vehicle photogrammetry point cloud, and utilizing ground points of LiDAR data to realize the normalization of the unmanned aerial vehicle photogrammetry point cloud.

And extracting the canopy width of the single tree by adopting a distance judgment and segmentation method based on point cloud. And (4) carrying out single-tree segmentation on the point cloud from high to low by utilizing the relative horizontal distance between the trees, wherein the elevation value of each point in the normalized point cloud represents the height of the point from the ground surface, and the highest point in the crown breadth is regarded as the top point of the tree. Because the spacing of the tops is greater than the spacing of the bottoms of the crowns, the point clouds of one tree are clustered together in order from top to bottom using the relative horizontal distance of the points in the point clouds from the top to be distinguished from the point clouds of other trees adjacent to it. In this way, the extraction of the single-wood crown is finally completed, as shown in fig. 1 and fig. 2.

fig. 1 is a schematic diagram of a segmentation result of a single image point cloud of a metasequoia forest, wherein fig. 1a, 1b, 1c, 1d, 1e, 1f, 1g, 1h, and 1i are schematic diagrams of a segmentation effect of a single image point cloud of 9 metasequoia plots, a black circle represents a fitted single canopy width, a central point is a top detected by single segmentation, and a green triangle is a top detected by LiDAR point cloud data; the single wood detection rate r is 0.83-0.92, the single wood detection accuracy rate p is 0.80-0.94, and the overall precision F is 0.82-0.93; the number N of the single trees is detected in 9 metasequoia sample plots_t399, the number of the single wood is missed₀58, the number of the single wood is excessively divided_cThe single-wood detection rate r is 0.87, the single-wood detection accuracy rate p is 0.88, and the overall accuracy F is 0.87. It can be seen that the overall accuracy of the single-wood segmentation is better.

FIG. 2 is a schematic diagram of a point cloud single tree segmentation result of a poplar forest image, wherein FIGS. 2a, 2b, 2c, 2d, 2e, 2f, 2g, 2h, and 2i are schematic diagrams of a point cloud single tree segmentation effect of 9 poplar sample images respectively, a black circle represents a fitted single tree crown width, a central point is a tree top detected by single tree segmentation, and a green triangle is a single tree top detected by LiDAR point cloud data; wherein the single wood detection rate r is 0.73-0.83, the single wood detection accuracy rate p is 0.70-0.91, and the total precision F is 0.76-087; the number N of the single trees is detected in 9 metasequoia sample plots_t336, the number of single trees is missed₀88, the number of the single wood is excessively divided_cThe single-wood detection rate r is 0.79, the single-wood detection accuracy rate p is 0.84, and the overall accuracy F is 0.81. The overall precision of the single-tree division of the poplar can be seen to be better.

And performing precision verification on the extraction of the single-wood crown, wherein the extraction precision of the single-wood crown is shown in table 1. As can be seen from Table 1, the detection rate, accuracy and overall precision of the single crown of metasequoia are all higher than those of poplar.

TABLE 1 summary table of single-wood crown extraction precision

	Detectivity%	The accuracy rate%	Overall precision%
				Metasequoia glyptostroboides (lour.) Merr	87.0	88.0	87.0
Poplar tree	79.0	84.0	81.0

(4) compiling a place level index table:

a. Guidance curve equation: calculating the average height of the dominant trees of the reference age on each index level by fitting a position-level index guide curve, and forming a position-level index curve cluster, wherein one position-level index curve represents an average tree height growth curve of the dominant tree height of the forest stand along with the change of the forest age under the condition of medium standing;

b. Determination of reference age and place-level index distance: the reference age is generally the reference age of the tree species (a0) when the tree of a certain age grows at a gradual growth rate and the local quality difference can be reflected sensitively. The method comprises the steps of firstly fitting a ground level index guide preselection model to determine a ground level index guide curve, then bringing the tree species reference age into the guide curve to obtain a superior tree height value on the guide curve, and unfolding according to ground level index distances by taking the superior tree height value as a reference to obtain each level of ground level index curves. The position-level index steering curve is as follows:

ln(H)＝ln(a)+bA

generally, the base age of a tree growing at a high rate is small, and the base age of a tree growing at a low rate is large. The rank index distance is generally determined according to the absolute variation range (Delta H) of the tree height of the tree species at a reference age and the level of forest farm management.

the calculation formula is as follows:

C＝ΔH/k

in the formula, C is a position exponential distance, Δ H is an absolute variation range of the tree height at a reference age, k is an exponential number, and generally, the position exponential distance is 1-4m, and the exponential number is about 10. The positions of the fir and the poplar in the example are in exponential order, and the exponent of C is 6, 8, 10, 12, 14, 16, 18, 20, 22 and 24, wherein the exponent of C is 2 m. The location-level indices for the two tree species are shown in tables 2 and 3.

TABLE 2 land level index table of metasequoia

TABLE 3 Poplar ground level exponent table

Claims

1. A method for extracting artificial forest trees and evaluating quality of forest trees based on unmanned aerial vehicle image point cloud is characterized in that firstly, an unmanned aerial vehicle is used for carrying a multispectral camera to collect multispectral photos, SFM technology is applied to match the unmanned aerial vehicle photos with coordinate information to generate photogrammetric point cloud data, and the photogrammetric point cloud data are processed to obtain normalized point cloud data; then, single-tree extraction and precision verification are carried out on the artificial forest by adopting a PCS method based on normalized point cloud; and finally, compiling a place level index table by the average height of the dominant trees and combining the reference age information of the class where the sample plot is located, and performing quality evaluation on the artificial forest.

2. The method for extracting the artificial forest single trees and evaluating the quality of the forest stand based on the unmanned aerial vehicle image point cloud according to claim 1, which is characterized by comprising the following steps:

(1) Carrying a high-resolution camera by means of an unmanned aerial vehicle to acquire high-resolution image data, and carrying out remote sensing data preprocessing: matching the unmanned aerial vehicle photos with coordinate information by using an SFM technology to generate photogrammetric point cloud, and processing photogrammetric point cloud data to obtain normalized point cloud data;

(4) And (5) compiling a status index table: the method comprises the steps of firstly fitting a ground level index guide preselection model to determine a ground level index guide curve, then bringing a tree species reference age into the guide curve to obtain a dominant tree height value on the guide curve, taking the dominant tree height value as a reference, expanding according to ground level index distances to obtain ground level index curves of all levels, and accordingly compiling a ground level index table to perform artificial forest quality evaluation.

3. The method for extracting single trees and evaluating quality of artificial forests based on the unmanned aerial vehicle image point cloud according to claim 1 or 2, wherein the artificial forests are Chinese fir and poplar.

4. The method for extracting single trees of artificial forests and evaluating quality of standing trees based on the unmanned aerial vehicle image point cloud according to claim 1 or 2, wherein during the acquisition of high-resolution images, sample plots are selected to be arranged in each orthoimage strip of the artificial forests of different ages, each sample plot is subjected to scale detection, the average height of the sample plots is calculated, 5 dominant trees are selected in each sample plot, and the average value of the heights of the dominant trees is used as the average height of the dominant trees; the patterns are circular patterns, and the radius of the circular patterns is 15 m.

5. The method for extracting the artificial forest single trees and evaluating the quality of the forest trees based on the unmanned aerial vehicle image point cloud according to claim 1 or 2, wherein the process of processing the image point cloud data to obtain normalized point cloud data is as follows: acquiring aerial photogrammetry point cloud data; then, giving a space coordinate to the image point cloud by using an automatic matching technology through a ground control point, converting the point cloud into an actual geographic coordinate system, and performing secondary coordinate correction on the unmanned aerial vehicle image point cloud by using LiDAR data to obtain a corrected unmanned aerial vehicle photogrammetric point cloud; and then, combining the corrected unmanned aerial vehicle photogrammetry point cloud, and utilizing the ground points of the LiDAR data to realize the normalization of the unmanned aerial vehicle photogrammetry point cloud. The elevation value of each point in the normalized point cloud represents the height of the point from the ground surface, and the highest point in the crown is considered as the top point of the tree.

6. The method for extracting the artificial forest single trees and evaluating the quality of the forest stand based on the unmanned aerial vehicle image point cloud according to claim 1 or 2, wherein the reference age is the age at which the tree grows gradually at a certain age in the process of tree growth, and the difference of the quality of the forest stand can be reflected sensitively, and the age is taken as the reference age of the tree species.

7. the method for extracting and evaluating quality of single trees in artificial forest based on unmanned aerial vehicle image point cloud according to claim 1 or 2, wherein the elevation value of each point in the normalized point cloud represents the height of the point from the ground surface, and the highest point within a certain threshold is regarded as the top point of the tree.

8. The method for extracting and evaluating quality of single trees in artificial forest based on unmanned aerial vehicle image point cloud according to claim 2, wherein the index grade of the position is determined according to the absolute variation range (Δ H) of the tree height of the tree species at the reference age and the operating level of the forest farm, and the calculation formula is as follows:

C＝ΔH/k

9. The method for extracting and evaluating quality of artificial forest trees based on the unmanned aerial vehicle image point cloud according to claim 2, wherein the geostationary index guide curve is as follows:

ln(H)＝ln(a)+bA