Method for constructing three-dimensional measurement model of forest sample plot by using image

Abstract

The application uses images to construct a forest sample plot measurement model. The three-dimensional reconstruction of images obtained by using a smart phone has the advantages of low measurement cost and simple actual operation suitable for field operation. The improved DBSCAN algorithm improves the accuracy of clustering, so that the constructed three-dimensional measurement model can be used for obtaining forest structure parameters, and the measurement accuracy of the structure parameters is much higher than the national forest second-class survey standard.

Description

TECHNICAL FIELD

[0001] The application applies artificial intelligence technology and proposes a method for constructing a three-dimensional measurement model of a sample plot of a forest stand by using images, aiming to meet the demand for accurate measurement of forest stand structure parameters in the fields of forestry carbon sink measurement, forest resource information science, and ecosystem research. BACKGROUND

[0002] A sample plot of a forest stand is essentially a small piece of forest, usually with a size of 25m x 25m, and the trees in the sample plot are usually representative.

[0003] Forest stand structure parameters, including tree height, diameter at breast height, crown width, forest layer, leaf area index, density, and volume, are used as input quantities for ecological models such as ecological physics models, global change models of ecosystem function, and allometric models of stand volume, to study the laws of energy flow and material circulation in ecosystems.

[0004] Among them, the diameter at breast height refers to the diameter of the tree trunk at 1.3 meters from the ground, and the tree height refers to the distance between the rootstock and the tree top. The lack of direct and accurate measurement conditions for structure parameters other than diameter at breast height and tree height is the main reason for the large error and low efficiency of forest stand structure parameter extraction. Therefore, constructing a high-precision three-dimensional measurement model of a forest stand is the only way to realize the visualization analysis of forest stand structure and improve the extraction accuracy of forest stand structure parameters.

[0005] Currently, unmanned aerial vehicles or ground-based laser scanners are widely used to obtain three-dimensional point cloud data of the outer contour of trees. This measurement method is limited in its widespread application due to factors such as expensive equipment and inconvenience in carrying the equipment. With the continuous improvement of the performance of image acquisition equipment and the development of computer vision technology, three-dimensional reconstruction of the outer contour of the measurement target has begun to appear, generating point clouds from the surface data of the target. Therefore, the present application proposes to generate three-dimensional point clouds of a sample plot of a forest stand by using images, to realize single-tree segmentation by improving the DBSCAN algorithm, and to construct a three-dimensional measurement model of a sample plot of a forest stand by combining tree height and crown width data.

[0006] DBSCAN is a representative density-based clustering algorithm, which has obvious advantages in clustering point clouds with high density. However, the selection of the scanning radius Eps parameter of the DBSCAN algorithm is set by humans, which makes the clustering results unable to meet the measurement requirements. The present application improves the DBSCAN algorithm to realize a method for automatically obtaining the scanning radius Eps parameter, which can effectively improve the clustering accuracy and maximize the restoration of tree structure.

[0007] This invention utilizes images to construct a forest stand measurement model. Using smartphones to acquire images for 3D reconstruction offers advantages such as low measurement cost and ease of operation, making it suitable for fieldwork. An improved DBSCAN algorithm enhances clustering accuracy, enabling the constructed 3D measurement model to be used for obtaining forest stand structure parameters. The measurement accuracy of these parameters far exceeds the national Class II forest survey standard. Summary of the Invention

[0008] This invention discloses a method for constructing a three-dimensional measurement model of a forest stand plot using images. The method involves acquiring forest stand plot data through video recording using photographic equipment, converting the video recordings into images by sampling at 25-frame intervals, and inputting the images into the Pix4Dmapper software to obtain a three-dimensional point cloud. The point cloud is then segmented into individual tree point clouds using an optimized DBSCAN clustering algorithm. Combining the measured tree height and diameter at breast height (DBH) of individual trees, the vertical and horizontal scaling factors between the three-dimensional forest stand model and the forest stand measurement model are calculated. Based on these scaling factors, the point cloud coordinates are adjusted to construct the three-dimensional measurement model of the forest stand plot. The steps are as follows:

[0009] Step 1: Data Acquisition: Use a smartphone to acquire a video recording containing a panoramic view of the forest stand plot, set the sampling interval to 25 frames to obtain the forest stand plot image, and input it into the Pix4Dmapper software to obtain a 3D point cloud; measure the height and diameter at breast height (DBH) of a single tree A in the forest stand plot in the field.

[0010] Step 2: Use the DBSCAN clustering algorithm to segment the forest stand point cloud into individual tree point clouds:

[0011] In a point cloud with n points, calculate the value of any arbitrary point P. i (x i ,y i ,z i ) to another point P j (x j ,y j ,z j The distance d between them ij The calculation formula is as follows:

[0012]

[0013] Where i = 1, 2, ..., n, j = 1, 2, ..., n, and i ≠ j; at point P i Within a sphere centered at d and with Eps as its radius, when d ij When ≤Eps, then P j If a point is located within the spherical space, and the number of points contained within the spherical space is not less than 4, then P is... i This is called the core point, denoted by P. k It is represented by k≤n; with core point P kk spherical space is constructed with the center and Eps as the radius, the point cloud is divided into k classes, if the spherical space intersects, the points in the spherical space are merged into a class, and the points in the same class constitute a tree.

[0014] The DBSCAN clustering algorithm is optimized to realize automatic acquisition of the Eps parameter:

[0015] ① Initialization of the Eps parameter

[0016] 60 random numbers between 0.5 and 10 are generated and represented by u k , k = 1, 2, …, 60; the specific calculation steps are as follows:

[0017] Step 1, let Eps = u k , classify the point cloud, and the point cloud is divided into c classes;

[0018] Step 2, according to formula (1), the distance between each internal point P i and other points in the same class is calculated, and the average value is represented by d s (p i ), each point P i corresponds to a d s (p i );

[0019] Step 3, according to formula (1), the average distance between point P i and points in each different class is calculated, represented by d h (P i ), a total of c-1 different class distance averages are calculated, and the minimum value is represented by min{d h (P i )}. Each point P i corresponds to a min{d h (P i )};

[0020] Step 4, the silhouette coefficient Sil i of point P i is calculated, and the calculation formula of the silhouette coefficient is

[0021]

[0022] where max{d s (P i ), min{d h (P i )}} represents the larger value of min{d h (P i )} and d s (p i );

[0023] Step5 Total n P in point cloud i Point, the average profile coefficient of n points is calculated as

[0024]

[0025] Then each Eps = u k Corresponding to an average profile coefficient Sil k ;

[0026] Step6 Calculate the cosmic expansion rate NI k For

[0027] NI k = 1-Sil k (4)

[0028] Total 60 times from step1 to step6 calculation, each u k Corresponding to a NI k , according to the order from small to large NI k , corresponding to the completion of the order of u k , the minimum value of NI k in 60 times of calculation corresponds to u1, which is the current optimal value of Eps, and Eps initialization is completed.

[0029] ② Update of Eps parameter

[0030] Generate 60 new random numbers u k Based on u1, complete an update, and the update formula is

[0031]

[0032] Where r1∈[0,1], r2∈[0,1], r3∈[0,1] are 60 random numbers, t is the current iteration number, the initial value is 1, and the iteration number is increased by 1 after completing an update.

[0033] ③ Obtain the optimal value of Eps parameter

[0034] After updating, 60 random numbers are obtained and represented by u k , total 60 times from step1 to step6 calculation, each u k Corresponding to a NI k , according to the order from small to large NI k , corresponding to the completion of the order of u k , the minimum value of NI kThe minimum value of u1 of Eps is the current optimal value of Eps, and then the updating of Eps parameter according to formula ② is completed, and the iteration number is increased by 1 after one updating; when the iteration number is equal to 1000, the calculation is completed, and Eps = u1 is the optimal value of the parameter.

[0035] Step 3, calculate the proportional coefficient, adjust the point cloud coordinates:

[0036] Select the point cloud corresponding to the single tree A with the measured height and diameter at breast height in step 1, the highest point P H (x H ,y H ,z H ), the lowest point P L (x L ,y L ,z L ), and the actual height value of the tree A is H (unit: decimeter), then the vertical direction proportional coefficient R z is

[0037]

[0038] According to the vertical direction proportional coefficient, the point cloud coordinates are adjusted to P i (x i ,y i ,Z i ), wherein Z i =R z ·z i , and then the diameter at breast height value d 13 (P i ) of the single tree A is measured, and the actual diameter at breast height of the tree A is B (unit: decimeter), then the horizontal direction proportional coefficient R xy is

[0039]

[0040] According to the horizontal direction proportional coefficient, the point cloud coordinates are adjusted to P i (X i ,Y i ,Z i ), wherein X i =R xy ·x i , Y i =R xy ·y i , and the construction of the three-dimensional measurement model of the sample plot is completed.

[0041] The photographic equipment is a smart phone or an electronic device with a camera function. DETAILED DESCRIPTION

[0042] The specific embodiments of the present application are illustrated by taking a set of data from an experiment in situ at the Northeast Forestry University Experimental Forest as an example.

[0043] The experimental equipment used in the experiment is an iPhone 12, and the video shooting performance index is 1080p (1920x1080, 60 frames / second) video recording.

[0044] The actual number of trees in the experimental plot of this election is 22, and the correct segmentation is 22. This experiment divides 23 trees into two groups: the first group selects 5 trees that meet one of the conditions of being located at the edge of the plot, being in the center, and having a height higher than the average value. The second group is composed of the remaining 17 trees from this experiment. The proportion obtained from the first group is applied to estimate the actual height and diameter at breast height of the trees in the second group, and compared with the true tree height and diameter at breast height.

[0045] 1. Compare the height of the trees after clustering and segmentation in the model with the actual tree height, and obtain the proportion value of the relationship between the two as 5.45 to estimate the absolute error of the height measurement of the second group of trees between 0.05-0.37m, with the highest relative error of 2.03% and the average relative error of 1.53%.

[0046] 2. Compare the diameter at breast height of the trees after clustering and segmentation in the model with the actual diameter at breast height, and obtain the proportion value of the relationship between the two as 15.99 to estimate the absolute error of the diameter at breast height measurement of the second group of trees between 0.003-0.148m, with the highest relative error of 7% and the average relative error of 2%.

[0047] The present application combines the DBSCAN clustering method, based on image three-dimensional reconstruction technology and machine vision technology, and proposes a method for constructing a forest three-dimensional measurement model based on DBSCAN algorithm using images. This method has a wide measurement range, collects a large amount of standing tree information, realizes tree height measurement for a row of trees and forest, effectively reduces the measurement workload, has low cost, portable equipment, simple operation, and can be completed by a single person.

Claims

1. A method for constructing a three-dimensional measurement model of a sample plot of a forest stand using images, characterized by: The forest sample plot data is obtained by using a video recording device, the video recording is converted into images by sampling with an interval of 25 frames, the three-dimensional point cloud is obtained by inputting the images into software Pix4Dmapper, the point cloud is segmented into single-tree point cloud by optimizing the DBSCAN clustering algorithm, the vertical and horizontal direction proportional coefficients of the forest three-dimensional model and the forest measurement model are calculated by combining the measured height and diameter at breast height of the single tree, and the construction of the three-dimensional measurement model of the forest sample plot is obtained based on the proportional coefficients and the point cloud coordinates. Step 1, obtaining data: using a smart phone to obtain a video recording containing a panorama of the forest sample plot, setting the sampling interval to 25 frames to obtain images of the forest sample plot, and inputting the images into software Pix4Dmapper to obtain three-dimensional point cloud; measuring the height and diameter at breast height of a single tree A in the forest sample plot; Step 2, using the DBSCAN clustering algorithm to segment the forest sample point cloud into single-tree point cloud: In a point cloud of a number of points , the distance between an optional point and another point is calculated , the formula of which is: （1） in , ,and ; at point With the center as the center, Within a sphere of space with a parameter of radius, when At that time, If a point is located within the spherical space, and the number of points contained within the spherical space is not less than four, then... Called the core point, using It means that among them ; with core point Centered on, with Constructed by radius A spherical space divides the point cloud into If the spherical spaces intersect, then the points within the spherical spaces merge into one class, and the points of the same class constitute a standing tree. Optimizing DBSCAN clustering algorithm implementation Automatic acquisition of parameters: Initialization of parameters Generate 60 random numbers between [0.5, 10] for use Indicates, ; The specific calculation steps are as follows: Step 1 Let , perform point cloud classification, the point cloud is divided into classes; Step 2 Calculate the distance between each internal point and other points in the same class according to formula (1) and take the average value as the distance between each internal point and other points in the same class, denoted as ; each point corresponds to a ; Step 3: Calculate the points according to formula (1). The average distance to each different class of points is used It indicates that a total of [number] calculations were performed. The average distance between 1 distinct classes, where the minimum value is represented by . Represents; each point Corresponding to one ; Step 4 Calculation Point profile coefficient The calculation formula of the profile coefficient is （2） wherein represents the value of the greater of and the greater of Step 5: Total in point cloud indivual Point, calculation The average contour coefficient of the points is （3） then each corresponding to an average profile coefficient ; Step 6 Calculate the rate of cosmic expansion To （4） Total 60 times from step1 to step6, each Corresponding to one , according to In ascending order, corresponding to the completion of The sorting, where the minimum value of Corresponding to The current optimal value, complete Initialization;​ updating of parameters With Generate 60 new random numbers One update is done, update formula is （5） wherein are each 60 random numbers, is the current iteration number, the initial value is 1, and the iteration number is incremented by 1 after completing one update. acquiring parameter optimum value After updating, 60 random numbers are obtained , and the total number of calculations from step 1 to step 6 is 60, each corresponding to one , according to arranged in ascending order, corresponding to the completion of ordering, among the 60 calculations The minimum value of is the current optimal value, and then according to After updating the parameters, the number of iterations is increased by 1; when the number of iterations is equal to 1000, the calculation is completed, which is the optimal value of the parameters; Step 3, calculating the proportional coefficient and adjusting the point cloud coordinates: Select the point cloud corresponding to the single tree A in step 1, whose height and diameter at breast height are measured in the field, the highest point , the lowest point , the actual height value of the tree A has been measured as H, H is in decimeter, then the vertical direction scale factor is （6） According to the vertical direction proportional coefficient, the point cloud coordinates are adjusted as wherein Then the single tree A diameter at breast height value is measured by the point cloud It is measured that the actual diameter at breast height of the tree A is B, and B unit is decimeter, then the horizontal direction proportional coefficient is （7） The point cloud coordinates are adjusted according to a horizontal direction proportional coefficient wherein , , the construction of the three-dimensional measurement model of the sample plot is completed.

2. The method for constructing a three-dimensional mensuration model of a plot of a stand using images as recited in claim 1, wherein: The video recording device is a smart phone or an electronic device with a camera function.

Images