Sample plot scale broad-leaved forest real leaf area index inversion method

Abstract

The invention belongs to the technical field of ground-based laser point cloud data processing, and particularly relates to a sample plot scale broad-leaved forest real leaf area index inversion method. According to the method, point geometric feature information of target point cloud data is utilized, and a mean value of a point expansion distance reps and a nearest neighbor point distance rngb of each point is taken as a point radius Rp of each point; then, according to the point radius of each point, the area Sp occupied by the corresponding point is calculated, the sum of the areas occupied by all the points is carried out to obtain the single-face leaf area Sleaf of the target area, and then the real leaf area index LAIt is solved. And furthermore, the broad-leaved forest point cloud data is calculated in parallel through layering, so that the efficiency is improved. According to the method, the internal parameters are used for rapid calculation and solution, and the real leaf area index is accurately inverted; the problems that an existing method is complex in process, accurate external parameter input is needed, and wood components, aggregation effects and leaf dip angle distribution cannot be solved are solved, and support is provided for forest monitoring through laser radar.

Description

Technical Field

[0001] The present invention belongs to the technical field of ground-based lidar point cloud data processing, and specifically relates to a method for inverting the true leaf area index at the plot scale for broad-leaved forests. Background Art

[0002] Broad-leaved forests are an important part of the ecosystem and play an important role in carbon storage, biodiversity, and climate regulation. In broad-leaved forests, leaves are the main medium for many physiological activities, such as intercepting rainfall, absorbing light, and gas exchange. Therefore, the measurement of leaves is of great significance for evaluating the physiological state of forest ecosystems. To quantify the size and density of leaves, researchers usually use the Leaf Area Index (LAI), which is defined as half of the total surface area of leaves per unit ground area.

[0003] During the large-scale measurement of LAI, since the leaves are not randomly distributed in the canopy, the measurement results are affected by woody components, aggregation effects, and leaf inclination distributions. Therefore, the obtained LAI is the Effective Leaf Area Index (LAIe), rather than the True Leaf Area Index (LAIt). LAIt can more accurately quantify the leaf area and reflect the photosynthetic capacity of broad-leaved forests. Therefore, eliminating the influence of woody components, aggregation effects, and leaf inclination distributions is crucial for measuring the true leaf area index and further accurately monitoring broad-leaved forests.

[0004] Lidar emits high-density lasers to quickly and accurately obtain the leaf distribution information inside the broad-leaved forest canopy, providing a feasible way to invert the true leaf area index. The most common method for lidar to invert the leaf area index is the gap fraction inversion method. This method assumes that the leaves are evenly and randomly distributed in the forest canopy. By calculating the gap fraction of lidar in the forest canopy and inverting the LAI according to the Beer-Lambert law, and then using the aggregation index and the leaf inclination distribution model to correct the obtained LAI, the LAIt is finally obtained. However, the gap fraction-based method faces problems such as complex calculation of the aggregation index, inaccurate correction of the leaf inclination distribution model, and inability to remove the influence of woody components, resulting in a low accuracy of the obtained LAIt.

[0005] To remove the influence of woody components, a branch and leaf separation operation is used to preprocess the forest point cloud to obtain the forest leaf point cloud, and the leaf area index is calculated based on the forest leaf point cloud to obtain the LAIt.

[0006] Hosoi et al. proposed a method for processing forest leaf point clouds using a voxel model. This method divides space into a three-dimensional space composed of cubic voxels, and calculates the true leaf area index by statistically analyzing the interaction frequency between the laser point cloud and the voxels. The voxel model uses voxels smaller than the leaf size as the model for calculating leaf area, eliminating the influence of the leaf aggregation effect; by using a leaf inclination distribution model, the influence of the leaf inclination distribution is weakened.

[0007] You et al. used the Alpha-shape envelope model to process forest leaf point clouds. This method constructs triangular patches on the point cloud with a sphere of alpha radius, calculates the area of all triangular patches, and thus obtains LAIt. The Alpha-shape model uses triangular patches as the model for calculating leaf area, eliminating the influence of the leaf aggregation effect; by dividing the leaf into multiple triangular patches, the influence of the leaf inclination distribution is weakened.

[0008] However, both the voxel model and the Alpha-shape model require the input of external parameters to ensure the accurate construction of the model. The accurate inversion of LAIt depends on the accurate input of external parameters. Summary of the Invention

[0009] In view of the above problems or deficiencies, to solve the problem of complex inversion process and the need for accurate parameter input in the existing inversion of the true forest leaf area index, the present invention provides a method for inverting the true leaf area index of broad-leaved forests at the plot scale. This method is based on the point radius of each point in the leaf point cloud, calculates the area represented by each point through the point radius, and thus obtains the true forest leaf area index.

[0010] A method for inverting the true leaf area index of broad-leaved forests at the plot scale, the specific steps are as follows:

[0011] Step 1. Data preprocessing:

[0012] Use a statistical filtering method to filter the initial point cloud data of the plot, removing the outliers in the initial broad-leaved forest point cloud. Then perform a branch and leaf separation operation to separate the leaf point cloud from the branch point cloud, obtaining the preprocessed broad-leaved forest leaf point cloud.

[0013] Step 2. Calculate the point radius of each point in the point cloud set: Taking each point as the center of a sphere, a sphere is established at this point, and the area of the great circle of the sphere is used as the area occupied by this point. The radius of the sphere is calculated from the point radius of this point.

[0014] Calculate the point dilation distance r epμ and the distance from the point to its nearest neighbor point r ngb , and the point dilation distance r eps and the nearest neighbor point distance r ngbThe mean value is used as the point radius R of this point p ;

[0015]

[0016] If only the point dilation distance r eps is used as the point radius, since the distance between some point clouds obtained by multi-station splicing may be too small, resulting in an underestimated point radius and thus an underestimated inversion of the leaf area index; if only the distance r ngb to the nearest neighbor point is used as the point radius, the leaf area index will be overestimated due to the excessive overlap between multiple spheres caused by the too large point radius. Therefore, the present invention selects the mean value of the point dilation distance r eps and the distance r ngb to the nearest neighbor point as the point radius of this point.

[0017] Calculation principle of the point dilation distance r eps : A gradually expanding virtual sphere is established with each point as the center. The initial radius r eps of all spheres is 0, and the radius of each sphere starts to expand and grow at the same rate. When a sphere collides (is tangent) with another sphere, the growth of this sphere stops until the growth of all spheres stops. The final radius size of each sphere is used as the dilation distance of the point where the sphere is located.

[0018] Calculation principle of the distance r ngb from a point to its corresponding nearest neighbor point: A KDTREE is established in each layer of the point cloud set P j , each point in the point cloud set is traversed, and the nearest neighbor point of each point is searched using the KDTREE, and the geometric distance between this point and the nearest neighbor point is calculated.

[0019] Step 3. Calculate the true leaf area index of broad-leaved forests:

[0020] Based on the point radius R p calculated in step 2, calculate the area S p occupied by each point:

[0021]

[0022] Statistically sum up the areas S p occupied by all points to obtain the single-sided leaf area S leaf of the target area;

[0023]

[0024] Divide the single-sided leaf area S leaf by the floor area S ground to calculate the true leaf area index LAIt:

[0025] LAIt = Sleaf / S ground 。

[0026] Further, the filtering in step 1 is specifically as follows: it is necessary to set the number of nearest neighbors and the standard deviation multiple as initial parameters; first, for each point, select the n nearest neighbor points according to the distance, and then calculate the average distance from each point to all its corresponding n neighbor points. And the standard deviation. Assume that the average distance conforms to a Gaussian distribution, and regard the points whose average distance of neighbor points is not within the range of the standard deviation multiple as noise points, and retain the points that meet the standard deviation multiple.

[0027]

[0028] Among them, d i represents the distance from the current point to its i-th neighbor point, i takes values among n, μ is the average distance of the points that conform to the Gaussian distribution, std is the standard deviation of the points that conform to the Gaussian distribution, and σ is the standard deviation multiple.

[0029] Further, for the original broad-leaved forest leaf point cloud set P obtained after the preprocessing in step 1, layer the point cloud set P according to the height information of the points themselves, process each layer of point cloud data in parallel, and calculate the point radius of each point in the point cloud set P. Since the number of broad-leaved forest point clouds is huge, layer the points according to their own height information and process each layer of point cloud data in parallel to improve the calculation efficiency.

[0030] The original coordinates of the leaf point cloud are p k (x k , y k , z k ), N is the number of leaf point cloud points, L is the number of layers to be divided, and layer the broad-leaved forest leaf point cloud evenly according to L. The height of each layer is:

[0031]

[0032] Among them: Z max represents the maximum value of the Z coordinate value in the broad-leaved forest leaf point cloud data, and Z min represents the minimum value of the Z coordinate value in the broad-leaved forest leaf point cloud data. After layering, the point cloud set P j for each layer is:

[0033] P j = {p|Z min +(j - 1)×H < p z < Z min + j×H, p ∈ P}.

[0034] Further, the branch and leaf separation operation in step 1 adopts a branch and leaf separation method based on graph theory.

[0035] Further, the preprocessing in step 1 is carried out after normalizing the initial point cloud data of the sample plot.

[0036] The present invention first performs preprocessing of filtering and branch and leaf separation on the broad-leaved forest point cloud to obtain the broad-leaved forest leaf point cloud; then, according to the height information of the point cloud, the obtained leaf point cloud is stratified to obtain multi-layer leaf point clouds, and the multi-layer leaf point clouds are processed in parallel, thereby improving the operation efficiency; for each layer of leaf point cloud, the point dilation distance r of each point in the point cloud is calculated eps and the distance r to the nearest neighbor point ngb , and the mean of the two is taken as the radius of the area occupied by each point (point radius R p ); then, according to the point radius of each point, the area S occupied by the corresponding point is calculated p ; finally, the areas occupied by all points are counted and divided by the floor area to obtain the true leaf area index LAIt of the target broad-leaved forest.

[0037] In summary, the present invention utilizes the point geometric feature information of the target point cloud data, and takes the mean of the point dilation distance r eps and the distance r to the nearest neighbor point ngb as the point radius R of each point p ; then, according to the point radius of each point, the area S occupied by the corresponding point is calculated p , and then the sum of the areas occupied by all points is obtained to get the single-sided leaf area S of the target area leaf , and then the true leaf area index LAIt is solved. And further, by performing stratified parallel calculation on the broad-leaved forest point cloud data, the rapid calculation of the true leaf area index is realized. The present invention performs rapid operation and solution with internal parameters, accurately inverses the true leaf area index; overcomes the problems that the existing methods have complex calculation processes, require accurate external parameter input, and cannot solve the problems of wood components, aggregation effects, and leaf inclination distributions, and provides support for lidar monitoring of forests. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a schematic flow chart of the present invention;

[0039] Figure 2 is a schematic diagram of a broad-leaved forest in an embodiment;

[0040] Figure 3 is a schematic diagram of the stratification of a broad-leaved forest in an embodiment.

[0041] Figure 4 is a diagram of the inversion accuracy of a single tree and a forest in an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The following examples further illustrate the present invention in detail by fitting forest data obtained through Poisson distribution in conjunction with the accompanying drawings; the development environment is PyCharm, the programming language is Python 3.7, and the software packages used are mainly open3d and numpy.

[0043] Original data acquisition: This embodiment acquires simulated broadleaf forest point cloud data based on the LESS platform. LESS is a three-dimensional radiation transfer model based on ray tracing that can simulate remote sensing data and images of large-scale and real 3D scenes. Simply input a predetermined tree model and plan the distribution of the broadleaf forest, select different lidar platforms and their corresponding parameters in LESS, and you can acquire high-precision three-dimensional point cloud data.

[0044] Four citrus tree models (CISI1-CISI4) in the HET14_WCO_UND scene of RAMI VI were selected as tree models in the broad-leaved forest. The size of the broad-leaved forest scene is 10m*10m. Based on the principle that any two trees are at least two meters apart and the tree positions are distributed in a Poisson distribution, the pre-selected citrus tree models are randomly placed in the broad-leaved forest scene to generate the broad-leaved forest example of the embodiment.

[0045] The VZ-1000 ground-based laser radar was selected to perform peripheral and central scanning of the broad-leaved forest scene. The peripheral scanning is divided into four stations, which scan the broad-leaved forest scene from the four directions of southeast, northwest, and northeast. The ground-based laser radar is placed 10m away from the broad-leaved forest, and it is ensured that each scan can completely scan the broad-leaved forest. The central scanning is to place the ground-based laser radar in the center of the broad-leaved forest scene and not blocked by a single broad-leaved tree, and use panoramic scanning to scan the broad-leaved forest scene. The zenith angle resolution and horizontal angle resolution of the scan are both 0.04°, ensuring that the scanned point cloud data is complete. The multi-site cloud data obtained are spliced ​​to obtain the initial broad-leaved forest point cloud data, and the LESS platform is used to calculate the true leaf area index in this scene.

[0046] A plot-scale inversion method for the true leaf area index of broad-leaved forests, the specific steps are as follows:

[0047] Step 1: Data preprocessing: Statistical filtering is used to filter and denoise the original point cloud. Two times the standard deviation and 10 adjacent points are selected as filtering criteria in the statistical filtering. The point cloud data is separated from branches and leaves using the labels provided by the point cloud data to obtain the leaf point cloud data of the target broad-leaved forest.

[0048] Step 2, broad-leaved forest point cloud stratification: In this embodiment, 5 is selected as the total number of stratification layers, and the broad-leaved forest leaf point cloud of the embodiment is stratified. The stratification result is as follows: Figure 3 shown.

[0049] Step 3: Calculation of the distance between point cloud points: For each layer of point cloud data, traverse each point to calculate the dilation distance of each point and the distance between the point and its neighboring points respectively. Calculate the point radius of each point based on the point dilation distance and the distance between the point and its neighboring points.

[0050] Step 4: Calculation of the true leaf area index of broad-leaved forests: Based on the point radius calculated in Step 3, calculate the area occupied by each point, and count the areas occupied by all points within 5 layers. Divide the total leaf area obtained by the area of the sample plot to obtain the true leaf area index of the sample plot.

[0051] Accuracy evaluation:

[0052] Calculate the accuracy of the retrieved LAIt and the true LAIt as the evaluation index. The calculation process is as follows:

[0053]

[0054] where LAI i represents the retrieved LAIt, and LAIt i represents the true LAIt. The accuracy of the inversion of the true leaf area index of broad-leaved forests at the scale of a 10m×10m sample plot in this embodiment is Figure 4 shown as 83.94%.

[0055] As can be seen from the above embodiments, the present invention realizes the inversion of the true leaf area index of broad-leaved forests at the sample plot scale, and the accuracy test result reaches. The method provided by the present invention weakens the influence of the aggregation effect, wood components, and leaf inclination distribution on the inversion of the true leaf area index without inputting additional parameters, and quickly and accurately inverses the true leaf area index by means of hierarchical parallel processing of point clouds, providing support for lidar monitoring of forests.

Claims

1. A plot-scale broad-leaved forest true leaf area index inversion method, characterized in that: The specific steps are as follows: Step 1. Data preprocessing: The statistical filtering method is used to filter the initial point cloud data of the sample plot to remove the outliers of the initial broad-leaved forest point cloud; then the leaf point cloud is separated from the branch point cloud by the branch separation operation to obtain the pre-processed broad-leaved forest leaf point cloud; P is the original broad-leaved forest leaf point cloud set obtained after preprocessing; Step 2. Calculate the point radius of each point in the point cloud collection: Calculate the point dilation distance r for each point eps , the distance between a point and its nearest neighbor point is r ngb , expand the point by a distance r eps The distance r from the nearest neighbor point ngb The mean of the points is taken as the point radius R of the point p ; Step 3. Calculate the real leaf area index of broad-leaved forest: Based on the point radius R calculated in step 2 p , calculate the area S occupied by each point p : Count the area S occupied by all points p Get the single-sided leaf area S of the target area leaf ; The single-sided blade area S leaf Divide by the floor area S ground Calculate the true leaf area index LAIt: Milk leaf / S ground .

2. The method for inverting the real leaf area index of broad-leaved forests at a plot scale as claimed in claim 1, characterized in that: The filtering in step 1 is specifically as follows: Set the number of nearest neighbor points and the standard deviation multiple as initial parameters; first, for each point, select the n nearest neighbor points according to the distance, and then calculate the average distance from each point to all its corresponding n neighbor points and standard deviation; assuming that the average distance conforms to the Gaussian distribution, the points whose average distance to neighboring points is not within the range of multiples of the standard deviation are regarded as noise points, and the points that meet the multiples of the standard deviation are retained; Among them, d i It represents the distance from the current point to its i-th neighbor point, i is a value among n, μ is the average distance of points that satisfy the Gaussian distribution, std is the standard deviation of the points that satisfy the Gaussian distribution, and σ is the standard deviation multiple.

3. The method for inverting the real leaf area index of broad-leaved forests at a plot scale according to claim 1, characterized in that: The original broad-leaved forest leaf point cloud set P obtained after the preprocessing in step 1 is layered according to the height information of the point itself, and the point cloud data of each layer is processed in parallel to calculate the point radius of each point in the point cloud set P; The original coordinates of the leaf point cloud are p k (x k ,y k ,z k ), N is the total number of leaf point clouds, L is the number of layers to be layered, and the broad-leaved forest leaf point cloud is evenly layered according to L, and the height of each layer is: Where: Z max Indicates the maximum value of the Z coordinate value in the broad-leaved forest leaf point cloud data, Z min Represents the minimum value of the Z coordinate value in the broad-leaved forest leaf point cloud data; the point cloud set P of each layer after stratification j for: P j ={p|Z min +(j-1)×H<p z <Z min +j×H,p∈P}。 4. The method for inverting the real leaf area index of broad-leaved forests at a plot scale according to claim 1, characterized in that: The branch-leaf separation operation in step 1 adopts a branch-leaf separation method based on graph theory.

5. The method for inverting the real leaf area index of broad-leaved forests at a plot scale according to claim 1, characterized in that: The preprocessing in step 1 is performed after normalizing the initial point cloud data of the sample plot.

Images