A method for estimating vegetation aggregation index based on Poisson distribution and point cloud clustering

Abstract

The present invention belongs to the technical field of laser radar remote sensing data processing, and specifically relates to a vegetation concentration index estimation method based on Poisson distribution model and point cloud clustering. The present invention uses laser radar to obtain three-dimensional point cloud data of forest vegetation canopy, and based on the principle that the ideal state of leaf distribution in the forest canopy satisfies the Poisson distribution, the point cloud is aligned, denoised, filtered and normalized; the degree of deviation between the cluster center and the Poisson point is quantitatively expressed, and the concentration index is calculated by the ratio between the number of cluster centers that have not been offset and the number of all Poisson points, and a laser radar vegetation concentration index estimation method based on Poisson distribution and point cloud clustering is established. The present invention overcomes the intermediate error caused by the prior art that the canopy gap rate must be inverted first, and at the same time avoids the defects of being unable to collect data around the clock due to limitations such as usage time and weather.

Description

Technical Field

[0001] The invention belongs to the technical field of laser radar remote sensing data processing, and in particular relates to a vegetation concentration index estimation method based on a Poisson distribution model and point cloud clustering. Background Art

[0002] The geometric structure of the forest canopy is very important for studying the transmission and distribution of radiation in the plant canopy and the photosynthesis of leaves. Vegetation canopy structure parameters can describe the canopy structure more accurately and quantitatively, and are also input parameters for many ecological models. When calculating the collision between light incident on the canopy from a certain angle and the leaves, the distribution of leaves must be taken into account. Leaves are not randomly distributed in space, but show a certain degree of aggregation. Compared with the random distribution of leaves, the aggregation of leaves on branches will expand the gaps in the canopy, so we need specific parameters to describe the spatial distribution of leaves and find a suitable method to indirectly and efficiently measure such parameters.

[0003] LiDAR (Light Detection and Ranging) is an active remote sensing technology that integrates multiple technologies. Its greatest advantage is that it can efficiently and accurately obtain three-dimensional spatial information without contacting the target, especially the vertical structure information of the canopy that cannot be obtained by passive remote sensing. These advantages make the inversion of canopy structural parameters more accurate and rapid, and more and more researchers are beginning to use LiDAR technology to study the inversion of canopy structural parameters.

[0004] The Clumping Index (CI) is an important canopy structural characteristic parameter that describes the canopy radiation transfer process. It can describe the aggregation of the forest canopy and plays a very important role in studying the radiation transfer process of vegetation and the indirect measurement of the leaf area index. The study of the vegetation radiation transfer process is usually based on Beer's law. According to Beer's law, the leaf distribution pattern in the canopy is assumed to be randomly distributed; in real three-dimensional scenes, due to the highly complex structure in the canopy, there must be different degrees of aggregation between leaves. Nilson (1971) introduced a correction parameter to describe the aggregation of the canopy, which is still used today:

[0005]

[0006] Where P(θ) is the canopy gap ratio in the θ zenith angle direction, G(θ) is the projection ratio, and λ 0 is the correction factor; if the blades are distributed in a clustered manner, λ 0 <1, if the leaves are randomly distributed, λ 0 =1, if the leaves are regularly distributed, λ 0>1. Later, Chen et al. (1991) defined the correction coefficient as the aggregation index, which is used to describe the difference between the true three-dimensional spatial distribution of vegetation canopy and the Poisson distribution, namely:

[0007]

[0008] Where LAI e It is defined as the effective leaf area index, and LAI is the real leaf area index, so the aggregation index Ω can correct the aggregation distribution of leaf elements in the canopy. The main methods currently used to calculate the leaf aggregation index include:

[0009] (1) Logarithmic mean method based on canopy gap ratio CI LX : In order to solve the problem of large gaps between vegetation due to non-random distribution of leaves, Lang proposed a method to calculate the aggregation index based on the logarithmic mean of the canopy gap rate. This method divides the study area into several small areas. It is generally believed that the canopy cannot be regarded as uniform at the plot scale, but it can be approximately regarded as uniform at the scale of each small area. The aggregation index is calculated by the ratio of the logarithm of the gap rate mean to the average of the logarithm of the gap rate.

[0010]

[0011] Where P is the gap ratio, θ is the zenith angle, The limitation of this method is that the division of regions needs to be extremely cautious, because if the region is too small, the gap rate of the region is often zero, and the logarithm of zero cannot be defined. If the region is too large, it is impossible to distinguish different regions, especially the large gaps between vegetation or tree crowns.

[0012] (2) Based on the gap size distribution method CI CC:Chen proposed a method to derive the aggregation index using the canopy gap size distribution. This method eliminates the assumptions about the spatial distribution patterns of leaf elements and tree crowns. The canopy gap distribution with aggregation effects was measured under the canopy using the Tracing Radiation and Architecture of Canopies (TRAC). By using the core idea of ​​eliminating large light spots, larger gaps were iteratively deleted until the distribution of smaller gaps conformed to the random model. Small gaps were then interpreted as gaps within leaf clusters, and large gaps were interpreted as gaps between leaf clusters. The canopy contraction before and after elimination, the change in leaf area index, and the gap size distribution function before and after iteration were used to calculate the aggregation index. Subsequently, Leblanc pointed out that Chen's derivation of the aggregation index lacked a normalization factor, and revised the canopy gap size distribution theory based on the canopy analyzer. This method can be applied to all types of plant canopies without making spatial pattern assumptions about canopy units (CI CC ).

[0013]

[0014] where F m (0,θ) is the measured cumulative gap fraction greater than zero, i.e., the canopy gap fraction, F mr (0,θ) is the gap fraction of the crown when large gaps that are impossible when assuming a random distribution of crown elements are removed, given the LAI and leaf element width.

[0015] In further research by Leblanc, combined with CI LX and CI CC Two methods, using gap distribution theory, a new method (CI CLX ), which solves the segment size-dependent problem in the logarithmic gap average method when the large gaps are not uniform, and the intra-segment heterogeneity problem using the gap distribution theory, but the large gaps within the segments will affect the accuracy of this method in calculating the aggregation index.

[0016]

[0017] in Is using CI CC The element aggregation index of segment k of the method, and is the gap fraction of segment k.

[0018] These two mainstream traditional methods often have the defects of relying on specific measuring instruments in use, requiring the inversion of the gap ratio before converting the result into the aggregation index, increasing the error caused by the intermediate variables, and the results are greatly affected by irregular large gaps, and certain assumptions need to be made about the spatial distribution pattern of the canopy in advance. Summary of the invention

[0019] In view of the above-mentioned problems or shortcomings, and to solve the problems of high dependence on measuring instruments and relatively low accuracy in the existing vegetation concentration index estimation, the present invention provides a vegetation concentration index estimation method based on Poisson distribution model and point cloud clustering, which quantitatively expresses the difference between the three-dimensional spatial distribution of the real forest canopy and the Poisson distribution through cluster analysis, and more accurately inverts the canopy concentration index without the need to invert the canopy gap rate.

[0020] A vegetation concentration index estimation method based on Poisson distribution and point cloud clustering includes the following steps:

[0021] Step 1: Point cloud data acquisition and preprocessing:

[0022] The data source of the target plot is obtained by scanning with a laser scanner, and preprocessed, exported into a text format, and the points below the scanner height are removed. The preprocessing is based on research needs, first registering the plot, then resampling, filtering, and normalizing the point cloud data to establish a canopy height model CHM.

[0023] Furthermore, the data source is a single data source or a fused data source. The present invention can be used under the condition of a single data source, or can select collaborative inversion of multiple data sources: when using a single data source such as a ground-based laser radar or an airborne laser radar, the sample site data source is directly preprocessed; when using a fused data source for collaborative inversion, it is necessary to simultaneously obtain the ground-based and airborne laser radar point cloud data on the target sample site, and select a sample site data source of not less than 10m*10m in size in the overlapping part of the acquisition range of the two data.

[0024] Step 2: Data segmentation:

[0025] After generating CHM from the normalized point cloud in step 1, the point cloud segmentation algorithm based on the gradient change of the tree vertex cloud is used to perform coarse segmentation on the point cloud to obtain point cloud units without inter-crown gaps.

[0026] Since the principle of this method is to approximately simulate the distribution of canopy leaves by generating random points of Poisson distribution, so as to judge the degree of clustering of leaves in the canopy, and it is necessary to make the spatial analysis object the largest target object that conforms to the Poisson distribution, and extract the convex hull unit (i.e., point cloud unit) that does not contain large gaps between tree crowns, it is necessary to perform a coarse segmentation on the preprocessed plot data in step 1 so that each part of the point cloud after segmentation does not contain large gaps, which will affect the simulation effect of the Poisson distribution.

[0027] Step 3: Poisson point generation:

[0028] The point cloud unit obtained in step 2 is enclosed by a cube, with the center of the canopy bottom as the origin and the boundary of the block data in the horizontal plane as the interval, and Poisson points are generated in the point cloud unit according to the density not greater than the original point cloud of the canopy. These Poisson points are the reference points of the spatial distribution of the canopy point cloud under ideal conditions. Since the parameter λ of the Poisson distribution is equal to its mathematical expectation (i.e., mean), the value of λ is determined by the size of the point cloud unit.

[0029] Step 4: K-means clustering:

[0030] The Poisson points obtained in step 3 are used as the initial sample centers of clustering. The K-means algorithm is used to cluster the sample point clouds, and the corresponding relationship between the point clouds and leaves is established. The records of the movement of the cluster centers during the iteration of the K-means algorithm are used as quantitative indicators of the degree of deviation between the spatial distribution of leaves and the Poisson distribution.

[0031] Since there is no one-to-one correspondence between point clouds and leaves, point clouds need to be clustered when studying the spatial distribution of leaves. The K-means algorithm is a clustering analysis algorithm that achieves sample classification through continuous iterative solutions. K-means clusters samples according to similarity. Its basic idea is to make each sample point closer to the center of the cluster. When using the K-means algorithm for clustering, there is often a problem that the initial sample center setting has a great influence on the clustering results. This method uses Poisson points as the initial sample center of clustering, which just solves this problem. The K-means algorithm is used to cluster the sample point clouds and establish the corresponding relationship between point clouds and leaves.

[0032] Step 5: Calculate the aggregation index:

[0033] Analyze the offset of the cluster center position after clustering obtained in step 4 compared with the original Poisson point position, set a threshold, and take the proportion of points in the generated Poisson points whose position offset is less than the set threshold as the aggregation index result of a point cloud unit.

[0034]

[0035] Where Ω is the aggregation index, N 未偏移 N is the number of cluster centers whose offset distance is less than the set threshold during the clustering process. 全部 is the total number of Poisson points.

[0036] Then, the projection area of ​​different point cloud units in the whole plot on the horizontal plane is used as the weight, and the weighted average of different point cloud units is performed according to their projection area to obtain the distribution of the aggregation index of the plot in the horizontal direction. Finally, the canopy is divided in the vertical direction according to the required height interval, and the distribution of the aggregation index of the whole plot at different heights is calculated.

[0037] Furthermore, the threshold value is set in step 5 as follows: for the result of clustering a single data source, 0 is used as the threshold value; for the result of clustering a fused data source, based on the correlation between the results of different data sources in the fused data source, the minimum threshold value when the correlation is higher is selected as the threshold value, wherein the evaluation of the correlation is determined based on the determination coefficient between the clustering results of different data sources, and when the determination coefficient is greater than 0.65, it is considered that the correlation between the two is high.

[0038] Step 2, step 3 and step 4 of the present invention and the principles involved:

[0039] Canopy Height Model (CHM) is an important model for forestry applications. It is a surface model that represents the height of vegetation from the ground and can provide the horizontal and vertical distribution of the canopy. To obtain CHM, the original point cloud data needs to be filtered to separate the ground points and object points, and then interpolated to obtain the digital surface model (DSM) and digital elevation model (DEM) of the sample site. Finally, the difference between the two is calculated to obtain CHM.

[0040] CHM essentially builds a two-dimensional grid at a given interval on the sample plane, projects the point cloud into the grid, takes the difference between the highest and lowest values ​​of the point cloud elevation in the grid, interpolates the values ​​of the adjacent grids for the grid without point cloud, and obtains a two-dimensional matrix with canopy height as the content. By setting the threshold, find the larger value in the matrix as the tree vertex, find its gradient in all directions, and then find the local minimum value, and use the boundary formed by the connection of the minimum value as the boundary to roughly segment the point cloud data.

[0041] Poisson distribution is a discrete probability distribution that represents the probability of a certain number of events occurring at a known constant average rate in a specific time or space, and is independent of the interval between the occurrences of the previous event. It can also be used to represent the number of events occurring within a specific interval, such as distance, area or volume. Under limited conditions, the number of events occurring within a fixed time interval can be expressed as a random number with a Poisson distribution. If a discrete random variable X obeys a Poisson distribution with parameter λ, then its probability density function is:

[0042]

[0043] Among them, k represents the number of occurrences. For a Poisson point process, it usually needs to be completed in a bounded area. The most important steps are to create a number of random points and randomly distribute the points in an appropriate manner. The number of random points needs to be determined by the scale of the point cloud unit. The number of points should be equal to the product of the volume of the point cloud unit and the required point cloud density. The distribution of random points depends on the value of the Poisson distribution parameter λ. For the Poisson distribution, λ is equal to the mean and variance of the Poisson distribution at the same time. Therefore, in order to ensure the simulation effect of Poisson points on the canopy point cloud in the point cloud unit, the mean of the Poisson distribution, that is, λ, needs to be consistent with the geometric center of the point cloud unit.

[0044] K-means algorithm, also known as K-means algorithm, is a clustering algorithm that is solved through continuous iteration. Its core idea is: first randomly select K samples from the sample set as cluster centers, and calculate the distance between all samples and these K "cluster centers". For each sample, divide it into the cluster with the "cluster center" closest to it, and then calculate the new "cluster center" of each cluster for the new cluster. In general, the calculation process of K-means algorithm can be summarized as: the selection of the number of clusters K, the distance from each sample point to the "cluster center", and the update of the "cluster center" according to the newly divided clusters. Repeat the last two steps until the cluster center changes tend to be stable, and obtain K clusters. It can also be seen that for the effect of K-means algorithm, there are three most important factors, namely how to determine the appropriate K value, how to set the initial cluster center, and confirm the conditions for the termination of iteration. The K value can be determined according to the number of Poisson points; the initial cluster center can be set as the Poisson point coordinate, so as to quantitatively describe the deviation of the canopy point cloud from the Poisson distribution state by comparing the spatial movement distance of the cluster center before and after clustering; the condition for the termination of the iteration is set to stop the iteration when the sum of the distances of all point cloud data from the cluster center of the cluster is the smallest.

[0045] The present invention uses laser radar to obtain three-dimensional point cloud data of forest vegetation canopy, and based on the principle that the ideal state of leaf distribution in the forest canopy satisfies the Poisson distribution, performs registration, denoising, filtering and normalization processing on the point cloud; based on the CHM generated by the normalized point cloud, the point cloud data is roughly segmented; Poisson points are generated according to demand in the segmented point cloud unit; K-means clustering is performed on the point cloud data based on the Poisson points, and the moving distance of the cluster center during the clustering process is recorded; the threshold for judging whether the cluster center has not shifted is determined by calculating the determination coefficient and the mean square error between the ground-based and airborne data; the aggregation index of the sample plot is obtained by calculating the ratio between the number of cluster centers that have not shifted and the total number of Poisson points; the data is vertically layered according to the required height intervals, and finally the vertical distribution result of the aggregation index of the study area based on the method is obtained. The present invention quantitatively expresses the degree of deviation between cluster centers and Poisson points, calculates the aggregation index by the ratio between the number of cluster centers without deviation and the number of all Poisson points, and establishes a lidar vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering. The process is as follows: Figure 1 shown.

[0046] To sum up, the present invention takes into account the principle that the leaf distribution in the canopy satisfies the Poisson distribution under ideal conditions. By simulating Poisson points and point cloud clustering methods, it can overcome the intermediate error caused by the need to first invert the canopy gap rate in other inversion methods based on lidar point clouds when estimating the aggregation index; it also avoids the defects of being unable to collect data around the clock due to limitations such as usage time and weather when using other traditional methods such as digital hemispherical photography methods; ultimately, the broad-leaved forest canopy aggregation index is effectively estimated based on lidar data. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of the present invention;

[0048] Figure 2 A schematic diagram of sample data for implementing the present invention;

[0049] Figure 3 A schematic diagram of a canopy height model of an embodiment;

[0050] Figure 4 A schematic diagram of data segmentation results in an embodiment;

[0051] Figure 5 A schematic diagram of a Poisson point simulation of an embodiment;

[0052] Figure 6 This is a schematic diagram of the K-means clustering process principle;

[0053] Figure 7 Schematic diagram of the principle of inverting the aggregation index of the present invention;

[0054] Figure 8 Result diagram of the offset distance of the cluster center in the point cloud unit of the embodiment;

[0055] Fig. 9 It is a line graph of the results of inverting the aggregation index using this method under different thresholds;

[0056] Fig.10 This is a line chart comparing the results of the inversion of the concentration index using the ground-based and airborne lidar data by this method;

[0057] Fig.11 This is a line graph comparing the results of the aggregation index estimated by this method and based on the laser point area model. DETAILED DESCRIPTION

[0058] The present invention is further described in detail below through an example embodiment in conjunction with the accompanying drawings.

[0059] In order to quantify the difference between this real situation and the Poisson distribution, the present invention determines a reference object that is close to the leaf spatial distribution pattern under the ideal state. A large number of random points that conform to the three-dimensional Poisson distribution are generated in a limited spatial unit as the Poisson standard points of the leaf spatial distribution. These Poisson standard points are used as clustering centers to cluster the leaf point cloud and analyze its real spatial distribution. In order to facilitate the comparison of the changes in the clustering centers before and after clustering, we record the moving distance of the clustering center during the clustering process and use it as a quantitative indicator of the difference between the real spatial distribution of the leaf and the Poisson distribution under the ideal state. When the moving distance is between 0 and the threshold, it can be considered that the spatial distribution of the leaf approximately satisfies the Poisson distribution; when the moving distance exceeds the threshold range, it can be considered that the leaf is clustered in space. By calculating the number of points whose cluster center offset distance does not exceed the threshold in the limited spatial unit, the ratio of the number of all Poisson points can obtain the aggregation index of the canopy within a certain range.

[0060] First, the spatial analysis object needs to be the largest target object that conforms to the Poisson distribution, and the convex hull unit without large gaps needs to be extracted. The canopy point cloud data needs to be roughly segmented. After generating the canopy height model (CHM) from the normalized point cloud, the watershed algorithm is used to roughly segment the point cloud by the slope of the tree top to obtain the block point cloud data. Then the roughly segmented point cloud unit is enveloping with a cube, with the center of the canopy bottom as the origin and the boundary of the block data in the horizontal plane as the interval, and random points that conform to the three-dimensional Poisson distribution are generated according to the density of one point per square meter. Then the K-means algorithm is used to cluster the point cloud to achieve the correspondence between the point cloud and the leaves. The record of the movement of the sample center during the iteration process of the K-means algorithm can be used as a quantitative indicator of the degree of deviation between the spatial distribution of leaves and the Poisson distribution. Finally, the Euclidean distance between the sample center and the Poisson point after clustering is calculated. If the sample center does not deviate after clustering, it is considered that the leaf distribution in this area conforms to the Poisson distribution. The ratio of the number of sample centers that have not been offset to the number of Poisson points within each point cloud unit is the aggregation index result of the area. The projection area of ​​different point cloud units in the entire sample plot on the horizontal plane is used as the weight, and the weighted average of different point cloud units is performed according to their projection area to obtain the aggregation index of the sample plot.

[0061] The development environment of this embodiment is Microsoft Visual Studio 2013, and the programming language is C.

[0062] A vegetation concentration index estimation method based on Poisson distribution and point cloud clustering, the specific steps are as follows:

[0063] Step 1. Verification data selection The German forest multi-platform lidar public dataset released by H. Weiser et al. in 2022 was selected. In the BR03 plot in Bretten Forest, a 50m*50m plot was intercepted from the overlapping part of the ground-based and airborne laser data. The sensor scanning parameters and the distribution of trees in the plot are shown in Tables 1 and 2 respectively. According to step 1 of the technical solution, the plot was resampled, filtered and normalized (the dataset has been registered) to facilitate the subsequent establishment of CHM for the point cloud data.

[0064] Figure 2 Schematic diagram of sample site data for the embodiment, where (a) is the canopy point cloud data acquired by ground-based laser radar, and (b) is the canopy point cloud data acquired by airborne laser radar. Figure 3 Schematic diagram of the canopy height model.

[0065] After processing, the minimum distance between point clouds is 0.05m. Since the main research subject is the forest canopy, the point cloud below 8.9m is subjected to elevation filtering according to the point cloud height distribution curve.

[0066] Table 1. Airborne laser scanning (ALS) and ground-based laser scanning (TLS) sensors and acquisition parameters. The specifications of the sensors are shown in the data sheets (RIEGL Laser Measurement Systems, 2017, 2019, 2020c).

[0067]

[0068]

[0069] Table 2 Number of trees per plot and various data sources. Most of these trees were measured using different platforms and at different times. ULS data were available for 1173 trees with leaf loss. Of these, 133 trees were measured in the fall of 2019, 537 trees were measured in the spring of 2020, and 503 trees were measured in both the fall of 2019 and the spring of 2020.

[0070]

[0071]

[0072] Step 2: Data segmentation: After the sample point cloud is normalized to generate CHM, the watershed algorithm is used to roughly segment the point cloud according to the slope of the tree top. Finally, it is segmented into 65 point cloud units. The segmented point cloud units do not contain large gaps. Figure 4 This is a schematic diagram of the data segmentation results.

[0073] Step 3, Poisson point generation: Envelope the roughly segmented point cloud unit with a cube, take the center of the canopy bottom as the origin, and the boundary of the horizontal plane of the segmented data as the interval, and generate Poisson points at a density of one point per square meter (when the density is large, the calculation efficiency of this method will be greatly reduced due to the influence of computer computing power, so it is not advisable to set too high Poisson point density). Figure 5 Schematic diagram of Poisson point simulation.

[0074] Step 4, K-means clustering: The Poisson point obtained in step 3 is used as the cluster center to achieve the correspondence between the point cloud and the leaves. The K-means algorithm records the movement of the sample center during the iteration process as a quantitative indicator of the degree of deviation between the leaf spatial distribution and the Poisson distribution. Figure 6 Schematic diagram of the K-means clustering process.

[0075] Step 5, calculate the concentration index: set six different thresholds of 0, 0.5, 1, 1.5, 2, and 5 to measure whether the sample center deviates from the Poisson point; by comparing the determination coefficient and mean square error between the ground-based and airborne data inversion results under different thresholds, the optimal threshold that meets the conditions should be able to make the inversion results of the two data have a high correlation; at the same time, in order to more sensitively distinguish the degree of deviation between the blade point cloud and the Poisson point, the threshold should be as close to 0 as possible. Calculate the determination coefficient and mean square error between the ground-based and airborne data inversion results, and select the minimum threshold of 1.5 (R 2 =0.6791) is used as the threshold to determine whether the cluster center deviates from the Poisson distribution.

[0076] The aggregation index was calculated according to the determined optimal threshold of 1.5, and the vertical distribution of the aggregation index of the sample plot was generated based on the inversion results of the airborne and ground-based data at intervals of 1 m.

[0077] Figure 7 Schematic diagram of the principle of inverting the aggregation index using this method. Figure 8 This is the offset distance result diagram of the cluster center within the point cloud unit. Fig. 9 The line graphs are the results of inverting the concentration index using this method under different thresholds, where (a) is the result obtained based on ground-based lidar data, and (b) is the result obtained based on airborne lidar data. Fig.10 This is a line graph comparing the results of inverting the concentration index using this method for ground-based and airborne lidar data. Fig.11 This is a line graph comparing the results of the aggregation index estimated by this method and based on the laser point area model.

[0078] It can be seen from the above examples that this example collects and calculates the overlapping part of the ground and airborne data in the BR03 sample plot in Bretten Forest, and compares and analyzes the aggregation index results obtained by inversion based on the laser point area model and the finite length averaging method, and verifies the accuracy, and obtains a good correlation (R 2 =0.9716RMSE=0.0742), which fully illustrates the applicability of the method of the present invention. Since the principle of the laser point area method is to simulate the scanning principle of the ground-based laser radar scanner, the simulation of the real laser point area is limited by the scanner parameters, while the method based on Poisson distribution and point cloud clustering is based on the principle of statistical model and unsupervised classification, and its application range is wider, and the inversion sample area is also larger, making it better than the inversion method based on the laser point area model when processing airborne data and performing collaborative inversion.

Claims

1. A vegetation concentration index estimation method based on Poisson distribution and point cloud clustering, characterized in that: The following steps are involved: Step 1: Point cloud data acquisition and preprocessing: The data source of the target plot is obtained by scanning with a laser scanner, and after preprocessing, it is exported into a text format and the points below the height of the scanner are removed; the preprocessing is to first align the plot according to the research needs, and then resample, filter and normalize the point cloud data to establish a canopy height model CHM; Step 2: Data segmentation: After generating CHM from the normalized point cloud through step 1, the point cloud segmentation algorithm based on the gradient change of the tree vertex cloud is used to perform rough segmentation on the point cloud to obtain point cloud units without inter-crown gaps; Step 3: Poisson point generation: The point cloud unit obtained after rough segmentation in step 2 is enclosed by a cube, with the center of the canopy bottom as the origin and the boundary of the block data in the horizontal plane as the interval, and Poisson points are generated in the point cloud unit according to the density not greater than the original point cloud of the canopy. These Poisson points are the reference points of the spatial distribution of the canopy point cloud under ideal conditions; the parameter λ of the Poisson distribution is equal to its mathematical expectation, and the value of λ is determined by the size of the point cloud unit; Step 4: K-means clustering: The Poisson points obtained in step 3 are used as the initial sample centers of clustering. The K-means algorithm is used to cluster the sample point clouds, and the corresponding relationship between the point clouds and leaves is established. The records of the movement of the cluster centers during the iteration of the K-means algorithm are used as quantitative indicators of the degree of deviation between the leaf spatial distribution and the Poisson distribution. Step 5: Calculate the aggregation index: Analyze the offset of the cluster center position after clustering obtained in step 4 compared with the original Poisson point position, set a threshold, and take the proportion of points in the generated Poisson points whose position offset is less than the set threshold as the aggregation index result of a point cloud unit; Where Ω is the aggregation index, N 未偏移 N is the number of cluster centers whose offset distance is less than the set threshold during the clustering process. 全部 is the number of all Poisson points; Then, the projection area of ​​different point cloud units in the whole plot on the horizontal plane is used as the weight, and the weighted average of different point cloud units is performed according to their projection area to obtain the distribution of the aggregation index of the plot in the horizontal direction; Finally, the canopy is divided vertically according to the required height intervals, and the distribution of the aggregation index of the entire sample plot at different heights is calculated.

2. The vegetation concentration index estimation method based on Poisson distribution and point cloud clustering as claimed in claim 1, characterized in that: The data source in step 1 is a single data source or a fused data source. When a single data source such as ground-based lidar or airborne lidar is used, the sample site data source is directly preprocessed; when a fused data source is used for collaborative inversion, it is necessary to simultaneously obtain ground-based and airborne lidar point cloud data on the target sample site, and select a sample site data source of not less than 10m*10m in size from the overlapping part of the acquisition range of the two data.

3. The vegetation concentration index estimation method based on Poisson distribution and point cloud clustering as claimed in claim 2, characterized in that: The threshold value is set in step 5 as follows: for the result of clustering a single data source, 0 is used as the threshold value; for the result of clustering a fused data source, based on the correlation between the results of different data sources in the fused data source, the minimum threshold value when the correlation is higher is selected as the threshold value, wherein the evaluation of the correlation is determined based on the determination coefficient between the clustering results of different data sources, and when the determination coefficient is greater than 0.65, it is considered that the correlation between the two is high.

Images