CN116503727A - Vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering - Google Patents
Vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering Download PDFInfo
- Publication number
- CN116503727A CN116503727A CN202310221090.6A CN202310221090A CN116503727A CN 116503727 A CN116503727 A CN 116503727A CN 202310221090 A CN202310221090 A CN 202310221090A CN 116503727 A CN116503727 A CN 116503727A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- poisson
- clustering
- distribution
- canopy
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 230000002776 aggregation Effects 0.000 title claims abstract description 45
- 238000004220 aggregation Methods 0.000 title claims abstract description 44
- 238000001914 filtration Methods 0.000 claims abstract description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 230000011218 segmentation Effects 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 238000003064 k means clustering Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000013139 quantization Methods 0.000 claims description 5
- 238000011160 research Methods 0.000 claims description 5
- 238000000638 solvent extraction Methods 0.000 claims description 5
- 238000012804 iterative process Methods 0.000 claims description 4
- 238000012952 Resampling Methods 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 abstract description 4
- 230000007547 defect Effects 0.000 abstract description 3
- 238000012545 processing Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 11
- 230000005855 radiation Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 235000013405 beer Nutrition 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 238000005056 compaction Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000029553 photosynthesis Effects 0.000 description 1
- 238000010672 photosynthesis Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/88—Lidar systems specially adapted for specific applications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention belongs to the technical field of laser radar remote sensing data processing, and particularly relates to a vegetation aggregation index estimation method based on a Poisson distribution model and point cloud clustering. The method utilizes the laser radar to acquire three-dimensional point cloud data of the forest vegetation canopy, and carries out registration, denoising, filtering and normalization processing on the point cloud based on the principle that the ideal state of the blade distribution in the forest canopy meets poisson distribution; quantitatively expressing the deviation degree of the clustering centers and the poisson points, calculating the aggregation index by using the ratio of the number of the clustering centers which are not deviated to the number of all poisson points, and establishing a laser radar vegetation aggregation index estimation method based on poisson distribution and point cloud clustering. The invention overcomes the defects that the prior art needs to invert the canopy gap rate firstly to cause intermediate errors, and meanwhile, the invention avoids the limitation of factors such as service time, weather and the like, and the data acquisition cannot be carried out all the day.
Description
Technical Field
The invention belongs to the technical field of laser radar remote sensing data processing, and particularly relates to a vegetation aggregation index estimation method based on a Poisson distribution model and point cloud clustering.
Background
The geometry of the forest canopy is important for studying the transmission and distribution of radiation in the plant canopy and photosynthesis of the leaves. The parameters of the vegetation canopy structure can describe the canopy structure more accurately and quantitatively, and are also input parameters of a plurality of ecological models. The distribution of the blade is taken into account when calculating the impingement of light impinging the canopy from a certain angle with the blade. The leaves are not randomly distributed in space, but rather are gathered to a certain extent, and the gathering of the leaves on the branches enlarges the gaps in the tree crown compared with the random distribution of the leaves, so that specific parameters are needed to describe the spatial distribution of the leaves, and a proper method is found to indirectly and efficiently measure the parameters.
The laser radar technology (Light Detection and Ranging, liDAR) is an active remote sensing technology integrated by multiple technologies, and has the biggest advantage of being capable of efficiently and accurately acquiring three-dimensional space information of a detected target without contacting the detected target, and particularly acquiring vertical structure information of a canopy which cannot be acquired by a passive remote sensing mode. These advantages have made inversion of the canopy structure parameters more accurate and rapid, and more researchers are beginning to study inversion of canopy structure parameters using lidar technology.
The aggregation Index (CI) is an important canopy structure characteristic parameter for describing the canopy radiation transmission process, can describe the aggregation condition of forest canopy, and plays a very important role in researching the radiation transmission process of vegetation and indirectly measuring leaf area Index. Research on vegetation radiation transmission processes is generally conducted based on Beer's law, according to which the blade distribution pattern in the canopy is assumed to be randomly distributed; in a real three-dimensional scene, the aggregation conditions of different degrees are necessarily existed among the blades due to the high complexity of the structure in the crown layer. Nilson (1971) introduced a correction parameter to describe the aggregation of the canopy, and was used as yet:
wherein P (θ) is the cap layer gap ratio in the direction of the zenith angle, G (θ) is the projection ratio, λ 0 Is a correction coefficient; lambda if the blades are in a concentrated distribution 0 < 1, lambda if the blades are randomly distributed 0 =1, λ if the blades are regularly distributed 0 >1. The correction factor is hereinafter defined by Chen (1991) et al as an aggregation index describing the degree of difference between the true three-dimensional spatial distribution of vegetation canopy and poisson distribution, namely:
LAI in e Defined as the effective leaf area index, LAI is the true leaf area index, so the aggregation index Ω can correct the aggregation distribution of the leaf elements in the canopy. The main methods currently used to calculate the blade concentration index include:
(1) Logarithmic average method CI based on canopy gap rate LX : in order to solve the problem that there is a large gap between vegetation due to non-random distribution of leaves, lang proposed a method for calculating a concentration index based on the logarithmic mean of the gap rate of the canopy, which divides the investigation region into several small regions, and generally considers that the canopy cannot be regarded as uniform on the scale of the plot, but can be regarded as approximately uniform on the scale of each small region, and calculates the concentration index by the ratio of the logarithmic mean of the gap rate to the logarithmic mean of the gap rate.
Wherein P is the clearance rate, θ is the zenith angle,is azimuth. The limitation of this method is that the division of the region needs to be carefully considered, because if the region division is too small, the gap rate of the region is often zero, and the logarithm of zero cannot be definedIf the area division is too large, it is impossible to distinguish between different areas, in particular large gaps between vegetation or crowns.
(2) CI based on gap size distribution method CC : chen proposes a method of deriving the aggregation index using the coronal gap size distribution, which eliminates the assumption of spatial distribution patterns of leaf elements and crowns, measures the coronal gap distribution with aggregation effects under the canopy using a canopy radiation and structure analyzer (Tracing Radiation and Architecture of Canopies, TRAC), iteratively removes larger gaps by eliminating the core idea of large spots until the distribution of smaller gaps conforms to a random model, then small gaps are interpreted as gaps within the leaf cluster gaps, large gaps are interpreted as gaps between the leaf cluster gaps, and the aggregation index is calculated by eliminating the canopy compaction before and after, the change in leaf area index, and the iterative front and back gap size distribution function. Leblanc thereafter indicated that the derivation of Chen aggregation index lacks normalization factors, corrects the canopy gap size distribution theory based on canopy analyzer, and is applicable to all types of plant canopy without spatial pattern assumption for canopy units (CI CC )。
Wherein F is 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 the large gaps, which were impossible when assuming random distribution of crown elements, are removed given the LAI and leaf element width.
In a further study of Leblanc, CI was incorporated LX And CI (CI) CC Two methods, using the gap distribution theory, have proposed a new method (CI CLX ) The problem of segment size correlation in the logarithmic gap averaging method and the problem of intra-segment heterogeneity using gap distribution theory when large gaps are non-uniform are solved, but large gaps in segments can affect the accuracy of the method in calculating the aggregation index.
Wherein the method comprises the steps ofIs to use CI CC Element concentration index of segment k of method, and +.>Is the gap fraction of segment k.
The two main stream traditional methods often have the defects that a specific measuring instrument is relied on in use, the result is converted into an aggregation index through inversion of the gap rate, errors caused by intermediate variables are increased, the result is greatly influenced by irregular large gaps, a certain assumption is needed to be made on the spatial distribution mode of the canopy in advance, and the like.
Disclosure of Invention
Aiming at the problems or the shortcomings, the invention provides a vegetation aggregation index estimation method based on a Poisson distribution model and point cloud clustering, which quantitatively expresses the difference between the three-dimensional spatial distribution of a real forest canopy and the Poisson distribution in a clustering analysis mode, and more accurately inverts the aggregation index of the canopy under the condition of not inverting the canopy gap rate.
A vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering comprises the following steps:
step 1, acquiring and preprocessing point cloud data:
and scanning by a laser scanner to obtain a data source of the target sample, preprocessing, deriving the data source into a text format, and removing points lower than the height of the scanner. The preprocessing is to register the sample areas according to research requirements, and then to establish a canopy height model CHM for point cloud data after resampling, filtering and normalizing operations.
Further, the data source is a single data source or a fusion data source. The invention can be used under the condition of single data source, and can also select multi-data source collaborative inversion: when a single data source such as a ground-based lidar or an airborne lidar is used, the sample-site data source is directly preprocessed; when the fusion data source is used for collaborative inversion, foundation and airborne laser radar point cloud data are required to be acquired on a target sample, and sample data sources with the size not smaller than 10m and not smaller than 10m are selected at the overlapping part of the acquisition range of the two data.
Step 2, data partitioning:
after CHM is generated on the normalized point cloud through the step 1, the point cloud is subjected to rough segmentation by using a point cloud segmentation algorithm based on the gradient change of the point cloud at the top of the tree, and a point cloud unit without the inter-crown gap is obtained.
According to the principle of the method, the distribution of the canopy blades is approximately simulated by generating random points of the Poisson distribution, so that the aggregation degree of the blades in the canopy is judged, a space analysis object is required to be the largest target object conforming to the Poisson distribution, and a convex hull unit (namely a point cloud unit) which does not contain big gaps among crowns is extracted, so that the preprocessed sample data in the step 1 is required to be roughly segmented, each part of segmented point cloud does not contain big gaps, and the simulation effect of the Poisson distribution is further affected.
Step 3, poisson point generation:
and (3) enveloping the roughly-segmented point cloud units obtained in the step (2) by using a cube, taking the bottom center of the canopy as an origin, taking the boundary of the segmented data on the horizontal plane as an interval, and generating poisson points in the point cloud units according to the density not greater than the original point cloud of the canopy. The poisson points are the reference points of the spatial distribution of the canopy point cloud in an ideal state. Wherein the value of λ is determined by the size of the point cloud unit, since the parameter λ of the poisson distribution is equal to its mathematical expectation (i.e. the mean value).
Step 4, K-means clustering:
and (3) taking the poisson points obtained in the step (3) as an initial sample center of clustering, clustering sample site clouds by using a K-means algorithm, establishing a corresponding relation between the point clouds and the blades, and taking a record of movement of the clustering center by the K-means algorithm in an iterative process as a quantization index of the deviation degree of the spatial distribution and the poisson distribution of the blades.
Because the point clouds and the blades are not in one-to-one correspondence, the point clouds need to be clustered when the spatial distribution of the blades is studied. The K-means algorithm is a cluster analysis algorithm, classification of samples is achieved through continuous iterative solution, the K-means clusters the samples according to similarity, and the basic idea is that the distance from each sample point to the center of the cluster is smaller. When the K-means algorithm is used for clustering, the problem that the initial sample center setting has a large influence on a clustering result often exists, and the method uses poisson points as the initial sample centers of the clustering, so that the problem is exactly solved, the K-means algorithm is used for clustering sample site clouds, and the corresponding relation between the point clouds and the blades is established.
Step 5, calculating a concentration index:
and (3) comparing the clustered central position obtained in the step (4) with the offset of the original poisson point, setting a threshold value, and taking the proportion of the generated poisson point with the position offset smaller than the set threshold value in the whole as an aggregation index result of a point cloud unit.
Wherein Ω is a aggregation index, N Unbiased N is the number of clustering centers with offset distance smaller than a set threshold value in the clustering process All of which Is the number of total poisson points.
And taking the projection areas of different point cloud units in the whole sample plot on the horizontal plane as weights, and carrying out weighted average on the different point cloud units according to the projection areas to obtain the aggregation index distribution of the sample plot in the horizontal direction. Finally, the canopy is divided in the vertical direction according to the required height interval, and the aggregation index distribution of the whole sample plot on different heights is calculated.
Further, the setting a threshold in the step 5 specifically includes: for the result of clustering the single data source, taking 0 as a threshold value; and selecting a minimum threshold value when the correlation is higher as a threshold value according to the correlation of the results among different data sources in the fusion data sources for the result of the fusion data source clustering, wherein the evaluation of the correlation is determined according to the decision coefficient among the clustering results among the different data sources, and considering that the correlation is higher when the decision coefficient is larger than 0.65.
Step 2, step 3 and step 4 of the present invention and the principles involved:
the canopy height model (Canopy Height Model, CHM) is an important model for forestry applications, and is a surface model that represents the vegetation height from the ground, which can provide horizontal and vertical distribution of canopy. To obtain the CHM, filtering the original point cloud data to separate ground points and ground feature points, performing interpolation operation to obtain a digital ground model (DSM) and a Digital Elevation Model (DEM) of the sample area, and performing difference operation to obtain the CHM.
The CHM is essentially to set up a two-dimensional grid on a sample ground plane at given intervals, project a point cloud in the grid, take the difference between the highest value and the lowest value of the point cloud elevation in the grid, interpolate the values of the adjacent grids taken by the non-point cloud grid, and obtain a two-dimensional matrix with the canopy height value as the content. And (3) by setting a threshold value, searching a larger value in the matrix as a tree vertex, solving the gradient of the larger value in each direction, searching a local minimum value, and performing rough segmentation on the point cloud data by taking a boundary formed by connecting the minimum values as a limit.
A poisson distribution is a discrete probability distribution that represents the probability that a certain number of events occur at a known constant average rate in a particular time or space, and is independent of the time between the occurrence of the previous event. It may also be used to represent the number of events occurring within a particular interval, such as distance, area or volume. Under defined conditions, the number of events occurring within a fixed time interval may be represented as a random number with a poisson distribution. If a discrete random variable X obeys a poisson distribution with parameter λ, the probability density function is:
where k represents the number of occurrences. For a poisson dot process, it is often necessary to do so in a bounded area, with the most important steps being creating a random number of dots and randomly laying out the dots in an appropriate manner. The random number of points needs to be determined by the size of the point cloud unit, and the number of points should be equal to the product of the volume of the point cloud unit and the required density of the point cloud. The distribution mode of the random points depends on the value of the poisson distribution parameter lambda. For poisson distribution, lambda is equal to the mean value and variance of poisson distribution at the same time, so that in order to ensure the simulation effect of the crown layer point cloud in the poisson point-to-point cloud unit, the mean value of poisson distribution, namely lambda, needs to be consistent with the geometric center of the point cloud unit.
The K-means algorithm is also known as a K-means algorithm, is a clustering algorithm which is solved by continuous iteration, and has the following core ideas: k samples are randomly selected from a sample set to serve as cluster centers, distances between all samples and the K cluster centers are calculated, each sample is divided into clusters where the cluster center closest to the sample is located, and then new cluster centers of all the clusters are calculated for new clusters. In general, the calculation of the K-means algorithm can be summarized as: and selecting the number K of clusters, the distance from each sample point to the cluster center, updating the cluster center according to the newly divided clusters, and repeating the last two steps until the change of the cluster center tends to be stable, thereby obtaining K clusters. It can also be seen that the most important influencing factors for the effect of the K-means algorithm are three points, namely how to determine the appropriate K value, how to set the initial cluster center, and the condition to confirm the termination of the iteration. The determination of the K value can be determined according to the number of poisson points; the initial clustering center can be set to be poisson point coordinates so as to quantitatively describe the deviation condition of the canopy point cloud and the poisson distribution state by comparing the space movement distances of the clustering centers before and after clustering; the condition for ending the iteration is set to stop the iteration when the sum of the cluster center distances of all the point cloud data from the clusters is minimum.
According to the invention, three-dimensional point cloud data of a forest vegetation canopy is obtained by utilizing a laser radar, and point cloud registration, denoising, filtering and normalization are carried out on the basis of the principle that the ideal state of blade distribution in the forest canopy meets poisson distribution; performing rough segmentation on point cloud data based on CHM generated by normalized point cloud; generating poisson points in the segmented point cloud units according to requirements; k-means clustering is carried out based on the Poisson point-to-point cloud data, and the moving distance of a clustering center in the clustering process is recorded; determining a threshold value for judging that the clustering center is not deviated by calculating a decision coefficient and a mean square error between the foundation and the airborne data; obtaining an aggregation index of the sample plot by calculating the ratio of the number of cluster centers which are not shifted to the total number of poisson points; and vertically layering the data according to the required height interval to finally obtain the aggregation index vertical distribution result of the research area based on the method. The invention establishes a laser radar vegetation aggregation index estimation method based on poisson distribution and point cloud clustering by quantitatively expressing the offset degree of the clustering centers and the poisson points and calculating the aggregation index by using the ratio of the number of the clustering centers which are not offset to the number of all the poisson points, wherein the flow is shown in figure 1.
In summary, the principle that the distribution of the blades in the canopy meets poisson distribution under an ideal state is considered, and the intermediate error caused by inversion of canopy gap rate of other inversion methods based on laser radar point cloud can be overcome when the aggregation index is estimated by simulating poisson points and a point cloud clustering method; meanwhile, the defects that data acquisition cannot be carried out all weather and the like due to the limitation of the use time, weather and other factors when other traditional methods such as a digital hemispherical photography method are used are avoided; finally, the aggregation index of the broad-leaved forest canopy is effectively estimated based on the laser radar data.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of example pattern data;
FIG. 3 is a schematic diagram of a canopy height model of an embodiment;
FIG. 4 is a diagram illustrating a data partitioning result according to an embodiment;
FIG. 5 is a schematic diagram of a Poisson Point simulation of an embodiment;
FIG. 6 is a schematic diagram of a K-means clustering process;
FIG. 7 is a schematic diagram of the principle of inversion of the concentration index according to the present invention;
FIG. 8 is a graph of the offset distance results for the class center in a point cloud unit according to an embodiment;
FIG. 9 is a plot of the results of the inversion of the concentration index of the present method at different thresholds;
FIG. 10 is a comparison line graph of results of the present method for performing concentration index inversion on foundation and airborne lidar data, respectively;
FIG. 11 is a comparison line graph of the present method and the results of the concentration index estimated based on the laser spot area model.
Detailed Description
The invention will be described in further detail below by way of an example of a sample and with reference to the accompanying drawings.
In order to quantify the difference between the real situation and the poisson distribution, a reference object which is close to the spatial distribution mode of the blade in an ideal state is determined. A large number of random points conforming to three-dimensional poisson distribution are generated in a limited space unit, the random points are used as poisson standard points of the space distribution of the blade, the poisson standard points are used as clustering centers, the blade point cloud is clustered, and the real space distribution condition of the blade point cloud is analyzed. In order to facilitate comparison of the change condition of the clustering centers before and after clustering, the moving distance of the clustering centers in the clustering process is recorded and used as a quantization index of the difference between the real space distribution of the blades and the Poisson distribution in an ideal state. When the moving distance is between 0 and the threshold value, the spatial distribution of the blade can be considered to approximately satisfy the poisson distribution; when the distance of movement is outside the threshold range, the blades may be considered to be in a clustered distribution in space. And calculating the number of points with the center offset distance of the aggregation center in the defined space unit not exceeding a threshold value, and obtaining the aggregation index of the canopy within a certain range by the ratio of the number of the points to the total poisson points.
Firstly, a space analysis object is required to be the largest target object conforming to poisson distribution, convex hull units without large gaps are extracted, and coarse segmentation is required to be carried out on canopy point cloud data. After a Crown Height Model (CHM) is generated for the normalized point cloud, the point cloud is roughly segmented through the gradient of the tree top by using a watershed algorithm, and segmented point cloud data are obtained. And then enveloping the roughly-segmented point cloud units with cubes, taking the bottom center of the canopy as an origin, taking the boundary of the segmented data on the horizontal plane as an interval, and generating random points conforming to three-dimensional Poisson distribution according to the density of one point per square meter. And clustering the point cloud by using a K-means algorithm to realize that the point cloud corresponds to the blade, wherein the record of the movement of the sample center by the K-means algorithm in the iterative process can be used as a quantization index of the deviation degree of the spatial distribution and the poisson distribution of the blade. And finally, calculating Euclidean distance between the clustered sample center and the poisson point, and considering that the regional blade distribution accords with the poisson distribution if the clustered sample center is not deviated. The ratio of the number of sample centers which are not deviated in each point cloud unit to the number of poisson points is the aggregation index result of the area, the projection areas of different point cloud units in the whole sample plot on the horizontal plane are taken as weights, and the aggregation index of the sample plot is obtained by carrying out weighted average on the different point cloud units according to the projection areas.
The development environment of this embodiment is Microsoft Visual Studio 2013 and the programming language is C.
A vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering comprises the following specific steps:
step 1, verifying data selection is carried out on a German forest multi-platform laser radar public data set published in 2022 by H.Weiser et al, a sample plot with the size of 50m and 50m is intercepted by selecting a superposition part of foundation and airborne laser data in a BR03 sample plot positioned in a Brayton forest, and the scanning parameters of a sensor and the tree distribution conditions in the sample plot are shown in tables 1 and 2 respectively. The sample plot is resampled, filtered and normalized (the dataset has been registered) to facilitate subsequent CHM establishment of point cloud data according to claim step 1.
Fig. 2 is a schematic diagram of exemplary plot data, where (a) is canopy point cloud data acquired by a ground-based lidar, and (b) is canopy point cloud data acquired by an airborne lidar. FIG. 3 is a schematic illustration of a canopy height model.
After treatment, the minimum distance between the point clouds was 0.05m. Because the research main body is a forest canopy, the point cloud is subjected to elevation filtering at a part below 8.9m according to the point cloud height distribution curve.
Table 1 on-board laser scanning (ALS) and foundation laser scanning (TLS) sensors and acquisition parameters. The specifications of the sensor are shown by the data sheet (RIEGL laser measurement systems, 2017,2019,2020 c).
Table 2 number of trees per plot and various data sources. Most of these trees are measured by different platforms and at different times. There are 1173 leaves falling tree ULS data. Of these, 133 trees were measured in autumn in 2019, 537 trees were measured in spring in 2020, and 503 trees were measured in both autumn in 2019 and spring in 2020.
Step 2, data partitioning: after the CHM is generated through normalization of the sample site cloud, the point cloud is roughly segmented through the gradient of the tree top by using a watershed algorithm, and finally the point cloud is segmented into 65 point cloud units, wherein the segmented point cloud units do not contain large gaps. FIG. 4 is a diagram illustrating the data partitioning result.
Step 3, poisson point generation: the point cloud unit after rough segmentation is enveloped by a cube, the center of the bottom of a canopy is taken as an origin, the boundary of the data after segmentation on a horizontal plane is taken as an interval, and poisson points are generated according to the density of one point per square meter (when the density is large, the influence of the computing force of a computer is caused, the computing efficiency of the method is greatly reduced, and therefore, the method is not suitable for setting the too high poisson point density). Fig. 5 is a schematic diagram of poisson point simulation.
Step 4, K-means clustering: and (3) taking the poisson points obtained in the step (3) as a clustering center, realizing that the point cloud corresponds to the blades, and taking the record of the movement of the sample center as a quantization index of the deviation degree of the spatial distribution and the poisson distribution of the blades in the iterative process by a K-means algorithm. FIG. 6 is a schematic diagram of the K-means clustering process.
Step 5, calculating a concentration index: setting 0,0.5,1,1.5,2,5 six different thresholds to measure whether the sample center deviates from the poisson point; the optimal threshold meeting the conditions can enable the inversion results of the two data to have higher correlation by comparing the decision coefficients and the mean square errors between the ground under different thresholds and the inversion results of the airborne data; meanwhile, in order to more sensitively distinguish the offset degree between the blade point cloud and the poisson point, the threshold value is as close to 0 as possible. Calculating the decision coefficient and mean square error between the ground and the inversion result of the airborne data, and selecting the minimum threshold value of 1.5 (R 2 = 0.6791) as a threshold for determining whether the cluster center deviates from the poisson distribution.
And (3) calculating the aggregation index according to the determined optimal threshold value of 1.5, and generating the vertical distribution of the sample aggregation index according to the inversion result of the airborne and foundation data with the height of 1m as the interval.
FIG. 7 is a schematic diagram of the method for inverting the concentration index. Fig. 8 is a graph of the offset distance results of the cohesive centers in the point cloud unit. Fig. 9 is a plot of the results of the inversion of the concentration index for the method at different thresholds, where (a) is based on the results of the ground-based lidar data acquisition and (b) is based on the results of the airborne lidar data acquisition. FIG. 10 is a comparison line graph of results of the present method for performing concentration index inversion on foundation and airborne lidar data, respectively. FIG. 11 is a comparison line graph of the present method and the results of the concentration index estimated based on the laser spot area model.
As can be seen from the above examples, the present example performs data acquisition and calculation on the overlapping portion of the foundation and the airborne data in the BR03 sample in the forest of the cloth Lei Teng, and performs comparative analysis and accuracy verification on the overlapping portion and the airborne data, and performs inversion based on the laser spot area model and the finite length averaging method to obtain a better correlation (R) 2 = 0.9716 rmse=0.0742), fully illustrating the applicability of the method of the invention. The principle of the laser spot area method is that a scanning principle of a foundation laser radar scanner is simulated, the real laser spot area is simulated and limited by scanner parameters, and the method based on poisson distribution and point cloud clustering is based on a statistical model and an unsupervised classification principle, so that the application range is wider, the inversion sample area range is also wider, and the effect of the method is superior to that of an inversion method based on the laser spot area model when the airborne data is processed and the collaborative inversion is carried out.
Claims (3)
1. A vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering is characterized by comprising the following steps:
step 1, acquiring and preprocessing point cloud data:
scanning by a laser scanner to obtain a data source of a target sample area, preprocessing, exporting the data source into a text format, and removing points lower than the height of the scanner; the preprocessing is to register the sample areas according to research requirements, and then to establish a canopy height model CHM for point cloud data after resampling, filtering and normalizing operation;
step 2, data partitioning:
after CHM is generated on the normalized point cloud through the step 1, the point cloud is subjected to rough segmentation by using a point cloud segmentation algorithm based on the gradient change of the point cloud at the top of the tree, so that a point cloud unit without inter-crown gaps is obtained;
step 3, poisson point generation:
enveloping the roughly-segmented point cloud units obtained in the step 2 by using a cube, taking the center of the bottom of the canopy as an origin, taking the boundary of the segmented data on the horizontal plane as an interval, and generating poisson points in the point cloud units according to the density not greater than the original point cloud of the canopy, wherein the poisson points are reference points of the spatial distribution of the canopy point cloud in an ideal state; wherein, the parameter lambda of poisson distribution is equal to the mathematical expectation, and the value of lambda is determined by the size of the point cloud unit;
step 4, K-means clustering:
taking the poisson points obtained in the step 3 as initial sample centers of clustering, clustering sample site clouds by using a K-means algorithm, establishing a corresponding relation between the point clouds and blades, and taking a record of movement of the clustering centers by the K-means algorithm in an iterative process as a quantization index of the deviation degree of the spatial distribution and the poisson distribution of the blades;
step 5, calculating a concentration index:
analyzing the deviation of the clustered central position obtained in the step 4 compared with the original poisson point position, setting a threshold value, and taking the proportion of the generated poisson point with the position deviation smaller than the set threshold value as the aggregation index result of a point cloud unit;
wherein Ω is a aggregation index, N Unbiased N is the number of clustering centers with offset distance smaller than a set threshold value in the clustering process All of which Number of total poisson points;
taking the projection areas of different point cloud units in the whole sample plot on the horizontal plane as weights, and carrying out weighted average on the different point cloud units according to the projection areas to obtain aggregation index distribution of the sample plot in the horizontal direction;
finally, the canopy is divided in the vertical direction according to the required height interval, and the aggregation index distribution of the whole sample plot on different heights is calculated.
2. The vegetation concentration index estimation method based on poisson distribution and point cloud clustering as claimed in claim 1, wherein: the data source in the step 1 is a single data source or a fusion data source, and when the single data source such as a foundation laser radar or an airborne laser radar is used, the sample area data source is directly preprocessed; when the fusion data source is used for collaborative inversion, foundation and airborne laser radar point cloud data are required to be acquired on a target sample, and sample data sources with the size not smaller than 10m and not smaller than 10m are selected at 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, wherein:
the threshold value setting in the step 5 specifically comprises the following steps: for the result of clustering the single data source, taking 0 as a threshold value; and selecting a minimum threshold value when the correlation is higher as a threshold value according to the correlation of the results among different data sources in the fusion data sources for the result of the fusion data source clustering, wherein the evaluation of the correlation is determined according to the decision coefficient among the clustering results among the different data sources, and considering that the correlation is higher when the decision coefficient is larger than 0.65.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310221090.6A CN116503727A (en) | 2023-03-09 | 2023-03-09 | Vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310221090.6A CN116503727A (en) | 2023-03-09 | 2023-03-09 | Vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116503727A true CN116503727A (en) | 2023-07-28 |
Family
ID=87323748
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310221090.6A Pending CN116503727A (en) | 2023-03-09 | 2023-03-09 | Vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116503727A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116881609A (en) * | 2023-09-06 | 2023-10-13 | 合肥工业大学 | Calculation method for gap rate of forest canopy in universality |
-
2023
- 2023-03-09 CN CN202310221090.6A patent/CN116503727A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116881609A (en) * | 2023-09-06 | 2023-10-13 | 合肥工业大学 | Calculation method for gap rate of forest canopy in universality |
| CN116881609B (en) * | 2023-09-06 | 2023-11-21 | 合肥工业大学 | Calculation method for gap rate of forest canopy in universality |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107274417B (en) | Single tree segmentation method based on airborne laser point cloud aggregation relation | |
| CN106529469B (en) | Unmanned aerial vehicle-mounted LiDAR point cloud filtering method based on self-adaptive gradient | |
| CN110287932B (en) | Road blocking information extraction method based on deep learning image semantic segmentation | |
| Ma | DEM generation and building detection from lidar data | |
| Briese | Three-dimensional modelling of breaklines from airborne laser scanner data | |
| CN110333554B (en) | NRIET rainstorm intelligent similarity analysis method | |
| Jin et al. | A point-based fully convolutional neural network for airborne lidar ground point filtering in forested environments | |
| CN111340723B (en) | Terrain-adaptive airborne LiDAR point cloud regularization thin plate spline interpolation filtering method | |
| CN109961470B (en) | Living standing tree She Shuxing accurate estimation method based on laser point cloud | |
| CN112068153B (en) | Crown clearance rate estimation method based on foundation laser radar point cloud | |
| CN111598780B (en) | Terrain adaptive interpolation filtering method suitable for airborne LiDAR point cloud | |
| CN115761303A (en) | Ground object classification method based on airborne laser radar point cloud and remote sensing image data | |
| CN112446844B (en) | Point cloud feature extraction and registration fusion method | |
| CN110988909A (en) | TLS-based vegetation coverage determination method for sandy land vegetation in alpine and fragile areas | |
| CN116503727A (en) | Vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering | |
| Li et al. | An iterative-mode scan design of terrestrial laser scanning in forests for minimizing occlusion effects | |
| CN115272876A (en) | Remote sensing image ship target detection method based on deep learning | |
| CN116698691B (en) | Atmospheric fine particle AOD inversion method and device, electronic equipment and storage medium | |
| CN111814666B (en) | Single tree parameter extraction method, system, medium and equipment under complex forest stand | |
| CN117636157A (en) | Method for determining physical change of ginkgo in autumn defoliation period by using ground laser scanning | |
| CN111060922B (en) | Tree point cloud extraction method based on airborne laser radar point cloud spatial distribution characteristics | |
| CN117035174A (en) | Method and system for estimating biomass on single-woodland of casuarina equisetifolia | |
| CN113724381B (en) | Dynamic three-dimensional scene rapid reconstruction method based on high-resolution remote sensing image | |
| CN113096048B (en) | Method and system for radiation correction of generalized cloud-driven and geometric collaborative remote sensing image | |
| CN112381029B (en) | Method for extracting airborne LiDAR data building based on Euclidean distance |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |