CN110598707B - Single plant tree crown segmentation method for airborne laser point cloud water pouring spreading and energy function control - Google Patents
Single plant tree crown segmentation method for airborne laser point cloud water pouring spreading and energy function control Download PDFInfo
- Publication number
- CN110598707B CN110598707B CN201910632238.9A CN201910632238A CN110598707B CN 110598707 B CN110598707 B CN 110598707B CN 201910632238 A CN201910632238 A CN 201910632238A CN 110598707 B CN110598707 B CN 110598707B
- Authority
- CN
- China
- Prior art keywords
- crown
- water
- tree
- spreading
- surface model
- 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.)
- Active
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 156
- 238000000034 method Methods 0.000 title claims abstract description 61
- 230000011218 segmentation Effects 0.000 title claims abstract description 28
- 241000196324 Embryophyta Species 0.000 title claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 49
- 238000001514 detection method Methods 0.000 claims abstract description 27
- 239000013598 vector Substances 0.000 claims description 14
- 230000007423 decrease Effects 0.000 claims description 9
- 230000001360 synchronised effect Effects 0.000 claims description 9
- 238000012804 iterative process Methods 0.000 claims description 4
- 238000013461 design Methods 0.000 claims description 3
- 238000011084 recovery Methods 0.000 abstract description 7
- 238000004590 computer program Methods 0.000 abstract description 5
- 244000166124 Eucalyptus globulus Species 0.000 abstract 3
- 241000219927 Eucalyptus Species 0.000 description 50
- 241000894007 species Species 0.000 description 14
- 230000000694 effects Effects 0.000 description 11
- 238000009499 grossing Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 238000011160 research Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 238000005259 measurement Methods 0.000 description 5
- 239000002023 wood Substances 0.000 description 4
- 230000001276 controlling effect Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000011065 in-situ storage Methods 0.000 description 3
- 238000011835 investigation Methods 0.000 description 3
- 238000000926 separation method Methods 0.000 description 3
- 239000002689 soil Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000000630 rising effect Effects 0.000 description 2
- DSSYKIVIOFKYAU-XCBNKYQSSA-N (R)-camphor Chemical compound C1C[C@@]2(C)C(=O)C[C@@H]1C2(C)C DSSYKIVIOFKYAU-XCBNKYQSSA-N 0.000 description 1
- 244000283070 Abies balsamea Species 0.000 description 1
- 235000007173 Abies balsamea Nutrition 0.000 description 1
- 241000723346 Cinnamomum camphora Species 0.000 description 1
- 244000050510 Cunninghamia lanceolata Species 0.000 description 1
- 244000103090 Eucalyptus robusta Species 0.000 description 1
- 241000218378 Magnolia Species 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229960000846 camphor Drugs 0.000 description 1
- 229930008380 camphor Natural products 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000035558 fertility Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- -1 locust Chemical compound 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 235000015816 nutrient absorption Nutrition 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000005507 spraying Methods 0.000 description 1
- 235000005221 swamp mahogany Nutrition 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Length Measuring Devices By Optical Means (AREA)
- Image Processing (AREA)
Abstract
The patent provides a crown segmentation algorithm based on airborne laser radar data, which utilizes a water spreading method on a Digital Surface Model (DSM) to delineate the boundary of a crown region, combines the gradient direction on the DSM with the crown boundary unit and the height difference between the last two crown vertices, establishes an energy function for controlling the water spreading, compensates the limitation of synchronously realizing the water spreading in each crown by a computer program, and optimizes the segmentation result of an overlapped region between adjacent crowns. The experimental result shows that the average detection rate (recovery) of fir in the mixed forest is 0.90, the accuracy (precision) of the detection rate is 0.71, the overall accuracy (f) of the detected tree is 0.80, and for eucalyptus, recovery= 94.76%, precision=0.81, and f=0.86. In all plots, the average detection precision of the crown diameters of the fir is lower than that of eucalyptus, and the detection precision of the heights of the fir is higher than that of the eucalyptus by 0.48%.
Description
Technical Field
The invention relates to a method for extracting single tree profiles from airborne laser radar data.
Background
The airborne laser radar is adopted to scan the forest tree, so that a large amount of tree data can be obtained. The singular data are data of the whole tree, and profile data of the single tree are often required when mechanical operations such as spraying the target with the accuracy are performed. How to process the laser point cloud data acquired by the airborne laser radar to obtain the profile data of the single tree is difficult to solve.
Disclosure of Invention
The invention aims to provide a single plant tree crown segmentation method for controlling the water pouring spreading and energy function of an airborne laser point cloud, which can obtain real and effective contour data of a single plant tree from laser radar data processing.
The invention discloses a single plant crown segmentation method for controlling the water pouring spreading and energy function of an airborne laser point cloud, which comprises the following steps:
a. converting tree point cloud data acquired by an airborne laser radar into a digital surface model C; rasterizing the digital surface model C of the stand, thereby converting the digital surface model C into a plurality of units C i,j Is a combination of (a); then using local maxima algorithm from the digital surface model CSearching for a unit c with a local maximum corresponding to the z value i,j Further, the detection of the crown vertex is completed;
b. pouring water into the whole crown forest stand by taking the crown top as a seed point of a water pouring spreading algorithm, and synchronously spreading water in each crown to draw the boundary of the tree;
the specific process of the step b of the single tree crown segmentation method is as follows:
inverting the digital surface model C of the tree, treating each spatial crown as a pit, and layering the inverted digital surface model C according to the height interval;
pouring water is started, and the water naturally flows to the lowest layer of the pit;
continuously pouring water to cause water level rise and synchronous region growth, dividing the edges of all crowns by the water to accurately separate the upper intersecting regions of adjacent crowns according to the edge information corresponding to the lower information filled with water as guidance;
and (3) completing the water pouring process, and drawing the spatial outline of each tree.
The specific process of water spreading in the single tree crown segmentation method is as follows:
after the division of the digital surface model C into different levels by height, the region growing starts with the highest level l detected 1 Crown apex t of (2) k These crown vertices t k Unit C located on digital surface model C i,j In, i.e. c i,j ==t k ,k∈[1,2,...r]&t k ∈l 1 The method comprises the steps of carrying out a first treatment on the surface of the Then combining the phenotype characteristic of the shape of the crown, namely, gradually reducing the z value of the scanning laser spot on the surface of the crown from the same polarity at the top of the crown to all directions, and adopting a neighborhood searching strategy to simulate water spreading; at the time of detecting crown apex t k And its location on the DSM, find a current layer/ 1 Each of which is associated with a cell c as crown apex i,j Adjacent units of (a)And satisfies the phenotype characteristic of crown shape +.>To aid in determining the water flow process; />Representing each cell c i,j The corresponding z value; then, the water region of each crown continues to spread according to the phenotypic characteristics of the crown shape, from +.>Start and synchronously spread to +.> wherein />Representation->And satisfy->Next, the iterative process is repeated until all cells c that are at the vertices of the crown i,j Partnership unit and belongs to current layer l 1 Is spread by water;
at a height interval l 1 After the inner water spreading is completed, the corresponding area of each pit immersed in water on the digital surface model C is inherited to the next height interval l 2 In (a) and (b); then, at the height interval l 2 Crowns with newly added crown vertices will also add to the region growth, unfinished water level growth and crown vertices present at the upper layer l 1 Will continue their region growing and the overlap area between two crowns will be correctly divided according to the water spread situation inherited from the upper layer; the water spreading pattern within each pit will be applied to the next level intervals in the level sequence until all level layers are processed.
According to the single tree crown segmentation method, the gradient direction information of the digital surface model C and the height difference information of each unit in the digital surface model C are combined, and the water spreading sequence is controlled, namely:
design energy function Boundarynernergy (t) k ) To evaluate the correctness of the water spreading sequence in each crown, during the water spreading process, boundaryenergy (t) k ) The value should continue to decrease until the minimum value is reached;
indicating that the measurement is made in a clockwise manner from vector pointing north to vector +.>Is included in the plane of the first part;
q represents the current boundary element c belonging to each crown in each step of water propagation i,j Is the number of (3);
representing the position from the crown vertex t k To the current cell c i,j Is a vector of (2); />Representing the difference in height between the two crown vertices where each crown water boundary unit should be closest to as the pouring process continuesThe reciprocal of the height difference is continuously increased, while the reciprocal of the height difference is continuously decreased, and beta is a corresponding weight coefficient.
Thus, we constructed the energy function Boundaryanegy (t) k ) The calculated value is continuously reduced in the water pouring process and the water level is continuously expanded in each tree crown, namely, the energy function is correspondingly an accurate single plant separation result when the energy function is reduced to a minimum value. Meanwhile, if the energy function value suddenly increases (breaks through the original descending trend) in the water pouring process of a single crown, the water level is represented to have error propagation in a certain crown and extends into other adjacent crowns. For such a case, the water level spreading sequence of the wrong crown (corresponding to the abrupt increase of the energy function) may be arranged to the end by adjusting the sequence of spreading of the water level in each crown, thereby allowing the overlapped crown areas to be correctly classified into each tree body.
The invention has the beneficial effects that: in the patent, a crown segmentation algorithm based on airborne laser radar data is provided, and a water spreading method on a Digital Surface Model (DSM) is utilized to describe a treeA border of the crown region; by combining the gradient direction on the DSM with the height difference between the crown boundary unit and the last two crown vertexes, an energy function for controlling the water spreading is established, which compensates for the limitation of the synchronous realization of the water spreading in each crown by the computer program, and optimizes the segmentation result of the overlapped area between the adjacent crowns. The present patent evaluates the efficiency of the present method with 12 plots of two tree species (fir and eucalyptus) having different crown sizes and tree heights and compares the tree parameters obtained according to the present patent method with measured values. The result shows that the average detection rate (recovery) of the fir in the mixed forest is 0.90, the accuracy (precision) of the detection rate is 0.71, the overall precision (f) of the tree is detected to be 0.80, and the fir is lower than various indexes (recovery= 94.76%, precision=0.81, and f=0.86) of the eucalyptus. Average detection accuracy (R 2 =0.79, rmse=0.25 m, rrmse=13.56%) is lower than the average detection accuracy (R 2 =0.77, rmse=0.24 m, rrmse=12.21%). High detection accuracy (R) 2 =0.87, rmse=0.89 m, rrmse=5.82%) higher than eucalyptus detection accuracy 0.48% (R) 2 =0.91, rmse=1.01 m, rrmse=6.30%). The method has the advantages of relatively simple calculation process, high material rate and good accuracy.
Drawings
FIG. 1 is a workflow diagram of a single tree profile extracted from airborne lidar data;
FIG. 2 is a schematic diagram of a water spreading expansion process in an inverted tree model;
FIG. 3 is a program operational diagram of the propagation process of water in corresponding DSMs and crowns;
FIG. 4 is a schematic diagram of control criteria for determining crown boundary design;
fig. 5 is a graph showing the ascending and descending trend of the spreading sequence and the energy function value of the structure on each crown.
FIG. 6 is a graph of the results of analysis of individual crowns of six selected plots.
FIG. 7 is a graph comparing the crown radius values measured in situ with crown radius values obtained according to the method of the present patent.
Fig. 8 is a graph comparing the tree height measured in situ with the tree height obtained by the method of the present patent.
Fig. 9 is a graph of the effect of height spacing on crown radius rRMSE and computation time.
Detailed Description
1 study area
The research area is located in peak forest, a subtropical artificial forest in Guangxi province in south China, the range is from 108 degrees 7' of east longitude to 108 degrees 38 degrees of east longitude, 22 degrees 49' of north latitude to 23 degrees 5' of north latitude, and the research area is located at the south side of the northern regression line, and belongs to the wind moist climate in subtropical seasons, the sunshine is sufficient, the rainfall is abundant, the frost is little, snow is not produced, the climate is mild, the annual average air temperature is about 21.6 degrees, the precipitation is 1304 millimeters, and the altitude is 78-468 meters. The area of investigation was 52 square kilometers. The terrain is mainly on a gentle slope, the gradient is about 20-47 m, and the average gradient is 33. The main tree species in the study area are eucalyptus (Eucalyptus robusta) and fir (Cunninghamia lanceolata).
2. Materials and methods
2.1 procedure
The workflow illustrates a technique for extracting individual tree profiles from airborne lidar data (fig. 1). (1) Rasterizing a Digital Surface Model (DSM) of a stand, converting from a digital surface model C to a plurality of cells C i,j Is a combination of (a) and (b).
From each grid cell c i,j Calculating the highest scanning point of the z value in the ground LiDAR data, correspondingly calculating the z value of the unit cell, and marking asMeanwhile, the point clouds in the vertical direction within each cell are stored in the corresponding grid cell. (2) Crown vertices are detected from the DSM using a smoothing window that adapts based on the size of the tree height by using a local maxima algorithm. (3) The entire crown Lin Fenjin row is poured with water and the boundary of the tree is delineated with simultaneous spread of water within each crown. (4) The accuracy of the crown boundary is optimized and judged by combining the gradient direction on the DSM and the height difference between the crown boundary unit and the nearest two crown vertices. (5) And the estimated result of the canopy boundary is compared with the measured data, so that the effectiveness of the algorithm is verified.
The workflow shown with reference to fig. 1 illustrates a technique for extracting individual tree profiles from airborne lidar data. (1) Converting LiDAR data scanned in a research area into DSM (namely C) by adopting a rasterization method; (2) local maxima algorithm is applied to the DSM to detect crown vertices (i.e., units c with local maxima z-values i,j ) The method comprises the steps of carrying out a first treatment on the surface of the (3) Positioning the boundary of each crown by adopting a water pouring method of water spreading in each crown; (4) optimizing and guiding the division of the crown boundary by utilizing the gradient direction and the height difference; (5) and (5) completing crown segmentation.
2.2 laser data
Laser radar data acquisition, using a Riegl LMS-Q680i laser scanner, with a flight height of 750 meters and a flight speed of 180 km/h, the side loop of one route is 65%. The sensor records the echo signal of a laser pulse with a time interval of 3 ns. The scanner emits 1550nm wavelength at 300kHz pulse repetition frequency, the scanning frequency is 80Hz, the scanning angle is + -30 DEG, and the field of view is 60 deg. The beam divergence was 0.5mrad and the beam footprint size was 37.5cm. The average distance between the ground scanning points is 0.45m, and the pulse density is about 9.58 points/m 2 . The final extracted point cloud is stored in LAS 1.2 format.
2.3 manually measured data for a sample plot
Site data in the investigation region were manually measured in the field, and 12 square plots (side length=20 meters) were built up at the data acquisition points in total according to forest type, tree growth density and complex terrain. All plots were classified into eucalyptus (n=6) and fir (n=6) according to tree species composition. The center of the field plots was identified using a Trimble GeoXH6000 Global Positioning System (GPS) (Trimble, sunnyvale, CA, USA), and these points were corrected using high precision real-time differential signals received by a Jiangsu continuous operation reference station (jscor), thus obtaining sub-meter precision. The present study measured heights greater than 2 meters and used a vertex V-type altimeter @Sweden) measured crown apex height. Crown width is defined by the crown apex position along two perpendicular directionsTo the average of the two values measured. These plots can be classified into low growth density ( plots 1,4, 10 and 11. Low density trees range from 0 to 25 per plot), medium growth density ( plots 2, 5, 7, 8 and 12. Medium density trees range from 25 to 35 per plot), and high growth density ( plots 3,6 and 9. High density trees range from more than 35 per plot). In addition, in the research area of fir, the phenomenon of tree thinning occurs in recent decades, which results in mixed growth of various broad-leaved tree species such as camphor, locust, elm, magnolia and the like. As a fast-growing tree species, eucalyptus has stronger soil nutrient absorption capacity, so that other vegetation in the area dies, and a pure artificial forest which is not interfered by other tree species is formed.
2.4 DSM generation and crown vertex detection
First, point cloud data acquired by an airborne laser radar is classified into ground points and above-ground points (scanned tree points) with a morphology-based filter. Then, each cell c is used i,j The maximum height of all scan tree points within generates a DSM with a grid cell size d. Cell c without any point cloud i,j Generated by high interpolation of its neighboring cells, where i, j represents c i,j The ith row, jth column of cells in the generated set of cells C (i.e., DSM).
In the last decade, the phenomenon of tree thinning occurs in the fir wood artificial forest research area. Many broadleaf trees grow in these woodlands and are mixed with fir, resulting in inconsistent forest canopy composition and crown shape characterization. Broad leaf crowns have many wild branches, which makes it difficult to detect crown vertices. Therefore, a variable-scale Gaussian kernel smoothing filter is adopted to carry out smoothing treatment on DSM, and interference generated by mixed growth of other different broadleaf tree species in the current land is eliminated. According to the prior knowledge of the growth characteristics of the fir and the eucalyptus, the higher the fir and the eucalyptus, the larger the crown size. We therefore analyze the DSM height and use the DSM height as an important indicator for selecting the size of the smoothing filter window. For forest plots with larger average tree heights, a smoothing filter with larger window sizes is used to morphologically filter the corresponding forest plots, and vice versa. Here, the window size of the smoothing filter is set to 3x3 unit size for plots with smaller tree heights; the 5x5 unit size is set for plots with larger tree heights. After our DSM smoothing process, a local maximum lookup algorithm is used to identify the location of each crown vertex in the filtered DSM and use it as a seed point for the pour spread algorithm to achieve crown segmentation.
2.5. Water spread concept and energy function control
2.5.1 initial segmentation based on the Water pouring idea
In order to accurately extract the boundary parameters of each tree, an algorithm based on the water pouring idea is studied, and the spatial distribution of the single tree is obtained from the DSM. According to the concept of pouring water into a pit (inverting the profile of a crown), the water flowing into the pit is synchronously controlled according to the height of the tree. The schematic diagram of fig. 2 shows the detailed spreading process of the water flowing into each pit. To facilitate implementation of our algorithm using computer programming, the inverted DSM of each forest land is based on the height spacing H interval Is vertically divided into s layers, and during pouring, water first enters the lowest layer until the lowest layer is filled with water. When the water level rises, the water diffuses to an adjacent layer (upper layer). In this order, the pouring process is continued until the highest layer is filled with water. The water first enters the upper part of the forest canopy, i.e. the lower layer of the pits, there is less overlap between the upper crowns of the crowns, and the individual crown profile can be clearly outlined by the edges of the water in each pit. As the water pouring process proceeds, the recorded outline of the upper end of the crown and the lower layer morphological information of each pit are inherited, so as to guide the segmentation of the upper layer overlapping area of the crown (fig. 2).
Fig. 2 illustrates the detailed steps of the algorithm. Step one, inverting the DSM of the tree and treating each spatial crown as a pit, and layering the inverted DSM according to the height interval. And secondly, pouring water, wherein the water naturally flows to the lowest layer of the pit. Step three: the water pouring process is continued to lead to the rising of the water level and the synchronous growth of the areas, so that the edges of all crowns are divided, and the upper intersecting areas of adjacent crowns are accurately separated according to the edge information corresponding to the water filled lower-layer information as guidance. And fifthly, finishing the water pouring process and accurately drawing the spatial outline of each tree.
The expansion process of water spreading in the inverted tree model is described with reference to the schematic diagram of the algorithm shown in fig. 2 to accurately extract individual crowns. Take as an example a side view of three trees of different species. (b) A side view of a 180 degree inverted tree appears to be three pits. (c) Water was poured into the three pits, and the boundaries of the crown were delineated by the boundaries of the water. (d) And (e) a first step of layering the DSM according to a given height interval and pouring water into all the pits (inverted tree crowns), the water flowing first into the lowest layer of the pit. (f) And (g) along with the water pouring process, the water level rises in each crown and synchronously grows along with the region, the edges of each crown are divided, and the intersection region at the upper part of the adjacent crowns is accurately separated according to the edge information corresponding to the water filled lower layer information as guidance.
Our water spreading process is actually operated on DSM and the mathematical expression of our process is set out below. After the division of the DSM into different levels by height, region growing starts with the highest level/detected 1 Crown vertices of (a) located at element C of the digital surface model C i,j In, i.e. c i,j ==t k ,k∈[1,2,...r]&t k ∈l 1 . Then, combining the phenotype characteristic of the crown shape, namely that the z value of the scanning points on the crown surface gradually decreases from the top of the crown to all directions, adopting a neighborhood search strategy to simulate water spreading. At the time of detecting crown apex t n And its location on the DSM, find a current layer/ 1 Each of the inner and c i,j Adjacent units of (a)And satisfies the phenotype characteristic of crown shape +.>To help determine the water flow process. Then, the water region of each crown continues to spread according to the phenotypic characteristics of the crown shape, from +.>Start and synchronously spread to +.> wherein />Representation->And satisfy->Next, the iterative process is repeated until all vertices c of the crown i,j Partnership and belongs to the current layer l 1 Is spread by the water (fig. 3 a).
At a height interval l 1 After the inner water has spread, the corresponding area of each pit in the DSM immersed in water will be inherited to the next height interval l 2 Is a kind of medium. Then, at the height interval l 2 Crowns with newly added crown vertices will also add to the region growth, unfinished water level growth and crown vertices present at the upper layer l 1 Will continue their region growing and the overlap area between two crowns will be correctly divided according to the water spread situation inherited from the upper layer. The water spreading pattern within each pit will be applied to the next level intervals in the level sequence until all level layers are processed, as shown in particular in fig. 3.
The propagation of water in the corresponding DSM and crown is illustrated by reference to the program run chart shown in fig. 3. With simultaneous pouring of water from the highest level to the lowest level to each crown, (a 1), (b 1), (c 1), (d 1), (e 1), and (f 1) represent side views of LiDAR data. (a2) The term "b 2", "c 2", "d 2", "e 2" and "f 2" are plan views of the water expansion in the DSM in synchronization with the pouring processes shown in (a 1), (b 1), (c 1), (d 1), (e 1) and (f 1), respectively. The lidar data and the corresponding water level expanded region on the DSM are set to the same gray scale. Bold lines represent the boundaries of the crown submerged in water for each step of the algorithm.
2.5.2 control criteria for the Water spread sequence of the crowns of the trees
It is impossible to actually realize the synchronous spreading of water in each crown by using a computer program. The computer can only treat the spread of water in each crown in a random order in turn at the same height interval. This uncertainty sequence may lead to false water level spreads when adjacent trees have overlapping crown areas in the same height interval. For example, the water propagation delays in multiple crowns, since the randomness of the computer program gives priority to the water propagation operations to crown B, the intersection region that would otherwise belong to crown a is occupied by adjacent crown B. Thus, there is an urgent need for a criterion to control the water spreading sequence in each crown and to provide a criterion to evaluate the rationality of the sequence. Therefore, we combine the gradient direction information of the DSM with the height difference information of each unit on the DSM to constitute the control criteria for water spread sequence management. Specifically, the following is described.
A calculated gradient vector representing each cell on the DSM,/->Indicating that the measurement is made in a clockwise manner from vector pointing north to vector +.>Is included in the bearing. Then, an energy function was designed to evaluate the correctness of the water spreading sequence in each crown, as follows.
Wherein Q represents the current boundary element c belonging to each crown at each step of water propagation i,j Is a number of (3).Representing the position from the crown vertex t k North-pointing vectors. />Representation vector->To vector->Is a clockwise angle of (c). When boundary unit c i,j Belonging to the tree crown with real crown vertex t k Boundary element c when crown of (2) i,j Angle value of (2)Approximately equal to the calculated gradient-related angle +.>Thus, the correct water spread in each crown will result in the equation +.>Reaching a minimum. Conversely, when an improper water spreading sequence results in an improper crown boundary division, from the wrong crown apex t k+1 Calculated +.>With the current boundary c i,j Is not uniform in gradient direction, which results in +.>The value of (2) becomes larger.
In addition, the optimal segmentation result of the crown is guided by the height difference in the energy function. According to the idea, the difference in height from the point on the mountain's concave route between two mountains to the highest peak of the two mountains should be the largest. Thus, the units c on the crown boundary due to water spreading i,j Height difference between adjacent two nearest crown verticesWill gradually increase as the water spreads. On the real boundary of the crown formed by the spread of water, the difference in height between the unit and the adjacent crown apex +.>The maximum should be reached. The second term to the right of equation (3) will thus gradually decrease with the water spread, and in the course of the water spread finding the true boundary of each crown, the value of equation (3) for each crown profile should continue to decrease until a minimum value is reached, which represents the water spread reaching the true boundary of the crown, according to the correct segmentation procedure. In contrast, the value of equation (3) increases during the water propagation, which suggests that there is a segmentation error in the cross crown area, resulting in a backtracking of the program. A specific illustration is shown in fig. 4.
The control criteria designed for determining the crown boundary are illustrated with reference to the schematic diagram shown in fig. 4. (a) a computational graph representing the direction of the DSM gradient. (b) is a close-up of the rectangle in (a). When any unitBelongs to the vertex t of the true crown k Gradient direction during crown of (2)>Should be approximately equal to the peak t of the real crown from the peak t of the real crown k To the current unit->Direction of (i.e.)And unit->Is a crown vertex t k Rather than crown vertex t k+1 So (2) crown of the treeThis destroys the downward trend of the energy function value and suggests the occurrence of false water spread in the intersection area between crowns based on the rise of the energy function. (c) Illustrating the evaluation of the accuracy of the crown segmentation in combination with the height difference between the depicted crown boundary element and the two nearest crown vertices (i.e., element c on the true boundary of the crown i,j The height difference from the nearest two crown vertices is compared with any other element c middle The corresponding height differences are all large). (d) The boundary elements that are two closest to the crown vertices are determined using nearest neighbor rules.
2.5.3 energy function to guide crown segmentation
In order to make up the limitation of the computer program on the synchronous water spreading of each crown, the water spreading sequence is regulated by using the constructed energy function value, and then the segmentation result of the crossed crown area is optimized. From the description of equation (3) in section 2.5.2, a continuous decrease in the energy function value is reasonable in the water spread iteration step of each crown. If an unexpected increase in the energy function occurs during the water spreading process, this means that the water in the corresponding crown spreads to the adjacent crown area. Therefore, the corresponding water spreading step of the tree is canceled, the priority of water spreading is given to other subsequent crowns, the water spreading sequence of the current tree is adjusted, and the global optimization of the crown segmentation result is ensured. In the 1 st (eucalyptus) plot, the water spread in 29 crowns is shown in fig. 5, where the water spread boundaries in the 100 th, 500 th and 1168 th iteration steps are represented in different grey scales, respectively. Meanwhile, in the iterative process of our algorithm, the energy function value calculated according to the boundary divided by the water spread in each crown is shown in fig. 5, and the global optimality of water level expansion is ensured through the adjustment of our algorithm.
Referring to fig. 5, the spreading sequence of our water spreading algorithm on each crown is judged by the rising and falling trend of the constructed energy function values. (a) Each crown water spread and water boundary during an iteration is described by a line of different grey scale. (b) The descending trend of the energy function value of each crown is taken as the order of evaluating the spreading of water in each crown, and the correct segmentation result of the intersection area between crowns is ensured.
2.6 evaluation of accuracy
And (3) comparing the result obtained in the section 2.5 with measured data, and verifying the accuracy of the algorithm. And calculating the accuracy comparison of the crown top point of the detected tree with the crown top point actually measured in the research area by using a formula.
r (recovery) is the tree detection rate, p (precision) is the accuracy of the tree being detected, f is the overall accuracy of the tree being detected, TP (true positive) is the number of correctly detected trees, FN (false negative) is the number of undetected trees (missing errors), FP (false positive) is the number of trees that are not present in the region but added in error (misclassification errors).
In order to verify the matching degree of the outline of the single plant crown and the outline of the artificial crown, we compare the actual measured crown radius with the crown radius detected by our algorithm to verify the accuracy of the outline of the crown. We selected the correctly detected crowns for crown radius comparison. The accuracy of crown radius and height is assessed by R square, root Mean Square Error (RMSE) and relative root mean square error (relative RMSE, i.e., the ratio of RMSE to observed mean).
3. Results
Our algorithm was applied to 12 plots including fir and eucalyptus woods in this study. Fig. 6 shows the results of individual crowns of six selected plots based on DSM. The algorithm is evaluated, and the accuracy of the algorithm is verified. We consider three typical eucalyptus plots (plots 1-3) and three fir plots (plots 4-6) of different tree growth densities. FIG. 6 shows the position of a set of three square plots (1, 2, 3) of growing eucalyptus having a length of 20 meters; the other group is the location of three square plots (4, 5, 6) of growing fir wood of length 20 meters. Eucalyptus tree growth density (plant/m 2 ) From low to high, 0.06 (plot 1), 0.09 (plot 2), and 0.17 (plot 3), respectively. Tree growth density of fir plot (plant/m 2 ) From low to high, 0.06 (plot 4), 0.07 (plot 5), and 0.09 (plot 6) are sequentially present. Subsequently, 12 plots (including the 6 selected plots) were used for accuracy assessment. Table 1 shows the accuracy of our algorithm for these 12 graphs. The algorithm performance is good, and the average detection rate (r) is 0.93; furthermore, the highest accuracy (f) of the 12 graphs is 0.92. In the eucalyptus group (plots 1-3, 7-9), plot 2 has the greatest r value (0.97), plot 3 has the greatest p value (0.89), and plot f has the greatest f value (0.92). Among the fir wood (plots 4-6, 10-12), plot 4 has the greatest r value (0.96), the greatest p value (0.73), and the greatest f value (0.83). As is evident from table 1, the misclassification errors in the graph are much larger than the omission errors. The land block No. 6 has larger misclassification errors and omission errors. The problem of excessive segmentation of fir is more serious due to the severe needle-broad mixing and high standing wood density in the forest.
FIG. 7 shows the range of RMSE values for crown radii from 0.13m to 0.31 m. The average value of RMSE of eucalyptus plots is 0.24m, the rMSE value of crown radius ranges from 11.76% to 13.29%, and the average value is 12.21%. Land parcel 3 (rmse=0.17 m, tree growth density=0.17, f=0.92) had a minimum rRMSE of 11.83%. The range of RMSE values for the crown width of fir was 0.17-0.37 m with an average value of 0.24m. The rMSE value of the crown radius ranges from 12.05% to 14.05%, and the average value is 13.56%. Land parcel 6 (rmse=0.19 m, tree growth density=0.09, f=0.75) had a minimum rRMSE of 13.01%. In all plots, RMSE was less than 0.37m at maximum and rrmse was less than 14.05% at maximum. The result shows that the algorithm has stronger tree crown outline depicting capability. The average rMSE value of fir was Yu Anshu forest. The result shows that the algorithm has certain advantages in the aspect of eucalyptus canopy precision. FIG. 8 shows that the RMSE values for eucalyptus plot tree heights range from 0.82m to 1.71m, with an average of 1.01m, and the tree height rMSE values range from 5.65% to 6.81%, with an average of 6.30%. Land parcel 9 (rmse=0.92 m, tree growth density=0.12, f=0.88) had a minimum rRMSE of 5.65%. The width RMSE value of the fir crown is between 0.87m and 0.94m, and the average value is 0.89m. Tree height rRMSE values range from 4.28% to 6.96% with an average of 5.83%. Land block 11 (rmse=0.88 m, tree growth density=0.06, f=0.81%) rRMSE was the lowest, 4.28%. In all plots, RMSE was at most 2.46m and rrmse was at most 10.58%. The result shows that the algorithm can accurately obtain the tree heights of eucalyptus and fir. The data shows that the difference in rRMSE values for two tree species at tree height is only 1%. The result shows that the algorithm has certain universality in the aspect of tree height extraction.
TABLE 1 evaluation of accuracy of tree segmentation in test sample soil
TP number of correctly detected trees. FP: number of additional trees not present in the area (misclassification error). FN: number of trees not detected (missing error). And r, tree detection rate. p, correctness of the detected tree. f, overall accuracy of the detected tree.
Referring to fig. 6, a crown is detected for six selected plots using our algorithm. (a), (b), (c), (d), (e) and (f) represent individual crown divisions of plots 1,2,3, 4,5 and 6, respectively. Classification by tree growth density of eucalyptus ( plots 1,2, 3) and fir ( plots 4,5, 6), black (gray) border indicates detected crowns. The black dots represent the crown vertices measured manually and the black crosses represent crown vertices detected with our algorithm.
Referring to fig. 7, the crown radius values measured in situ are compared to crown radius values detected in our algorithm for 12 areas. (a), (b), (c), (d), (f), (g), (h), (i), (j), (k) and (l) represent crown radius comparison graphs of plots 1,2,3, 4,5,6, 7, 8, 9, 10, 11 and 12, respectively.
Referring to fig. 8, the tree height measured in the field is compared to the tree height detected by our algorithm estimated 12 plots. (a), (b), (c), (d), (e), (f), (g), (h), (i), (j), (k) and (l) represent tree height comparisons of 1,2,3, 4,5,6, 7, 8, 9, 10, 11 and 12 plots, respectively.
4. Discussion of the invention
4.1 Algorithm efficiency analysis
First, we analyzed the detection rate of crown vertices in the study area. In fig. 6, we divide the 6 plots selected into three categories based on tree growth density, including low growth density (plots 1, 4), medium growth density (plots 2, 5), and high growth density (plots 3, 6). For eucalyptus, the overall accuracy of the tree (f) detected is highest, and the tree growth density of the plots is also high (0.17 tree/m 2 ). As the tree growth density increases, the detection accuracy tends to increase. This is because eucalyptus plots are planted uniformly and have a certain regularity. Meanwhile, the growth density of the land parcel tree is increased, and the smooth window with smaller size is selected from the uniform land parcel, so that the smooth effect on the DSM form is good. As the growth density of the fir tree increases, the measured crown vertex accuracy decreases. The result shows that the algorithm has a good segmentation effect on the fir low-tree-growth-density land block. In plots 4 and 6, when the tree growth density increased by 0.03 plants/m 2 At this time, the total accuracy (f) of the fir detected in the land parcels is reduced by 0.08. This result shows that the overall accuracy of the tree detected in the graph is more sensitive to the tree growth density. When the tree growth density is too high (parcel 6=0.09 plants/m 2 ) Over-segmentation is severe, which is the growth of firCommon errors when the density is high. The overall accuracy (f) of the detection result of the fir is lower than that of the eucalyptus due to the difference of the characteristics and the human factors of the two tree species of the fir and the eucalyptus. Eucalyptus has stronger soil fertility competitiveness and is an economic tree species. Every few years, eucalyptus is cut to be replaced by economic value, and new seedlings are planted; thus, few other plants are present in the lower layer of eucalyptus. Fir has a weaker competing factor on the forest, and national policies define fir as a nationally protected non-fellable plant. Some secondary plants also grew under fir; thus, the growth environment of fir is more complex than that of eucalyptus under the same climatic conditions. The above causes the overall accuracy of fir to be lower than that of eucalyptus. The result shows that (1) the tree detection precision of the southern fir is reduced along with the increase of the tree growth density; (2) The detection precision of eucalyptus is increased along with the increase of the tree growth density; and (3) the detection precision of the fir is lower than that of eucalyptus.
Next, we analyzed the accuracy of the crown radius. The accuracy of crown radius detected by our algorithm increases with increasing tree growth density of the same species. In fig. 7 and 8, we analyzed the crown radius and tree height and compared with the field measurements in plots 1-6. The accuracy of eucalyptus and fir crown radius values increases with increasing tree growth density. In eucalyptus and fir, the rRMSE values for high and low tree growth densities are 1.46% (plots 1, 3) and 1.05% (plots 4, 6), respectively. The results show that an increase in the tree growth density has an effect on the accuracy of the crown radius. In plot 1 and plot 4, the densities of the two plots are similar, and the canopy radius accuracy of eucalyptus is 0.76% higher than that of fir. This is because the shape of the fir crown varies more than that of the eucalyptus crown, increasing the difficulty of crown radius calculation. Because of the high calculation difficulty, the eucalyptus precision is slightly higher than that of fir at the same tree growth density. In our algorithm, we segment the crown according to the crown vertices that are correctly detected. At higher tree growth densities, crowns of both tree species are more uniform (few unique shapes) and the effect of secondary trees is reduced. Under the condition, the synchronous searching mode is utilized to search each height interval of each crown, and the algorithm effect is good. The synchronous spreading of water makes the boundary area of each crown more regular, and makes the determination of the crown boundary more accurate.
For the tree heights of six plots, all plots have higher accuracy in detecting the tree heights. The results show that the variation of the tree growth density has no significant effect on the accuracy of estimating the tree height. After detecting crown vertices, our algorithm shows that the tree height is independent of the accuracy of crown radius detection. rRMSE values were below 8.75% for all plot heights. The result shows that the algorithm has a good tree height acquisition effect, and the tree species and land block tree growth density have no obvious influence on the tree height acquisition.
4.3 height spacing H interval Influence on algorithm
In addition to smoothing windows and point cloud density, altitude spacing is a major parameter affecting algorithm accuracy. The height interval may control the accuracy of the boundary and the steps of the search process during the search. The control of each water run is based on the altitude interval. As the height spacing decreases, the steps performed in the program increase, thereby increasing the computation time of the algorithm. To further investigate the effect of fir and eucalyptus on the height spacing, we adjusted the height spacing threshold from 0.2 meters to 2 meters, changing 0.2 meters each time, and analyzed the variation of crown radii and calculated time rRMSE values for eucalyptus (plots 1-3) and fir (plots 4-6). Fig. 9 shows the effect of height spacing on crown radius rRMSE values. In fig. 9, the rRMSE values of the crown radii detected by eucalyptus and fir continue to increase as the height spacing increases. The smaller height interval can distinguish the peak positions of the crowns belonging to different height layers, thereby ensuring better synchronous water level spreading. The larger height interval makes the data volume of the search result with increased area large in each step, the interference is more, and the overall accuracy is reduced. Fig. 9 illustrates that setting the height spacing to 0.2m results in the lowest rRMSE values for eucalyptus and fir. Fig. 9 shows that a small height spacing increases the accuracy of detection of the crown radius. The variation of rMSE value distribution of fir crown radius with algorithm height interval adjustment is 0.5% higher than that of eucalyptus. Thus, the rRMSE of fir is more sensitive than eucalyptus in the height interval.
The calculated time for eucalyptus and fir decreases rapidly with increasing height spacing, as shown in fig. 9 c-d. The height difference of the investigation region is in the range of 20-40m, and the height interval is set to 0.2m for better crown radius accuracy. While lower height spacing may improve the accuracy of fir and eucalyptus crown radii, using our algorithm for larger woodland areas requires more computation time. The selection of the height interval has important influence on the accuracy and calculation time of crown radius estimation.
See the effect of the height spacing on crown radius rRMSE and computation time shown in fig. 9. (a) Distribution of eucalyptus detected crown radius rRMSE calculation values with height interval. (b) The radius rRMSE detected by fir crown varies with the height interval. (c) Distribution of calculated time of eucalyptus individual plant isolation with height interval. (d) The calculation time of fir single plant separation varies with the height interval.
5. Conclusion(s)
The accurate single plant separation algorithm can effectively improve the sustainable accurate management quality of forests. Our method is largely divided into three parts. First, tree top information is extracted from the on-board laser point cloud by a local maxima algorithm and gaussian kernel smoothing filtering. Secondly, we simulate the physical principle of pouring water into an inverted tree surface model to achieve segmentation of the initial crown. Thirdly, the gradient direction and the height difference on the DSM are utilized to construct an energy function to assist in determining the crown boundary, so that the problem that the crown overlapping area cannot be accurately segmented is solved, and the defect that water synchronously spreads in each crown is overcome. And finally, comparing the calculated result with the actual measurement result of the subtropical zone research area of the south high peak woodland. The result shows that the algorithm has higher precision in the aspect of tree top detection and crown radius detection of eucalyptus and fir. The accuracy of eucalyptus crown detection is higher (recovery=0.94, precision=0.81, f=0.86), and fir is secondary (recovery=0.90, precision=0.71, f=0.80). At the same time, the average accuracy of the fir crown radius (R 2 =0.79, rmse=0.25 m, rrmse=13.56%) is lower than eucalyptus (R 2 =0.77, rmse=0.24 m, rrmse=12.21%). The fir and eucalyptus trees have similar high precision (less than 1%). Meanwhile, fir (crown radius rRMSE value range of 3.30%) is sensitive to high intervalsThe degree is higher than that of eucalyptus (the range of the crown radius rMSE is 2.73%), and more interference is caused by more needle-broad mixing phenomenon caused by the intermediate afforestation strategy of fir-like plots. But the overall accuracy achieved by our method meets the requirements of fine measurement of forest stands.
Claims (2)
1. The single plant tree crown segmentation method for controlling the water pouring spreading and energy function of the airborne laser point cloud is characterized by comprising the following steps of: it comprises the following steps:
a. converting tree point cloud data acquired by an airborne laser radar into a digital surface model C; rasterizing the digital surface model C of the stand, thereby converting the digital surface model C into a plurality of units C i,j Is a combination of (a); then, using local maximum algorithm, searching the unit C corresponding to the local maximum of the z value from the digital surface model C i,j Further, the detection of the crown vertex is completed;
b. pouring water into the whole crown forest stand by taking the crown top as a seed point of a water pouring spreading algorithm, and synchronously spreading water in each crown to draw the boundary of the tree, wherein the concrete process is as follows:
inverting the digital surface model C of the tree, treating each spatial crown as a pit, and layering the inverted digital surface model C according to the height interval;
pouring water is started, and the water naturally flows to the lowest layer of the pit;
continuously pouring water to cause water level rise and synchronous region growth, dividing the edges of all crowns by the water to accurately separate the upper intersecting regions of adjacent crowns according to the edge information corresponding to the lower information filled with water as guidance;
completing the water pouring process, and drawing the space outline of each tree;
the specific process of water spread is as follows:
after the division of the digital surface model C into different levels by height, the region growing starts with the highest level l detected 1 Crown apex t of (2) k These crown vertices t k Unit C located on digital surface model C i,j In, i.e. c i,j ==t k ,k∈[1,2,...r]&t k ∈l 1 The method comprises the steps of carrying out a first treatment on the surface of the Then combining the phenotype characteristic of the shape of the crown, namely, gradually reducing the z value of the scanning laser spot on the surface of the crown from the same polarity at the top of the crown to all directions, and adopting a neighborhood searching strategy to simulate water spreading; at the time of detecting crown apex t k And its location on the DSM, find a current layer/ 1 Each of which is associated with a cell c as crown apex i,j Adjacent units of (a)And satisfies the phenotype characteristic of crown shape +.>To aid in determining the water flow process; then, the water region of each crown continues to spread according to the phenotypic characteristics of the crown shape, from +.>Start and synchronously spread to +.> wherein />Representation->And satisfy->Next, the iterative process is repeated until all cells c that are at the vertices of the crown i,j Partnership unit and belongs to current layer l 1 Is spread by water;
at a height interval l 1 After the inner water spreading is completed, the corresponding area of each pit immersed in water on the digital surface model C is inherited to the next height interval l 2 In (a) and (b); then, at the heightInterval l 2 Crowns with newly added crown vertices will also add to the region growth, unfinished water level growth and crown vertices present at the upper layer l 1 Will continue their region growing and the overlap area between two crowns will be correctly divided according to the water spread situation inherited from the upper layer; the water spreading pattern within each pit will be applied to the next level intervals in the level sequence until all level layers are processed.
2. The method for dividing the single tree crowns according to claim 1, wherein the method comprises the following steps:
the gradient direction information of the digital surface model C and the height difference information of each unit in the digital surface model C are combined to control the water spreading sequence, namely:
design energy function Boundarynernergy (t) k ) To evaluate the correctness of the water spreading sequence in each crown, during the water spreading process, boundaryenergy (t) k ) The value should continue to decrease until the minimum value is reached;
representing complianceMeasuring from north-pointing vector to vector in a clockwise manner>Is included in the plane of the first part;
q represents the current boundary element c belonging to each crown in each step of water propagation i,j Is the number of (3);
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910632238.9A CN110598707B (en) | 2019-07-13 | 2019-07-13 | Single plant tree crown segmentation method for airborne laser point cloud water pouring spreading and energy function control |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910632238.9A CN110598707B (en) | 2019-07-13 | 2019-07-13 | Single plant tree crown segmentation method for airborne laser point cloud water pouring spreading and energy function control |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110598707A CN110598707A (en) | 2019-12-20 |
CN110598707B true CN110598707B (en) | 2023-06-06 |
Family
ID=68852794
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910632238.9A Active CN110598707B (en) | 2019-07-13 | 2019-07-13 | Single plant tree crown segmentation method for airborne laser point cloud water pouring spreading and energy function control |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110598707B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20230350065A1 (en) * | 2020-09-30 | 2023-11-02 | Nanjing Maoting Information Technology Co., Ltd. | Method of individual tree crown segmentation from airborne lidar data using novel gaussian filter and energy function minimization |
CN115131678B (en) * | 2022-07-01 | 2023-03-07 | 杭州万林数链科技服务有限公司 | Block chain architecture construction method and system for digital intelligent forestry |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009022278A (en) * | 2007-06-19 | 2009-02-05 | Tokyo Electric Power Co Inc:The | Method for recognizing treetop, device for recognizing treetop, and program for recognizing treetop |
CN107832681A (en) * | 2017-10-16 | 2018-03-23 | 福州大学 | The high evaluation method of forest list ebon of joint LiDAR point cloud and synchronous remote sensing image |
-
2019
- 2019-07-13 CN CN201910632238.9A patent/CN110598707B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009022278A (en) * | 2007-06-19 | 2009-02-05 | Tokyo Electric Power Co Inc:The | Method for recognizing treetop, device for recognizing treetop, and program for recognizing treetop |
CN107832681A (en) * | 2017-10-16 | 2018-03-23 | 福州大学 | The high evaluation method of forest list ebon of joint LiDAR point cloud and synchronous remote sensing image |
Also Published As
Publication number | Publication date |
---|---|
CN110598707A (en) | 2019-12-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Liu et al. | Three decades of ephemeral gully erosion studies | |
CN107063197B (en) | Reservoir characteristic curve extraction method based on spatial information technology | |
CN108227041B (en) | Horizontal visibility forecasting method based on site measured data and mode result | |
CN104063718B (en) | The method with selection remotely-sensed data and sorting algorithm in area reckoning is recognized in crop | |
EP3574746A1 (en) | Method and system for expressing airborne cloud seeding line considering cloud water | |
CN102628944B (en) | Stratus cloud and convective cloud automatic recognition method based on Doppler radar data | |
CN104463164B (en) | It is a kind of based on umbrella frame method and crown height than trees canopy structure information extracting method | |
CN108981616B (en) | Method for inverting effective leaf area index of artificial forest by unmanned aerial vehicle laser radar | |
CN110598707B (en) | Single plant tree crown segmentation method for airborne laser point cloud water pouring spreading and energy function control | |
CN105891833A (en) | Method of identifying warm cloud precipitation rate based on Doppler radar information | |
CN107330898A (en) | Altitudinal vegetation zone quantitatively delineates computational methods and system | |
Rowe et al. | Investigation of microphysical processes occurring in organized convection during NAME | |
Zhang et al. | Remote sensing monitoring of gullies on a regional scale: a case study of Kebai region in Heilongjiang Province, China | |
CN113866770A (en) | Hail cloud early identification method and storage medium | |
CN106372362B (en) | Loessial gulch retrogressive erosion horizon prediction method based on remote sensing technology | |
CN113739707B (en) | Sea ice information extraction method | |
Grabiec et al. | Distribution of snow accumulation on some glaciers of Spitsbergen | |
CN116541658B (en) | Urban near-ground kilometer height wind profile measurement analysis method and device | |
Mukhamedjanov et al. | The use of satellite data for monitoring rivers in the Amu Darya basin | |
Sokol et al. | Nowcasting of hailstorms simulated by the NWP model COSMO for the area of the Czech Republic | |
CN117129997A (en) | Dual-polarization radar differential phase shift rate foot identification method based on convection storm | |
RU2769575C1 (en) | Method for generating soil degradation maps | |
CN115435674A (en) | Method for jointly inverting north sea ice accumulated snow depth through field observation and satellite remote sensing | |
Zlatanović et al. | Comparison of an automated and manual method for calculating storm runoff response in ungauged catchments in Serbia | |
da Silveira et al. | Observational Investigation of the Statistical Properties of Surface-Layer Turbulence in a Suburban Area of São Paulo, Brazil: Objective Analysis of Scaling-Parameter Accuracy and Uncertainties |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |