Leaf vein space distribution feature extraction method based on broadleaf plant leaf point cloud

Abstract

The invention belongs to the field of plant leaf physiological structure phenotypic feature extraction methods, and provides a leaf vein spatial distribution feature extraction method based on broadleaf plant leaf point cloud. Comprising the steps of leaf point cloud extraction, spatial clustering, search radius setting, local point cluster screening, average normal calculation, point cloud directional projection, digital surface model construction, local elevation iterative optimization, local dip angle calculation, vein candidate pixel set extraction based on morphological filtering and pattern spot skeleton analysis. According to the method, the in-situ lossless extraction of the leaf vein distribution structure and morphological parameters is realized by calculating the height difference change of the leaf veins and the neighborhood leaf surfaces.

Description

Technical Field

[0001] The invention relates to the field of plant leaf physiological structure phenotype feature extraction methods, and in particular to a method for extracting leaf vein spatial distribution features based on broad-leaved plant leaf point clouds. Background Art

[0002] The veins of broad-leaved plants refer to the vascular system in the leaves, which dominates the transportation of water, minerals and organic matter in the leaves. The vessels in the veins are responsible for transporting water and minerals from the roots of the plant to the leaves. This water is dissipated from the stomata of the leaves through transpiration, thereby maintaining the water balance in the plant body and driving the roots to absorb more water and nutrients. The sieve tubes in the veins are responsible for transporting organic matter produced by photosynthesis, such as glucose and sucrose, from the leaves to other parts of the plant body, including roots, stems and fruits. These organic substances provide energy and raw materials for plant growth, storage and reproduction. The veins transport water, maintain the water potential in the leaves, and regulate the opening and closing of the stomata. This regulatory mechanism is achieved through transpiration, which helps the leaves retain appropriate water, while promoting the absorption of carbon dioxide and supporting photosynthesis. While transporting organic matter, the veins are also responsible for the distribution of energy and nutrients between different organs of the plant. This is crucial for the growth and development of plants, the formation of storage organs, and the reproduction process. In addition, leaf veins provide support for broadleaf leaves to maintain a certain shape and rigidity, allowing the leaves to better capture light for photosynthesis, and also allowing the leaves to remain intact and not easily damaged in the face of environmental factors such as wind and rain. Leaf veins are also involved in the disease defense of leaves. When leaves are damaged, the vessels and sieve tubes in the leaf veins can redistribute resources, repair damaged tissues, help wound healing, and enhance the survival rate of plants. According to the distribution of leaf veins, the leaf veins of broadleaf plants are usually divided into types such as reticular veins, parallel veins, palmate veins, and pinnate veins. Reticular veins are the most common type of leaf veins, and the main feature is that the leaf veins form a mesh structure within the leaf. The main veins and lateral veins are interwoven into a complex mesh system, usually with a main vein in the center, the midrib, and the lateral veins branching from the midrib further branch out to form a mesh structure. Parallel leaf veins start from the base of the leaf and extend to the top of the leaf. Parallel veins are common in monocots, but also occur in some broadleaf plants, such as some grasses. Palmate veins emanate from the base of the petiole and extend radially outward, with multiple main veins extending in all directions from the edge of the leaf. Pinnate veins have a distinct main vein that runs through the entire leaf, with many feather-like lateral veins branching out from both sides of the midrib. The above distribution pattern leads to significant differences in the density of vein distribution in different plant species. The density of vein distribution, that is, the number or length of veins per unit area, can reveal the physiological characteristics, adaptation strategies and ecological functions of plants. In the process of agricultural breeding, vein density is an important selective trait that is related to the drought resistance, heat resistance and disease resistance of crops. By studying the vein density of different plants, the environmental adaptation strategy of plant phenotypic characteristics can be inferred for crop growth modeling. Understanding the distribution of vein density of crops can help agricultural managers optimize water resources and nutrient application strategies, and can also help develop bionic materials with excellent strength and flexibility for use in fields such as construction, textiles and biomedicine.By using high-precision multi-angle LiDAR or structured light imaging and stitching, we can obtain high-precision point clouds of leaves and depict the millimeter-level fine structure of the vein distribution area on the leaf surface, including the shape, distribution pattern and direction of the veins. We can also analyze the three-dimensional shape of the veins, including bending, undulations, etc. With certain computer vision analysis algorithms and deep learning models, we can realize the recognition of vein distribution patterns based on the digital surface model of the leaf surface, which can be used for plant classification, health status assessment and ecological research.

[0003] Existing structural analysis based on leaf point clouds mainly focuses on the extraction of structural parameters related to leaf area and morphology, but there is still a lack of available solutions for the extraction of the structure and distribution patterns of fine-scale leaf organs such as veins. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method for extracting the spatial distribution features of leaf veins based on point clouds of broad-leaved plants, and to realize in-situ non-destructive extraction of leaf vein distribution structure and morphological parameters by calculating the height difference between leaf veins and neighboring leaf surfaces.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for extracting spatial distribution features of veins of broad-leaved plants based on leaf point clouds, comprising:

[0007] Collecting a broad-leaved plant point cloud, and performing leaf separation on the broad-leaved plant point cloud to obtain a leaf point cloud extraction result;

[0008] Performing spatial clustering on the leaf point cloud extraction results to obtain a single broadleaf point cloud;

[0009] Setting a search radius according to the mean value of the distance data;

[0010] Using the search radius to filter the distance data to obtain a local point cluster;

[0011] Using the local point cluster to construct a local discrete point space coordinate matrix, and calculating and averaging the point cluster normals according to the local discrete point space coordinate matrix to obtain an average normal;

[0012] Projecting the single broadleaf point cloud onto a plane perpendicular to the average normal and parallel to the spatial growth direction of the target leaf to obtain a projection image;

[0013] Constructing a digital surface model after directional projection of the point cloud using a voxel method according to the projection image;

[0014] The digital surface model is optimized and iterated locally using a 3*3 square sliding window to obtain a local elevation difference lifting model;

[0015] Calculating the local inclination angle of each pixel in the local height difference lifting model to obtain leaf orientation data;

[0016] Extracting a set of candidate leaf vein pixels in the local height difference lifting model using the Freeman algorithm according to the leaf orientation data;

[0017] Using a disc-shaped filter window with a radius of 1 pixel to perform an expansion operation on the candidate leaf vein pixel set, and using a disc-shaped filter window with a radius of 2 pixels to perform a morphological closing operation on the expanded candidate leaf vein pixel set to obtain linearly continuously distributed leaf vein pixels;

[0018] Performing a pattern skeleton analysis on the linear continuously distributed leaf vein pixels to obtain a leaf vein pattern skeleton;

[0019] The vein pattern skeleton is subjected to vein distribution density calculation and multi-type and multi-scale analysis to obtain vein distribution characteristics.

[0020] Preferably, the projection formula of the projected image is: proj =P-(P×n)n; where P proj is the projection coordinate; P is the coordinate of any point in the single broadleaf point cloud; n is the average normal.

[0021] Preferably, the calculation formula of the local height difference lifting model is:

[0022]

[0023] Among them, DSM LS is the local height difference lifting model; DSM is the digital surface model; DSM Mean is the digital surface model after mean filtering; N is the number of iterative analysis; and i represents the current number of calculations.

[0024] Preferably, the calculation formula of the local inclination angle is: Wherein, θ is the local inclination angle; G x G is the height component of the pixel in the X direction in the local height difference lifting model; y It is the height component of the pixel in the Y direction in the local height difference lifting model.

[0025] Preferably, constructing a digital surface model after directional projection of the point cloud using a voxel method according to the projection image comprises:

[0026] Randomly extract 1% of the points of the single broadleaf point cloud to obtain a point cloud subset;

[0027] Calculating the Euclidean distance of the nearest neighbor point of each point in the point cloud subset in the three-dimensional space, and calculating the average of the Euclidean distances of all the nearest neighbor points to obtain an average Euclidean distance;

[0028] setting twice the average Euclidean distance as the spatial resolution;

[0029] Dividing the projection image into a plurality of elevation statistical units according to the spatial resolution;

[0030] The point cluster elevation mean of each elevation statistical unit is calculated to obtain the digital surface model.

[0031] Preferably, extracting a set of candidate leaf vein pixels in the local height difference lifting model using the Freeman algorithm according to the leaf orientation data includes:

[0032] Calculate the slope aspect and local height difference between each pixel and 8 neighboring pixels in the local height difference lifting model to obtain slope change data;

[0033] Using the slope change data to determine the edge position, moving to the next pixel according to the edge position, and recording the moving direction to obtain leaf vein growth direction data;

[0034] A weight calculation is performed using a recursive method according to the leaf vein growth direction data to obtain a leaf vein candidate pixel weight grid;

[0035] Calculating the pixel mean of the leaf vein candidate pixel weight grid to obtain a pixel threshold;

[0036] The pixel threshold is used to binarize the leaf vein candidate pixel weight grid to obtain the candidate leaf vein pixel set.

[0037] Preferably, the calculation formula for the leaf vein distribution density is: pixel =S YM / S P ; where ρ pixel is the leaf vein distribution density; S YM is the number of leaf vein pixels in the search window in the leaf vein pattern skeleton; S P It is the total number of pixels in the retrieval window in the leaf vein pattern skeleton.

[0038] The present invention discloses the following technical effects:

[0039] The present invention provides a method for extracting the spatial distribution features of leaf veins based on leaf point clouds of broad-leaved plants. By calculating the height difference between leaf veins and neighboring leaf surfaces, the defect of the traditional structural analysis method based on leaf point clouds, which lacks the ability to extract leaf veins, is solved, and the in-situ non-destructive extraction of leaf vein distribution structure and morphological parameters is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0041] Figure 1 A schematic diagram of a process for extracting spatial distribution features of veins of broad-leaved plants based on leaf point clouds provided by an embodiment of the present invention;

[0042] Figure 2 An exemplary reference diagram provided for an embodiment of the present invention;

[0043] Figure 3 An implementation flow chart provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The purpose of the present invention is to provide a method for extracting the spatial distribution features of leaf veins based on point clouds of broad-leaved plants, and to achieve in-situ lossless extraction of leaf vein distribution structure and morphological parameters by calculating the height difference between leaf veins and neighboring leaf surfaces.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Figure 1 A schematic diagram of a process for extracting spatial distribution features of veins of broad-leaved plants based on leaf point clouds provided by an embodiment of the present invention, Figure 2 The exemplary reference diagrams provided for the embodiments of the present invention are as follows: Figure 1 and Figure 2 As shown, the present invention provides a method for extracting spatial distribution features of leaf veins based on point clouds of broad-leaved plants, comprising:

[0048] Collecting broad-leaved plant point clouds, and performing leaf separation on the broad-leaved plant point clouds to obtain leaf point cloud extraction results;

[0049] Perform spatial clustering on the leaf point cloud extraction results to obtain a single broadleaf point cloud;

[0050] Set the search radius according to the mean of the distance data;

[0051] Use the search radius to filter the distance data and obtain local point clusters;

[0052] The local point cluster is used to construct a local discrete point space coordinate matrix, and the point cluster normal is calculated and averaged according to the local discrete point space coordinate matrix to obtain the average normal;

[0053] Project the single broadleaf point cloud onto a plane perpendicular to the average normal and parallel to the spatial growth direction of the target leaf to obtain a projection image;

[0054] According to the projection image, a digital surface model after directional projection of the point cloud is constructed by using the voxel method;

[0055] The local elevation optimization and iteration of the digital surface model are performed using a 3*3 square sliding window to obtain a local elevation difference lifting model;

[0056] Calculate the local inclination angle of each pixel in the local height difference lifting model to obtain the leaf orientation data;

[0057] According to the leaf orientation data, the Freeman algorithm is used to extract the candidate leaf vein pixel set in the local height difference lifting model;

[0058] A disc-shaped filter window with a radius of 1 pixel is used to dilate the candidate leaf vein pixel set, and a disc-shaped filter window with a radius of 2 pixels is used to perform a morphological closing operation on the dilated candidate leaf vein pixel set to obtain linearly continuously distributed leaf vein pixels.

[0059] Performing pattern skeleton analysis on linear continuously distributed leaf vein pixels to obtain leaf vein pattern skeleton;

[0060] The vein distribution density is calculated and multi-type and multi-scale analysis is performed on the vein pattern skeleton to obtain the vein distribution characteristics.

[0061] Specifically, the projection formula of the projected image is: proj =P-(P×n)n; where P proj are projection coordinates; P is the coordinate of any point in a single broadleaf point cloud; n is the average normal.

[0062] Furthermore, the calculation formula of the local height difference lifting model is:

[0063]

[0064] Among them, DSM LS is the local height difference lifting model; DSM is the digital surface model; DSM Mean is the digital surface model after mean filtering; N is the number of iterative analysis; i represents the current number of calculations.

[0065] Preferably, the calculation formula of the local inclination angle is: Where θ is the local inclination angle; G x is the height component of the pixel in the X direction in the local height difference lifting model; G y It is the height component of the pixel in the Y direction in the local height difference lifting model.

[0066] Furthermore, a digital surface model after directional projection of the point cloud is constructed using a voxel method according to the projection image, including:

[0067] Randomly extract 1% of the points in a single broadleaf point cloud to obtain a point cloud subset;

[0068] Calculate the Euclidean distance of the nearest neighbor point in the three-dimensional space for each point in the point cloud subset, and calculate the average of the Euclidean distances of all the nearest neighbor points to obtain the average Euclidean distance;

[0069] Set twice the average Euclidean distance as the spatial resolution;

[0070] Divide the projection image into a number of elevation statistical units according to the spatial resolution;

[0071] The mean elevation of the point cluster of each elevation statistical unit is calculated to obtain the digital surface model.

[0072] Specifically, the Freeman algorithm is used to extract the candidate leaf vein pixel set in the local height difference lifting model according to the leaf orientation data, including:

[0073] Calculate the slope aspect and local height difference between each pixel and the eight neighboring pixels in the local height difference lifting model to obtain the slope change data;

[0074] The edge position is determined by using the slope change data, and the next pixel is moved to according to the edge position. The moving direction is recorded to obtain the leaf vein growth direction data.

[0075] According to the leaf vein growth direction data, the weight is calculated using a recursive method to obtain the leaf vein candidate pixel weight grid;

[0076] Calculate the pixel mean of the leaf vein candidate pixel weight grid to obtain the pixel threshold;

[0077] The pixel threshold is used to binarize the leaf vein candidate pixel weight grid to obtain the candidate leaf vein pixel set.

[0078] Preferably, the calculation formula for leaf vein distribution density is: pixel =S YM / S P ; where ρ pixel is the density of leaf vein distribution; S YMis the number of leaf vein pixels in the retrieval window in the leaf vein pattern skeleton; S P It is the total number of pixels in the retrieval window in the leaf vein pattern skeleton.

[0079] Specifically, the leaf point cloud is preprocessed. Under natural growth conditions, the overall orientation of broadleaf leaves is variable and not necessarily parallel to the ground. In order to analyze the distribution pattern of leaf veins, it is necessary to project the leaf point cloud according to its overall orientation onto a plane parallel to the growth orientation, and construct a digital surface model (DSM) of the leaf. In the point cloud, each point is defined by its X, Y, and Z coordinates to define its three-dimensional spatial position. To characterize the orientation of the local point cluster, the normal of the point cluster spatial coordinate fitting plane can be extracted. The normal refers to the direction vector perpendicular to a point on the surface of the point cloud. The normal vectors of the local fitting plane of the point cloud are averaged point by point, and the result is called the average normal direction. The average normal direction reflects the overall distribution direction of the leaf point cloud and is used to indicate the projection surface of the leaf.

[0080] Furthermore, the point cloud projection process converts the three-dimensional point cloud to a two-dimensional plane. Projection according to the average normal direction means projecting each point in the point cloud onto a plane perpendicular to the average normal direction. The projected data retains the direction information of the point cloud parallel to the projection plane and eliminates the influence of other non-projection directions. The projection formula is:

[0081] P proj =P-(P×n)n

[0082] A local normal vector is calculated for each point in the point cloud using a neighborhood search algorithm (such as K nearest neighbor or radius neighborhood search). Each point in the point cloud is projected along the average normal direction onto a plane perpendicular to that direction, which will be used to construct the DSM of the leaf point cloud in the projection direction, providing a data source for vein extraction and distribution pattern analysis.

[0083] Specifically, the leaf surface model is constructed. After the point cloud projection is completed, the range interval of the enclosing rectangle of the leaf point cloud in the projection plane is determined according to the projection results. Based on this interval as a mask, the voxel method is used to construct the digital surface model (Digital Surface Model, DSM) of the leaf point cloud after projection. The voxel method divides the projected point cloud into several elevation statistical units according to the resolution W on the projection surface, and takes the mean elevation of the point cluster in each unit as the elevation of the unit. In order to determine W, points accounting for 1% of the total number are randomly extracted from the point cloud, and the Euclidean distance D of each point to its nearest neighbor in the three-dimensional space is counted. Then, twice the mean of D corresponding to all retrieval points is calculated as the spatial resolution W of the DSM. The generated DSM is used as the key data source for subsequent leaf vein extraction.

[0084] Furthermore, considering that the height difference between the vein distribution area and the surrounding leaf surface may not be obvious, it is necessary to iteratively use grid arithmetic filtering to increase the height difference between the vein distribution area and the surrounding leaf surface to improve the accuracy of vein extraction. First, the DSM is processed using a 3*3 window mean filter to obtain the DSM Mean For the central pixel P of an analysis window in DSM, the mean filtering result is the mean of the neighboring pixels in the eight directions of P. Then, calculate DSM and DSM Mean The difference is calculated and added to the DSM, and iterated N times. Mean , then the sum is lower than DSM Mean Then, the iterative cumulative result is divided by N to obtain the DSM after the local height difference is increased. LS , to avoid leaf edge extraction errors.

[0085]

[0086] Considering the operation efficiency and the rationality of height difference analysis, N is set to 3 here.

[0087] Specifically, leaf vein pixel extraction. Using the DSM after projection of high-precision leaf point cloud LS , extract the pixels covered by leaf veins. Since the local elevation of the leaf vein distribution area is lower than the surrounding non-leaf vein area, the local pixel elevation change can be used to extract a set of low-value pixels with valley-like distribution, namely, the pixels covered by leaf veins. First, use DSM LS Analyze the leaf orientation within the local 3*3 window. For each pixel, use the elevation values ​​of the 8 neighboring pixels to calculate the inclination of the local neighborhood of the pixel. The leaf orientation is determined by the inclination components of the two directions (such as the X and Y directions) obtained in the local leaf inclination calculation. The local inclination angle θ of the central pixel of the window is calculated using the following formula:

[0088]

[0089] The result is the inclination angle of the central pixel, based on which the DSM is calculated. LS The local leaf orientation of each pixel neighborhood in .

[0090] Furthermore, after completing the leaf orientation analysis, the local orientation change information is used to extract the set of candidate pixels for leaf veins. In the extraction process, the Freeman algorithm is used. The growth of leaf veins on the leaf surface has the characteristics of spatial continuity and bifurcated growth. The growth path of leaf veins is always in the direction with the largest change in local slope of the leaf surface, that is, growing from the leaf edge to the petiole. The Freeman algorithm describes the distribution of gullies on a spatially continuous surface by recording the direction of a pixel relative to its previous pixel. Similar to the process of extracting slope water flow paths based on digital elevation models, in the process of extracting candidate pixels for leaf veins based on the Freeman algorithm, for DSM LS For each pixel in the DSM, the growth path of the vein from the leaf edge to the petiole is calculated based on its slope and local height difference. Usually, the directions of the 8 pixels in the neighborhood of the central pixel are represented by integers from 0 to 7 (including horizontal, vertical and diagonal directions). LS Start from one point on the edge of the image and move step by step along the edge, recording the direction from the current point to the next point each time, until you return to the starting point or reach the DSM. LS If the contour is closed, it will eventually return to the starting point; otherwise, the contour will be in the DSM LS The open point of the low-value elevation zone ends. For broad-leaved leaves, this low-value elevation open point is often the petiole. For each pixel, the chain code of the possible extension path of the leaf vein is recorded according to the orientation code.

[0091] Furthermore, once the DSM is determined LS The direction of leaf vein growth at each pixel can be analyzed by tracing back and accumulating the leaf vein growth path. A recursive method is usually used to calculate the weight of each pixel from downstream to upstream to extract candidate pixels for leaf vein coverage. The weight of the pixel close to the petiole is equal to the sum of the weights of the upper pixels plus its own weight. LS The leaf vein candidate pixel weight grid YM is obtained. In YM, the high-value pixel distribution area is likely to have leaf veins. The pixel mean in YM is further used as the threshold to binarize YM and obtain the candidate leaf vein pixel set.

[0092] Specifically, leaf vein extraction optimization. In the above extraction results, the extraction results of leaf vein candidate pixels in the local flat leaf surface area may be patchy and broken. Here, the morphological opening and closing operations on the binary grayscale image are used to further optimize the leaf vein extraction results, improve the continuity of the extracted leaf veins and the subsequent extraction accuracy of the leaf vein distribution density. BL ), firstly, the morphological dilation operation is used to increase the edge width of the independent foreground patch, so that the adjacent vein candidate patches that are partially broken are spatially connected. Here, a disc-shaped filter window with a radius of 1 pixel is set for YM BLThen, the morphological closing operation is used to dilate the YM BL Perform corrosion to accelerate the stripping of YM BL The redundant edge pixels of the leaf vein pixels that have been connected and optimized are removed. The neighborhood range of the closing operation is defined using a disk-shaped filter window with a radius of 2 pixels, and the result is a linear and continuously distributed leaf vein pixel.

[0093] Furthermore, using the leaf vein pixels after connectivity optimization, the leaf vein pixel pattern is reduced to its center line using the pattern skeleton analysis while retaining the topological structure of the leaf vein pixel. The skeletonization analysis reduces the shape of the object by iteratively stripping the edge pixels of the object to generate a thinnest connected path that maintains the topological structure of the object. In the results, the leaf vein pixels in YM BL In each iteration, the algorithm identifies and marks the edge pixels of the current object. For each marked edge pixel, it is further checked whether it can be deleted without destroying the topological structure of the image. If deleting this pixel will cause the object to be disconnected or the number of holes to change, the pixel will not be deleted. The above process will continue to iterate, gradually thinning the leaf vein pattern from the outside to the inside, until there are no more pixels to delete. Finally, the remaining set of pixels is the leaf vein pattern skeleton, which is used as the LS The final result of extracting leaf vein pixels is used as the data source for leaf vein distribution pattern extraction and local density analysis.

[0094] Specifically, leaf vein distribution density analysis. Statistical leaf projection DSM LS The number of pixels N in P and the number of leaf vein pixels extracted N YM , use the following formula to calculate the overall distribution density of leaf veins:

[0095] ρ=N YM / N P

[0096] Since the veins are not evenly distributed inside the leaf, the local window statistical method is used to further analyze the changes in vein density within the leaf. LS The coverage of the DSM is set to S. LS The number of leaf vein pixels S in the search window is traversed from the upper left corner YM For an analysis window, the vein distribution density ρ within it pixel for:

[0097] ρ pixel =S YM / S P

[0098] Through the above analysis, the changing trend of vein distribution density in the leaf projection surface can be obtained. LS , a density distribution map is generated and visualized on the leaf surface, which can intuitively depict the distribution characteristics of leaf veins in various parts of the leaf.

[0099] Furthermore, the leaves were divided into different radial segments (from petiole to leaf tip) and transverse segments (approximately parallel to the main vein direction). The vein density of each segment was calculated, and the trend of vein density changes at different developmental positions of the leaves was analyzed based on this. The leaves were then divided into several concentric elliptical areas along the vein extension direction (usually from the midrib to the leaf margin), and the midpoint of the ellipse was the leaf DSM. LS The center point of the major axis of the outer rectangle and the spacing between the rings refer to the spacing between the intersections of the lateral veins and the main veins, and the vein distribution density in the near-center ellipse and the multi-layer outer rings is analyzed respectively. On this basis, the areas with or without vein distribution are extracted to achieve multi-type and multi-scale analysis of the vein distribution density on the leaf surface, which is used for interactive modeling of leaf structural phenotypic characteristics and physiological processes.

[0100] Specifically, the specific implementation steps are as follows:

[0101] S1: Use RGB-D camera or LiDAR to collect millimeter-level high-precision point clouds of broad-leaved plants, and manually separate the leaf and branch point clouds in the plants in a desktop visualization software platform (for example, Cloudcompare or Leica Cyclone software). You can also use a local point cloud distribution pattern analysis algorithm or use additional attribute information of the point cloud (such as point-by-point surface reflectivity, target RGB color, and echo intensity attributes) to assist in separating the leaf and branch point clouds.

[0102] S2: Based on the leaf point cloud extraction results, spatial clustering and segmentation are performed according to the change in point cloud density between leaves to obtain a single broad-leaf point cloud. The search radius is set based on the Euclidean distance between adjacent points to obtain a local point cluster in a single broad-leaf point cloud. The point cluster is used to construct a local discrete point spatial coordinate matrix, and the normal of the point cluster is calculated. Each point in the point cloud is projected onto a plane parallel to the spatial growth direction of the leaf along the average normal direction, the spatial resolution W is set, and a DSM of the leaf point cloud coverage range is constructed. Then, the non-zero pixel coverage area in the DSM is used as a mask to extract the DSM after the leaf point cloud projection.

[0103] S3: Use a 3*3 square sliding window to traverse and analyze the entire DSM. Calculate the elevation change rate and mean of the local leaf surface pixels in each analysis window. If the elevations in the analysis window are the same, abandon the extraction and continue with the next analysis. Then, iteratively optimize the local elevation of the DSM. Under the premise of maintaining the local elevation fluctuation pattern of the leaf surface, increase the difference between the high-value area and the low-value area in the local leaf surface DSM to construct the DSM LSTo highlight the distribution of leaf vein pixels, so as to facilitate the subsequent extraction of DSM LS Spatially continuous leaf vein pixels.

[0104] S4: Extract DSM through local slope analysis and leaf vein weight calculation based on Freeman algorithm LS The candidate leaf vein pixel set in . Without destroying the connectivity of leaf veins, the iterative morphological dilation and thinning method is used to set disk-shaped sliding analysis units, and the dilation and thinning analysis window sizes with radii of 1 and 2 are set to increase the spatial connectivity of candidate leaf vein patterns, and to strip off the edge points of blocky clustered leaf vein patterns, and obtain a skeletonized leaf vein pixel set.

[0105] S5: Statistics DSM LS The number of pixels in the image and the number of vein pixels extracted from it are used to calculate the overall distribution density of the veins. The local vein density changes within the leaf are further analyzed using the local window statistical method. LS The local vein density distribution map of the leaf surface obtained by analysis quantitatively displays the distribution characteristics and change trends of veins in various parts of the leaf. The leaf is divided into multiple horizontal and vertical sections, and the DSM-based LS The vein distribution map extracted is used to further analyze the trend of vein density changes in each section. Then the leaf is extended along the vein direction (usually from the midrib to the leaf edge), referring to the spacing between the intersections of the veins, and the DSM LS Divide into several concentric elliptical areas, and analyze the vein distribution density in the near-central ellipse and the multi-layer outer ring. Through the above analysis, the total length of the veins inside the target leaf, the local vein density changes in the leaf surface space, and the vein density changes in the segmented and partitioned areas can be obtained, so as to quantitatively express the relevant leaf structure phenotypic characteristics, which can be used for modeling related to leaf physiological processes and three-dimensional digital display of leaf fine structure.

[0106] The beneficial effects of the present invention are as follows:

[0107] (1) The solution proposed by the present invention can be used to extract the complex three-dimensional spatial distribution pattern of leaf veins in situ and losslessly based on the high-precision leaf point cloud. On this basis, the structure and distribution density information of the leaf veins can be quantitatively analyzed.

[0108] (2) Based on the leaf vein extraction results of the leaf point cloud, we can combine the measured values ​​of leaf physiological indicators to construct a thematic model to simulate the growth and development process of plant organs in the leaves, thus providing reliable data for building a refined digital crop model.

[0109] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0110] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for extracting the spatial distribution features of leaf veins of broad-leaved plants based on leaf point clouds, characterized in that: include: Collecting a broad-leaved plant point cloud, and performing leaf separation on the broad-leaved plant point cloud to obtain a leaf point cloud extraction result; Performing spatial clustering on the leaf point cloud extraction results to obtain a single broadleaf point cloud; Setting a search radius according to the mean value of the distance data; Using the search radius to filter the distance data to obtain a local point cluster; Using the local point cluster to construct a local discrete point space coordinate matrix, and calculating and averaging the point cluster normals according to the local discrete point space coordinate matrix to obtain an average normal; Projecting the single broadleaf point cloud onto a plane perpendicular to the average normal and parallel to the spatial growth direction of the target leaf to obtain a projection image; Constructing a digital surface model after directional projection of the point cloud using a voxel method according to the projection image; The digital surface model is optimized and iterated locally using a 3*3 square sliding window to obtain a local elevation difference lifting model; Calculating the local inclination angle of each pixel in the local height difference lifting model to obtain leaf orientation data; Extracting a set of candidate leaf vein pixels in the local height difference lifting model using the Freeman algorithm according to the leaf orientation data; Using a disc-shaped filter window with a radius of 1 pixel to perform an expansion operation on the candidate leaf vein pixel set, and using a disc-shaped filter window with a radius of 2 pixels to perform a morphological closing operation on the expanded candidate leaf vein pixel set to obtain linearly continuously distributed leaf vein pixels; Performing a pattern skeleton analysis on the linear continuously distributed leaf vein pixels to obtain a leaf vein pattern skeleton; The vein pattern skeleton is subjected to vein distribution density calculation and multi-type and multi-scale analysis to obtain vein distribution characteristics.

2. The method for extracting the spatial distribution features of veins of broad-leaved plants based on leaf point cloud according to claim 1, characterized in that: The projection formula of the projected image is: proj =P-(P×n)n; where P proj is the projection coordinate; P is the coordinate of any point in the single broadleaf point cloud; n is the average normal.

3. The method for extracting the spatial distribution features of veins of broad-leaved plants based on leaf point cloud according to claim 1, characterized in that: The calculation formula of the local height difference lifting model is: Among them, DSM LS is the local height difference lifting model; DSM is the digital surface model; DSM Mean is the digital surface model after mean filtering; N is the number of iterative analysis; and i represents the current number of calculations.

4. The method for extracting the spatial distribution features of veins of broad-leaved plants based on leaf point cloud according to claim 1, characterized in that: The calculation formula of the local inclination angle is: Wherein, θ is the local inclination angle; G x G is the height component of the pixel in the X direction in the local height difference lifting model; y It is the height component of the pixel in the Y direction in the local height difference lifting model.

5. The method for extracting the spatial distribution features of veins of broad-leaved plants based on leaf point cloud according to claim 1, characterized in that: Constructing a digital surface model after directional projection of the point cloud using a voxel method according to the projection image, including: Randomly extract 1% of the points of the single broadleaf point cloud to obtain a point cloud subset; Calculating the Euclidean distance of the nearest neighbor point of each point in the point cloud subset in the three-dimensional space, and calculating the average of the Euclidean distances of all the nearest neighbor points to obtain an average Euclidean distance; setting twice the average Euclidean distance as the spatial resolution; Dividing the projection image into a plurality of elevation statistical units according to the spatial resolution; The point cluster elevation mean of each elevation statistical unit is calculated to obtain the digital surface model.

6. The method for extracting the spatial distribution features of veins of broad-leaved plants based on leaf point cloud according to claim 1, characterized in that: Extracting a candidate leaf vein pixel set in the local height difference lifting model using the Freeman algorithm according to the leaf orientation data includes: Calculate the slope aspect and local height difference between each pixel and 8 neighboring pixels in the local height difference lifting model to obtain slope change data; Using the slope change data to determine the edge position, moving to the next pixel according to the edge position, and recording the moving direction to obtain leaf vein growth direction data; A weight calculation is performed using a recursive method according to the leaf vein growth direction data to obtain a leaf vein candidate pixel weight grid; Calculating the pixel mean of the leaf vein candidate pixel weight grid to obtain a pixel threshold; The pixel threshold is used to binarize the leaf vein candidate pixel weight grid to obtain the candidate leaf vein pixel set.

7. The method for extracting the spatial distribution features of veins of broad-leaved plants based on leaf point cloud according to claim 1, characterized in that: The calculation formula of the leaf vein distribution density is: pixel =S YM / S P ; Among them, ρ pixel is the leaf vein distribution density; S YM is the number of leaf vein pixels in the search window in the leaf vein pattern skeleton; S P It is the total number of pixels in the retrieval window in the leaf vein pattern skeleton.

Images