A method for measuring leaf geometric traits of indoor mature corn single plant point cloud

Abstract

The application belongs to the technical field of three-dimensional computer vision and plant phenotype information extraction, and specifically discloses a leaf geometric character measurement method for indoor mature corn single plant point cloud, which comprises the following steps: obtaining three-dimensional point cloud data of the indoor mature corn single plant and performing pretreatment to obtain normalized point cloud; performing enhanced skeleton extraction on the pretreated point cloud; performing skeleton decomposition and stem-leaf coarse segmentation on the extracted enhanced skeleton to obtain stem point set and leaf point set; constructing a corn plant self-coordinate system according to the obtained stem points and leaf points, and transforming the stem points and leaf points to the plant coordinate system; performing multi-constraint fine segmentation on the stem point set and leaf point set obtained through coarse segmentation to obtain a coordinate-corrected skeleton graph, stem point cloud and leaf point cloud; and performing triangulation reconstruction on the obtained leaf point cloud and calculating the total area of a single leaf. The application realizes high-precision automatic segmentation of corn stems and leaves, and stably obtains key phenotype parameters such as leaf area.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional computer vision and plant phenotypic information extraction technology, and in particular to a method for measuring the leaf geometric characteristics of a single mature maize plant point cloud indoors. Background Technology

[0002] With the development of technologies such as LiDAR, structured light, and multi-view reconstruction, indoor high-throughput phenotyping platforms can now acquire high-precision 3D point cloud data of individual maize plants. However, achieving stable organ-level segmentation and high-precision trait measurement from complex, dense, and occluded maize point clouds still faces many technical challenges, mainly in the following aspects:

[0003] 1. The skeleton topology is susceptible to breakage due to sparse regions: Under indoor potted plant scanning conditions, the point cloud density at the base of the corn stem is often significantly reduced or even missing due to factors such as flowerpot obstruction and limited scanning angle. Traditional skeleton extraction methods based on Laplacian contraction or median transformation are prone to skeleton breakage when local neighborhood constraints are weak, causing subsequent skeleton-based stem and leaf decomposition to fail.

[0004] 2. Point clouds at the top of the ear are easily misidentified as leaves: In mature maize, the ear is usually located in the highest region of the plant, and its geometry overlaps with that of unfolded leaves. If no structural constraints for height are introduced, the point clouds at the top of the ear are often mistakenly classified as leaves, resulting in discontinuous stems and affecting the calculation of parameters such as plant height.

[0005] 3. The complex structure of the stem-leaf junction area makes it easy to cause segmentation confusion: The geometric shapes of the corn stalk and leaf sheath are similar, especially when the top heart leaf has not fully unfolded. The normal direction of its point cloud is highly consistent with the height direction of the stalk. Traditional segmentation methods that rely solely on Euclidean distance or simple normal vector differences are difficult to distinguish accurately, often leading to leaves being misjudged as stems or stalks being misjudged as leaves. This can easily cause missegmentation in the stem-leaf junction area and the ear area.

[0006] 4. Uneven point cloud density on blades leads to area measurement bias: Due to the influence of scanning angle and occlusion, the point cloud in the central axis region of the blade is often relatively sparse. Directly using single-scale triangulation or the assumption of a regular shape for area calculation can easily lead to a systematic underestimation of the blade surface area. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method for measuring the leaf geometric characteristics of a single mature maize plant using point clouds in an indoor environment.

[0008] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0009] A method for measuring the leaf geometric characteristics of a single mature maize plant in an indoor environment using point clouds, comprising the following steps:

[0010] S1. Obtain the three-dimensional point cloud data of a single mature maize plant indoors and preprocess it to obtain a standardized point cloud.

[0011] S2. Enhanced skeleton extraction is performed on the preprocessed point cloud;

[0012] S3. Perform skeleton decomposition and coarse segmentation of stems and leaves on the extracted enhanced skeleton to obtain stem point sets and leaf point sets;

[0013] S4. Construct the corn plant's own coordinate system based on the obtained stem points and leaf points, and transform the stem points and leaf points to the plant coordinate system;

[0014] S5. Perform multi-constraint fine segmentation on the stem point set and leaf point set obtained from the coarse segmentation to obtain the coordinate-corrected skeleton map, stem point cloud and leaf point cloud;

[0015] S6. Triangulate and reconstruct the obtained leaf point cloud and calculate the total area of ​​a single leaf.

[0016] Furthermore, the preprocessing in step S1 includes:

[0017] Let the original point cloud be represented as:

[0018] ;

[0019] in: Represents the three-dimensional spatial coordinates of the i-th point; N represents the total number of points in the point cloud;

[0020] First, an outlier filtering method based on statistical distribution is used to calculate the distance from each point to its nearest neighbor. If the average distance between the nearest neighbor points deviates from the global mean by more than a preset threshold, the point is identified as a noise point and removed, thereby eliminating measurement errors and environmental noise.

[0021] Finally, point cloud downsampling is performed: the voxel mesh downsampling method is adopted to divide the three-dimensional space into cubic voxels with a side length of v. Only one representative point is retained in each voxel to reduce the size of the point cloud and maintain the overall geometric structure features.

[0022] After the above processing, the normalized point cloud is obtained:

[0023] ;

[0024] Where: M < N.

[0025] Further, step S2 specifically includes:

[0026] S21, Laplace contraction

[0027] Perform a Laplace shrinkage operation on the preprocessed point cloud P': Let the nearest neighbor set of the i-th point be... Then its Laplace operator is defined as:

[0028] ;

[0029] in: This represents the Laplace displacement vector at the i-th point; For point The number of nearest neighbors; For point The position update formula for the j-th nearest neighbor is:

[0030] ;

[0031] in: , Let represent the coordinates of the point after the t-th and t+1-th iterations, respectively; This is the shrinkage coefficient, used to control the intensity of point cloud aggregation towards the central axis;

[0032] S22, Skeleton Node Sampling and Graph Construction

[0033] The farthest point sampling method is used to select the skeleton candidate node set on the shrunk point cloud:

[0034] ;

[0035] Where: U represents the set of skeleton nodes; K is the number of skeleton nodes. An undirected weighted skeleton graph G is constructed using the skeleton nodes as vertices:

[0036] ;

[0037] in: E represents the set of nodes in the graph; E represents the set of skeleton edges, when the node and When the Euclidean distance between them is less than a given threshold, establish an edge. The edge weight is defined as:

[0038] .

[0039] Furthermore, step S3 specifically includes:

[0040] S31, Stem skeleton extraction

[0041] In the skeleton graph G, identify all nodes with a degree of 1 as the set of endpoints:

[0042] ;

[0043] Select the endpoint with the smallest height as the root node at the base of the stem. And calculate the shortest path distance from it to all nodes:

[0044] ;

[0045] in This represents the shortest path length from node v to the base of the stem; the node with the largest distance is selected. And the path: It is defined as the main framework of the stem, with the remaining branches serving as the leaf framework;

[0046] S32, Coarse segmentation based on nearest skeleton

[0047] For any point cloud point Find its nearest skeleton node:

[0048] ;

[0049] If the most recent skeleton node Then the point If a point is classified as a stem point, it is classified as a leaf point belonging to the corresponding leaf skeleton, thus completing the initial organ allocation of the point cloud.

[0050] Furthermore, step S4 specifically includes:

[0051] Let the set of skeleton node coordinates be:

[0052] ;

[0053] in, Let M be the 3D coordinates of the j-th skeleton node, and M be the number of skeleton nodes; let the set of stem skeleton node indices be... The root node index at the base of the stem is , Obtained by skeleton decomposition; take the set of nodes corresponding to the stem skeleton:

[0054] ;

[0055] (1) Determine the direction of the stem axis: for In a one-dimensional principal component analysis, the direction vector of the first principal component is denoted as... And normalize it: To maintain consistency in direction, when The third component is inverted when it is positive, so that it satisfies the predetermined orientation;

[0056] (2) Determine the X / Y axes in a plane perpendicular to the stem axis: take the root node at the base of the stem as the origin of the coordinate system:

[0057] ;

[0058] Translate all skeleton nodes to Coordinates of the origin:

[0059] ;

[0060] And In Projecting onto a vertical plane:

[0061] ;

[0062] right Performing 2D PCA yields two principal direction vectors in the plane. Then, orthogonalization and normalization are performed to obtain the orthogonal basis. The three orthogonal axes of the plant coordinate system are ultimately defined as follows: ;

[0063] (3) Rotate the point cloud / skeleton to the plant coordinate system: Construct the rotation matrix:

[0064] ;

[0065] Then any three-dimensional point The coordinates after transformation to the plant coordinate system are:

[0066] ;

[0067] in These are the coordinates of the transformed point.

[0068] Further, step S5 specifically includes:

[0069] S51, Stem principal axis fitting and radial constraint

[0070] The stem point set obtained from the preliminary division Perform principal component analysis to extract the first principal axis direction:

[0071] ;

[0072] in: The normalized stem direction vector; any point The radial distance to the stem axis is defined as:

[0073] ;

[0074] in: The center point of the stem point set, when When the value exceeds a preset threshold, the point is determined to be a non-stalk point and removed from the stalk set;

[0075] S52, Density Constraints and Height-Priority Reassignment

[0076] For each point in the stem point set Calculate its local average neighborhood distance:

[0077] ;

[0078] in: For point The One nearest neighbor point; Let the number of neighboring points be , if If the distance is significantly greater than the global average neighborhood distance, the point is considered to belong to an abnormally sparse region, is marked as an undetermined point, and is reassigned to the most appropriate leaf organ category in subsequent steps;

[0079] Sort the set of undetermined points U in descending order of height:

[0080] ;

[0081] For each traverse all organ categories Calculate the consistency distance Choose the smallest one as the final owner:

[0082] ;

[0083] If the local neighborhood of all organs is empty, then there is no comparable region. Then the rollback strategy will be executed: That is, the undetermined point is eventually returned to the stem to ensure the continuity of the main structure;

[0084] S53, Height Constraints

[0085] Obtain the point cloud height sequence:

[0086] ;

[0087] in: For point The z-component. Define the upper and lower bounds of the height:

[0088] ;

[0089] Make the height span Let the top scale parameter be... Then the threshold for the ear is:

[0090] ;

[0091] When satisfied At that time, directly force the label of that point to be set to stem: ;

[0092] S54, Local Geometric Consistency Constraints

[0093] For each undetermined point Evaluate the geometric consistency of the organ with the stem and each leaf in their local neighborhood, assuming the organ category number is . , The set of points for this organ has been identified as follows:

[0094] ;

[0095] In each organ point set In China, with Centered on, with radius Take a local block within the sphere's neighborhood:

[0096] ;

[0097] in: Adaptive estimation based on point cloud density;

[0098] Define the point block to be evaluated With candidate organ blocks Sinkhorn OT cost, construct the cost matrix:

[0099] ;

[0100] And take the weight vector:

[0101] ;

[0102] With regularization coefficient Perform Sinkhorn calculations to obtain the OT cost:

[0103] ;

[0104] This cost is then used as the local geometric consistency distance.

[0105] Further, step S6 specifically includes:

[0106] S61, Multi-scale Adaptive Triangulation Reconstruction

[0107] The point cloud set of a single leaf obtained by fine segmentation First, the point cloud normal vectors are estimated, and the normal directions are unified through global consistency constraints; then, the average nearest neighbor distance of the leaf point cloud is calculated.

[0108] ;

[0109] Subsequently, a multi-scale adaptive rolling ball triangulation strategy was employed and the resulting mesh was merged.

[0110] The first stage uses a global-scale rolling sphere radius range to capture the continuous curved surface of the main body of the blade; the second stage uses a smaller local-scale rolling sphere radius range for sparse point cloud regions such as the leaf vein axis to fill the holes created by the main mesh in the leaf vein region; the triangular meshes generated in the two stages are combined to obtain the triangular mesh model of the blade.

[0111] ;

[0112] in: For the set of grid vertices, It is a set of triangular facets;

[0113] S62, Blade Area Calculation

[0114] For any triangular facet Let the coordinates of its three vertices be respectively Its semi-perimeter is:

[0115] ;

[0116] The area of ​​the triangular facet is calculated using Heron's formula:

[0117] ;

[0118] The total surface area of ​​a single blade is defined as:

[0119] ;

[0120] in: A set of triangular facets for blades; This represents Heron's formula for calculating the area of ​​a triangle based on the coordinates of its three vertices.

[0121] Compared with existing technologies, this invention achieves high-precision automatic segmentation of corn stalks and leaves through enhanced skeleton construction, skeleton-driven coarse segmentation, multi-constraint fine segmentation, and adaptive triangulation measurement strategies, and stably obtains key phenotypic parameters such as leaf area. Specifically, it is reflected in:

[0122] 1. Significantly improved stem and leaf segmentation accuracy: By innovatively integrating geometric features, spatial distribution, and local density information, precise segmentation of stems and leaves is achieved. Compared with existing technologies, this method provides clearer segmentation at stem-leaf boundaries, effectively solving the ambiguity problem in the stem-leaf junction area, especially at the junction of leaf sheath and stem, achieving more accurate organ differentiation.

[0123] 2. More accurate blade area measurement: By using the adaptive ball-pivoting method to reconstruct the blade point cloud at multiple scales and calculating the blade area using Heron's formula, the simplistic assumptions about blade shape made by traditional methods are avoided. By accurately calculating and summing the areas of each triangle, the true shape of the blade can be accurately reflected, unaffected by blade bending and deformation. This significantly improves the accuracy of blade area calculation and effectively reduces the impact of uneven point cloud distribution on the measurement results.

[0124] 3. High robustness of the skeleton structure: By introducing a large neighborhood search during the Laplace contraction phase and combining it with a connected component forced connection strategy, the skeleton breakage problem caused by the sparse point cloud at the base of the maize stem is effectively alleviated, ensuring the integrity of the topological structure of the entire plant.

[0125] 4. Fine segmentation has biological rationality: In the fine segmentation stage, axial radius constraints, point cloud density constraints and local neighborhood optimal matching mechanisms are introduced to make the point cloud assignment more consistent with the spatial distribution characteristics of real plant organs.

[0126] 5. High adaptability and high degree of automation: This method can complete the entire process from point cloud input to organ segmentation and trait output without manual interaction, and is suitable for point cloud data of maize single plants with different growth stages and different numbers of leaves. Attached Figure Description

[0127] Figure 1 This is a schematic diagram of the overall process for measuring the leaf geometric characteristics of a single mature maize plant in the indoor field, according to the present invention.

[0128] Figure 2 A performance comparison of multi-scale skeleton extraction algorithms and a visualization of key features.

[0129] Figure 3 This diagram illustrates the principle of constructing a coordinate system for maize plants and transforming point cloud space based on principal component analysis.

[0130] Figure 4 This is a schematic diagram of multi-constraint reclassification in fine point cloud segmentation.

[0131] Figure 5 A schematic diagram illustrating the principle of adaptive triangulation reconstruction and area calculation for leaf point clouds. Detailed Implementation

[0132] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0133] like Figure 1As shown, this embodiment proposes a method for measuring leaf geometric traits from point clouds of mature maize plants indoors. This embodiment uses point clouds of potted maize plants collected by an indoor high-throughput phenotyping platform as an example, but the invention is not limited to this application scenario. The main steps of the invention are as follows:

[0134] S1. Obtain three-dimensional point cloud data of mature maize plants indoors (which can be obtained directly using a ground-based lidar scanner) and preprocess them to obtain a standardized point cloud.

[0135] Let the original point cloud be represented as:

[0136] ;

[0137] in: Represents the three-dimensional spatial coordinates of the i-th point; N represents the total number of points in the point cloud;

[0138] First, an outlier filtering method based on statistical distribution is used to calculate the distance from each point to its nearest neighbor. If the average distance between the nearest neighbor points deviates from the global mean by more than a preset threshold, the point is identified as a noise point and removed, thereby eliminating the impact of measurement errors and environmental noise on subsequent processing.

[0139] Finally, point cloud downsampling is performed: the voxel mesh downsampling method is adopted to divide the three-dimensional space into cubic voxels with a side length of v. Only one representative point is retained in each voxel to reduce the size of the point cloud and maintain the overall geometric structure features.

[0140] After the above processing, the normalized point cloud is obtained:

[0141] ;

[0142] Where: M < N.

[0143] S2. Enhanced skeleton extraction is performed on the preprocessed point cloud;

[0144] S21, Laplace contraction

[0145] Perform a Laplace shrinkage operation on the preprocessed point cloud P': Let the nearest neighbor set of the i-th point be... Then its Laplace operator is defined as:

[0146] ;

[0147] in: This represents the Laplace displacement vector at the i-th point; For point The number of nearest neighbors; For point The position update formula for the j-th nearest neighbor is:

[0148] ;

[0149] in: , Let represent the coordinates of the point after the t-th and t+1-th iterations, respectively; The contraction coefficient is used to control the strength of point cloud aggregation towards the central axis. By expanding the nearest neighbor search range during the Laplacian contraction process, the sparse areas of the point cloud, such as the stem base, can still maintain sufficient geometric constraints, thereby effectively avoiding the breakage of the skeleton structure.

[0150] S22, Skeleton Node Sampling and Graph Construction

[0151] The farthest point sampling method is used to select the skeleton candidate node set on the shrunk point cloud:

[0152] ;

[0153] Where: U represents the set of skeleton nodes; K is the number of skeleton nodes. An undirected weighted skeleton graph G is constructed using the skeleton nodes as vertices:

[0154] ;

[0155] in: E represents the set of nodes in the graph; E represents the set of skeleton edges, when the node and When the Euclidean distance between them is less than a given threshold, establish an edge. The edge weight is defined as:

[0156] .

[0157] The specific process in this step is shown in Figure 2, which shows the original point cloud, the skeleton points after Laplacian shrinkage, the skeleton points extracted by FPS, and the skeleton map after edge collapse processing, intuitively presenting the effect of multi-scale skeleton extraction.

[0158] S3. Perform skeleton decomposition and coarse segmentation of stems and leaves on the extracted enhanced skeleton to obtain stem point sets and leaf point sets;

[0159] S31, Stem skeleton extraction

[0160] In the skeleton graph G, identify all nodes with a degree of 1 as the set of endpoints:

[0161] ;

[0162] Select the endpoint with the smallest height as the root node at the base of the stem. And calculate the shortest path distance from it to all nodes:

[0163] ;

[0164] in This represents the shortest path length from node v to the base of the stem; the node with the largest distance is selected. And the path: It is defined as the main framework of the stem, with the remaining branches serving as the leaf framework;

[0165] S32, Coarse segmentation based on nearest skeleton

[0166] For any point cloud point Find its nearest skeleton node:

[0167] ;

[0168] If the most recent skeleton node Then the point If a point is classified as a stem point, it is classified as a leaf point belonging to the corresponding leaf skeleton, thus completing the initial organ allocation of the point cloud.

[0169] S4. Construct the corn plant's own coordinate system based on the obtained stem points and leaf points, and transform the stem points and leaf points to the plant coordinate system;

[0170] Let the set of skeleton node coordinates be:

[0171] ;

[0172] in, Let M be the 3D coordinates of the j-th skeleton node, and M be the number of skeleton nodes; let the set of stem skeleton node indices be... The root node index at the base of the stem is , Obtained by skeleton decomposition; take the set of nodes corresponding to the stem skeleton:

[0173] ;

[0174] (1) Determine the direction of the stem axis: for One-dimensional principal component analysis (PCA) is performed, and the direction vector of the first principal component is denoted as... And normalize it: To maintain directional consistency, this embodiment adopts a directional reduction consistent with the code: when The third component is inverted when it is positive, so that it satisfies the predetermined orientation;

[0175] (2) Determine the X / Y axes in a plane perpendicular to the stem axis: take the root node at the base of the stem as the origin of the coordinate system:

[0176] ;

[0177] Translate all skeleton nodes to Coordinates of the origin:

[0178] ;

[0179] And In Projecting onto a vertical plane:

[0180] ;

[0181] right Performing 2D PCA yields two principal direction vectors in the plane. Then, orthogonalization and normalization are performed to obtain the orthogonal basis. The three orthogonal axes of the plant coordinate system are ultimately defined as follows: ;

[0182] (3) Rotate the point cloud / skeleton to the plant coordinate system: Construct the rotation matrix:

[0183] ;

[0184] Then any three-dimensional point The coordinates after transformation to the plant coordinate system are:

[0185] ;

[0186] in These are the coordinates of the transformed point. The specific process is as follows: Figure 3 As shown, this demonstrates how to determine the growth principal axis direction by calculating the centroid of the point cloud and PCA, establish an orthogonal coordinate system with the principal axis as the Z-axis, and rotate the point cloud to this coordinate system.

[0187] S5. Perform multi-constraint fine segmentation on the stem point set and leaf point set obtained from the coarse segmentation to obtain the coordinate-corrected skeleton map, stem point cloud and leaf point cloud;

[0188] S51, Stem principal axis fitting and radial constraint

[0189] The stem point set obtained from the preliminary division Perform principal component analysis (PCA) to extract the first principal axis direction:

[0190] ;

[0191] in: The normalized stem direction vector; any point The radial distance to the stem axis is defined as:

[0192] ;

[0193] in: The center point of the stem point set, when When the value exceeds a preset threshold, the point is determined to be a non-stalk point and removed from the stalk set;

[0194] S52, Density Constraints and Height-Priority Reassignment

[0195] For each point in the stem point set Calculate its local average neighborhood distance:

[0196] ;

[0197] in: For point The One nearest neighbor point; Let the number of neighboring points be , if If the distance is significantly greater than the global average neighborhood distance, the point is considered to belong to an abnormally sparse region, is marked as an undetermined point, and is reassigned to the most appropriate leaf organ category in subsequent steps;

[0198] To improve the stability of the top structure, the set of undetermined points U is sorted from highest to lowest height:

[0199] ;

[0200] For each traverse all organ categories Calculate the consistency distance Choose the smallest one as the final owner:

[0201] ;

[0202] If the local neighborhood of all organs is empty, then there is no comparable region. Then the rollback strategy will be executed: That is, the undetermined point is eventually returned to the stem to ensure the continuity of the main structure;

[0203] S53, Height Constraints

[0204] To avoid misidentifying the top of the corn ear at maturity as leaves, this embodiment employs a forced constraint based on the actual point cloud height. Point cloud height sequence is obtained as follows:

[0205] ;

[0206] in: For point The z-component. Define the upper and lower bounds of the height:

[0207] ;

[0208] Make the height span Let the top scale parameter be... Then the threshold for the ear is:

[0209] ;

[0210] When satisfied At that time, directly force the label of that point to be set to stem: ;

[0211] S54, Local Geometric Consistency Constraints

[0212] For each undetermined point Evaluate the geometric consistency between the organ and each organ category (stem and leaf) in their local neighborhoods. Let the organ category number be . (For example The set of points for this organ has been determined as follows:

[0213] ;

[0214] In each organ point set In China, with Centered on, with radius Take a local block within the sphere's neighborhood:

[0215] ;

[0216] in: Adaptive estimation based on point cloud density;

[0217] Define the point block to be evaluated (Single-point block) and candidate organ block Sinkhorn OT cost, construct the cost matrix:

[0218] ;

[0219] And take the weight vector:

[0220] ;

[0221] With regularization coefficient Perform Sinkhorn calculations to obtain the OT cost:

[0222] ;

[0223] And this cost is used as the local geometric consistency distance;

[0224] The specific process is as follows: Figure 4As shown, based on the coarse segmentation of the point cloud of a single maize plant, various geometric constraints are introduced for fine reclassification, targeting easily confused points in the stem-leaf connection region, the sparse top region, and the ear region. The figure illustrates the gradual correction process of each constraint on misclassified points at different stages using a faceted approach.

[0225] S6, such as Figure 5 As shown, the obtained leaf point cloud is triangulated and reconstructed, and the total area of ​​a single leaf is calculated:

[0226] S61, Multi-scale Adaptive Triangulation Reconstruction

[0227] The point cloud set of a single leaf obtained by fine segmentation First, the point cloud normal vectors are estimated, and the normal directions are unified through global consistency constraints; then, the average nearest neighbor distance of the leaf point cloud is calculated.

[0228] ;

[0229] Subsequently, a multi-scale adaptive ball piping strategy was employed and the meshes were merged to generate the mesh:

[0230] The first stage uses a global-scale rolling sphere radius range to capture the continuous curved surface of the main body of the blade; the second stage uses a smaller local-scale rolling sphere radius range for sparse point cloud regions such as the leaf vein axis to fill the holes created by the main mesh in the leaf vein region; the triangular meshes generated in the two stages are combined to obtain the triangular mesh model of the blade.

[0231] ;

[0232] in: For the set of grid vertices, This is a set of triangular facets. This strategy effectively overcomes the problem of underestimation of surface area caused by uneven point cloud density in the leaves.

[0233] S62, Blade Area Calculation

[0234] For any triangular facet Let the coordinates of its three vertices be respectively Its semi-perimeter is:

[0235] ;

[0236] The area of ​​the triangular facet is calculated using Heron's formula:

[0237] ;

[0238] The total surface area of ​​a single blade is defined as:

[0239] ;

[0240] in: A set of triangular facets for blades; This represents Heron's formula for calculating the area of ​​a triangle based on the coordinates of its three vertices.

[0241] The specific process is as follows: Figure 5 As shown, the triangulation process of a single leaf point cloud, the method for calculating the area of ​​triangles, and the summation process of the total leaf area are illustrated, intuitively presenting the principle of leaf area calculation. Through the aforementioned multi-scale adaptive triangulation and area calculation methods, the systematic underestimation of leaf area caused by uneven leaf point cloud density or insufficient sampling of the leaf vein region is effectively avoided, thereby significantly improving the stability and accuracy of leaf geometric shape measurement.

[0242] In summary, this invention achieves high-precision automatic segmentation of corn stalks and leaves and stably obtains key phenotypic parameters such as leaf area by using enhanced skeleton construction, skeleton-driven coarse segmentation, multi-constraint fine segmentation, and adaptive triangulation measurement strategies.

[0243] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. A method for measuring the leaf geometric characteristics of a single mature maize plant in an indoor environment using point clouds, characterized in that, Includes the following steps: S1. Obtain the three-dimensional point cloud data of a single mature maize plant indoors and preprocess it to obtain a standardized point cloud. S2. Enhanced skeleton extraction is performed on the preprocessed point cloud; S3. Perform skeleton decomposition and coarse segmentation of stems and leaves on the extracted enhanced skeleton to obtain stem point sets and leaf point sets; S4. Construct the corn plant's own coordinate system based on the obtained stem points and leaf points, and transform the stem points and leaf points to the plant coordinate system; S5. Perform multi-constraint fine segmentation on the stem point set and leaf point set obtained from the coarse segmentation to obtain the coordinate-corrected skeleton map, stem point cloud and leaf point cloud; S6. Triangulate and reconstruct the obtained leaf point cloud and calculate the total area of ​​a single leaf.

2. The method for measuring leaf geometric traits of a single mature maize plant in an indoor environment according to claim 1, characterized in that, The preprocessing in step S1 includes: Let the original point cloud be represented as: ； in: Represents the three-dimensional spatial coordinates of the i-th point; N represents the total number of points in the point cloud; First, an outlier filtering method based on statistical distribution is used to calculate the distance from each point to its nearest neighbor. If the average distance between the nearest neighbor points deviates from the global mean by more than a preset threshold, the point is identified as a noise point and removed, thereby eliminating measurement errors and environmental noise. Finally, point cloud downsampling is performed: the voxel mesh downsampling method is adopted to divide the three-dimensional space into cubic voxels with a side length of v. Only one representative point is retained in each voxel to reduce the size of the point cloud and maintain the overall geometric structure features. After the above processing, the normalized point cloud is obtained: ； Where: M < N.

3. The method for measuring leaf geometric traits of a single mature maize plant in an indoor environment using point clouds, as described in claim 2, is characterized in that... Step S2 specifically includes: S21, Laplace contraction Perform a Laplace shrinkage operation on the preprocessed point cloud P': Let the nearest neighbor set of the i-th point be... Then its Laplace operator is defined as: ； in: This represents the Laplace displacement vector at the i-th point; For point The number of nearest neighbors; For point The position update formula for the j-th nearest neighbor is: ； in: , Let represent the coordinates of the point after the t-th and t+1-th iterations, respectively; This is the shrinkage coefficient, used to control the intensity of point cloud aggregation towards the central axis; S22, Skeleton Node Sampling and Graph Construction The farthest point sampling method is used to select the skeleton candidate node set on the shrunk point cloud: ； Where: U represents the set of skeleton nodes; K is the number of skeleton nodes. An undirected weighted skeleton graph G is constructed using the skeleton nodes as vertices: ； in: E represents the set of nodes in the graph; E represents the set of skeleton edges, when the node and When the Euclidean distance between them is less than a given threshold, establish an edge. The edge weight is defined as: 。 4. The method for measuring leaf geometric traits of a single mature maize plant in an indoor location based on point clouds, as described in claim 1, is characterized in that... Step S3 specifically includes: S31, Stem skeleton extraction In the skeleton graph G, identify all nodes with a degree of 1 as the set of endpoints: ； Select the endpoint with the smallest height as the root node at the base of the stem. And calculate the shortest path distance from it to all nodes: ； in This represents the shortest path length from node v to the base of the stem; the node with the largest distance is selected. And the path: It is defined as the main framework of the stem, with the remaining branches serving as the leaf framework; S32, Coarse segmentation based on nearest skeleton For any point cloud point Find its nearest skeleton node: ； If the most recent skeleton node Then the point If a point is classified as a stem point, it is classified as a leaf point belonging to the corresponding leaf skeleton, thus completing the initial organ allocation of the point cloud.

5. The method for measuring leaf geometric traits of a single mature maize plant in an indoor location cloud as described in claim 1, characterized in that, Step S4 specifically includes: Let the set of skeleton node coordinates be: ； in, Let M be the 3D coordinates of the j-th skeleton node, and M be the number of skeleton nodes; let the set of stem skeleton node indices be... The root node index at the base of the stem is , Obtained by skeleton decomposition; take the set of nodes corresponding to the stem skeleton: ； (1) Determine the direction of the stem axis: for In a one-dimensional principal component analysis, the direction vector of the first principal component is denoted as... And normalize it: To maintain consistency in direction, when The third component is inverted when it is positive, so that it satisfies the predetermined orientation; (2) Determine the X / Y axes in a plane perpendicular to the stem axis: take the root node at the base of the stem as the origin of the coordinate system: ； Translate all skeleton nodes to Coordinates of the origin: ； And In Projecting onto a vertical plane: ； right Performing 2D PCA yields two principal direction vectors in the plane. Then, orthogonalization and normalization are performed to obtain the orthogonal basis. The three orthogonal axes of the plant coordinate system are ultimately defined as follows: ; (3) Rotate the point cloud / skeleton to the plant coordinate system: Construct the rotation matrix: ； Then any three-dimensional point The coordinates after transformation to the plant coordinate system are: ； in These are the coordinates of the transformed point.

6. The method for measuring leaf geometric traits of a single mature maize plant in an indoor location based on point clouds, as described in claim 1, is characterized in that... Step S5 specifically includes: S51, Stem principal axis fitting and radial constraint The stem point set obtained from the preliminary division Perform principal component analysis to extract the first principal axis direction: ； in: The normalized stem direction vector; any point The radial distance to the stem axis is defined as: ； in: The center point of the stem point set, when When the value exceeds a preset threshold, the point is determined to be a non-stalk point and removed from the stalk set; S52, Density Constraints and Height-Priority Reassignment For each point in the stem point set Calculate its local average neighborhood distance: ； in: For point The One nearest neighbor point; Let the number of neighboring points be , if If the distance is significantly greater than the global average neighborhood distance, the point is considered to belong to an abnormally sparse region, is marked as an undetermined point, and is reassigned to the most appropriate leaf organ category in subsequent steps; Sort the set of undetermined points U in descending order of height: ； For each traverse all organ categories Calculate the consistency distance Choose the smallest one as the final owner: ； If the local neighborhood of all organs is empty, then there is no comparable region. Then the rollback strategy will be executed: That is, the undetermined point is eventually returned to the stem to ensure the continuity of the main structure; S53, Height Constraints Obtain the point cloud height sequence: ； in: For point The z-component. Define the upper and lower bounds of the height: ； Make the height span Let the top scale parameter be... Then the threshold for the ear is: ； When satisfied At that time, directly force the label of that point to be set to stem: ; S54, Local Geometric Consistency Constraints For each undetermined point Evaluate the geometric consistency of the organ with the stem and each leaf in their local neighborhood, assuming the organ category number is . , The set of points for this organ has been identified as follows: ； In each organ point set In China, with Centered on, with radius Take a local block within the sphere's neighborhood: ； in: Adaptive estimation based on point cloud density; Define the point block to be evaluated With candidate organ blocks Sinkhorn OT cost, construct the cost matrix: ； And take the weight vector: ； With regularization coefficient Perform Sinkhorn calculations to obtain the OT cost: ； This cost is then used as the local geometric consistency distance.

7. The method for measuring leaf geometric traits of a single mature maize plant in an indoor environment according to claim 1, characterized in that, Step S6 specifically includes: S61, Multi-scale Adaptive Triangulation Reconstruction The point cloud set of a single leaf obtained by fine segmentation First, the point cloud normal vectors are estimated, and the normal directions are unified through global consistency constraints; then, the average nearest neighbor distance of the leaf point cloud is calculated. ； Subsequently, a multi-scale adaptive rolling ball triangulation strategy was employed and the resulting mesh was merged. The first stage uses a global-scale rolling sphere radius range to capture the continuous curved surface of the main body of the blade; the second stage uses a smaller local-scale rolling sphere radius range for sparse point cloud regions such as the leaf vein axis to fill the holes created by the main mesh in the leaf vein region; the triangular meshes generated in the two stages are combined to obtain the triangular mesh model of the blade. ； in: For the set of grid vertices, It is a set of triangular facets; S62, Blade Area Calculation For any triangular facet Let the coordinates of its three vertices be respectively Its semi-perimeter is: ； The area of ​​the triangular facet is calculated using Heron's formula: ； The total surface area of ​​a single blade is defined as: ； in: A set of triangular facets for blades; This represents Heron's formula for calculating the area of ​​a triangle based on the coordinates of its three vertices.

Images