Method for extracting multi-dimensional phenotypic traits of fruit tree root system based on three-dimensional point cloud

Abstract

The application discloses a fruit tree root system multi-dimensional phenotype trait extraction method based on a three-dimensional point cloud, and comprises the following steps: obtaining a dense three-dimensional root system point cloud obtained by a three-dimensional reconstruction method and performing pretreatment; extracting geometric traits from the pretreated three-dimensional root system point cloud; simultaneously performing skeletonization processing on the pretreated three-dimensional root system point cloud to obtain a one-dimensional root system skeleton graph, and extracting topological traits from the skeleton graph; along each skeleton point of the skeleton graph, performing circular cross-section fitting on surrounding point clouds on a cross-section perpendicular to a local direction of the skeleton point, estimating a local diameter, and extracting diameter traits; according to size traits, spatial range traits, distribution traits, topological traits and diameter traits, derivative traits and root crown coordination traits are extracted; and all the extracted traits are integrated to output a multi-dimensional standardized phenotype vector, and the application provides a comprehensive and quantitative data basis for fruit tree seedling root system quality evaluation and transplanting survival prediction.

Description

Technical Field

[0001] This invention relates to the fields of agricultural information and plant phenomics, specifically to a method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds. Background Technology

[0002] The structure and functional characteristics of fruit tree roots directly determine the seedlings' ability to acquire water and nutrients after transplanting, and are key factors affecting transplant survival. A comprehensive and accurate phenotypic analysis of the root system is fundamental to root quality assessment and transplant survival prediction.

[0003] However, traditional methods for obtaining root phenotypes often rely on destructive measurements, such as root dry weight weighing or root growth potential bioassays. These operations are not only time-consuming and labor-intensive but also irreversible, as the measured seedlings cannot be used for field planting, making it difficult to meet the quality screening requirements of testing before planting. While two-dimensional imaging methods using flatbed scanning combined with image analysis software can extract basic parameters such as root length, root diameter, and projected area, they completely lose the true three-dimensional spatial information of the root system in the soil. This makes it impossible to accurately characterize key functional traits such as root volume, three-dimensional spatial distribution, branching topology, and soil exploration ability. Numerous studies have shown that the three-dimensional configuration of the root system has a crucial impact on transplant survival, and two-dimensional projection cannot capture this important information.

[0004] Even with the emergence of 3D imaging technologies such as CT and MRI in recent years, current studies typically extract only one or two traits, such as root mass or root length. They lack a systematic characterization of root geometry, topology, diameter composition, spatial distribution, and root-crown coordination, resulting in extracted traits that fail to fully reflect the multidimensional functional characteristics of the root system. This limits the accuracy of transplant survival prediction models. Furthermore, despite the continuous development of 3D reconstruction technology, a complete technical pipeline for systematically extracting multidimensional phenotypic traits from 3D point cloud data remains lacking. Most existing 3D root analysis tools are designed for the root systems of small plants such as Arabidopsis and rice, and a standardized extraction process suitable for the root systems of large fruit tree seedlings has not yet been provided. Moreover, the definitions and calculation methods of traits vary between different studies, making cross-study comparisons and validations difficult.

[0005] Therefore, there is an urgent need to establish a complete technical solution that can automatically and systematically extract multi-dimensional root phenotypic traits from three-dimensional point clouds, so as to provide a comprehensive and quantitative data foundation for accurate assessment of root quality and prediction of transplant survival. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: This application provides a method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds, including: Obtain a dense 3D root system point cloud obtained by a 3D reconstruction method and preprocess the 3D root system point cloud; Size characteristics, spatial range characteristics, and distribution characteristics were extracted from the preprocessed 3D root system point cloud. Simultaneously, skeletonization processing is performed on the preprocessed 3D root system point cloud to obtain a 1D root system skeleton map, and topological features are extracted from the skeleton map. Along each skeleton point of the skeleton diagram, a circular cross-section is fitted to the surrounding point cloud on a cross-section perpendicular to the local direction of the skeleton point to estimate the local diameter and extract the diameter characteristics; Based on the aforementioned size traits, spatial range traits, distribution traits, topological traits, and diameter traits, derived traits and root-crown coordination traits are extracted; The extracted size traits, spatial range traits, distribution traits, topological traits, diameter traits, derived traits, and root-crown coordination traits are integrated to output a multidimensional standardized phenotypic vector.

[0007] In one possible implementation, the acquisition of the dense 3D root system point cloud obtained by the 3D reconstruction method and the preprocessing of the 3D root system point cloud include: Obtain the dense 3D root point cloud obtained from 3D reconstruction, calculate the average Euclidean distance from each point to multiple nearest neighbors, and calculate the mean and standard deviation of the average Euclidean distance of all points. Outliers whose average Euclidean distance exceeds a preset value are removed, and a three-dimensional coordinate system is established with the root position as the origin. The coordinates of all point clouds are translated and rotated in the three-dimensional coordinate system to complete coordinate normalization and output the preprocessed standard three-dimensional root system point cloud.

[0008] In one possible implementation, dimensional properties are extracted from the preprocessed 3D root point cloud, including: An alpha shape surface mesh is constructed from the preprocessed 3D root system point cloud. The alpha shape parameters are determined based on the average nearest neighbor distance of each point in the 3D root system point cloud, using the following formula: in, The average nearest neighbor distance of each point in the 3D root system point cloud is used; a closed surface mesh surrounding the root system point cloud is generated according to the alpha shape parameter; the total root volume is obtained by summing the volume of the directed tetrahedron formed by the mesh patch and the reference point; and the convex hull volume surrounding all root system point cloud points is calculated using the Quickhull algorithm. The total root surface area is obtained by calculating the area of ​​each triangular facet of the alpha-shaped surface mesh. The formula is as follows: in, The total root surface area, For the first The area of ​​each grid patch This represents the total number of grid patches; The surface area to volume ratio is calculated based on the total root surface area and the total root volume, using the following formula: in, It is the surface area to volume ratio. This is the total root volume; Simultaneously, the root system compactness is calculated based on the total root volume and the convex hull volume, using the following formula: in, For root system compactness, Let be the convex hull volume.

[0009] In one possible implementation, spatial extent characteristics are extracted from the preprocessed 3D root point cloud, including: The vertical coordinates and horizontal distances to the central axis of all three-dimensional root point cloud points are calculated. The vertical distance from the root collar to the lowest root point is taken as the maximum root depth, and the maximum horizontal distance from all points to the central axis is taken as the maximum radial spread. The calculation formulas are as follows: in, For the maximum root depth, The Z-coordinate of the lowest root point; For maximum radial span, and The first The coordinate components of a point cloud point in the horizontal plane; The 90th percentile of the absolute value of the root point cloud depth is taken as the 90% biomass depth, and the 90th percentile of the horizontal radial distance of the point cloud points is taken as the 90% biomass radial range. The calculation formulas are as follows: in, For 90% biomass depth, The radial range is 90% biomass. Represents the 90th percentile. For the first Vertical coordinates of each point cloud point; The depth-to-spread ratio is calculated based on the maximum root depth and the maximum radial spread, using the following formula: in, For deeper development; The maximum root depth; This represents the maximum radial span.

[0010] In one possible implementation, the distribution characteristics are extracted from the preprocessed 3D root point cloud, including: The vertical centroid is calculated based on the average vertical coordinates of all 3D root point cloud points. The calculation formula is as follows: in, It is the vertical centroid; This represents the total number of points in the point cloud. For the first Vertical coordinates of each point cloud point; The 3D root system point cloud is divided into multiple fan-shaped regions with preset angles on a horizontal plane. The standard deviation and mean of the point cloud point count ratio in each fan-shaped region are calculated. The horizontal uniformity is obtained based on the standard deviation and the mean. The calculation formula is as follows: in, For horizontal uniformity; and These are the standard deviation and mean of the percentage of points in the sector region, respectively. The vertical skewness is calculated based on the vertical coordinate distribution skewness of all three-dimensional root system point cloud points. The calculation formula is as follows: in, Vertical skewness; The average vertical coordinate of all point cloud points; An exponential decay model is fitted based on the density variation of 3D root point cloud points in the vertical direction with depth, and the decay coefficient is used as the root density gradient. The calculation formula is as follows: in, For depth Root point cloud density at the location; This represents the initial root point cloud density; The root density gradient; The exploration efficiency is calculated based on the convex hull volume and the hemispherical volume with the maximum 3D root distance as its radius. The calculation formula is as follows: in, To explore efficiency, Let convex hull volume be the volume. The maximum three-dimensional distance from a point in the point cloud to the origin of the root collar; Pi; and The first The coordinate components of a point cloud point in the horizontal plane For the first The vertical coordinates of each point cloud point.

[0011] In one possible implementation, the process of performing skeletonization on the preprocessed 3D root point cloud to obtain a 1D root skeleton map, and extracting topological features from the skeleton map, includes: A Laplacian-based point cloud shrinkage algorithm is used to iteratively shrink the preprocessed 3D root system point cloud, so that the point cloud converges towards the central axis and maintains the topological structure. After the contraction is completed, the connection threshold is adaptively determined based on the local point cloud density and the contracted point set is converted into a graph structure. Redundant nodes are deleted by edge folding while retaining branch points and endpoints. Then, a Gaussian filter is used to smooth the skeleton node positions to obtain a one-dimensional root system skeleton graph. The total root length is calculated by summing the Euclidean lengths of all edges in the skeleton graph, using the following formula: in, The total root length; For the skeleton edge set; For the skeleton edge Length; The number of nodes with a preset degree in the skeleton graph is taken as the root tip number, and the number of nodes with a degree greater than or equal to the preset value is taken as the branch point number. The branch density is calculated based on the number of branch points and the total root length, using the following formula: in, Branch density; Number of branch points; The total root length; For each branch point, determine the direction vector of the parent root and the direction vector of the child root, calculate the angle between each child root and the parent root, and average all the angles to obtain the average branch angle. The calculation formula is as follows: in, The average branch angle; The total number of included angles; For the first Angle between the mother root and the daughter root; Starting from the root neck node, traverse the skeleton graph and count the maximum number of branch points along any path from the root neck to the root tip. Use the maximum number of branch points as the topological depth. The branching order of each root tip is determined according to the root segment hierarchy, and the average branching order of all root tips is calculated as the average branching order. The calculation formula is as follows: in, The average branch order; Root number; For the first The branch order corresponding to each root tip; Further grouping and summing the skeleton edge lengths according to the branch order yields the first-order root length, second-order root length, and third-order and above root lengths. The calculation formula is as follows: in, , and These are the first-order root length, the second-order root length, and the third-order and above root length, respectively. , and These are sets of skeleton edges of order one, two, and three and above, respectively. For the skeleton edge The length.

[0012] In one possible implementation, along each skeleton point of the skeleton map, a circular cross-section is fitted to the surrounding point cloud on a cross-section perpendicular to the local direction of the skeleton point, the local diameter is estimated, and the diameter characteristics are extracted, including: For each skeleton point in the skeleton diagram, determine the local skeleton direction vector; In a cross-sectional plane perpendicular to the local skeleton direction vector, obtain neighboring point cloud points, project the neighboring point cloud points onto the cross-sectional plane, and fit a circle using the least squares method. The diameter of the fitted circle is used as the local root diameter at the skeleton point. When the number of projection points is less than the preset number or the circle fitting residual exceeds the preset value, this local root diameter is marked as unreliable and corrected by interpolation of adjacent reliable estimates. Using the length of each skeleton edge as a weight, a weighted average is calculated for the local root diameter at each skeleton point to obtain the average root diameter. The calculation formula is as follows: in, The average root diameter; For the skeleton edge Length; For the skeleton edge The average diameter of the two end skeleton points; Obtain the total length of the fine root segments whose diameters are less than a preset value at both ends of the skeleton edge, and calculate the fine root ratio based on the total length of the fine root segments and the total root length. The calculation formula is as follows: in, For the proportion of fine roots; The total length of the fine root segments with a diameter of less than 2 mm; The total root length.

[0013] In one possible implementation, derived traits and root-crown coordination traits are extracted based on the size traits, spatial extent traits, distribution traits, topological traits, and diameter traits, including: The root-to-root ratio is calculated based on the total root volume in the size trait and the total root length in the topological trait. The calculation formula is as follows: in, It is longer than the root; The total root length; This is the total root volume; The exploration-utilization index is calculated based on the convex hull volume, total root volume, and convex hull volume scaling term in the dimensional properties, and the total root length in the topological properties. The calculation formula is as follows: in, For exploration and utilization index; Let convex hull volume be the volume of the convex hull. This is the total root volume; The total root length; The feature length scale is obtained by converting the convex hull volume; The root-to-crown volume ratio is calculated based on the total root volume, trunk cross-sectional area, and plant height, using the following formula: in, The root-to-crown volume ratio; The cross-sectional area of ​​the tree trunk; Plant height; The functional balance index is calculated based on the total root surface area, trunk cross-sectional area, and number of lateral branches in the size traits. The calculation formula is as follows: in, It is a functional balance index; Estimate the surface area of ​​the total root surface area or the surface area of ​​the fine roots; The cross-sectional area of ​​the tree trunk; This represents the number of lateral branches.

[0014] Compared with the prior art, the beneficial effects of this application are as follows: All traits in this application are extracted from three-dimensional point cloud data without damaging the seedlings. This enables a seedling quality screening process of testing before planting, fully preserving three-dimensional spatial information and avoiding information loss caused by two-dimensional projection.

[0015] This application encompasses five major categories—geometric structure, topological architecture, diameter composition, functional derivation, and root-crown coordination—and a total of 31 traits. It systematically characterizes the size, shape, spatial distribution, branching complexity, diameter composition, and functional features of root systems. The method is applicable to three-dimensional point clouds of root systems in bare-root seedlings of apple, pear, peach, cherry, or other fruit tree species. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds, provided for an embodiment of this application; Figure 2 This is a schematic diagram of alpha shape surface mesh reconstruction provided in an embodiment of this application; Figure 3 This is a schematic diagram of volume calculation provided for an embodiment of this application; Figure 4 This is a schematic diagram of the three-dimensional spatial distribution characteristics of the root system provided in an embodiment of this application; wherein, Figure 4 (a) in the diagram is a schematic diagram of a three-dimensional cross-section of the root system. Figure 4 (b) in the diagram is a polar coordinate orientation distribution map. Figure 4 (c) in the diagram is the root density gradient profile. Figure 4 (d) in the figure represents the vertical skewness distribution. Figure 5 This is a schematic diagram of topological feature extraction provided in an embodiment of this application; Figure 6 This is a schematic diagram of diameter distribution provided for an embodiment of this application; Figure 7 This is a schematic diagram illustrating the proportion of fine roots provided in an embodiment of this application. Figure 8 This is a schematic diagram illustrating the extraction of derived traits and root-crown coordination in an embodiment of this application. Figure 9 The correlation network diagram provided for the embodiments of this application; Figure 10 The gravel diagram provided for the embodiments of this application; Figure 11 The accumulated variance principal component analysis plot provided for the embodiments of this application; Figure 12 Principal component score plots provided for embodiments of this application; Figure 13 Load heatmap provided for embodiments of this application. Detailed Implementation

[0017] The present invention will be described below with reference to the accompanying drawings and specific embodiments. The following examples use 118 two-year-old bare-root Fuji / M.26 apple seedlings as experimental material, but the scope of this application is not limited to apples and can be extended to other fruit tree species such as pears, peaches, and cherries. All calculations were performed in a Python 3.10 environment, using libraries such as Open3D, NetworkX, SciPy, and NumPy.

[0018] Figure 1 A flowchart illustrating a method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds, provided in this application embodiment, is shown below. Figure 1 This embodiment provides a method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds, comprising: S101: Obtain the dense 3D root system point cloud obtained by the 3D reconstruction method and preprocess the 3D root system point cloud.

[0019] In this embodiment, a dense 3D root point cloud is obtained using a 3D reconstruction method for fruit tree roots based on root adaptive 3D Gaussian sputtering. The point cloud for each tree has 90,000-160,000 points, and each point contains 3D coordinates. and color information .

[0020] The average Euclidean distance from each point to its 50 nearest neighbors is calculated. The mean and standard deviation of the average Euclidean distance of all points are also calculated. Outliers whose average Euclidean distance exceeds the mean plus twice the standard deviation are removed. Typically, 0.5%-2% of the points are removed.

[0021] The root collar position is automatically detected and defined as the minimum Z-coordinate of the point cloud, i.e., the highest point. Since the seedlings are photographed upside down, the root collar is used as the origin of the coordinate system, and the Z-axis direction downward is the positive depth direction. All point cloud coordinates are translated to this coordinate system to complete the coordinate normalization and output the preprocessed standard three-dimensional root system point cloud.

[0022] S102, extract size characteristics, spatial range characteristics and distribution characteristics from the preprocessed 3D root system point cloud.

[0023] In this embodiment, 15 geometric traits are extracted from the preprocessed 3D root point cloud, categorized into size traits, spatial extent traits, and distribution traits. Size traits include total root volume, convex hull volume, root compactness, total surface area, and surface area-to-volume ratio. Spatial extent traits include maximum root depth, maximum radial spread, 90% biomass depth, 90% biomass radial range, and depth-to-spread ratio. Distribution traits include vertical centroid, horizontal evenness, vertical skewness, root density gradient, and exploration efficiency.

[0024] See Figure 2 and Figure 3 In this embodiment, when extracting dimensional characteristics, an alpha shape surface mesh is constructed from the preprocessed 3D root point cloud. The alpha shape parameter is determined based on the average nearest neighbor distance of each point in the 3D root point cloud, and the calculation formula is as follows: in, This represents the average nearest neighbor distance for each point in the 3D root system point cloud. A closed surface mesh enclosing the root system point cloud is generated based on the alpha shape parameter. The total root volume is obtained by summing the volumes of the directed tetrahedrons formed by the mesh patches and the reference points, and the convex hull volume enclosing all root system point cloud points is calculated using the Quickhull algorithm. This alpha shape parameter is used to generate a surface mesh that can enclose fine roots without excessively crossing voids. The total root volume is obtained by summing the volumes of the directed tetrahedrons formed by the mesh patches and the reference points. The convex hull volume is calculated using the Quickhull algorithm. .

[0025] The total root surface area is obtained by summing the areas of the triangular faces of the alpha-shaped mesh, using the following formula: in, This represents the total root surface area; For the first The area of ​​each grid patch; This represents the total number of grid faces.

[0026] The ratio of surface area to volume is calculated based on the total root surface area and total root volume. The formula is as follows: in, A higher value indicates a finer root system and more branches; This represents the total root surface area; Let be the total root volume. The root system compactness is calculated based on the total root volume and the convex hull volume, using the following formula: in, Root system compactness reflects the degree to which the root system fills the convex hull space; The convex hull volume represents the volume of soil space explored by the root system.

[0027] When extracting the spatial extent characteristics, the vertical coordinates and horizontal distances to the central axis of all three-dimensional root point cloud points are calculated. The vertical distance from the root collar to the lowest root point is taken as the maximum root depth, and the maximum horizontal distance is taken as the maximum radial spread. The calculation formulas are as follows: in, The maximum root depth; The Z-coordinate of the lowest root point; Maximum radial span; and The first The coordinate components of a point cloud point in the horizontal plane.

[0028] The 90th percentile of the absolute value of point cloud depth is taken as the 90% biomass depth, and the 90th percentile of the horizontal radial distance of point cloud points is taken as the 90% biomass radial range. The calculation formulas are as follows: in, 90% biomass depth; The radial range is 90% of the biomass; Represents the 90th percentile; For the first The vertical coordinates of each point cloud point.

[0029] The depth-to-spread ratio is calculated based on the maximum root depth and maximum radial spread. The calculation formula is as follows: in, It is used to characterize the proportional relationship between the root system's ability to penetrate vertically and its ability to spread horizontally.

[0030] Figure 4 This is a schematic diagram of the three-dimensional spatial distribution characteristics of the root system provided in an embodiment of this application, wherein, Figure 4 (a) in the diagram is a three-dimensional cross-section of the root system, showing the vertical centroid and exploration efficiency. Figure 4 (b) in the diagram is a polar coordinate azimuth distribution map, showing the horizontal uniformity. Figure 4 (c) in the diagram is a root density gradient profile, showing the decay of the root density gradient with depth. Figure 4 (d) in the diagram represents the vertical skewness distribution, showing the degree of horizontal aggregation and offset of roots in each soil layer. When extracting the distribution characteristics, the arithmetic mean of the Z-coordinates of all point cloud points is used as the vertical centroid, calculated using the following formula: in, It is the vertical centroid; This represents the total number of points in the point cloud. For the first The vertical coordinates of each point cloud point.

[0031] The standard skewness statistic of the Z-coordinate is taken as the vertical skewness, and the calculation formula is: in, Vertical skewness; The average vertical coordinate of all point cloud points; A negative value indicates that the root point cloud is more concentrated in the deeper regions, while a positive value indicates that it is more concentrated in the near-surface regions.

[0032] The 3D root system point cloud is divided into 12 30° sector regions on a horizontal plane. The standard deviation and mean of the proportion of points in each sector region are calculated, and the horizontal uniformity is calculated accordingly. The calculation formula is as follows: in, and These are the standard deviation and mean of the percentage of points in the sector region, respectively. The closer to 1, the more uniform the radial distribution.

[0033] The root density gradient is obtained by fitting an exponential decay model of root point cloud density with depth. The calculation formula is as follows: in, For depth Root point cloud density at the location; This represents the initial root point cloud density; This represents the root density gradient.

[0034] The exploration efficiency is calculated based on the convex hull volume and the hemispherical volume with the maximum 3D root distance as its radius. The calculation formula is as follows: in, To explore efficiency, we characterize the actual utilization of potential explorable soil space by the root system; Let convex hull volume be the volume of the convex hull. The maximum three-dimensional distance from a point in the point cloud to the origin of the root collar; Pi; For the first The vertical coordinates of each point cloud point.

[0035] S103, simultaneously, skeletonization processing is performed on the preprocessed 3D root system point cloud to obtain a 1D root system skeleton map, and topological features are extracted from the skeleton map.

[0036] See Figure 5In this embodiment, the preprocessed point cloud is subjected to skeletonization processing to extract a one-dimensional root system skeleton diagram. Then, 10 topological traits are extracted from the skeleton diagram, including total root length, number of root tips, number of branch points, branch density, average branch angle, topological depth, average branch order, first-order root length, second-order root length, and third-order and above root length.

[0037] A Laplacian-based point cloud shrinkage algorithm is used to iteratively shrink the point cloud. The initial smoothing weights and attraction weights are as follows: in, For Laplacian smoothing weights, Apply constraint weights to the point cloud. The number of shrinkage iterations is set to 20. After shrinkage, the connection threshold is adaptively determined based on the local point cloud density to construct the graph structure. Redundant nodes are removed through iterative edge folding while retaining branch points and endpoints. The Gaussian smoothing filter smooths the skeleton node positions, where This represents the standard deviation of the Gaussian smoothing filter.

[0038] The total root length is calculated by summing the Euclidean lengths of all edges in the skeleton graph, using the following formula: in, The total root length; For the skeleton edge set; For the skeleton edge The length.

[0039] The number of nodes with degree 1 in the skeleton graph is used as the root tip number. The number of nodes with a degree greater than or equal to 3 is used as the number of branch points. The branch density is calculated based on the number of branch points and the total root length. The formula is as follows: in, Branch density; Number of branch points; The total root length.

[0040] For each branch point, calculate the angle between the parent root direction and the child root direction, and take the average of all angles as the average branch angle. The calculation formula is as follows: in, The average branch angle; The total number of included angles; For the first Angle between the mother root and the daughter root.

[0041] Starting from the root node, use breadth-first search to traverse the skeleton graph, count the number of branches along each path from the root to the root tip, and use the maximum value as the topology depth. .

[0042] Assign values ​​to the root segments according to the branching order rules, and calculate the average branching order of all root tips as the average branching order. The calculation formula is as follows: in, The average branch order; Root number; For the first The branching order corresponding to each root tip.

[0043] Simultaneously, the lengths of the skeleton edges are grouped and summed according to the branch order to obtain the first-order root length, second-order root length, and third-order and above root length. The calculation formula is as follows: in, , and These are the first-order root length, the second-order root length, and the third-order and above root length, respectively. , and These are sets of skeleton edges of order one, two, and three or higher, respectively.

[0044] S104. Along each skeleton point in the skeleton diagram, perform circular cross-section fitting on the surrounding point cloud on a cross-section perpendicular to the local direction of the skeleton point, estimate the local diameter, and extract the diameter characteristics.

[0045] See Figure 6 and Figure 7 In this embodiment, a local skeleton direction vector is determined for each skeleton point in the skeleton diagram. Point cloud points within a cross-sectional plane perpendicular to this direction vector are selected and projected onto the cross-sectional plane. A circle is fitted using the least squares method, and the diameter of the fitted circle is the local root diameter at that skeleton point. When fewer than 10 projection points are involved in the fitting or the circle fitting residual exceeds 1 mm, the local root diameter estimate is marked as unreliable, and interpolation correction is performed using reliable diameter estimates from adjacent skeleton points. Furthermore, the average root diameter and the proportion of fine roots are calculated based on the local root diameter estimation results.

[0046] The average root diameter is obtained by weighting the local root diameters using the lengths of each skeleton side as weights. The calculation formula is as follows: in, The average root diameter; For the skeleton edge Length; For the skeleton edge The average diameter of the two skeleton points.

[0047] The proportion of fine roots is determined by using a diameter less than 2mm as the criterion. The ratio of the total length of fine root segments to the total root length is calculated using the following formula: in, For the proportion of fine roots; The total length of the fine root segments with a diameter of less than 2 mm; The total root length.

[0048] S105, based on size traits, spatial range traits, distribution traits, topological traits, and diameter traits, derived traits and root-crown coordination traits are extracted.

[0049] See Figure 8 In this embodiment, based on the aforementioned basic traits, derived indices and root-crown coordination indices are calculated, and four traits are further extracted: specific root length, exploration-utilization index, root-crown volume ratio, and functional balance index.

[0050] The specific root length is calculated based on the total root length and total root volume, using the following formula: in, It is longer than the root; The total root length; This represents the total root volume.

[0051] The exploration-utilization index is calculated based on the convex hull volume, total root volume, total root length, and the convex hull volume scaling term. The calculation formula is as follows: in, The exploration-utilization index comprehensively reflects the spatial exploration range of the root system and the local resource absorption efficiency. Let convex hull volume be the volume of the convex hull. This is the total root volume; The total root length; The feature length scale is obtained by converting the convex hull volume.

[0052] The root-to-crown volume ratio is calculated based on the total root volume, trunk cross-sectional area, and tree height. The formula is as follows: in, A higher value indicates that the underground portion is larger relative to the above-ground portion; The cross-sectional area of ​​the trunk corresponding to the trunk diameter. Plant height.

[0053] The functional balance index is calculated based on the total root surface area, trunk cross-sectional area, and number of lateral branches. The formula is as follows: in, It is a functional balance index used to characterize the functional balance relationship between absorptive root capacity and transpiration demand; Estimate the surface area of ​​the total root surface area or the surface area of ​​the fine roots; The cross-sectional area of ​​the tree trunk; This represents the number of lateral branches.

[0054] S106 integrates the extracted size traits, spatial range traits, distribution traits, topological traits, diameter traits, derived traits, and root-crown coordination traits to output a multidimensional standardized phenotypic vector.

[0055] In this embodiment, the original values ​​of the above 31 traits are integrated, and each trait is standardized by Z-score. The calculation formula is as follows: in, For the first Tree No. Standardized values ​​of individual traits; For the first Tree No. Raw values ​​of personality traits; and The first Output the mean and standard deviation of a trait across all trees. Output a 31-dimensional standardized phenotypic vector for each tree, along with the original values ​​and unit information.

[0056] Descriptive statistical analysis was performed on the extraction results of 118 seedlings to verify the effectiveness of the extraction method and the degree of phenotypic variation: Table 1 presents the descriptive statistical results of 31 root phenotypic traits. As shown in Table 1, all traits were successfully extracted from the 3D point clouds of 118 trees, achieving a 100% success rate. The coefficient of variation ranged from 14.9% to 39.2%, indicating sufficient variation in each trait within commercial seedlings and the potential to differentiate seedlings of varying quality. Branch complexity-related traits ( , The largest variation (CV>36%) was observed in shape, which describes the trait class ( The variation was the smallest (CV=14.9%).

[0057] In addition, the association between traits and transplant survival was verified. The 31 extracted traits were compared with the 365-day transplant survival results of 118 trees by univariate association analysis to verify the biological significance and predictive value of the traits.

[0058] Figure 9 The correlation network diagram in the image is used to illustrate the connections between strongly correlated traits. The thickness of the lines represents the correlation strength, and it can reveal the synergistic changes among size traits, topological traits, distribution traits, diameter traits, and derived traits. Figure 9 It is possible to identify trait combinations with high redundancy and key trait clusters closely related to transplant survival. Positive correlations represent changes in the same direction for two traits. For example, a larger total root volume corresponds to a larger surface area, which is a result of synergistic growth. Negative correlations represent changes in opposite directions for two traits, such as specific root length and average root diameter, reflecting the biological principle that fine roots are longer than root length and coarse roots are shorter than root length. Figure 10 The data is used to obtain the number of principal components and the variance explained by each principal component. The eigenvalues ​​of the first four principal components in the figure are all greater than 1. Figure 11 shows the distribution of different seedling samples in the first two principal component spaces, as well as the separation characteristics of the surviving and dead groups. The score vectors of each indicator in the figure show that all root morphology indicators, such as branch density, exploration efficiency, maximum root depth, number of root tips, total root length, fine root ratio, total surface area, and total volume, are oriented towards the surviving sample area represented by the green dots. This indicates that root branching configuration, spatial expansion ability, and the characteristics of fine roots and total root volume are the core indicators for distinguishing the survival status of transplanted seedlings. The confidence ellipses of the two groups of samples have only a small overlap, indicating that there is a significant difference between the surviving and dead groups in the two-dimensional principal component space composed of PC1 and PC2. The confidence ellipses of the surviving and dead groups have only a small overlap, indicating that there is a significant difference between the two types of seedlings in the PC1-PC2 dimension. The confidence intervals of the two groups of samples have only a small overlap, indicating that there is a significant difference between the surviving and dead groups in the PC1-PC2 dimension. Figure 12 further illustrates the differentiation characteristics of seedling samples in the third principal component dimension. The green dots represent surviving samples mainly distributed in the positive PC1 region, and also distributed in both the positive and negative PC3 axes. The index vectors of total volume, branch density, depth-to-spread ratio, and vertical skewness all point towards the surviving sample region, indicating that total root volume, branching development characteristics, and root vertical expansion characteristics are key traits that are conducive to seedling transplant survival. The confidence ellipses of the two groups of samples have only a small overlap. The dead group samples are concentrated on the negative PC1 side, while the surviving samples are biased towards the positive PC1 region. The two types of seedlings show obvious differentiation in the two-dimensional principal component space of PC1-PC3. Figure 13The correlation heatmap of variables in the figure is used to obtain the contribution of each phenotypic trait to each principal component. As can be seen from the figure, PC1 is significantly positively correlated with the root size index, representing the overall root size dimension; PC2 is strongly correlated with the vertical distribution index, representing the root depth distribution dimension; PC3 is correlated with the branching configuration index, representing the root topology dimension; and PC4 is correlated with the diameter distribution index, representing the root thickness dimension. Figure 13 shows the correlation heatmap of each root trait with the first four principal components, used to quantify the contribution of different phenotypes to each principal component. PC1 represents the overall root size dimension; PC2 reflects the vertical and horizontal spatial distribution characteristics of the root system; PC3 represents the root branching topology; and PC4 reflects the proportion of fine roots and the root thickness composition characteristics. Figures 10-13 Principal component analysis results can screen out core traits that differentiate root quality from transplant survival, providing a basis for phenotypic dimensionality reduction, seedling quality grading, and survival prediction modeling.

[0059] Univariate association analysis was performed on the 31 extracted traits and the 365-day transplant survival results of 118 trees to verify the biological significance and predictive value of the traits: 22 traits showed significant differences between the survival group (n=87) and the death group (n=31) (p<0.05), of which 18 traits remained significant after Benjamini-Hochberg correction.

[0060] The four traits with the largest effect sizes: (Cohen's d=1.18, AUC=0.782), (d=1.14, AUC=0.774) (d=1.04, AUC=0.758) and (d=1.02, AUC=0.752), both showing large effect sizes (|d|>0.8).

[0061] Six traits were not significantly associated with survival: , , , , and These are mainly shape-descriptive traits, indicating that the amount of root system is more important than its shape.

[0062] Principal component analysis showed that the first four principal components explained 72.3% of the total variance. PC1 (30.8% of the variance) was dominated by size-related traits. On PC1, the survival tree was significantly biased towards the positive end, and the death tree was biased towards the negative end, proving that the overall root size and complexity are the phenotypic dimensions most strongly associated with survival.

[0063] The above results fully verify that the 31 traits extracted in the embodiments of this application have clear biological significance and strong predictive value for transplant survival.

[0064] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0065] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0066] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds, characterized in that, include: Obtain a dense 3D root system point cloud obtained by a 3D reconstruction method and preprocess the 3D root system point cloud; Size characteristics, spatial range characteristics, and distribution characteristics were extracted from the preprocessed 3D root system point cloud. Simultaneously, skeletonization processing is performed on the preprocessed 3D root system point cloud to obtain a 1D root system skeleton map, and topological features are extracted from the skeleton map. Along each skeleton point of the skeleton diagram, a circular cross-section is fitted to the surrounding point cloud on a cross-section perpendicular to the local direction of the skeleton point to estimate the local diameter and extract the diameter characteristics; Based on the aforementioned size traits, spatial range traits, distribution traits, topological traits, and diameter traits, derived traits and root-crown coordination traits are extracted; The extracted size traits, spatial range traits, distribution traits, topological traits, diameter traits, derived traits, and root-crown coordination traits are integrated to output a multidimensional standardized phenotypic vector.

2. The method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds according to claim 1, characterized in that, The acquisition of the dense 3D root system point cloud obtained by the 3D reconstruction method and the preprocessing of the 3D root system point cloud include: Obtain the dense 3D root point cloud obtained by the 3D reconstruction method, calculate the average Euclidean distance from each point to multiple nearest neighbors, and calculate the mean and standard deviation of the average Euclidean distance of all points. Outliers whose average Euclidean distance exceeds a preset value are removed, and a three-dimensional coordinate system is established with the root position as the origin. The coordinates of all point clouds are translated and rotated in the three-dimensional coordinate system to complete coordinate normalization and output the preprocessed standard three-dimensional root system point cloud.

3. The method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds according to claim 1, characterized in that, Size characteristics are extracted from the preprocessed 3D root point cloud, including: An alpha shape surface mesh is constructed from the preprocessed 3D root system point cloud. The alpha shape parameters are determined based on the average nearest neighbor distance of each point in the 3D root system point cloud, using the following formula: in, This represents the average nearest neighbor distance of each point in the 3D root system point cloud; A closed surface mesh surrounding the root point cloud is generated based on the alpha shape parameter. The total root volume is obtained by summing the volumes of the directed tetrahedrons formed by the mesh patches and the reference points. The convex hull volume surrounding all root point cloud points is calculated using the Quickhull algorithm. The total root surface area is obtained by calculating the area of ​​each triangular facet of the alpha-shaped surface mesh. The formula is as follows: in, The total root surface area, For the first The area of ​​each grid patch This represents the total number of grid patches; The surface area to volume ratio is calculated based on the total root surface area and the total root volume, using the following formula: in, It is the surface area to volume ratio. This is the total root volume; Simultaneously, the root system compactness is calculated based on the total root volume and the convex hull volume, using the following formula: in, For root system compactness, Let be the convex hull volume.

4. The method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds according to claim 1, characterized in that, Spatial extent characteristics are extracted from the preprocessed 3D root system point cloud, including: The vertical coordinates of all 3D root system point cloud points and the horizontal distances of all 3D root system point cloud points to the central axis are calculated. The vertical distance from the root collar to the lowest root point is taken as the maximum root depth, and the maximum horizontal distance of all points to the central axis is taken as the maximum radial spread. The calculation formulas are as follows: in, For the maximum root depth, The Z-coordinate of the lowest root point; For maximum radial span, and The first The coordinate components of a point cloud point in the horizontal plane; The 90th percentile of the absolute value of the root point cloud depth is taken as the 90% biomass depth, and the 90th percentile of the horizontal radial distance of the point cloud points is taken as the 90% biomass radial range. The calculation formulas are as follows: in, For 90% biomass depth, The radial range is 90% biomass. Represents the 90th percentile. For the first Vertical coordinates of each point cloud point; The depth-to-spread ratio is calculated based on the maximum root depth and the maximum radial spread, using the following formula: in, For deeper development; The maximum root depth; This represents the maximum radial span.

5. The method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds according to claim 1, characterized in that, Distribution characteristics were extracted from the preprocessed 3D root point cloud, including: The vertical centroid is calculated based on the average vertical coordinates of all 3D root point cloud points. The calculation formula is as follows: in, It is the vertical centroid; This represents the total number of points in the point cloud. For the first Vertical coordinates of each point cloud point; The 3D root system point cloud is divided into multiple fan-shaped regions with preset angles on a horizontal plane. The standard deviation and mean of the point cloud point count ratio in each fan-shaped region are calculated. The horizontal uniformity is obtained based on the standard deviation and the mean. The calculation formula is as follows: in, For horizontal uniformity; and These are the standard deviation and mean of the percentage of points in the sector region, respectively. The vertical skewness is calculated based on the vertical coordinate distribution skewness of all three-dimensional root system point cloud points. The calculation formula is as follows: in, Vertical skewness; The average vertical coordinate of all point cloud points; An exponential decay model is fitted based on the density variation of 3D root point cloud points in the vertical direction with depth, and the decay coefficient is used as the root density gradient. The calculation formula is as follows: in, For depth Root point cloud density at the location; This represents the initial root point cloud density; The root density gradient; The exploration efficiency is calculated based on the convex hull volume and the hemispherical volume with the maximum 3D root distance as its radius. The calculation formula is as follows: in, To explore efficiency, Let convex hull volume be the volume. The maximum three-dimensional distance from a point in the point cloud to the origin of the root collar; Pi; and The first The coordinate components of a point cloud point in the horizontal plane For the first The vertical coordinates of each point cloud point.

6. The method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds according to claim 1, characterized in that, The process of performing skeletonization on the preprocessed 3D root point cloud to obtain a 1D root skeleton map, and extracting topological features from the skeleton map, includes: A Laplacian-based point cloud shrinkage algorithm is used to iteratively shrink the preprocessed 3D root system point cloud, so that the point cloud converges towards the central axis and maintains the topological structure. After the contraction is completed, the connection threshold is adaptively determined based on the local point cloud density and the contracted point set is converted into a graph structure. Redundant nodes are deleted by edge folding while retaining branch points and endpoints. Then, a Gaussian filter is used to smooth the skeleton node positions to obtain a one-dimensional root system skeleton graph. The total root length is calculated by summing the Euclidean lengths of all edges in the skeleton graph, using the following formula: in, The total root length; For the skeleton edge set; For the skeleton edge Length; The number of nodes with a preset degree in the skeleton graph is taken as the root tip number, and the number of nodes with a degree greater than or equal to the preset value is taken as the branch point number. The branch density is calculated based on the number of branch points and the total root length, using the following formula: in, Branch density; Number of branch points; The total root length; For each branch point, determine the direction vector of the parent root and the direction vector of the child root, calculate the angle between each child root and the parent root, and average all the angles to obtain the average branch angle. The calculation formula is as follows: in, The average branch angle; The total number of included angles; For the first Angle between the mother root and the daughter root; Starting from the root neck node, traverse the skeleton graph and count the maximum number of branch points along any path from the root neck to the root tip. Use the maximum number of branch points as the topological depth. The branching order of each root tip is determined according to the root segment hierarchy, and the average branching order of all root tips is calculated as the average branching order. The calculation formula is as follows: in, The average branch order; Root number; For the first The branch order corresponding to each root tip; Further grouping and summing the skeleton edge lengths according to the branch order yields the first-order root length, second-order root length, and third-order and above root lengths. The calculation formula is as follows: in, , and These are the first-order root length, the second-order root length, and the third-order and above root length, respectively. , and These are sets of skeleton edges of order one, two, and three and above, respectively. For the skeleton edge The length.

7. The method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds according to claim 1, characterized in that, Along each skeleton point of the skeleton map, a circular cross-section is fitted to the surrounding point cloud on a cross-section perpendicular to the local direction of the skeleton point to estimate the local diameter and extract the diameter characteristics, including: For each skeleton point in the skeleton diagram, determine the local skeleton direction vector; In a cross-sectional plane perpendicular to the local skeleton direction vector, nearby point cloud points are obtained. The nearby point cloud points are projected onto the cross-sectional plane and fitted into a circle using the least squares method. The diameter of the fitted circle is used as the local root diameter at the skeleton point. When the number of projection points is less than the preset number or the circle fitting residual exceeds the preset value, this local root diameter is marked as unreliable and corrected by interpolation of adjacent reliable estimates. Using the length of each skeleton edge as a weight, a weighted average is calculated for the local root diameter at each skeleton point to obtain the average root diameter. The calculation formula is as follows: in, The average root diameter; For the skeleton edge Length; For the skeleton edge The average diameter of the two end skeleton points; Obtain the total length of the fine root segments whose diameters are less than a preset value at both ends of the skeleton edge, and calculate the fine root ratio based on the total length of the fine root segments and the total root length. The calculation formula is as follows: in, For the proportion of fine roots; The total length of the fine root segments with a diameter of less than 2 mm; The total root length.

8. The method for extracting multidimensional phenotypic traits of fruit tree roots based on three-dimensional point clouds according to claim 1, characterized in that, Based on the aforementioned size traits, spatial range traits, distribution traits, topological traits, and diameter traits, derived traits and root-crown coordination traits are extracted, including: The root volume is calculated based on the total root volume in the dimensional trait and the root-to-total ratio in the topological trait. The calculation formula is as follows: in, It is longer than the root; The total root length; This is the total root volume; The exploration-utilization index is calculated based on the convex hull volume, total root volume, and convex hull volume scaling term in the dimensional properties, and the total root length in the topological properties. The calculation formula is as follows: in, For exploration and utilization index; Let convex hull volume be the volume of the convex hull. This is the total root volume; The total root length; The feature length scale is obtained by converting the convex hull volume; The root-to-crown volume ratio is calculated based on the total root volume, trunk cross-sectional area, and plant height, using the following formula: in, The root-to-crown volume ratio; The cross-sectional area of ​​the tree trunk; Plant height; The functional balance index is calculated based on the total root surface area, trunk cross-sectional area, and number of lateral branches in the size traits. The calculation formula is as follows: in, It is a functional balance index; Estimate the surface area of ​​the total root surface area or the surface area of ​​the fine roots; The cross-sectional area of ​​the tree trunk; This represents the number of lateral branches.

Images