betaFeedback

WorksComputer Science

Procedural Volumetric Modeling of Plant Branching Structures for Finite Element Analysis

Abstract

Precision agriculture, smart breeding, and agricultural robotics require accurate and automated plant modeling. These models provide high-fidelity three-dimensional (3D) representations of plant architecture. They provide the geometric foundation for simulations of water and nutrient transport, light interception, structural loading, and crop lodging. Unlike static plant modeling pipelines, procedural modeling frameworks not only generate accurate 3D plant geometries but also support the generative modeling of crop diversity and the dynamic modeling of plant growth. While terrestrial laser scanning, LiDAR, photogrammetry, and neural reconstruction-based approaches have made 3D plant reconstruction possible, the resulting data are typically in the form of point clouds, which cannot be directly utilized for high-fidelity simulations. We present an automated volumetric procedural modeling framework for plant branching structures that generates analysis-suitable hexahedral meshes from input skeletons or 3D point clouds. The input skeleton is first converted into a rooted graph representation that captures the plant branching topology. Each graph edge is then represented by a smooth centerline B-spline curve, around which a cylindrical tensor-product B-spline volume is constructed. At each junction, incident B-spline volume control lattices are joined using blending operations. The resulting volumetric parameterization is evaluated to generate a smooth and conforming hexahedral mesh of the whole plant. We demonstrate the framework on three diverse plant datasets, namely mung bean, tomato, and walnut trees, generating meshes with both uniform and spatially varying branch radii. The framework also supports dynamic mesh generation suitable for modeling plant growth by locally updating newly added branches without reconstructing the full plant geometry.

Introduction

Accurate geometric modeling of plant architecture is a fundamental requirement in computational agricultural applications, such as digital twins - , functional-structural plant models - , precision agriculture , [ ](#ref-10], resource optimization , , genotype-to-phenotype modeling - , bio-design - , and crop breeding - . High-fidelity geometric models generated from advanced sensor datasets provide quantitative analysis of plant architecture traits under different genetic, environmental, and management conditions, improving agricultural productivity and resilience , , , . By integrating real-time sensor data and environmental parameters, plant models can provide a realistic virtual representation of plants, enabling non-invasive, dynamic, and bi-directional decision-making. In precision agriculture, they allow for accurate morphological evaluations, canopy architecture design, efficient plant breeding, and resource usage , . In simulation-driven crop design, high-fidelity crop models can be used in inverse design applications to generate idealized ideotypes for high productivity and improved resilience - . The goal is not only to generate plant models for visualization but also to use them directly for physics-based simulations, such as nutrient and water uptake, structural mechanics - , light-use efficiency, and ecophysiological and ecosystem modeling - .

The advent of advanced sensor technologies for three-dimensional (3D) plant reconstruction, such as light detection and ranging (LiDAR), terrestrial laser scanning (TLS), multiview stereo (MVS), structured light scanning, neural radiance fields (NeRF), and Gaussian splatting, has enabled non-invasive, high-throughput plant phenotyping and monitoring , - . Whole plant architecture is typically represented as depth maps, point clouds, or surface meshes. These reconstructions provide high-resolution quantitative and qualitative visual representations of plant architecture, but they are not directly suitable for physics-based simulations [19, 48, 49, 50, 51, 52]. This identifies a critical research gap, where we have access to large datasets of high-fidelity 3D plant reconstructions acquired from sensors. However, converting them into analysis-ready volumetric meshes remains computationally intensive, requires manual effort, and is not scalable for complex geometries and challenging applications.

Procedural modeling approaches encode plant geometry in a rule-based manner, mathematically embedding the latent plant branching architecture and topology. Grammar-based approaches, such as L-systems, encode branching topology and plant growth using string-based rules [53, 54, 55, 56, 57, 58, 59, 60]. Functional-structural plant models (FSPMs) utilize algorithmic rules to construct plant geometry and employ physiological processes to guide the development of plant morphology [5, 7, 38, 61, 62, 63, 64, 65]. Non-uniform rational B-splines (NURBS) have been used in procedural modeling frameworks to accurately generate high-fidelity plant representations [19, 50, 66, 67, 68]. These approaches allow for biologically realistic plant modeling, but they are limited to implicit or surface representations, which require considerable computational effort to generate analysis-suitable volumetric meshes.

Generating analysis-suitable volumetric meshes for branching structures is an especially challenging and non-trivial problem, particularly near junctions, where multiple branch segments connect at different angles. A robust volumetric modeling framework should capture the correct plant branching topology for all possible configurations while generating a regular and conforming mesh that is free of intersections, gaps, and distortions [69, 70]. Conventional volumetric meshing and NURBS-based modeling workflows for generating branching geometries can create highly accurate and high-quality models [71, 72, 73, 74, 75]. However, these approaches are static; modifying the branching topology, such as adding or deleting branches or changing the branch radius, requires considerable effort to remesh or even rebuild the entire geometry from scratch. This highlights a key limitation: there are currently no procedural volumetric meshing frameworks for automatically generating branching geometries while remaining compatible with changing plant topology.

We present an automated volumetric procedural modeling framework for generating analysis-suitable hexahedral (hex) meshes of plant branching structures from 3D point clouds and their corresponding plant skeletons [76, 77, 78, 79, 80, 81, 82]. The input skeleton is first converted into a rooted directional graph representation that captures the plant topology. B-spline curves are then fitted to each branch, rendering a smooth centerline geometry. Rotation-minimizing frames along the centerline curves are used to construct cylindrical B-spline volumes for individual branches. At bifurcation and trifurcation junctions, incident branch B-spline volumes are joined via control-point blending operations, enforcing seamless transitions. The resulting volumetric parameterization of the complete plant is then evaluated to generate a conforming and smooth hex mesh suitable for finite element analysis (FEA).

The proposed framework utilizes a compact set of rule-based generation steps instead of manual mesh construction. It generates plant geometry through a sequence of steps, progressing from graph and curve representations to volumetric meshes. The framework not only enables automatic volumetric mesh generation from plant skeleton data but also allows local updates without reconstructing the whole plant geometry as the plant topology changes. The main contributions of our work are:

  1. A procedural volumetric modeling framework that automatically converts 3D plant skeleton data obtained from sensors into hex meshes suitable for FEA.

  2. A branch-wise B-spline volumetric parameterization strategy that first represents each branch centerline as a B-spline curve. Rotation-minimizing frames are constructed along the centerline curves, and tensor-product cylindrical B-spline volumes are generated for each branch.

  3. A junction-blending procedure is applied to bifurcations and trifurcations, joining incident branch volumes with seamless transitions and producing conforming volumetric meshes for the entire plant geometry.

  4. Demonstration of analysis-suitable volumetric mesh generation for diverse plant species of increasing geometric complexity, which includes mung bean, tomato, and walnut trees.

  5. Robust generation of plant hex meshes with spatially varying branch radii from unstructured 3D point clouds.

  6. Demonstration of dynamic mesh generation for plant growth modeling, where newly added branches are locally parameterized and blended without reconstructing the whole plant geometry.

In recent years, there has been significant advancement in the field of 3D plant reconstruction technologies, enabling the capture of detailed structural and morphological plant traits for accurate phenotyping. These include active depth-based imaging technologies such as LiDAR [83], TLS [84, 85, 86, 87], structured-light sensors [88, 89], and time-of-flight cameras [90], which use light or infrared signals to generate depth maps. Passive 3D reconstruction methods, such as stereo vision, MVS [91, 92, 93, 94], and structure-from-motion(SfM) photogrammetry [95], offer ways to reconstruct depth maps without emitting signals. However, these methods generate point clouds that suffer from noise, occlusions, and missing data [23]. More recently, deep learning-based approaches such as NeRF [96, 97] and 3D Gaussian splatting [98] have emerged as phenotyping tools that generate high-resolution point clouds [49, 99, 100, 101, 102]. These approaches generate continuous volumetric scene representations, allowing them to handle thin structures and occlusions more easily [42, 103]. After 3D point cloud acquisition, the phenotyping process proceeds with image analysis steps, such as point cloud registration [104], surface reconstruction [19, 67, 105, 106], semantic segmentation [107], and skeletonization [108, 109], to extract key morphological traits [46, 89], yielding segmented point clouds, voxels, surfaces, or skeletons that describe plant geometry. Current approaches focus primarily on surface reconstruction for plant phenotyping, using triangulations [110, 111], B-splines, or NURBS [112, 113] to represent plant architecture at the organ, plant, or canopy scales [50, 106]. As a result, analysis-suitable volumetric modeling remains largely absent from plant phenotyping.

Procedural plant modeling frameworks utilize algorithmic rules rather than manual geometry generation to model 3D plant architecture. Using a small set of geometric parameters and a rule-based generation approach, geometric models can be generated more efficiently for realistic plant growth and phenotyping applications [114, 115]. Lindenmayer Systems (L-systems) are rule-based procedural modeling approaches used to generate realistic 3D plant structures through a recursive algorithmic generation of L-strings that encode plant topology, geometry, and growth [116]. Deterministic, stochastic, and parametric L-systems have been shown to model realistic 1D plant branching architectures and growth processes for plant visualization and botany [54, 117, 118]. Subsequent research extended these approaches to include interactive and sketch-based tree modeling and procedural reconstruction from images or point clouds [57, 119, 120, 121]. These approaches showcase the flexibility of the rule-based frameworks for both forward and inverse data-driven reconstruction. FSPM approaches utilize physiological processes along with plant geometry to model plant growth and function [7, 63, 122]. The geometric modeling components, such as plant organs and internodes, are combined with mechanistic modules for photosynthesis, carbon allocation, and resource competition in a unified simulation framework [8, 38, 64, 87, 123]. Both the L-system and FSPMs provide geometric outputs for visualization, trait analysis, and growth simulation; however, they cannot be directly used for physics-based simulations [63, 122].

Recently, data-driven approaches have utilized machine learning to encode key morphological parameters, such as plant topology and geometry, into compact representations [102, 124, 125, 126]. By training on large imaging and point cloud datasets, supervised and deep learning approaches, such as convolutional neural networks (CNNs) and multi-layer perceptrons (MLPs), recover parameters for morphological traits and plant growth, enabling automated trait estimation and complementing rule-based or mechanistic models [126]. They also provide low-dimensional embeddings of complex plant topology and geometry for inverse modeling and generative modeling of diverse plant geometries under varying environmental and genetic conditions. Computer-aided design (CAD) modeling allows for a smooth parametric representation of complex geometry. Spline-based modeling allows for precise control, enabling both local and global high-fidelity shape representation [113]. B-spline curves or generalized cylinders have been used to represent plant branches, crowns, roots, and plant organs [127, 128, 129]. A generalized cylinder-based approach for modeling plant branches with implicit ramiform junctions was introduced in [130]. In [131], plant branches were represented as 3D B-spline curves coupled with a 1D B-spline radius; however, junction blending was identified as an open problem. Generalized B-spline crown envelopes have been used to model the complex polymorphism of Chinese fir to support natural pruning simulations [132]. B-splines have also been coupled with L-systems for the visual modeling of rice root growth [133, 134]. Recent methods utilize B-spline centerline approaches along with parallel transport frames for plant geometric modeling. Procedural frameworks for NURBS-based surface fitting to point cloud data have been developed, with clear gaps remaining in analysis-suitable volumetric modeling [19, 50, 65, 135, 136, 137].

Volumetric hex meshing and spline-based volumetric parameterization (V-rep) methods have been developed for FEA and isogeometric analysis (IGA) for branching structures [73, 74, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149]. Trivariate tensor-product B-spline volumes were fitted to genus-0 surfaces using harmonic parametrization in the cylindrical coordinate system in [150]. In [151], trimmed tri-variate B-spline volumes using geometric primitives were glued along shared boundaries with C0C^{0}-continuity. Patient-specific NURBS volumetric modeling has been carried out for vascular structures using skeleton-based sweeps with template-based junction continuity enforcement [72]. Polycube and block-decomposition approaches have been used to generate high-quality, analysis-suitable volumetric models for complex geometries [152, 153, 154, 155, 156, 157, 158, 159]. These robust, analysis-suitable modeling approaches, however, are not tailored for plant phenotyping workflows and typically assume static geometric construction, with limited applicability for modeling dynamic plant branching topologies.

Methods

We present a procedural framework that generates hex meshes for FEA from input plant skeleton point clouds. These point clouds approximate the medial axis of the plant, capturing the branching topology and centerline geometry. We denote the 3D point cloud by X={xl}l=1N\mathbf{X} = \{\mathbf{x}_{l}\}_{l=1}^{N}, with xl∈R3\mathbf{x}_{l} \in\mathbb{R}^{3}, where NN denotes the total number of points. The input point cloud is first converted into a skeleton graph that encodes the branching topology and connectivity. The skeleton graph is then transformed into an analysis-suitable hex mesh through five steps: (1) extraction of a branch graph from the skeleton graph by collapsing maximal chains of degree-two vertices; (2) fitting a B-spline centerline curve to the coordinate sequence along each edge of the branch graph; (3) construction of rotation-minimizing reference frames along each branch centerline curve by parallel transport; (4) construction of a hollow cylindrical tensor-product B-spline volume for each branch; and (5) blending of control points at each junction node. The B-spline-based volumetric parameterization provides a smooth, locally controllable representation of the branch geometry. The compact support of B-spline basis functions confines junction blending to a small number of control point layers in the volume. The output of the framework is a conforming hex mesh obtained by evaluating and assembling the blended B-spline branch volumes. The detailed procedure for each of these steps is described in the remainder of this section.

Skeleton and branch graph representation

Let GS=(VS,ES)\mathcal{G}_{S} = (\mathcal{V}_{S}, \mathcal{E}_{S}) denote the skeleton graph, with nodes νl∈VS\nu_{l} \in\mathcal{V}_{S} and edges em∈ESe_{m} \in\mathcal{E}_{S}. We assume that the input skeleton graph has been cleaned to form a connected graph that captures the correct branching topology; skeleton extraction and cleanup from the raw point cloud are outside the scope of this work. Each node νl\nu_{l} is associated with an embedded coordinate Xl\mathbf{X}_{l} and a non-negative scalar Rl≥0R_{l} \ge0 representing the outer radius of the cross-section at that coordinate. From GS\mathcal{G}_{S}, we then construct a branch graph GB=(VB,EB)\mathcal{G}_{B} = (\mathcal{V}_{B}, \mathcal{E}_{B}) that is treated as a rooted tree, with all edges oriented from the root toward the leaves. The node set VB⊆VS\mathcal{V}_{B} \subseteq\mathcal{V}_{S} is defined by retaining all nodes in VS\mathcal{V}_{S} whose degree is not equal to two:

VB={νl∈VS∣deg⁡GS(νl)≠2}.(1)\mathcal{V}_{B} = \{\nu_{l} \in\mathcal{V}_{S} \mid\operatorname{deg}_{\mathcal{G}_{S}}(\nu_{l}) \ne2\}. \tag*{(1)}

Here, the degree of the node, deg⁡GS(νl)\operatorname{deg}_{\mathcal{G}_{S}}(\nu_{l}), is the number of edges in GS\mathcal{G}_{S} incident to νl\nu_{l}. Nodes with deg⁡GS(νl)=1\operatorname{deg}_{\mathcal{G}_{S}}(\nu_{l}) = 1 are terminal nodes. Among the terminal nodes, we select the root node, denoted by νroot\nu_{\mathrm{root}}, with VBroot={νroot}\mathcal{V}_{B}^{\mathrm{root}} = \{\nu_{\mathrm{root}}\} associated with either the basal skeleton vertex of the point cloud, such as the vertex corresponding to the lowest z-coordinate, or a user-defined vertex. The remaining terminal nodes form the leaf set VBleaf\mathcal{V}_{B}^{\mathrm{leaf}}. Nodes with deg⁡GS(νl)≥3\operatorname{deg}_{\mathcal{G}_{S}}(\nu_{l}) \ge3 are the junction nodes. Every junction node in the junction set VBjunc\mathcal{V}_{B}^{\mathrm{junc}} has one incoming edge and at least two outgoing edges in GB\mathcal{G}_{B}. The node set VB\mathcal{V}_{B} can thus be partitioned into the root, junction, and leaf sets as

VB=VBroot∪VBjunc∪VBleaf.(2)\mathcal{V}_{B} = \mathcal{V}_{B}^{\mathrm{root}} \cup\mathcal{V}_{B}^{\mathrm{junc}} \cup\mathcal{V}_{B}^{\mathrm{leaf}}. \tag*{(2)}

Each directed edge from a proximal node νβ\nu_{\beta} to a distal node να\nu_{\alpha} is denoted by bα∈EBb_{\alpha} \in\mathcal{E}_{B}. We identify each edge bαb_{\alpha} by its distal node να\nu_{\alpha}. Each edge bαb_{\alpha} stores an ordered coordinate sequence Qα=(q1α,…,qMαα)\mathbf{Q}^{\alpha} = (\mathbf{q}_{1}^{\alpha}, \ldots, \mathbf{q}_{M^{\alpha}}^{\alpha}), corresponding to the unique maximal path in GS\mathcal{G}_{S} from νβ\nu_{\beta} to να\nu_{\alpha}, whose interior nodes all have degrees equal to two. The endpoints satisfy q1α=xβ\mathbf{q}_{1}^{\alpha} = \mathbf{x}_{\beta} and qMαα=xα\mathbf{q}_{M^{\alpha}}^{\alpha} = \mathbf{x}_{\alpha}, where MαM^{\alpha} denotes the number of skeleton points along each edge bαb_{\alpha}.

B-spline centerline curve construction

We represent the centerline geometry of each edge bα∈EBb_{\alpha} \in\mathcal{E}_{B} by a smooth centerline B-spline curve, Cα:[0,1]→R3\mathbf{C}^{\alpha} : [0,1] \to\mathbb{R}^{3}, defined as:

Cα(u)=∑k=1nαNk,p(u)Pkα,u∈[0,1],(3)\mathbf{C}^{\alpha}(u) = \sum_{k=1}^{n^{\alpha}} N_{k,p}(u)\mathbf{P}_{k}^{\alpha}, \qquad u \in[0,1], \tag*{(3)}

where nαn^{\alpha} denotes the number of control points, Nk,p(u)N_{k,p}(u) denotes the kthk^{\mathrm{th}} univariate basis function of degree pp defined on the clamped knot vector Uα:={u1,u2,…,unα+p+1}\mathcal{U}^{\alpha} := \{u_1,u_2,\ldots,u_{n^{\alpha}+p+1}\}, and Pkα∈R3\mathbf{P}^{\alpha}_{k} \in\mathbb{R}^{3} is the corresponding B-spline control point [160]. We present two strategies to construct the centerline curves for the entire branch graph. The global construction strategy (Section 3.2.1) extracts the complete branch graph and fits an individual B-spline curve to each edge bαb_{\alpha}. In the hierarchical construction strategy (Section 3.2.2), the branch graph is initialized with a primary path (main stem) in the skeleton graph from the root node to a terminal leaf node. A B-spline curve is first fitted to this primary path to represent the centerline geometry. The complete branch graph is then grown recursively by attaching child branches one at a time. At each new junction, a knot insertion operation is performed on the B-spline curve to split the parent branch, create space for attaching the child branch, and enforce junction blending. In both strategies, the B-spline curve Cα\mathbf{C}^{\alpha} is fitted to a target coordinate sequence Dα=(d1α,…,dMDαα)\mathbf{D}^{\alpha} = (\mathbf{d}^{\alpha}_{1},\ldots,\mathbf{d}^{\alpha}_{M^{\alpha}_{D}}) of edge bαb_{\alpha}, consisting of MDαM^{\alpha}_{D} ordered coordinates.

In the global construction strategy, we set Dα=Q~α\mathbf{D}^{\alpha} = \widetilde{\mathbf{Q}}^{\alpha}, where Q~α\widetilde{\mathbf{Q}}^{\alpha} denotes the coordinate sequence obtained by pruning Qα\mathbf{Q}^{\alpha} to remove coordinates near each junction node (Section 3.2.1); in the hierarchical construction strategy, we instead set Dα=Qα\mathbf{D}^{\alpha} = \mathbf{Q}^{\alpha}. For each coordinate dκα\mathbf{d}^{\alpha}_{\kappa}, κ=1,…,MDα\kappa= 1,\ldots,M^{\alpha}_{D}, we first compute the corresponding B-spline parameter uˉκ\bar{u}_{\kappa} by chord-length parameterization [160]:

uˉκ=∑j=2κ∥djα−dj−1α∥∑j=2MDα∥djα−dj−1α∥,uˉ1=0,uˉMDα=1.(4)\bar{u}_{\kappa} = \frac{\sum_{j=2}^{\kappa}\left\lVert\mathbf{d}^{\alpha}_{j}-\mathbf{d}^{\alpha}_{j-1}\right\rVert}{\sum_{j=2}^{M^{\alpha}_{D}}\left\lVert\mathbf{d}^{\alpha}_{j}-\mathbf{d}^{\alpha}_{j-1}\right\rVert}, \qquad\bar{u}_{1}=0,\qquad\bar{u}_{M^{\alpha}_{D}}=1. \tag*{(4)}

The fitting loss is defined as the mean squared distance between the B-spline curve Cα\mathbf{C}^{\alpha} and the target coordinate sequence Dα\mathbf{D}^{\alpha}:

Lα(Cα)=1MDα∑κ=1MDα∥Cα(uˉκ)−dκα∥2.(5)\mathcal{L}^{\alpha}(\mathbf{C}^{\alpha}) = \frac{1}{M^{\alpha}_{D}}\sum_{\kappa=1}^{M^{\alpha}_{D}}\left\lVert\mathbf{C}^{\alpha}(\bar{u}_{\kappa})-\mathbf{d}^{\alpha}_{\kappa}\right\rVert^{2}. \tag*{(5)}

In both construction strategies, the B-spline control point positions, Pα\mathbf{P}^{\alpha}, are optimized to minimize the loss function Lα\mathcal{L}^{\alpha} using Adam optimization [161]. Figure 1 summarizes the complete pipeline from skeleton graph construction to B-spline centerline curve fitting.

Overview of the skeleton-to-centerline curve pipeline

Figure 1. Overview of the skeleton-to-centerline curve pipeline. (a) Skeleton graph of a mung bean plant. (b) Branch decomposition of the skeleton graph into a directed branch graph where each node is classified as a leaf node (green), junction node (red) or root node (orange) and each edge represents an individual branch. (c) Edges in the branch graph colored according to their assigned degrees of freedom (DOF), corresponding to the number of control points used for each B-spline centerline curve. (d) Fitted B-spline centerline curves for each branch along with their corresponding control points.

Global construction strategy

In the global construction strategy, the entire branch graph GB\mathcal{G}_{B} is extracted from GS\mathcal{G}_{S}. Before B-spline curve fitting, the ordered coordinate sequence Qα\mathbf{Q}^{\alpha} is pruned near each junction node to leave room for subsequent junction blending between adjacent B-spline volumes (Section 3.4). At each junction node νδ∈VBjunc\nu_\delta\in\mathcal{V}_{B}^{\mathrm{junc}}, we define a pruning sphere of radius RδpruneR_{\delta}^{\mathrm{prune}} centered at its embedded coordinate xδ\mathbf{x}_\delta. RδpruneR_{\delta}^{\mathrm{prune}} is set as the largest radius of the cross-section of all the branches incident at νδ\nu_\delta. Any coordinate of Qα\mathbf{Q}^{\alpha} that lies inside this sphere is removed, and the surviving ordered coordinate sequence forms the target coordinate set Q~α\widetilde{\mathbf{Q}}^{\alpha} used for B-spline curve fitting. The detailed steps for the pruning procedure are outlined in Algorithm 1.

Algorithm 1. Junction pruning procedure.

  1. Input: branch graph GB=(VB,EB)\mathcal{G}_{B} = (\mathcal{V}_{B}, \mathcal{E}_{B}), ordered coordinate sequence {Qα}bα∈EB\{\mathbf{Q}^{\alpha}\}_{b_{\alpha}\in\mathcal{E}_{B}}.

  2. For each junction node νδ∈VBjunc\nu_{\delta} \in\mathcal{V}_{B}^{\mathrm{junc}}

    • Set the radius RδpruneR_{\delta}^{\mathrm{prune}} of the pruning sphere centered at the embedded coordinate xδ\mathbf{x}_{\delta}.

    • For each edge bαb_{\alpha} incident to νδ\nu_{\delta}, remove from Qα\mathbf{Q}^{\alpha} every coordinate qα\mathbf{q}^{\alpha} with ∥qα−xδ∥≤Rδprune\lVert\mathbf{q}^{\alpha} - \mathbf{x}_{\delta} \rVert\leq R_{\delta}^{\mathrm{prune}}.

  3. Output: the surviving ordered coordinate sequence, denoted by {Q~α}bα∈EB\{\widetilde{\mathbf{Q}}^{\alpha}\}_{b_{\alpha}\in\mathcal{E}_{B}}.

For each edge bαb_{\alpha}, the number of control points for each centerline B-spline curve is assigned based on its branch length, with longer branches assigned more control points. The parameters of the B-spline curve are calculated for the target sequence Q~α\widetilde{\mathbf{Q}}^{\alpha} using (4). Finally, for each edge bαb_{\alpha}, the B-spline curve Cα\mathbf{C}^{\alpha} is fitted to Q~α\widetilde{\mathbf{Q}}^{\alpha} by minimizing the loss Lα\mathcal{L}^{\alpha} with respect to Pα\mathbf{P}^{\alpha} ((5)). In this construction strategy, an independent B-spline curve is fitted to each branch, so adjacent branch curves meeting at the junction do not share a common tangent direction. Tangential continuity is therefore not guaranteed and is enforced separately during junction blending between the branch volumes (Section 3.4). Algorithm 2 outlines the steps carried out in the global construction strategy.

Algorithm 2. Global construction strategy.

  1. Extract the complete branch graph GB=(VB,EB)\mathcal{G}_{B} = (\mathcal{V}_{B}, \mathcal{E}_{B}) from the skeleton graph GS\mathcal{G}_{S}.

  2. Compute the branch lengths {Lα}bα∈EB\{L^{\alpha}\}_{b_{\alpha}\in\mathcal{E}_{B}} and partition them into NbinN_{\mathrm{bin}} equal-count (quantile) bins.

  3. For each edge bα∈EBb_{\alpha} \in\mathcal{E}_{B}:

    • Assign nαn^{\alpha} according to the length-bin index of LαL^{\alpha}, with nmin⁡≤nα≤nmax⁡n^{\min} \leq n^{\alpha} \leq n^{\max}. Here, nmin⁡n^{\min} and nmax⁡n^{\max} refer to the minimum and maximum number of control points allowed for any branch, respectively.

    • Apply junction pruning to the ordered coordinate sequence Qα\mathbf{Q}^{\alpha} to obtain Q~α\widetilde{\mathbf{Q}}^{\alpha} (Algorithm 1).

    • Assign {uˉκ}\{\bar{u}_{\kappa}\} to the pruned coordinate sequence Q~α\widetilde{\mathbf{Q}}^{\alpha} using (4).

    • Perform B-spline curve fitting by optimizing the control point positions Pα\mathbf{P}^{\alpha} to minimize (5) using Adam optimization.

Hierarchical construction strategy

In the hierarchical construction strategy, we construct the complete branch graph recursively, starting from a primary path and progressively attaching child branches according to the proximal-to-distal branching order. A primary path, referred to as the main stem, is identified as the path from the root node to a selected terminal leaf node in GS\mathcal{G}_{S}. The branch graph GB\mathcal{G}_{B} for the main stem is initialized with the root node νroot\nu_{\mathrm{root}}, the terminal leaf node να\nu_{\alpha}, and a single directed edge bα∈EBb_{\alpha} \in\mathcal{E}_{B} connecting them. By default, the terminal leaf node να\nu_{\alpha} is selected as the leaf node at the end of the longest root-to-leaf path; although a user-defined leaf node can also be selected. Let Qα\mathbf{Q}^{\alpha} denote the ordered coordinate sequence along the edge bαb_{\alpha}. The coordinates in Qα\mathbf{Q}^{\alpha} are used as target coordinates for B-spline curve fitting (Eqs. (4)–(5)). Unlike the global construction strategy, nαn^{\alpha} for each edge is not set according to branch length-based binning, as not all the branch edges are available at initialization. Instead, nαn^{\alpha} is set equal to the number of coordinates in Qα\mathbf{Q}^{\alpha}.

Next, the child branches in the first branching order are attached to the main stem bαb_{\alpha}, whose fitted centerline B-spline curve is given as Cα\mathbf{C}^{\alpha}. Let νδ\nu_{\delta} be a junction node along the main stem path at which a path in GS\mathcal{G}_{S} branches off, and let νγ\nu_{\gamma} be the distal node of that path; together, they define the child edge bγb_{\gamma}. The node νδ\nu_{\delta} corresponds to the parameter uˉδ\bar{u}_{\delta} on the fitted B-spline curve Cα\mathbf{C}^{\alpha}. At two adjacent parameters on either side of uˉδ\bar{u}_{\delta}, uˉδ−\bar{u}_{\delta}^{-} and uˉδ+\bar{u}_{\delta}^{+}, the knot multiplicity of the knot vector Uα\mathcal{U}^{\alpha} is raised to p+1p+1 by repeated knot insertions. As a result, the B-spline curve Cα\mathbf{C}^{\alpha} is split into three curves defined in the parametric intervals [0,uˉδ−][0,\bar{u}_{\delta}^{-}], [uˉδ−,uˉδ+][\bar{u}_{\delta}^{-},\bar{u}_{\delta}^{+}], and [uˉδ+,1][\bar{u}_{\delta}^{+},1]. Because knot insertion is a geometry preserving operation [160, 162], the three curves together exactly reproduce the geometry of Cα\mathbf{C}^{\alpha}. The B-spline curve defined in the parametric interval [uˉδ−,uˉδ+][\bar{u}_{\delta}^{-},\bar{u}_{\delta}^{+}] is deleted to provide sufficient room for subsequent junction blending. This step serves a role similar to the junction pruning procedure carried out in Section 3.2.1. At initialization, the edge set in the branch graph GB\mathcal{G}_{B} is given as EB={bα}\mathcal{E}_{B}=\{b_{\alpha}\}, and it is updated as follows. A junction node νδ\nu_{\delta} is added to the main stem bαb_{\alpha}. The edge bαb_{\alpha} is now defined between the nodes νδ\nu_{\delta} and να\nu_{\alpha}, and its centerline B-spline curve is defined on the parameter interval [uˉδ+,1][\bar{u}_{\delta}^{+},1]. A new edge, bδb_{\delta}, is defined between the nodes νroot\nu_{\mathrm{root}} and νδ\nu_{\delta}, with its corresponding B-spline curve defined on the parameter interval [0,uˉδ−][0,\bar{u}_{\delta}^{-}]. A child edge bγb_{\gamma} is then defined from νδ\nu_{\delta} to νγ\nu_{\gamma} in the updated branch graph, with EB={bδ,bα,bγ}\mathcal{E}_{B}=\{b_{\delta},b_{\alpha},b_{\gamma}\}.

Algorithm 3. Hierarchical construction strategy.

  1. Identify the main stem in the input skeleton graph GS\mathcal{G}_{S} as the path from the root node to a terminal leaf node. The terminal leaf node is selected either by a user-defined strategy or, by default, as the leaf node of the longest root-to-leaf path. Initialize the branch graph GB\mathcal{G}_{B} with a single edge bαb_{\alpha} along this path, with the ordered coordinate sequence along this edge defined as Qα\mathbf{Q}^{\alpha}.

  1. Evaluate the B-spline curve parameters corresponding to the coordinates in Qα\mathbf{Q}^{\alpha} using Eq. (4), and fit the centerline B-spline curve Cα\mathbf{C}^{\alpha} by optimizing the control points Pα\mathbf{P}^{\alpha} to minimize Lα\mathcal{L}^{\alpha} (Eq. (5)). The number of control points is set equal to the number of coordinates in Qα\mathbf{Q}^{\alpha}.

  1. Traverse the remaining edges of GS\mathcal{G}_{S} in a depth-first manner to generate the complete branch graph, moving from each parent edge to its child edges. For each parent edge bαb_{\alpha} defined between its proximal node νβ\nu_{\beta} and distal node να\nu_{\alpha}:

  • Sort the children of bαb_{\alpha} in proximal-to-distal order, from the root node to the terminal leaf nodes.

  • For each child edge bγb_{\gamma}, attach it to bαb_{\alpha} at the junction node νδ\nu_{\delta} through the following steps: (i) Determine the parameter uˉδ\bar{u}_{\delta} in Uα\mathcal{U}^{\alpha} and the two adjacent parameters on either side of it, uˉδ−\bar{u}_{\delta}^{-} and uˉδ+\bar{u}_{\delta}^{+}, defined on the parent B-spline curve Cα\mathbf{C}^{\alpha}. (ii) Insert knots at uˉδ−\bar{u}_{\delta}^{-} and uˉδ+\bar{u}_{\delta}^{+} by repeated single knot insertion until each knot reaches multiplicity p+1p+1, creating three B-spline curves on the intervals [0,uˉδ−][0,\bar{u}_{\delta}^{-}], [uˉδ−,uˉδ+][\bar{u}_{\delta}^{-},\bar{u}_{\delta}^{+}], and [uˉδ+,1][\bar{u}_{\delta}^{+},1]. (iii) Remove the B-spline curve between [uˉδ−,uˉδ+][\bar{u}_{\delta}^{-},\bar{u}_{\delta}^{+}] to leave room for junction blending. (iv) bαb_{\alpha} is now defined between νδ\nu_{\delta} and να\nu_{\alpha}. A new branch edge, bδb_{\delta}, is defined between νβ\nu_{\beta} and νδ\nu_{\delta}. (v) The child edge bγb_{\gamma} is created between the nodes νδ\nu_{\delta} and νγ\nu_{\gamma}, and B-spline curve fitting is performed to evaluate the centerline curve Cγ\mathbf{C}^{\gamma}. (vi) The child edge bγb_{\gamma} is now defined as the parent edge for its children.

The remaining branches are added similarly to the branch graph by carrying out the knot insertion and child branch attachment steps. Each edge with a newly fitted B-spline curve subsequently serves as the parent edge for its descendants, and the sub-paths at the junction nodes along this path are attached as its child edges. The path traversal proceeds in a depth-first manner to generate the complete branch graph. Thus, when a child edge is added, all the child edges of that child are processed first before moving to the next child of the same parent. Each child branch attachment operation adds a junction node, νδ\nu_{\delta}, creates a new edge, bδb_{\delta}, updates the edge, bαb_{\alpha}, and finally adds a child edge, bγb_{\gamma}. Thus, the branch graph GB\mathcal{G}_{B} grows recursively from a single main stem edge into the complete branch graph, terminating once every point in the skeleton point cloud is fitted to the centerline B-spline curve. The advantage of the hierarchical construction strategy is that the knot insertion operation subdivides a B-spline curve without changing its geometry. Thus, regardless of how many times the parent B-spline curve has been subdivided to attach child branches, it still maintains its geometry and intrinsic smoothness while being represented as multiple B-spline curve segments compared to the global construction strategy. The hierarchical construction strategy is described in Algorithm 3.

B-spline volumetric parameterization for each branch

Each branch edge bα∈EB\mathbf{b}_{\alpha} \in\mathcal{E}_{B} of the branch graph GB\mathcal{G}_{B} is represented by a cylindrical trivariate tensor-product B-spline volume Vα:[0,1]3→R3\mathbf{V}^{\alpha} : [0,1]^{3} \to\mathbb{R}^{3} as

Vα(r,θ,u)=∑i=1ℓ∑j=1m∑k=1nαNi,s(r)Nj,q(θ)Nk,p(u)Pi,j,kα,(r,θ,u)∈[0,1]3,(6)\mathbf{V}^{\alpha}(r,\theta,u) = \sum_{i=1}^{\ell} \sum_{j=1}^{m} \sum_{k=1}^{n^{\alpha}} N_{i,s}(r)N_{j,q}(\theta)N_{k,p}(u)\mathbf{P}^{\alpha}_{i,j,k}, \qquad(r,\theta,u) \in[0,1]^{3}, \tag*{(6)}

where rr, θ\theta, and uu are the parametric coordinates, and ss, qq, and pp are the degrees of the B-spline basis functions in the radial, circumferential, and axial directions, respectively. The B-spline control lattice Pα\mathbf{P}^{\alpha} is of the size (ℓ×m×nα)(\ell\times m \times n^{\alpha}). ℓ\ell and mm are kept constant for all branches to facilitate better junction blending between the incident volumes. Each basis function Ni,s(r)N_{i,s}(r), Nj,q(θ)N_{j,q}(\theta), and Nk,p(u)N_{k,p}(u) is defined on the clamped knot vector R\mathcal{R}, the unclamped knot vector Θ\Theta, and the clamped knot vector Uα\mathcal{U}^{\alpha}, respectively. The circumferential knot vector, Θ\Theta, is unclamped in order to define periodic B-splines along the circumferential direction. To construct periodic B-splines, the last qq columns are overlapped with the first qq columns of the control lattice. Thus,

Pi,m−q+w,kα=Pi,w,kα,w=1,2,…,q,(7)\mathbf{P}^{\alpha}_{i,m-q+w,k} = \mathbf{P}^{\alpha}_{i,w,k}, \qquad w = 1,2,\ldots,q, \tag*{(7)}

ensures Cq−1C^{q-1} surface continuity in the circumferential direction [136]. The tensor-product B-spline volume for each edge Vα\mathbf{V}^{\alpha} is thus constructed by sweeping the (r,θ)(r,\theta) parameterized cross-sections along the centerline curve Cα\mathbf{C}^{\alpha} defined in Section 3.2 (Figure 2(a)).

B-spline volumetric parameterization of a single branch

Figure 2. B-spline volumetric parameterization of a single branch. (a) Fitted centerline B-spline curve. (b) Rotation-minimizing frames defined along the centerline curve, with tangent (red), normal (green), and binormal (blue) vectors defined at each Greville abscissa on the curve. (c) Circular cross-sections swept along the centerline curve. (d) Evaluated tensor-product B-spline volume with its corresponding control-point lattice. (e) Schematic showing the cylindrical parameterization of the B-spline volume and the radial, circumferential, and axial parametric directions (r,θ,u)(r,\theta,u).

A local orthonormal frame {Tk,Nk,Bk}k=1nα\{\mathbf{T}_{k}, \mathbf{N}_{k}, \mathbf{B}_{k}\}_{k=1}^{n^{\alpha}} is constructed along the centerline curve Cα\mathbf{C}^{\alpha} at each Greville abscissa {ξk}k=1nα\{\xi_{k}\}_{k=1}^{n^{\alpha}} (Figure 2(b)), defined on Uα\mathcal{U}^{\alpha} [160]. The unit tangent vector Tk\mathbf{T}_{k} is tangent to the centerline curve at Cα(ξk)\mathbf{C}^{\alpha}(\xi_{k}), whereas the normal and binormal vectors, Nk\mathbf{N}_{k} and Bk\mathbf{B}_{k}, span a cross-sectional plane that is perpendicular to Tk\mathbf{T}_{k}. The tangent vector is evaluated as Tk=C˙α(ξk)/∥C˙α(ξk)∥\mathbf{T}_{k} = \dot{\mathbf{C}}^{\alpha}(\xi_{k})/\lVert\dot{\mathbf{C}}^{\alpha}(\xi_{k})\rVert, where C˙α(ξk)=dCα(ξk)/du\dot{\mathbf{C}}^{\alpha}(\xi_{k}) = d\mathbf{C}^{\alpha}(\xi_{k})/du. The unit binormal vector is evaluated as the cross-product of the tangent and normal vectors, Bk=Tk×Nk\mathbf{B}_{k} = \mathbf{T}_{k} \times\mathbf{N}_{k}. The first unit normal vector along the centerline curve N1\mathbf{N}_{1} is obtained by projecting a fixed global up-vector, (0,0,1)(0,0,1), onto the plane perpendicular to T1\mathbf{T}_{1}. If T1\mathbf{T}_{1} is nearly parallel to the default up-vector, an alternative up-vector is used. The remaining normals along the centerline curve are propagated using the double-reflection parallel transport method [163], producing a rotation-minimizing frame by minimizing twist about the tangential direction.

The control points of Vα\mathbf{V}^{\alpha} are evaluated by sweeping circular cross-sections along the centerline curve Cα\mathbf{C}^{\alpha} at the Greville abscissae {ξk}k=1nα\{\xi_k\}_{k=1}^{n^\alpha}. At each ξk\xi_k, we define an angular location, ϕj\phi_j, evaluated via uniform sampling over the circumference of the cross section. Similarly, at each ξk\xi_k, a radial location is defined as ρi\rho_i, evaluated through uniform sampling between the inner and outer radii of the cross section. This is given as

ϕj=2π(j−1)m−q,j=1,…,m−q,(8)\phi_j = \frac{2\pi(j-1)}{m-q}, \quad j=1,\ldots,m-q, \tag*{(8)}
ρi=hk[η+(1−η)i−1ℓ−1],i=1,…,ℓ,(9)\rho_i = h_k\left[\eta+(1-\eta)\frac{i-1}{\ell-1}\right], \quad i=1,\ldots,\ell, \tag*{(9)}

where hkh_k denotes the outer branch radius at ξk\xi_k, obtained by interpolating the outer radius RlR_l at each skeleton node νl\nu_l along the centerline curve Cα\mathbf{C}^{\alpha}. The ratio of the inner branch radius to the outer branch radius, η∈(0,1)\eta\in(0,1), is fixed for all branches. Each control point in the control lattice Pα\mathbf{P}^{\alpha} is thus given as:

Pi,j,kα=Cα(ξk)+ρi(cos⁡(ϕj)Nk+sin⁡(ϕj)Bk),(10)\mathbf{P}^{\alpha}_{i,j,k}=\mathbf{C}^{\alpha}(\xi_k)+\rho_i\left(\cos(\phi_j)\mathbf{N}_k+\sin(\phi_j)\mathbf{B}_k\right), \tag*{(10)}

i=1,…,ℓi=1,\ldots,\ell, j=1,…,m−qj=1,\ldots,m-q, k=1,…,nαk=1,\ldots,n^\alpha.

The qq columns in the control-point lattice with circumferential index j=m−q+1,…,mj=m-q+1,\ldots,m are evaluated according to the periodicity condition in (7). By sweeping the cross-sections along the centerline and assembling the 3D control point lattice, we can then evaluate the corresponding tensor-product B-spline volume that produces the hollow cylindrical branch geometry shown in Figure 2(c-d).

Junction blending for continuity enforcement

As described in Section 3.3, each plant branch is represented by an independent B-spline volume constructed around its own centerline curve. Consequently, at a junction node νδ∈Vjunc\nu_\delta\in\mathcal{V}^{\mathrm{junc}}, the junction-facing boundary cross-sections of the incident branch volumes are generally not coincident. The incident branch volumes must therefore be blended at the junction so that the assembled volumetric mesh contains no gaps and maintains smooth transitions across the junctions. In this section, we describe how these continuity conditions are enforced on the branch control lattices before mesh evaluation. We first describe the G1G^1-continuity requirements for a pair of B-spline curves in Section 3.4.1 and subsequently describe the extension for a pair of B-spline volumes in Section 3.4.2. We then introduce a cross-section splitting technique for the three incident branch control lattices meeting at bifurcation junctions in Section 3.4.3, and finally extend this approach to trifurcation junctions in Section 3.4.4. The continuity conditions enforced on the control lattices at each junction are reflected in the evaluated hex mesh. While this approach can be generalized to higher-valence junctions, extensions beyond trifurcations are outside the scope of the present work.

Junction continuity between two B-spline curves

Consider two clamped B-spline curves, Cα\mathbf{C}^{\alpha} and Cβ\mathbf{C}^{\beta}, shown in Figure 3(a), with control point sequences P1α,…,Pnαα\mathbf{P}^{\alpha}_1,\ldots,\mathbf{P}^{\alpha}_{n^\alpha} and P1β,…,Pnββ\mathbf{P}^{\beta}_1,\ldots,\mathbf{P}^{\beta}_{n^\beta}, respectively. Here, nαn^\alpha and nβn^\beta denote the numbers of control points defining Cα\mathbf{C}^{\alpha} and Cβ\mathbf{C}^{\beta}, respectively. G0G^0-continuity is enforced by setting the terminal control point of Cα\mathbf{C}^{\alpha} equal to the first control point of Cβ\mathbf{C}^{\beta}. Thus, setting Pnαα=P1β\mathbf{P}^{\alpha}_{n^\alpha}=\mathbf{P}^{\beta}_1 closes the gap between the curves and enforces G0G^0-continuity, as shown in Figure 3(b). To achieve G1G^1-continuity, the tangent directions of the two curves must also be aligned at the junction. This can be imposed by making the shared control point and its immediate neighboring control points collinear [72, 160]. We impose both the continuity conditions simultaneously by placing the shared control point at the midpoint of the two adjacent control points:

$G^1$-continuity enforcement between two cubic B-spline curves

Figure 3. G1G^1-continuity enforcement between two cubic B-spline curves. (a) Before enforcement, the two curves are separated by a gap. (b) After G0G^0 enforcement, the last control point of the first curve and the first control point of the second curve are made coincident, resulting in a shared control point at the junction. (c) After G1G^1 enforcement, the shared control point is placed at P5α=P1β=12(P4α+P2β)\mathbf{P}^{\alpha}_{5}=\mathbf{P}^{\beta}_{1}=\frac{1}{2}(\mathbf{P}^{\alpha}_{4}+\mathbf{P}^{\beta}_{2}), making the three unique neighboring control points collinear: P4α\mathbf{P}^{\alpha}_{4}, P5α=P1β\mathbf{P}^{\alpha}_{5}=\mathbf{P}^{\beta}_{1}, and P2β\mathbf{P}^{\beta}_{2}.

Pnαα=P1β=12(Pnα−1α+P2β).(11)\mathbf{P}^{\alpha}_{n^\alpha}=\mathbf{P}^{\beta}_1=\frac{1}{2}\left(\mathbf{P}^{\alpha}_{n^\alpha-1}+\mathbf{P}^{\beta}_2\right). \tag*{(11)}

This ensures that the control points Pnα−1α\mathbf{P}^{\alpha}_{n^\alpha-1}, Pnαα=P1β\mathbf{P}^{\alpha}_{n^\alpha}=\mathbf{P}^{\beta}_1, and P2β\mathbf{P}^{\beta}_2 are collinear (Figure 3(c)), thereby providing a common tangent direction at the junction.

Junction continuity between two B-spline volumes

The G1G^1-continuity enforcement for curves in (11) extends naturally to a pair of adjacent trivariate tensor-product B-spline volumes. Consider two B-spline volumes, Vα\mathbf{V}^{\alpha} and Vβ\mathbf{V}^{\beta}, defined by the 3D control lattices Pα\mathbf{P}^{\alpha} and Pβ\mathbf{P}^{\beta}, respectively. For each fixed radial index, ii, and circumferential index, jj, the control points along the axial direction form an axial control column denoted by Pi,j,:α\mathbf{P}^{\alpha}_{i,j,:} or Pi,j,:β\mathbf{P}^{\beta}_{i,j,:}, as shown in Figure 4. The two B-spline volumes are joined by applying (11) column-wise to the axial control points for every fixed pair (i,j)(i,j). Specifically, G0G^0-continuity is enforced by making the terminal control point of each axial column in the control lattice of Vα\mathbf{V}^{\alpha} equal to the first control point of each corresponding axial column in the control lattice of Vβ\mathbf{V}^{\beta}. Thus, Pi,j,nαα=Pi,j,1β\mathbf{P}^{\alpha}_{i,j,n_{\alpha}}=\mathbf{P}^{\beta}_{i,j,1} for all (i,j)(i,j). This creates a shared boundary control cross-section between the two B-spline volumes, as shown in Figure 4(b). To enforce G1G^1-continuity, each control point on the shared boundary control cross-section is placed at the midpoint of its two neighboring axial control points, Pi,j,nα−1α\mathbf{P}^{\alpha}_{i,j,n_{\alpha}-1} and Pi,j,2β\mathbf{P}^{\beta}_{i,j,2}:

G1-continuity enforcement between two cubic trivariate tensor-product B-spline volumes

Figure 4. G1G^{1}-continuity enforcement between two cubic trivariate tensor-product B-spline volumes. The control lattices of the two B-spline volumes are denoted by Pα\mathbf{P}^{\alpha} and Pβ\mathbf{P}^{\beta}. A control point in Pα\mathbf{P}^{\alpha} is denoted by Pi,j,kα\mathbf{P}^{\alpha}_{i,j,k}, where ii, jj, and kk are the indices in the radial, circumferential, and axial directions, respectively. (a) Before enforcement, the two volumes are separated by a gap. (b) After G0G^{0} enforcement, the last control point in each axial control column of the first B-spline volume and the first control point in each axial control column of the second B-spline volume are made coincident, creating a shared boundary control cross-section, shown in green. (c) After G1G^{1} enforcement, each control point on the shared boundary control cross-section is placed according to Pi,j,5α=Pi,j,1β=12(Pi,j,4α+Pi,j,2β)\mathbf{P}^{\alpha}_{i,j,5}=\mathbf{P}^{\beta}_{i,j,1}=\frac{1}{2}\left(\mathbf{P}^{\alpha}_{i,j,4}+\mathbf{P}^{\beta}_{i,j,2}\right) for all (i,j)(i,j), so that the corresponding control points across the shared boundary cross section in the axial direction are collinear.

Pi,j,nαα=Pi,j,1β=12(Pi,j,nα−1α+Pi,j,2β),∀ (i,j).(12)\mathbf{P}^{\alpha}_{i,j,n_{\alpha}}=\mathbf{P}^{\beta}_{i,j,1}=\frac{1}{2}\left(\mathbf{P}^{\alpha}_{i,j,n_{\alpha}-1}+\mathbf{P}^{\beta}_{i,j,2}\right), \qquad\forall\,(i,j). \tag*{(12)}

Thus, three unique neighboring control points in each axial column become collinear, thereby providing a common axial tangent direction across the shared boundary between the two B-spline volumes.

Junction continuity at bifurcations

The pairwise continuity enforcement between two B-spline volumes described in Section 3.4.2 cannot be applied directly to bifurcations because three B-spline volumes meet at the junction. At a bifurcation junction node νδ∈VBjunc\nu_{\delta}\in\mathcal{V}^{\mathrm{junc}}_{B} in GB\mathcal{G}_{B}, one parent edge bδb_{\delta} meets two child edges: bγ1b_{\gamma_{1}} and bγ2b_{\gamma_{2}}. Let a∈{δ,γ1,γ2}a\in\{\delta,\gamma_{1},\gamma_{2}\} index the three branch edges incident at νδ\nu_{\delta}. For each incident edge bab_{a}, let Ca\mathbf{C}^{a} denote its centerline curve, Va\mathbf{V}^{a} denote its tensor-product B-spline volume, and Pa\mathbf{P}^{a} denote its corresponding control lattice. A control point in this lattice is denoted by Pi,j,ka\mathbf{P}^{a}_{i,j,k}, where ii, jj, and kk are the indices along the radial, circumferential, and axial directions, respectively. To enforce continuity at the bifurcation, the junction-facing control cross-section of each incident B-spline volume is first partitioned into two circumferential halves, as shown in Figure 5. A splitting line is defined on each junction-facing control cross-section to partition the axial control columns into two circumferential halves. Corresponding axial control columns from paired circumferential halves are then blended using (12). Because all incident control lattices have the same radial and circumferential resolution, corresponding circumferential halves can be paired across the three control lattices.

Splitting line construction at a bifurcation junction

Figure 5. Splitting line construction at a bifurcation junction. (a) The junction plane, shown in gray, is defined by the junction-facing endpoints of the parent and child centerline curves. The unit vector perpendicular to the junction plane n^\hat{\mathbf{n}} is shown. (b) On the junction-facing control-point cross-section of each incident branch, the radial line that is most closely aligned with the direction of n^\hat{\mathbf{n}} is selected as the splitting line (shown in magenta). (c) The splitting line is the radial diameter that partitions each control lattice cross-section into two halves. Each color indicates the pair of control-lattice axial columns associated with pairwise continuity enforcement between the branch edges (bδb_{\delta}–bγ1b_{\gamma_{1}} (blue), bδb_{\delta}–bγ2b_{\gamma_{2}} (red), bγ1b_{\gamma_{1}}–bγ2b_{\gamma_{2}} (green)).

We define a junction plane that passes through the endpoints of the three centerline curves near the junction, as shown in Figure 5(a). Let pδ=Cδ(1)\mathbf{p}^{\delta}=\mathbf{C}^{\delta}(1), pγ1=Cγ1(0)\mathbf{p}^{\gamma_{1}}=\mathbf{C}^{\gamma_{1}}(0), and pγ2=Cγ2(0)\mathbf{p}^{\gamma_{2}}=\mathbf{C}^{\gamma_{2}}(0) denote the junction-side endpoints of the parent and child curves, respectively. A unit normal vector to the junction plane, n^\hat{\mathbf{n}}, is given as

n^=(pγ1−pδ)×(pγ2−pδ)∥(pγ1−pδ)×(pγ2−pδ)∥(13)\hat{\mathbf{n}}=\frac{\left(\mathbf{p}^{\gamma_{1}}-\mathbf{p}^{\delta}\right)\times\left(\mathbf{p}^{\gamma_{2}}-\mathbf{p}^{\delta}\right)}{\left\lVert\left(\mathbf{p}^{\gamma_{1}}-\mathbf{p}^{\delta}\right)\times\left(\mathbf{p}^{\gamma_{2}}-\mathbf{p}^{\delta}\right)\right\rVert} \tag*{(13)}

This unit normal vector provides a common reference direction for selecting the splitting line on each junction-facing control-point cross-section. For each incident branch, a control cross-section at a given axial index is assigned a depth index dd based on its distance from the junction. Here, d=0d = 0 denotes the junction-facing cross-section on each control lattice. For each incident edge aa, the axial index of each control point on the cross-section at depth-dd is

kda={na−d,if a=δ1+d,if a∈{γ1,γ2},(14)k_{d}^{a} = \begin{cases} n^{a} - d, & \text{if } a = \delta\\ 1 + d, & \text{if } a \in\{\gamma_{1}, \gamma_{2}\}, \end{cases} \tag*{(14)}

where nan^{a} is the number of control points in the axial direction for the control lattice Pa\mathbf{P}^{a}. On the junction-facing control cross-section, with axial index k0ak_{0}^{a}, we define a radial line extending from P1,j,k0aa\mathbf{P}^{a}_{1,j,k_{0}^{a}} to Pℓ,j,k0aa\mathbf{P}^{a}_{\ell,j,k_{0}^{a}}. The splitting line is chosen as the radial diameter whose direction is most closely aligned with the normal vector n^\hat{\mathbf{n}}. The corresponding circumferential index jsplitaj_{\mathrm{split}}^{a} on the splitting line is thus given as:

jsplita=arg⁡max⁡j∈{1,…,m−q}∣(Pℓ,j,k0aa−P1,j,k0aa)∥Pℓ,j,k0aa−P1,j,k0aa∥⋅n^∣.(15)j_{\mathrm{split}}^{a} = \underset{j \in\{1,\ldots,m-q\}}{\arg\max}\left|\frac{\left(\mathbf{P}^{a}_{\ell,j,k_{0}^{a}}-\mathbf{P}^{a}_{1,j,k_{0}^{a}}\right)}{\left\|\mathbf{P}^{a}_{\ell,j,k_{0}^{a}}-\mathbf{P}^{a}_{1,j,k_{0}^{a}}\right\|}\cdot\hat{\mathbf{n}}\right|. \tag*{(15)}

The selected radial diameter partitions the control-lattice cross-section into two halves. Figure 5 shows the cross-sections partitioned by the splitting line on each control lattice cross-section, where matching colors indicate the corresponding pairs of axial columns used for pairwise continuity enforcement: bδ−bγ1b_{\delta}-b_{\gamma_{1}} (blue), bδ−bγ2b_{\delta}-b_{\gamma_{2}} (red), and bγ1−bγ2b_{\gamma_{1}}-b_{\gamma_{2}} (green). For each paired set of axial columns, the continuity rule in (12) is applied column-wise. The control points on the depth d=0d = 0 cross-section are placed at the midpoint of their corresponding depth d=1d = 1 neighboring control points, thereby enforcing G1G^{1}-continuity across that interface.

The splitting line itself is treated as a common three-way control-lattice interface among the three incident branches at the junction. For each radial index ii on the splitting line, we compute the centroid of the three adjacent depth d=1d = 1 control points as

P‾i=13∑a∈{δ,γ1,γ2}Pi,jsplita,k1aa.(16)\overline{\mathbf{P}}_{i} = \frac{1}{3}\sum_{a\in\{\delta,\gamma_{1},\gamma_{2}\}}\mathbf{P}^{a}_{i,j_{\mathrm{split}}^{a},k_{1}^{a}}. \tag*{(16)}

The control points on the splitting line of all three incident branches are then assigned to this centroid as

Pi,jsplita,k0aa=P‾i,∀ a∈{δ,γ1,γ2}.(17)\mathbf{P}^{a}_{i,j_{\mathrm{split}}^{a},k_{0}^{a}} = \overline{\mathbf{P}}_{i}, \qquad\forall\, a \in\{\delta,\gamma_{1},\gamma_{2}\}. \tag*{(17)}

This operation assigns a common column of control points to the shared splitting line among the parent and child branch control lattices while using the adjacent depth d=1d = 1 control columns to define the local tangent alignment at the junction.

Junction continuity at trifurcations

The enforcement of junction continuity at bifurcations, as described in Section 3.4.3, extends naturally to trifurcation junctions. At a trifurcation junction node vδ∈VBjuncv_{\delta} \in\mathcal{V}_{B}^{\mathrm{junc}} in GB\mathcal{G}_{B}, one parent edge bδb_{\delta} meets three child edges bγ1b_{\gamma_{1}}, bγ2b_{\gamma_{2}}, and bγ3b_{\gamma_{3}}. Let a∈{δ,γ1,γ2,γ3}a \in\{\delta,\gamma_{1},\gamma_{2},\gamma_{3}\} index the four branch edges incident at vδv_{\delta}. For each incident edge bab_{a}, let Ca\mathbf{C}^{a} denote its B-spline centerline curve, VaV^{a} denote its B-spline volume, and Pa\mathbf{P}^{a} denote its corresponding control lattice. As in the case of bifurcation junctions, the junction-facing cross-section of each incident control lattice is split into two halves by selecting a radial diameter, also known as the splitting line. The circumferential index of the splitting line, jsplitaj_{\mathrm{split}}^{a}, is determined using (15). The resulting partitioned control-lattice cross-sections are paired across the four incident edges as follows: bδ−bγ2b_{\delta}-b_{\gamma_{2}}, bγ2−bγ1b_{\gamma_{2}}-b_{\gamma_{1}}, bγ1−bγ3b_{\gamma_{1}}-b_{\gamma_{3}}, and bγ3−bδb_{\gamma_{3}}-b_{\delta}, as shown in Figure 6. For each paired set of axial columns in the control lattices, continuity is enforced using (12).

Volumetric hex mesh construction at a bifurcation junction and a trifurcation junction

Figure 6. Volumetric hex mesh construction at a bifurcation junction (top row) and a trifurcation junction (bottom row). The parent B-spline volume is shown in blue. For the bifurcation junction, the two child B-spline volumes are shown in green and yellow; for the trifurcation junction, the three child B-spline volumes are shown in red, yellow, and green. (a) Control lattices of the parent and child B-spline volumes before junction blending. (b) Evaluated B-spline branch volumes before junction blending, showing gaps between independently constructed incident volumes near the junction. (c) Control lattices after junction blending, where the control point positions on the junction-facing cross-sections are modified to enforce continuity between incident branches. (d) Evaluated B-spline volumes after junction blending, showing a small residual gap along the splitting-line interface. (e) Smooth, conforming hex mesh obtained after residual gap closure by averaging corresponding adjacent mesh vertices.

The continuity enforcement procedure at the trifurcation junction differs from that at the bifurcation junction in two ways: the construction of the unit normal vector n^\hat{\mathbf{n}} and the treatment of the shared splitting-line interface. At a bifurcation junction, the three junction-side endpoints of the centerline curves define a unique junction plane. However, at a trifurcation junction, the endpoints pa\mathbf{p}^{a} of the four centerline curves Ca\mathbf{C}^{a} are not always coplanar. Therefore, n^\hat{\mathbf{n}} is computed as the unit normal vector to the best-fit plane passing through these four endpoints. Let

pδ=Cδ(1),pγr=Cγr(0),r=1,2,3,(18)\mathbf{p}^{\delta} = \mathbf{C}^{\delta}(1), \qquad\mathbf{p}^{\gamma_{r}} = \mathbf{C}^{\gamma_{r}}(0), \qquad r = 1,2,3, \tag*{(18)}

denote the junction-side endpoints of the parent and child centerline curves. The centroid of these four endpoints is pˉ=14∑a∈{δ,γ1,γ2,γ3}pa\bar{\mathbf{p}} = \frac{1}{4}\sum_{a \in\{\delta,\gamma_{1},\gamma_{2},\gamma_{3}\}}\mathbf{p}^{a}. We define a 4×34 \times3 matrix E\mathbf{E} whose rows are (pa−pˉ)T(\mathbf{p}^{a} - \bar{\mathbf{p}})^{\mathsf{T}} for a∈{δ,γ1,γ2,γ3}a \in\{\delta,\gamma_{1},\gamma_{2},\gamma_{3}\}. Applying singular value decomposition (SVD) to E\mathbf{E}, we obtain n^\hat{\mathbf{n}} as the right singular vector corresponding to the smallest singular value. The circumferential index on the splitting line jsplitaj^{a}_{\mathrm{split}} is then determined for each control-lattice cross-section using (15).

At a trifurcation junction, the splitting line is treated as a common four-way control-lattice interface among the incident B-spline volumes. For each radial index ii on the splitting line, we compute the centroid of the four adjacent depth d=1d = 1 control points as

Pˉi=14∑a∈{δ,γ1,γ2,γ3}Pi,jsplita,k1aa.(19)\bar{\mathbf{P}}_{i} = \frac{1}{4}\sum_{a \in\{\delta,\gamma_{1},\gamma_{2},\gamma_{3}\}}\mathbf{P}^{a}_{i,j^{a}_{\mathrm{split}},k^{a}_{1}}. \tag*{(19)}

The control points on the shared splitting line of all four incident branches are then assigned to this centroid:

Pi,jsplita,k0aa=Pˉi,∀a∈{δ,γ1,γ2,γ3}.(20)\mathbf{P}^{a}_{i,j^{a}_{\mathrm{split}},k^{a}_{0}} = \bar{\mathbf{P}}_{i}, \qquad\forall a \in\{\delta,\gamma_{1},\gamma_{2},\gamma_{3}\}. \tag*{(20)}

However, the four adjacent depth d=1d = 1 splitting-line control points, Pi,jsplita,k1aa\mathbf{P}^{a}_{i,j^{a}_{\mathrm{split}},k^{a}_{1}}, where a∈{δ,γ1,γ2,γ3}a \in\{\delta,\gamma_{1},\gamma_{2},\gamma_{3}\}, need not be coplanar. For each radial index ii, we construct a 4×34 \times3 centered matrix whose rows are the offsets of these four points from their centroid:

Mi=(Pi,jsplitδ,k1δδ−PˉiPi,jsplitγ1,k1γ1γ1−PˉiPi,jsplitγ2,k1γ2γ2−PˉiPi,jsplitγ3,k1γ3γ3−Pˉi)(21)\mathbf{M}_i = \begin{pmatrix} \mathbf{P}^{\delta}_{i,j_{\mathrm{split}}^{\delta},k_1^{\delta}}-\bar{\mathbf{P}}_i \\ \mathbf{P}^{\gamma_1}_{i,j_{\mathrm{split}}^{\gamma_1},k_1^{\gamma_1}}-\bar{\mathbf{P}}_i \\ \mathbf{P}^{\gamma_2}_{i,j_{\mathrm{split}}^{\gamma_2},k_1^{\gamma_2}}-\bar{\mathbf{P}}_i \\ \mathbf{P}^{\gamma_3}_{i,j_{\mathrm{split}}^{\gamma_3},k_1^{\gamma_3}}-\bar{\mathbf{P}}_i \end{pmatrix} \tag*{(21)}

Applying SVD to Mi\mathbf{M}_i, we obtain the local unit normal vector m^i\hat{\mathbf{m}}_i as the right singular vector corresponding to the smallest singular value. The four adjacent depth d=1d=1 control points are then projected onto this best-fit plane. We first evaluate P~ia\tilde{\mathbf{P}}_i^a as

P~ia=Pi,jsplita,k1aa−[m^i⋅(Pi,jsplita,k1aa−Pˉi)]m^i,a∈{δ,γ1,γ2,γ3}.(22)\tilde{\mathbf{P}}_i^a=\mathbf{P}_{i,j_{\mathrm{split}}^{a},k_1^a}^a-\left[\hat{\mathbf{m}}_i\cdot\left(\mathbf{P}_{i,j_{\mathrm{split}}^{a},k_1^a}^a-\bar{\mathbf{P}}_i\right)\right]\hat{\mathbf{m}}_i,\qquad a\in\{\delta,\gamma_1,\gamma_2,\gamma_3\}. \tag*{(22)}

We then replace the original depth d=1d=1 control point Pi,jsplita,k1aa\mathbf{P}_{i,j_{\mathrm{split}}^{a},k_1^a}^a with its projected point Pi,jsplita,k1aa=P~ia\mathbf{P}_{i,j_{\mathrm{split}}^{a},k_1^a}^a=\tilde{\mathbf{P}}_i^a. This coplanarity condition serves as the multi-way analog of the collinearity condition in the pairwise case (Eq. 12), and enables common tangent-plane alignment across the four incident B-spline volumes at the junction. Eqs. (19)–(22) assign a common column of control points to the shared splitting line of the parent and child control lattices while enforcing coplanarity with the adjacent depth d=1d=1 control points that determine the local tangent alignment at the junction interface.

Hexahedral mesh generation

To generate the hex mesh vertices, each B-spline volume Vα\mathbf{V}^{\alpha} is evaluated on a structured 3D parametric grid following (6). The parametric grid is of the size Sr×Sθ×SuαS_r\times S_\theta\times S_u^{\alpha}, where SrS_r, SθS_\theta, and SuαS_u^{\alpha} denote the number of parametric coordinates along the radial, circumferential, and axial directions, respectively. The B-spline volume Vα\mathbf{V}^{\alpha} of each branch edge is then evaluated at these parametric coordinates to generate a structured grid of mesh vertices. By construction, the control points in the control lattice of each B-spline volume are obtained by uniform sampling along the radial and circumferential directions ((9)–(10)). Therefore, the uniform sampling of radial and circumferential parametric coordinates results in uniformly placed mesh vertices in each branch cross-section. SrS_r and SθS_\theta thus remain constant for all the branch edges, which is important for enforcing junction continuity. However, along the axial direction, the control points are not uniformly distributed across all the branch edges. Therefore, the parametric coordinates in the axial direction are chosen so that mesh vertices have approximately equal spacing along the branch centerline curve. The number of parametric coordinates along the axial direction SuαS_u^{\alpha} varies according to the branch length LαL^{\alpha}. For a fixed spacing between the mesh vertices along the axial direction, Δu\Delta_u, we calculate SuαS_u^{\alpha} as

Suα=max⁡(Smin⁡,⌈LαΔu⌉).(23)S_u^{\alpha}=\max\left(S_{\min},\left\lceil\frac{L^{\alpha}}{\Delta_u}\right\rceil\right). \tag*{(23)}

where Smin⁡S_{\min} is a prescribed lower bound for the number of axial parametric coordinates, ensuring that short branches still contain a minimum number of mesh vertices along the axial direction. The control lattices, after enforcing junction blending for bifurcations and trifurcations, are shown in fig:6c, and the corresponding evaluated B-spline volumes are shown in fig:6d. In the circumferential direction, periodic B-splines defined on an unclamped knot vector are non-interpolatory at the control points along the shared splitting-line interface between the incident branches. Therefore, a small residual gap may remain between the evaluated B-spline volumes along the splitting-line interface. This gap is closed during hex mesh construction by averaging corresponding adjacent mesh vertices, producing a conforming hex mesh across each junction, as shown in fig:6e.

Mesh quality study at junctions

We evaluate the performance of the junction continuity enforcement by varying the angles between the incident branches for bifurcation and trifurcation junctions. For each generated hex mesh, we report the element quality using the scaled Jacobian metric. For an element ee, let tA,1\mathbf{t}_{A,1}, tA,2\mathbf{t}_{A,2}, and tA,3\mathbf{t}_{A,3} denote the three edge vectors meeting at a corner AA, ordered consistently with the right-handed reference hexahedron. Let LA,1L_{A,1}, LA,2L_{A,2}, LA,3L_{A,3} denote the corresponding edge lengths. The Jacobian determinant at the corner AA is the signed volume of the parallelepiped spanned by these edge vectors, JA=det⁡[tA,1,tA,2,tA,3]J_A = \det[\mathbf{t}_{A,1}, \mathbf{t}_{A,2}, \mathbf{t}_{A,3}]. The scaled Jacobian at a corner AA of the element is evaluated by normalizing JAJ_A by the product of the three incident edge vectors. The scaled Jacobian of the element ee is thus given as

qe=min⁡A=1,…,8JALA,1LA,2LA,3.(24)q_e = \min_{A=1,\ldots,8} \frac{J_A}{L_{A,1}L_{A,2}L_{A,3}}. \tag*{(24)}

This metric satisfies ∣qe∣≤1|q_e| \le1, with qe=1q_e = 1 corresponding to an orthogonal, positively oriented element. qe≤0q_e \le0 indicates that at least one corner of the element is degenerate or inverted.

Figure 7 shows different bifurcation junction configurations generated by varying the angle between the three incident branches. For each configuration, the angle between each branch is set to be between 30∘30^\circ and 270∘270^\circ. For all the bifurcation junction configurations, we report positive scaled Jacobian values, as seen in Table 1. In addition, the 10th10^{\mathrm{th}} percentile of the element quality (P10P_{10}) satisfies P10≥0.925P_{10} \ge0.925, indicating that at least 90% of the mesh elements in each configuration have qe≥0.925q_e \ge0.925. Similarly, we perform the angle parametric study for trifurcation junctions, as shown in Figure 8, varying the angles between the four incident branches at the junction. For all the configurations, we report good mesh quality, where all the elements are non-inverted (qe>0q_e > 0). We maintain P10≥0.905P_{10} \ge0.905 for trifurcation junctions, indicating that at least 90% of the mesh elements in each configuration have qe≥0.905q_e \ge0.905. In all the junction configurations, the lowest qeq_e value is observed near the junction. Particularly, in one of the trifurcation junction configurations, the minimum qeq_e value is observed to be 0.165, as shown in Table 1. This is primarily due to increased localized curvature and extreme distortion near the junction introduced when the angles between the branches become very small.

Angle parametric study for bifurcation junction configurations

Figure 7. Angle parametric study for bifurcation junction configurations. For each configuration, the three listed angles denote the angles, in degrees, between consecutive incident branches around the junction. Each generated mesh is colored according to the element-wise scaled Jacobian value.

Angle parametric study for trifurcation junction configurations

Figure 8. Angle parametric study for trifurcation junction configurations. For each configuration, the four listed angles denote the angles, in degrees, between consecutive incident branches around the junction. Each generated mesh is colored according to the element-wise scaled Jacobian value.

Bifurcation JunctionTrifurcation Junction
ConfigurationMin(qeq_e)Mean(qeq_e)P10(qe)P_{10}(q_e)ConfigurationMin(qeq_e)Mean(qeq_e)P10(qe)P_{10}(q_e)
(135∘135^\circ, 90∘90^\circ, 135∘135^\circ)0.7400.9880.979(90∘90^\circ, 90∘90^\circ, 90∘90^\circ, 90∘90^\circ)0.6570.9820.954
(150∘150^\circ, 60∘60^\circ, 150∘150^\circ)0.6340.9850.979(135∘135^\circ, 45∘45^\circ, 45∘45^\circ, 135∘135^\circ)0.4600.9790.963
(165∘165^\circ, 30∘30^\circ, 165∘165^\circ)0.3050.9710.937(120∘120^\circ, 60∘60^\circ, 60∘60^\circ, 120∘120^\circ)0.5120.9800.962
(90∘90^\circ, 90∘90^\circ, 180∘180^\circ)0.5980.9850.974(60∘60^\circ, 60∘60^\circ, 60∘60^\circ, 180∘180^\circ)0.3590.9750.954
(60∘60^\circ, 240∘240^\circ, 60∘60^\circ)0.3280.9690.948(60∘60^\circ, 210∘210^\circ, 45∘45^\circ, 45∘45^\circ)0.3570.9620.905
(45∘45^\circ, 45∘45^\circ, 270∘270^\circ)0.2380.9640.925(150∘150^\circ, 30∘30^\circ, 30∘30^\circ, 150∘150^\circ)0.1650.9740.962

Table 1. Scaled Jacobian statistics for different bifurcation and trifurcation junction configurations. The minimum, mean, and 10th10^{\mathrm{th}} percentile (P10P_{10}) qeq_e values are reported for each configuration.

Results and Discussion

We evaluate the proposed procedural modeling framework with respect to volumetric mesh generation from point cloud data, dynamic mesh updating, and suitability for FEA. We test the framework on three different plant species with increasing geometric and topological complexity. We first demonstrate volumetric mesh generation from skeleton point clouds of mung bean (Section 4.1.1) and tomato (Section 4.1.2) plants. In Section 4.2, we demonstrate the performance of the framework for unstructured point clouds of walnut tree branches. Here, in addition to generating volumetric meshes from plant skeletons, we also estimate the branch radii from unstructured point clouds, enabling mesh generation for complex branching architectures with spatially varying radii. We then demonstrate the main advantage of the procedural framework over static mesh generation approaches: automated modeling and dynamic mesh updating during plant growth without reconstructing the entire mesh from scratch at each growth stage (Section 4.3). Finally, in Section 4.4, we solve a steady-state diffusion problem on the generated meshes to demonstrate their direct suitability for FEA.

Volumetric mesh generation from plant skeleton datasets

We demonstrate the procedural modeling framework by generating volumetric meshes from skeleton point clouds of mung bean and tomato plants. Since these input point clouds approximate only the centerline geometry of the plants and do not provide branch-radius information, we model each branch using a circular cross-section with a prescribed constant radius along its length. Specifically, we assign a constant outer branch radius RbR_b to every node νl∈VS\nu_l \in\mathcal{V}_S in the skeleton graph GS\mathcal{G}_S, such that Rl=RbR_l = R_b, and use a constant inner-to-outer branch radius ratio η\eta.

Mung bean plants (Vigna radiata)

Point clouds of several mung bean varieties were obtained via terrestrial laser scanning [50], and skeletons defining the plant medial axes were extracted from these point clouds using the method described in [164]. In all mung bean specimens, the resulting skeleton graphs contain only bifurcation junctions. Because the skeleton point clouds are relatively sparse, additional nodes are inserted along the skeleton graph edges by interpolation, so that the spacing between adjacent nodes does not exceed 1.0 mm. A uniform outer branch radius Rb=1.5R_b = 1.5 mm is assigned to all skeleton nodes, and η=0.5\eta= 0.5 provides an inner branch radius of 0.75 mm. In the volumetric parameterization for each branch, cubic B-spline basis functions are used in all three parametric directions. In the circumferential direction, we use periodic B-splines with 14 unique control points, whereas in the radial direction, we use 4 control points. The hex mesh vertices for each branch are evaluated on a parametric grid of Sθ=28S_\theta= 28 and Sr=4S_r = 4 coordinates, with an axial parametric spacing of Δu=1\Delta_u = 1 mm. We use the hierarchical construction strategy to preserve the geometric smoothness of the parent centerline B-spline curves during branch insertion, as outlined in Algorithm 3. For each branch edge bαb_\alpha, nαn^\alpha is set equal to the number of coordinates in Qα\mathbf{Q}^\alpha, as described in Section 3.2.2. We use the same parameters in the procedural framework for all the mung bean morphologies in the dataset, ranging from upright stem architectures to semi-erect, laterally sprawling branching architectures.

Figure 9(a) shows the volumetric meshes generated from the mung bean plant skeletons, along with the corresponding close-up views of selected bifurcation junctions colored by the element-wise scaled Jacobian qeq_e (Figure 9(b)). The minimum, mean, and 10th10^{\mathrm{th}} percentile (P10P_{10}) values of qeq_e are reported in Table 2 for each specimen. For all meshes, the mean qeq_e value is greater than 0.97, and the P10P_{10} value is greater than 0.95, indicating that at least 90% of the mesh elements have qe>0.95q_e > 0.95. The minimum qeq_e values are approximately between 0.1 and 0.27, and the lower-quality elements are localized to a small number of junctions. These junctions typically have child branches meeting the parent branch at very acute angles. The close-up views of these junctions show visually smooth parent-child branch interfaces without any kinks on the surface, consistent with the continuity conditions enforced during junction blending.

SpecimenMin(qeq_e)Mean(qeq_e)P10(qe)P_{10}(q_e)
Mung bean 10.26250.97730.9542
Mung bean 20.13830.98020.9622
Mung bean 30.10420.98160.9746
Mung bean 40.16510.97880.9567

Table 2. Scaled Jacobian statistics for mung bean plants.

Mung bean plant volumetric mesh generation

Figure 9. Mung bean plant volumetric mesh generation. (a) Rendering of the generated volumetric hex meshes for four mung bean plant specimens (Mung bean 1–4). (b) Close-up views of three representative bifurcation junctions from each plant specimen, colored by the element-wise scaled Jacobian qeq_e.

Tomato plants (Solanum lycopersicum)

We next apply the procedural framework to four skeleton point clouds of tomato plants from the TomatoWUR dataset [165]. The selected specimens have dataset serial numbers 101, 336, 345, and 98, corresponding respectively to Tomato 1–4 in Figure 10 and Table 3. Compared with the mung bean specimens, the tomato specimens contain more complex branching structures. In addition to plant stems, the skeletons also include large leaf-vein structures and smaller vein branches that attach near the distal ends of larger leaf branches. The knot insertion operation used in the hierarchical construction strategy (Section 3.2.2) requires nondegenerate parametric intervals on both sides of the child-branch attachment point on the parent B-spline curve. Specifically, if a child branch attaches to the parent B-spline

SpecimenMin(qeq_e)Mean(qeq_e)P10(qe)P_{10}(q_e)
Tomato 10.15980.95250.9109
Tomato 20.30440.95310.9088
Tomato 30.38930.95460.9141
Tomato 40.25160.95090.9026

Table 3. Scaled Jacobian statistics for tomato plants.

Tomato plant volumetric mesh generation and junction-level mesh quality

Figure 10. Tomato plant volumetric mesh generation. (a) A rendering of the generated volumetric hex meshes for four tomato plant specimens (Tomato 1–4). (b) Close-up views of three representative junctions from each plant model, colored by the element-wise scaled Jacobian value qeq_e.

centerline curve Cα(u)\mathbf{C}^{\alpha}(u) of branch bαb_{\alpha} at parameter uˉδ\bar{u}_{\delta}, the construction requires two adjacent parameters, uˉδ−\bar{u}_{\delta}^{-} and uˉδ+\bar{u}_{\delta}^{+}, such that uˉδ−<uˉδ<uˉδ+\bar{u}_{\delta}^{-} < \bar{u}_{\delta} < \bar{u}_{\delta}^{+}. When the attachment point lies too close to the distal end of the parent B-spline curve, u=1u = 1, there is insufficient parametric space to define a distinct neighboring parameter uˉδ+\bar{u}_{\delta}^{+} satisfying uˉδ<uˉδ+<1\bar{u}_{\delta} < \bar{u}_{\delta}^{+} < 1. Therefore, the hierarchical construction strategy is not suitable for this dataset because several smaller vein branches attach near the distal ends of their parent branches. We thus use the global construction strategy described in Section 3.2.1 to generate the B-spline centerline curves for the complete branch graph.

We increase the node density in each skeleton graph to achieve a maximum spacing of 1.0 mm between adjacent nodes. A constant outer branch radius Rb=2R_b = 2 mm is assigned to all the skeleton nodes, and η\eta is set to a constant value of 0.5, resulting in an inner branch radius of 1 mm. The volumetric parameterization for each branch uses cubic B-spline basis functions in all three parametric directions. The number of unique control points in the circumferential direction is set to 14, and the number of control points in the radial direction is set to 4. The number of axial control points is determined from the branch length using the global construction strategy (Algorithm 2). The hex mesh vertices are evaluated with Sθ=14S_{\theta} = 14 and Sr=3S_r = 3 parametric coordinates in each cross-section, and with an axial parametric spacing of Δu=2\Delta_u = 2 mm. Figure 10 shows the volumetric meshes for the tomato specimens, together with close-up views of selected junctions colored by the element-wise scaled Jacobian qeq_e. As shown in Table 3, the mean qeq_e value for each mesh is approximately 0.95, and P10(qe)P_{10}(q_e) is approximately 0.91, both of which are slightly lower than the corresponding values obtained for the mung bean meshes. The reduction in mesh quality relative to the mung bean dataset is consistent with the higher branching density and greater variation in branch lengths of the tomato specimens. Moreover, a larger number of junctions increases the number of mesh elements affected by junction blending, which can locally reduce mesh quality. However, since each junction only affects a small number of elements, the overall mesh quality decreases very slightly.

Walnut tree mesh generation with spatially varying branch radii

The meshes generated for the mung bean and tomato plants in Section 4.1 use a constant outer branch radius for all branches within each specimen, independent of branch order. Real plants, particularly trees, exhibit substantial variation in branch radius across branch orders and along individual branches. Thus, the proposed framework must also support the blending of incident branches with different radii at junctions. We demonstrate this capability using three walnut tree specimens from the tree dataset [166], which provides unstructured 3D point clouds of branching structures together with their corresponding ground-truth skeletons. At each node vlv_l on the skeleton graph, we estimate the outer radius of cross-section RlR_l from the neighboring points in the unstructured point cloud. To compute RlR_l, we first identify the points lying on the outer branch surface. A kk-nearest-neighbor graph of the point cloud is constructed, and points with one-sided neighborhoods are classified as outer-surface points. The radius RlR_l at each node is then computed as the average distance from vlv_l to its kk nearest outer-surface points. Finally, the radii RlR_l are interpolated along the fitted centerline B-spline curve Cα(u)\mathbf{C}^{\alpha}(u) of each branch to obtain the outer branch radius hkh_k at each Greville abscissa ξk\xi_k when constructing the control lattice in (10). In this way, the spatially varying branch radius along each centerline B-spline curve is incorporated directly into the volumetric parameterization, without requiring control-point optimization. The sparse plant skeletons provided in the dataset are resampled to a maximum internodal spacing of 1 mm, yielding a denser skeleton point cloud for more accurate B-spline centerline curve fitting. The centerline curves for the complete branch graph are constructed using the hierarchical construction strategy described in Section 3.2.2, preserving the geometric smoothness of the parent centerline B-spline curve each time a child branch is attached to it. The volumetric parameterization uses cubic basis functions in all three parametric directions, with 14 unique control points in the circumferential direction and 4 control points in the radial direction. The hex mesh vertices are computed by evaluating each B-spline volume on a cross-sectional parametric grid of size Sθ=42S_{\theta} = 42 and Sr=4S_r = 4, with the axial parametric spacing Δu\Delta_u set per specimen based on its branch-length distribution.

The reconstructed volumetric meshes of the walnut tree specimens are shown in fig:11a, with close-up views of representative junction regions in fig:11b. The input unstructured point cloud is overlaid on each mesh to visualize reconstruction accuracy, with each point colored by its reconstruction error, defined as the shortest Euclidean distance from the point to the mesh. Table 5 summarizes the reconstruction error statistics across all tree specimens. We report a mean error value of less than 3 mm and a maximum error value not exceeding 38.6 mm across all the specimens. As reported in Table 4, the meshes for walnut tree specimens achieve the highest overall mesh quality among the three datasets, with mean qeq_e and P10(qe)P_{10}(q_e) both greater than 0.99. This improvement relative to the meshes obtained in Section 4.1 is due to the finer circumferential discretization (Sθ=42S_{\theta} = 42 for walnut, versus 28 for mung bean and 14 for tomato specimens), together with a larger number of mesh elements per branch generated in the taller walnut

SpecimenMin(qeq_e)Mean(qeq_e)P10(qe)P_{10}(q_e)
Walnut 10.11810.99170.9941
Walnut 20.39220.99430.9953
Walnut 30.10520.99260.9953

Table 4. Scaled Jacobian statistics for walnut trees.

SpecimenMin ErrorMean ErrorMax Error
Walnut 10.12.738.6
Walnut 20.02.935.7
Walnut 30.22.931.8

Table 5. Reconstruction error statistics (in mm) for walnut trees.

Walnut tree volumetric mesh generation

Figure 11. Walnut tree volumetric mesh generation. (a) Volumetric hex mesh reconstructions of three walnut trees (Walnut 1–3). (b) Three representative junction regions from each tree, colored by the element-wise scaled Jacobian. The input point cloud is overlaid on the reconstructed volumetric mesh and colored by the shortest Euclidean distance to the mesh vertices.

trees. Because the mesh elements with lower quality remain confined to a localized region at each junction, increasing the number of elements per branch subsequently increases the proportion of high-quality elements away from the junctions and thus improves the overall mesh quality. The mesh elements corresponding to the minimum qeq_e values remain localized to only a small number of junction regions where branches meet at acute angles.

SpecimenElementsNodesSpecimenElementsNodesSpecimenElementsNodes
Mung bean 157,45677,136Tomato 137,88457,588Walnut 1354,690474,288
Mung bean 271,06495,332Tomato 243,90466,708Walnut 2692,118923,712
Mung bean 362,07683,244Tomato 326,32039,954Walnut 3292,824391,320
Mung bean 494,164126,132Tomato 445,92069,840

Table 6. Hexahedral mesh statistics in terms of number of elements and vertices for the reconstructed specimens of each dataset.

Dynamic plant growth modeling

Modeling plant growth and development is a dynamic process in which the meshes must robustly adapt to extending and bifurcating branches that continuously modify the underlying plant topology. The computational cost of static mesh generation pipelines can compound over multiple growth stages, thus becoming extremely prohibitive for the dynamic modeling of complex branching architectures. In our procedural framework, we first represent the plant centerline geometry as a graph and subsequently fit it to smooth B-spline centerline curves. A tensor-product B-spline volume is then constructed around the centerline curve of each branch and finally assembled to generate a hex mesh after junction-blending operations. This sequence of steps allows plant growth to be modeled using local graph updates. Newly introduced branch segments are first converted to B-spline branch volumes, and junction blending is carried out in a local manner while retaining the hex mesh generated in the previous growth stage. This provides a scalable approach for dynamic volumetric modeling of growing plants, enabling local mesh updates without constructing geometry from scratch.

We demonstrate dynamic mesh updating using the Mung bean 3 specimen (Section 4.1.1) by simulating growth through a breadth-first search (BFS) traversal of the skeleton graph, starting from the root node and moving systematically according to the branching order. At each growth stage, only the part of the skeleton that has grown so far, referred to as the active skeleton subgraph, is considered for volumetric mesh generation. The active skeleton subgraph contains all nodes whose BFS distance from the root node is less than or equal to the current growth level. Figure 12 shows the generated volumetric hex meshes at six representative growth stages. The hex mesh generated at previous growth stages is shown in blue, whereas newly added branches at the current growth stage are shown in orange. As reported in Table 7, the hex meshes generated at all the growth stages have positive scaled Jacobian values, indicating that there are no inverted mesh elements. The mean qeq_e value remains at least 0.9729 across all growth stages, indicating that the local mesh updates at each stage preserve global mesh quality. Table 8 compares the computation time required to rebuild the full plant geometry at each growth stage with the time required for dynamic updating by locally adding the new branches. We demonstrate that reconstructing the entire plant geometry from scratch requires 0.096 seconds at the final growth stage (Step 5). In contrast, the dynamic approach only requires between 0.03 and 0.05 seconds to generate the volumetric mesh at each growth stage after the initial step (Step 0). This confirms that the framework decouples local growth updates from global branching complexity, providing a scalable framework for dynamic plant modeling.

Volumetric reconstruction of a mung bean plant specimen at six representative growth stages

Figure 12. Volumetric reconstruction of a mung bean plant specimen at six representative growth stages via dynamic modeling. The meshes are generated by BFS-based frontier expansion of the skeleton graph from the root node. Newly added branches at each stage are shown in orange, while the plant geometry reconstructed in the previous stages is shown in blue. The fully mature plant geometry is overlaid as a gray silhouette.

Growth StageMin⁡(qe)\operatorname{Min}(q_e)Mean⁡(qe)\operatorname{Mean}(q_e)P10(qe)P_{10}(q_e)
Step 00.53070.97480.9402
Step 10.38790.97670.9517
Step 20.25110.97560.9547
Step 30.24120.97290.9450
Step 40.29450.97510.9524
Step 50.34270.97600.9559

Table 7. Scaled Jacobian statistics for plant growth modeling.

Growth StageFull Plant (s)Dynamic Update (s)
Step 00.0070.007
Step 10.0590.052
Step 20.0630.034
Step 30.0890.043
Step 40.0910.029
Step 50.0960.030

Table 8. Computation time (in seconds) for full plant reconstruction and dynamic mesh updating at each growth stage.

Diffusion simulation in volumetric meshes

To evaluate the analysis-suitability of the generated volumetric meshes, we solved a steady-state diffusion equation on each plant geometry. The purpose of these simulations is to demonstrate the direct solver-readiness of the volumetric meshes generated by the proposed procedural modeling framework. We thus verify that the generated hex meshes are conforming and have valid element orientations. This is important for downstream simulations of plant biomechanics and transport phenomena with volumetric meshes, where poor mesh quality could affect the reliability of FEA.

We define a steady-state nutrient concentration field c(x)c(\mathbf{x}) over the 3D plant domain Ω∈R3\Omega\in\mathbb{R}^{3}. We denote the unique branch incident to the root node νroot\nu_{\mathrm{root}}, which is the proximal node in the branch graph GB\mathcal{G}_{B}, as bζ∈EBb_{\zeta} \in\mathcal{E}_{B}. The B-spline volume Vζ\mathbf{V}^{\zeta} is defined for the branch bζb_{\zeta}. A concentration c(x)=1c(\mathbf{x}) = 1 is prescribed on the root (source) cross-section Γroot\Gamma_{\mathrm{root}}, defined as the proximal (u=0u = 0) face of Vζ\mathbf{V}^{\zeta}:

Γroot={Vζ(r,θ,0):r∈[0,1], θ∈[0,1]}.\Gamma_{\mathrm{root}} = \left\{ \mathbf{V}^{\zeta}(r,\theta,0) : r \in[0,1],\ \theta\in[0,1] \right\}.

This cross-section is referred to as the source boundary cross-section. A constant outward flux is prescribed on the terminal leaf (sink) cross-sections, defined as the distal (u=1u = 1) faces of all leaf branches. Let I={α:bα∈EB, να∈VBleaf}\mathcal{I} = \left\{ \alpha: b_{\alpha} \in\mathcal{E}_{B},\ \nu_{\alpha} \in\mathcal{V}_{B}^{\mathrm{leaf}} \right\} denote the index set of leaf branches. We then define the terminal leaf cross-sections as

Γleaf=⋃α∈I{Vα(r,θ,1):r∈[0,1], θ∈[0,1]}.\Gamma_{\mathrm{leaf}} = \bigcup_{\alpha\in\mathcal{I}} \left\{ \mathbf{V}^{\alpha}(r,\theta,1) : r \in[0,1],\ \theta\in[0,1] \right\}.

The lateral branch walls are the inner and outer cylindrical surfaces of every branch volume, which are the faces at r=0r = 0 (inner) and r=1r = 1 (outer) of each Vα\mathbf{V}^{\alpha}:

Γwall=⋃bα∈EB({Vα(0,θ,u):θ∈[0,1], u∈[0,1]}∪{Vα(1,θ,u):θ∈[0,1], u∈[0,1]}).\Gamma_{\mathrm{wall}} = \bigcup_{b_{\alpha}\in\mathcal{E}_{B}} \left(\left\{\mathbf{V}^{\alpha}(0,\theta,u):\theta\in[0,1],\ u\in[0,1]\right\}\cup\left\{\mathbf{V}^{\alpha}(1,\theta,u):\theta\in[0,1],\ u\in[0,1]\right\}\right).

The governing problem is modeled as the Laplace equation, with a Dirichlet boundary condition at Γroot\Gamma_{\mathrm{root}} and an inhomogeneous Neumann (flux) boundary condition at Γleaf\Gamma_{\mathrm{leaf}}. A homogeneous Neumann (no flux) boundary condition is applied at Γwall\Gamma_{\mathrm{wall}}. The diffusion problem is thus given as

∇2c(x)=0in Ω,c(x)=1on Γroot,∇c(x)⋅n=−100on Γleaf,∇c(x)⋅n=0on Γwall,(25)\begin{aligned} \nabla^{2}c(\mathbf{x}) &= 0 &&\text{in }\Omega,\\ c(\mathbf{x}) &= 1 &&\text{on }\Gamma_{\mathrm{root}},\\ \nabla c(\mathbf{x})\cdot\mathbf{n} &= -100 &&\text{on }\Gamma_{\mathrm{leaf}},\\ \nabla c(\mathbf{x})\cdot\mathbf{n} &= 0 &&\text{on }\Gamma_{\mathrm{wall}}, \tag*{(25)} \end{aligned}

where n\mathbf{n} is the outward unit normal defined on the domain boundary ∂Ω=Γroot∪Γleaf∪Γwall\partial\Omega=\Gamma_{\mathrm{root}}\cup\Gamma_{\mathrm{leaf}}\cup\Gamma_{\mathrm{wall}}. Although the steady-state diffusion problem does not completely capture the complexity of the nutrient transport process, it provides a useful test for ensuring the smooth propagation of the concentration solution field through multiple branching levels and across the blended junction regions.

We perform diffusion simulations on the hex meshes generated from three mung bean, tomato, and walnut tree specimens (Figure 13–Figure 15). The generated meshes span different levels of branching complexity, including meshes with uniform and spatially varying radii. The solver converges for every mesh, indicating that the meshes are valid for physics-based simulations. Across all specimens, the concentration field c(x)c(\mathbf{x}) decreases monotonically from the root cross section toward the terminal leaf cross-sections and remains continuous across all branch junction regions. These results demonstrate the capability of the proposed procedural volumetric modeling framework to not only generate valid meshes but also to support their direct use in FEA. The solver-readiness of the volumetric modeling framework

Steady-state diffusion simulation on mung bean volumetric meshes

Figure 13. Steady-state diffusion simulation on mung bean volumetric meshes. The computed concentration field is shown for a diffusion problem with a source prescribed at the root boundary cross-section Γroot\Gamma_{\mathrm{root}} and terminal conditions applied at the leaf cross-sections Γleaf\Gamma_{\mathrm{leaf}} for three mung bean plant specimens. The smooth and monotonic variation of the solution across multiple branching orders and blended junctions demonstrates the direct analysis suitability and solver-readiness of the generated hex meshes.

Steady-state diffusion simulation on walnut tree volumetric meshes

Figure 15. Steady-state diffusion simulation on walnut tree volumetric meshes. The computed concentration field is shown for a diffusion problem with a source prescribed at the root boundary cross-section Γroot\Gamma_{\mathrm{root}} and terminal conditions applied at the leaf cross-sections Γleaf\Gamma_{\mathrm{leaf}} for three tree-scale architectures featuring spatially varying branch radii. The smooth and monotonic variation of the solution across multiple branching orders and blended junctions demonstrates the direct analysis suitability and solver-readiness of the generated hex meshes.

Steady-state diffusion simulation on tomato volumetric meshes

Figure 14. Steady-state diffusion simulation on tomato volumetric meshes. The computed concentration field is shown for a diffusion problem with a source prescribed at the root boundary cross-section Γroot\Gamma_{\mathrm{root}} and terminal conditions applied at the leaf cross-sections Γleaf\Gamma_{\mathrm{leaf}} for three tomato plant specimens. The smooth and monotonic variation of the solution across multiple branching orders and blended junctions demonstrates the direct analysis suitability and solver-readiness of the generated hex meshes.

highlights its advantage over surface reconstruction methods, which often require additional mesh generation and cleanup pipelines to be carried out before physics-based simulations can be performed.

Conclusion and Future Work

We present a novel procedural framework that converts plant skeleton point clouds into analysis-suitable hexahedral meshes for plant branching structures. The pipeline extracts a branch graph from the skeleton, fits a B-spline centerline curve to each branch, constructs a cylindrical tensor-product B-spline volume along each centerline curve, and enforces smooth junction continuity across incident branch volumes at each junction via control point blending operations to generate the hex meshes. Mesh generation at several angle configurations for bifurcation and trifurcation junctions shows that mesh quality deteriorates only at extremely acute angles while remaining free of element inversions and intersections in all cases. In highly complex branching architectures, localized element inversions can occur near the junctions. However, these can be resolved by increasing the number of control points or by increasing the pruning radius near junctions to allow more room for junction blending. We demonstrate the generalizability of the framework across different plant species by generating analysis-suitable plant geometries with consistently good mesh quality. We also demonstrate that plant growth can be modeled dynamically by extending the skeleton graph while adding new branches at each growth stage and subsequently carrying out blending at the newly added junctions. The resulting meshes are solver-ready, as evidenced by a steady-state diffusion simulation in which the concentration field remains smooth across every junction for all the generated plant geometries. Together, these results demonstrate that the framework bridges the gap between surface-based plant reconstruction and analysis-ready procedural volumetric mesh generation pipelines. Several directions remain for future work. Our framework currently focuses on bifurcation and trifurcation junctions; however, some plant architectures exhibit higher-valence junctions. Extending the junction blending operations to such junctions would enable the modeling of more complex branching architectures. Moreover, while the present work uses B-spline volumes as a scaffold for generating conforming hexahedral meshes, we plan to develop a complete B-spline-based volumetric parameterization of the entire plant geometry for isogeometric analysis.

Acknowledgements

A. Moola and A. Pawar would like to acknowledge funding in part from the National Science Foundation under award number DMS-2533961 and the Translational AI Center at Iowa State University.

References

  1. [5]R. Zhang, W. Jing, Z. Zhang, H. Yan, H. Zhang, J. Gao, X. Zhou, N. Ma, An overview: Functional-structural plant model in simulating the growth and development of greenhouse horticultural crops, Horticultural Plant Journal.
  2. [7]R. Sievänen, C. Godin, T. M. DeJong, E. Niklmaa, Functional–structural plant models: A growing paradigm for plant studies, Annals of Botany 114 (4) (2014) 599–603.
  3. [8]S. Souliaou, Z. Wang, W. Sun, P. de Reffye, B. Collins, G. Louarn, Y. Song, Functional–structural plant models mission in advancing crop science: opportunities and prospects, Frontiers in Plant Science 12 (2021) 747142.
  4. [19]M. Hadadi, M. Saraeian, J. Godbersen, T. Z. Jubery, Y. Li, L. Attigala, A. Balu, S. Sarkar, P. S. Schnable, A. Krishnamurthy, B. Ganapathy subramanian, Procedural generation of 3D maize plant architecture from LiDAR data, Computers and Electronics in Agriculture 236 (2025) 110382.
  5. [23]E. Omia, E. Park, D. Semyalo, R. Joshi, B.-K. Cho, Advancements in 3D field-crop phenotyping using point clouds: A comparative review of sensor technology, target traits, and challenges under controlled and field conditions, Frontiers in Plant Science 17 (2026) 1731852.
  6. [38]X. Wang, J. Hua, M. Kang, H. Wang, P. de Reffye, Functional–structural plant model “GreenLab”: A state-of-the-art review, Plant Phenomics 6 (2024) 0118.
  7. [42]P. Ghose, A. Bashir, A. Zahid, Advancements in 3D reconstruction for plant phenotyping: technologies, applications, challenges, and future directions, Sensors 26 (9) (2026) 2730.
  8. [46]N. Harandi, B. Vandenberghe, J. Yankerschaver, S. Depuydt, A. Van Messem, How to make sense of 3D representations for plant phenotyping: A compendium of processing and analysis techniques, Plant Methods 19 (2023) 60.
  9. [48]H. R. Sankaramaddi, W. S. Lee, K. Kim, Y. Hong, 2D-to-3D image reconstruction in agriculture: A review of methods, challenges, and ai-driven opportunities, Sensors 26 (6) (2026) 1775.
  10. [49]S. Joshi, J. Di Salvo, Y. Shen, M. Hadadi, V. N. Boddepalli, Z. Jubery, S. Sarkar, A. Singh, B. Ganapathysubramanian, A. K. Singh, A. Krishnamurthy, Neural 3D reconstruction and immersive VR visualization of row crops across phenological growth stages, Smart Agricultural Technology (2026) 102024.
  11. [50]M. Hadadi, T. Z. Jubery, P. S. Schnable, A. Singh, B. Benes, A. Krishnamurthy, B. Ganapathyubramanian, FloraForge: Procedural generation of editable and analysis-ready 3D plant geometric models using LLM-assisted template design, Smart Agricultural Technology (2026) 102316.
  12. [51]P. Xie, R. Du, Z. Ma, H. Cen, Generating 3D multispectral point clouds of plants with fusion of snapshot spectral and RGB-D images, Plant Phenomics 5 (2023) 0040.
  13. [52]G. Rivera, R. Porras, R. Florencia, J. P. Sánchez-Solís, LiDAR applications in precision agriculture for cultivating crops: A review of recent advances, Computers and Electronics in Agriculture 207 (2023) 107737.
  14. [53]F. Boudon, C. Pradal, T. Cokelaer, P. Prusinkiewicz, C. Godin, L-Py: An L-system simulation framework for modeling plant architecture development based on a dynamic language, Frontiers in Plant Science 3 (2012) 76.
  15. [54]P. Prusinkiewicz, M. Cieslak, P. Ferraro, J. Hanan, Modeling plant development with L-systems, in: Mathematical Modelling in Plant Biology, Springer, 2018, pp. 139–169.
  16. [55]W. Herulambang, R. Wardoyo, 3D simulation of plant growth modeling using neuro-fuzzy, Lindenmayer system, and turtle geometry, JEECS (Journal of Electrical Engineering and Computer Sciences) 1 (2) (2016) 91–98.
  17. [56]D. Leitner, S. Klepsch, G. Bodner, A. Schnepf, A dynamic root system growth model based on L-systems: Tropisms and coupling to nutrient uptake from soil, Plant and Soil 332 (1) (2010) 177–192.
  18. [57]J. J. Lee, B. Li, B. Benes, Latent L-systems: Transformer-based tree generator, ACM Transactions on Graphics 43 (1) (2023) 1–16.
  19. [58]J. Bernard, I. McQuillan, Techniques for inferring context-free Lindenmayer systems with genetic algorithm, Swarm and Evolutionary Computation 64 (2021) 100893.
  20. [59]M. Allen, P. Prusinkiewicz, T. DeJong, Using L-systems for modeling the architecture and physiology of growing trees: The L-PEACH model, in: Proc. of the 4th Intern. Workshop on Functional-Structural Plant Models. Montpellier, France, 2004.
  21. [60]Y. Ong, K. Streit, M. Henke, W. Kurth, An approach to multiscale modelling with graph grammars, Annals of Botany 114 (4) (2014) 813–827.
  22. [61]T. M. DeJong, D. Da Silva, J. Vos, A. J. Escobar-Gutiérrez, Using functional–structural plant models to study, understand and integrate plant development and ecophysiology, Annals of Botany 108 (6) (2011) 987–989.
  23. [62]L. Feng, J.-C. Mailhol, H. Rey, S. Griffon, D. Auclair, P. De Reffye, Comparing an empirical crop model with a functional structural plant model to account for individual variability, European Journal of Agronomy 53 (2014) 16–27.
  24. [63]J. Vos, J. B. Evers, G. H. Buck-Sorlin, B. Andrieu, M. Chelle, P. H. B. De Visser, Functional–structural plant modelling: A new versatile tool in crop science, Journal of Experimental Botany 61 (8) (2010) 2101–2115.
  25. [64]G. Louarn, Y. Song, Two decades of functional–structural plant modelling: Now addressing fundamental questions in systems biology and predictive ecology, Annals of Botany 126 (4) (2020) 501–509.
  26. [65]P. Du, M. Xu, X. Zhu, J.-X. Wang, HUG-VAS: A hierarchical NURBS-based generative model for aortic geometry synthesis and controllable editing, arXiv preprint arXiv:2507.11474.arxiv.org/abs/2507.11474
  27. [66]W. Wu, Y. Hu, Y. Lu, Parametric surface modelling for tea leaf point cloud based on non-uniform rational basis spline technique, Sensors 21 (4) (2021) 1304.
  28. [67]R. Ando, Y. Ozawa, W. Guo, Robust surface reconstruction of plant leaves from 3D point clouds, Plant Phenomics.
  29. [68]H. Murata, K. Noshita, Three-dimensional leaf edge reconstruction combining two-and three-dimensional approaches, Plant Phenomics 6 (2024) 0181.
  30. [69]S. H. Cheikh, K. M. Moerman, D. C De Oliveira, J. Jacob, N. Mogulkoс, B. R. Davidson, S. Walker-Samuel, R. J. Shipley, Tube2FEM: A general-purpose highly automated pipeline for flow-related processes in (embedded) tubular objects, Royal Society Open Science 12 (8).
  31. [70]N. Pietroni, M. Campen, A. Sheffer, G. Cherchi, D. Bommes, X. Gao, R. Scateni, F. Ledoux, J. Remacle, M. Livesu, Hex-mesh generation and processing: A survey, ACM Transactions on Graphics 42 (2) (2022) 1–44.
  32. [71]A. Li, Y. J. Zhang, Intracellular material transport simulation in neurons using isogeometric analysis and deep learning, in: International Conference on Computational Science, Springer, 2023, pp. 486–493.
  33. [72]Y. Zhang, Y. Bazilevs, S. Goswami, C. L. Bajaj, T. J. R. Hughes, Patient-specific vascular NURBS modeling for isogeometric analysis of blood flow, Computer Methods in Applied Mechanics and Engineering 196 (29-30) (2007) 2943–2959.
  34. [73]A. Li, Y. J. Zhang, Isogeometric analysis-based physics-informed graph neural network for studying traffic jam in neurons, Computer Methods in Applied Mechanics and Engineering 403 (2023) 115757.
  35. [74]A. Li, Y. J. Zhang, Modeling material transport regulation and traffic jam in neurons using PDE-constrained optimization, Scientific Reports 12 (1) (2022) 3902.
  36. [75]A. Li, Y. J. Zhang, Modeling intracellular transport and traffic jam in 3D neurons using PDE-constrained optimization, Journal of Mechanics 38 (2022) 44–59.
  37. [76]C. Godin, E. Costes, H. Sinoquet, A method for describing plant architecture which integrates topology and geometry, Annals of Botany 84 (3) (1999) 343–357. doi:10.1006/anbo.1999.0923.DOI
  38. [77]N. Pfeifer, B. Gorte, D. Winterhalder, Automatic reconstruction of single trees from terrestrial laser scanner data, in: International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 35, 2004, pp. 114–119.
  39. [78]N. D. Cornea, D. Silver, P. Min, Curve-skeleton properties, applications, and algorithms, IEEE Transactions on Visualization and Computer Graphics 13 (3) (2007) 530–548. doi:10.1109/TVCG.2007.1002.DOI
  40. [79]J. Cao, A. Tagliasacchi, M. Olson, H. Zhang, Z. Su, Point cloud skeletons via Laplacian-based contraction, in: 2010 Shape Modeling International Conference, IEEE, 2010, pp. 187–197. doi:10.1109/SMI.2010.25.DOI
  41. [80]H. Huang, S. Wu, D. Cohen-Or, M. Gong, H. Zhang, G. Li, B. Chen, L1-medial skeleton of point cloud, ACM Transactions on Graphics 32 (4) (2013) 65:1–65:8. doi:10.1145/2461912.2461913.
  42. [81]A. Tagliasacchi, T. Delauné, M. Spagnuolo, N. Amenta, A. Telea, 3D skeletons: A state-of-the-art report, Computer Graphics Forum 35 (2) (2016) 573–597. doi:10.1111/cgf.12865.DOI
  43. [82]W. Middleton, Q. Shu, F. Ludwig, Representing living architecture through skeleton reconstruction from point clouds, Scientific Reports 12 (1) (2022) 1549.
  44. [83]E. E. Medici, B. Brede, K. Calders, M. Disney, M. Herold, E. R. Lines, M. H. Nunes, P. Raumonen, M. Rautiainen, N. Saarinen, I. Starck, Expanding forest research with terrestrial LiDAR technology, Nature Communications 16 (1) (2025) 8853.
  45. [84]T. Medic, J. Bömer, S. Paulus, Challenges and recommendations for 3D plant phenotyping in agriculture using terrestrial lasers scanners, ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences 10 (2023) 1007–1014.
  46. [85]J. Rodriguez-Sanchez, J. L. Snider, K. Johnsen, C. Li, Cotton morphological traits tracking through spatiotemporal registration of terrestrial laser scanning time-series data, Frontiers in Plant Science 15 (2024) 1436120.
  47. [86]Y. Liang, K. Zhou, L. Cao, An advanced three-dimensional phenotypic measurement approach for extracting ginkgo root structural parameters based on terrestrial laser scanning, Frontiers in Plant Science 15 (2024) 1356078.
  48. [87]V. Bekkers, J. Evers, A. Lau, Improving the 3D representation of plant architecture and parameterization efficiency of functional–structural tree models using terrestrial LiDAR data, AoB Plants 17 (2) (2025) plae071.
  49. [88]T. T. Nguyen, D. C. Slaughter, N. Max, J. N. Maloof, N. Sinha, Structured light-based 3D reconstruction system for plants, Sensors 15 (8) (2015) 18587–18612.
  50. [89]J. Wang, Y. Zhang, R. Gu, Research status and prospects on plant canopy structure measurement using visual sensors based on three-dimensional reconstruction, Agriculture 10 (10) (2020) 462.
  51. [90]P. Song, Z. Li, M. Yang, Y. Shao, Z. Pu, W. Yang, R. Zhai, Dynamic detection of three-dimensional crop phenotypes based on a consumer-grade RGB-D camera, Frontiers in Plant Science 14 (2023) 1097725.
  52. [91]Z. Wang, H. Zhang, L. Bian, L. Zhou, Y. Ge, Accurate plant 3D reconstruction and phenotypic traits extraction via stereo imaging and multi-view point cloud alignment, Frontiers in Plant Science 16 (2025) 1642388.
  53. [92]S. Daiki, K. Noshita, A comparative study of plant phenotyping workflows based on three-dimensional reconstruction from multi-view images, BioRxiv (2024) 2024–03.
  54. [93]J. C. Rose, S. Paulus, H. Kuhlmann, Accuracy analysis of a multi-view stereo approach for phenotyping of tomato plants at the organ level, Sensors 15 (5) (2015) 9651–9665.
  55. [94]J. Martinez-Guanter, Á. Ribeiro, G. G. Peinatenos, M. Pérez-Ruiz, R. Gerhards, J. M. Bengochea-Guevara, J. Machleb, D. Andújar, Low-cost three-dimensional modeling of crop plants, Sensors 19 (13) (2019) 2883.
  56. [95]A. Fernández-Sarría, I. López-Cortés, J. Martí, J. Estornell, Estimation of walnut structure parameters using terrestrial photogrammetry based on structure-from-motion (SfM), Journal of the Indian Society of Remote Sensing 50 (10) (2022) 1931–1944.
  57. [96]H.-B. Choi, J.-K. Park, S. H. Park, T. S. Lee, NeRF-based 3D reconstruction pipeline for acquisition and analysis of tomato crop morphology, Frontiers in Plant Science 15 (2024) 1439086.
  58. [97]H. Kang, K. Hu, W. Ying, Y. Pan, C. Chen, High-fidelity 3D reconstruction of plants using neural radiance field, Available at SSRN 4639584.
  59. [98]D. Zhang, J. Gajardo, T. Medic, I. Katricioglu, M. Boss, N. Kirchgessner, A. Walter, L. Roth, Wheat3DGS: In-field 3D reconstruction, instance segmentation and phenotyping of wheat heads with gaussian splatting, in: Proceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 5360–5370.
  60. [99]M. A. Arshad, T. Z. Jubery, J. Afful, A. J. Ignasu, A. Balu, B. Ganapathysubramanian, S. Sarkar, A. Krishnamurthy, Evaluating neural radiance fields for 3D plant geometry reconstruction in field conditions, Plant Phenomics 6 (2024) 0235.
  61. [100]S. Joshi, A. Rairdin, E. Tranel, T. Z. Jubery, S. Sarkar, A. Singh, B. Ganapathysubramanian, A. Krishnamurthy, Drone-based 3D reconstruction of plants in field conditions using neural radiance fields (NeRFs), in: The First International Workshop on AI in Agriculture, 2026.
  62. [101]L. A. Stuart, D. M. Wells, J. A. Atkinson, S. Castle-Green, J. Walker, M. P. Pound, High-fidelity wheat plant reconstruction using 3D gaussian splatting and neural radiance fields, GigaScience 14 (2025) giaf022.
  63. [102]S. Wu, C. Hu, B. Tian, Y. Huang, S. Yang, S. Li, S. Xu, A 3D reconstruction platform for complex plants using OB-NeRF, Frontiers in Plant Science 16 (2025) 1449626.
  64. [103]J. Li, X. Qi, S. H. Nabavi, M. Liu, D. Chen, Q. Sun, X. Zhang, X. Yin, Z. Li, A survey on 3D reconstruction techniques in plant phenotyping: From classical methods to neural radiance fields (NeRF), 3D gaussian splatting (3DGS), and beyond, Plant Phenomics (2025) 100137.
  65. [104]N. Chebrolu, F. Magistri, T. Läbe, C. Stachniss, Registration of spatio-temporal point clouds of plants for phenotyping, PLoS one 16 (2) (2021) e0247243.
  66. [105]L. Stausberg, B. Jost, L. Klingbeil, H. Kuhlmann, A 3D surface reconstruction pipeline for plant phenotyping, Remote Sensing 16 (24) (2024) 4720.
  67. [106]E. Kimara, M. Hadadi, J. Godbersen, A. Balu, T. Z. Jubery, Y. Li, A. Krishnamurthy, P. S. Schnable, B. Ganapathysubramanian, MaizeField3D: A curated 3D point cloud and procedural model dataset of field-grown maize from a diversity panel, Plant Phenomics (2025) 100108.
  68. [107]A. Zarei, B. Li, J. C. Schnable, E. Lyons, D. Pauli, K. Barnard, B. Benes, PlantSegNet: 3D point cloud instance segmentation of nearby plant organs with identical semantics, Computers and Electronics in Agriculture 221 (2024) 108922.
  69. [108]S. Wu, W. Wen, B. Xiao, X. Guo, J. Du, C. Wang, Y. Wang, An accurate skeleton extraction approach from 3D point clouds of maize plants, Frontiers in Plant Science 10 (2019) 248.
  70. [109]A. Chaudhury, C. Godin, Skeletonization of plant point cloud data using stochastic optimization framework, Frontiers in Plant Science 11 (2020) 773.
  71. [110]S. Jay, G. Rabatel, X. Hadoux, D. Moura, N. Gorretta, In-field crop row phenotyping from 3D modeling performed using structure from motion, Computers and Electronics in Agriculture 110 (2015) 70–77.
  72. [111]W. Gérard, P. Burger, P. Casadebaig, N. Langlade, P. Debaeke, M. Devy, A. Herbulot, 3D plant phenotyping in sunflower using architecture-based organ segmentation from 3D point clouds, in: 5th International Workshop on Image Analysis Methods for the Plant Sciences, 2016.
  73. [112]T. T. Santos, L. V. Koenigkan, J. G. A. Barb edo, G. C. Rodrigues, 3D plant modeling: Localization, mapping and segmentation for plant phenotyping using a single hand-held camera, in: European Conference on Computer Vision, Springer, 2014, pp. 247–263.
  74. [113]Y. Zhuo, F. You, MIRAGE: Biomechanically interpretable 3D generation and reconstruction of maize plants, Computers and Electronics in Agriculture 248 (2026) 111774.
  75. [114]O. Stáva, S. Pirk, J. Kratt, B. Chen, R. Měch, O. Deussen, B. Benes, Inverse procedural modelling of trees, Computer Graphics Forum 33 (6) (2014) 118–131. doi:10.1111/cgf.12282.DOI
  76. [115]B. Li, N. A. Schwarz, W. Pahubicki, S. Pirk, B. Benes, Interactive invigoration: Volumetric modeling of trees with strands, ACM Transactions on Graphics (TOG) 43 (4) (2024) 1–13.
  77. [116]P. Prusinkiewicz, A. Lindenmayer, The Algorithmic Beauty of Plants, Springer, 1990. doi:10.1007/978-1-4613-8476-2.DOI
  78. [117]P. Prusinkiewicz, Modeling of spatial structure and development of plants: A review, Scientia Horticulturae 74 (1-2) (1998) 113–149.
  79. [118]M. Cieslak, P. Prusinkiewicz, Gillespie-lindenmayer systems for stochastic simulation of morphogenesis, In Silico Plants 1 (1) (2019) diz009.
  80. [119]J. Guo, H. Jiang, B. Benes, O. Deussen, X. Zhang, D. Lischinski, H. Huang, Inverse procedural modeling of branching structures by inferring L-systems, ACM Transactions on Graphics (TOG) 39 (5) (2020) 1–13.
  81. [120]J. S. Magnusson, A. Hils mann, P. Eisert, Towards L-system captioning for tree reconstruction, arXiv preprint arXiv:2305.06483.arxiv.org/abs/2305.06483
  82. [121]J. Bernard, I. McQuillan, Stochastic L-system inference from multiple string sequence inputs: J. bernard, i. mcquillan, Soft Computing 27 (10) (2023) 6783–6798.
  83. [122]J. B. Evers, V. Letort, M. Renton, M. Kang, Computational botany: Advancing plant science through functional–structural plant modelling, Annals of Botany 121 (5) (2018) 767–772.
  84. [123]H. O’Sullivan, P. Raumonen, P. Kaitanemi, J. Perttunen, R. Sievänen, Integrating terrestrial laser scanning with functional–structural plant models to investigate ecological and evolutionary processes of forest communities, Annals of Botany 128 (6) (2021) 663–684.
  85. [124]M. Debbagh, Y. Liu, Z. Zheng, X. Jiang, S. Sun, M. Lefsrud, Generative plant growth simulation from sequence-informed environmental conditions, in: IAPR Workshop on Artificial Neural Networks in Pattern Recognition, Springer, 2024, pp. 308–319.
  86. [125]A. J. Zhai, X. Wang, K. Li, Z. Jiang, J. Zhou, S. Wang, Z. Jin, K. Guan, S. Wang, CropCraft: Inverse procedural modeling for 3D reconstruction of crop plants, arXiv preprint arXiv:2411.09693.
  87. [126]Y. Liu, J. Guo, B. Benes, O. Deussen, X. Zhang, H. Huang, TreePartNet: Neural decomposition of point clouds for 3D tree reconstruction, ACM Transactions on Graphics 40 (6) (2021) 1–16. doi:10.1145/3478513.3480486.DOI
  88. [127]W. Wen, B. Li, B.-j. Li, X. Guo, A leaf modeling and multi-scale remeshing method for visual computation via hierarchical parametric vein and margin representation, Frontiers in Plant Science 9 (2018) 783.
  89. [128]F. Okura, 3D modeling and reconstruction of plants and trees: A cross-cutting review across computer graphics, vision, and plant phenotyping, Breeding Science 72 (1) (2022) 31–47.
  90. [129]M. Åkerblom, P. Raumonen, M. Kassalainen, E. Casella, Analysis of geometric primitives in quantitative structure models of tree stems, Remote Sensing 7 (4) (2015) 4581–4603.
  91. [130]J. Bloementhal, Modeling the mighty maple, ACM SIGGRAPH Computer Graphics 19 (3) (1985) 305–311. doi:10.1145/325165.325249.
  92. [131]Z. Wu, M. Zhou, X. Wang, Interactive modeling of 3D tree with ball B-spline curves, International Journal of Virtual Reality 8 (2) (2009) 101–107.
  93. [132]Z. Cui, H. Zhang, Y. Liu, J. Zhang, T. Yang, Y. Zuo, K. Lei, Constructing coupling model of generalized B-spline curve and crown (CMGBCC) to explore the 3D modeling of Chinese fir polymorphism, Forests 14 (6) (2023) 1267.
  94. [133]L. Yang, Z. Tang, P. Wu, J. Peng, J. Shi, L. Long, Visual modeling of rice root growth based on B-spline curve, Wuhan University Journal of Natural Sciences 27 (1) (2022) 85–92.
  95. [134]L. Yang, J. Peng, P. Shao, Modeling and visualization of rice roots based on morphological parameters, IEEE Access 8 (2020) 23407–23416.
  96. [135]A. Moola, A. Balu, A. Krishnamurthy, A. Pawar, THB-Diff: A GPU-accelerated differentiable programming framework for THB-splines, Engineering with Computers 40 (6) (2024) 3477–3493.
  97. [136]A. Moola, A. M. Corpuz, M. J. Burkhart, C. J. Ross, A. Mir, H. M. Burkhart, C.-H. Lee, M.-C. Hsu, A. Pawar, VALVEFIT: An analysis-suitable B-spline-based surface fitting framework for patient-specific modeling of tricuspid valves, Computer Methods in Applied Mechanics and Engineering 448 (2026) 118466.
  98. [137]A. D. Prasad, A. Balun, H. Shah, S. Sarkar, C. Hegde, A. Krishnamurthy, NURBS-DIFF: A differentiable programming module for NURBS, Computer-Aided Design 146 (2022) 103199.
  99. [138]T. J. R. Hughes, A. Cottrell, Y. J. Bazilevs, Isogeometric analysis: CAD, finite elements, NURBS, exact geometry and mesh refinement, Computer Methods in Applied Mechanics and Engineering 194 (39–41) (2005) 4135–4195. doi:10.1016/j.cma.2004.10.008.DOI
  100. [139]J. A. Cottrell, T. J. R. Hughes, Y. Bazilevs, Isogeometric analysis: Toward integration of CAD and FEA, John Wiley & Sons, 2009.
  101. [140]A. Li, A. Barati Farimani, Y. J. Zhang, Deep learning of material transport in complex neurite networks, Scientific Reports 11 (1) (2021) 11280.
  102. [141]X. Wei, X. Li, K. Qian, T. J. Hughes, Y. J. Zhang, H. Casquero, Analysis-suitable unstructured T-splines: Multiple extraordinary points per face, Computer Methods in Applied Mechanics and Engineering 391 (2022) 114494.
  103. [142]K. Joshi, A. D. Bhatt, Building a T-spline-based tri-variate heterogeneous model of human airways: A reverse engineering approach, Engineering with Computers 38 (5) (2022) 4085–4097.
  104. [143]K. Joshi, A. D. Bhatt, Development of multi-branch T-spline templates and its applications in reverse engineering, Computer-Aided Design & Applications 17 (3).
  105. [144]Y. Yu, Y. J. Zhang, K. Takizawa, T. E. T ezduyar, T. Sasaki, Anatomically realistic lumen motion representation in patient-specific space–time isogeometric flow analysis of coronary arteries with time-dependent medical-image data, Computational Mechanics 65 (2) (2020) 395–404.
  106. [145]Y. Zheng, F. Chen, Volumetric parameterization with truncated hierarchical B-splines for isogeometric analysis, Computer Methods in Applied Mechanics and Engineering 401 (2022) 115662.
  107. [146]T. W. Sederberg, J. Zheng, A. Bakenov, A. Nasri, T-splines and T-NURCCs, ACM Transactions on Graphics 22 (3) (2003) 477–484. doi:10.1145/882262.882295.DOI
  108. [147]L. Liu, Y. Zhang, T. J. R. Hughes, M. A. Scott, T. W. Sederberg, Volumetric T-spline construction using Boolean operations, Engineering with Computers 30 (4) (2014) 425–439. doi:10.1007/s00366-013-0346-6.DOI
  109. [148]A. Collin, G. Sangalli, T. Takacs, Analysis-suitable G¹ multi-patch parametrizations for C¹ isogeometric spaces, Computer Aided Geometric Design 47 (2016) 93–113. doi:10.1016/j.cagd.2016.05.009.DOI
  110. [149]Y. Bazilevs, V. M. Calo, Y. Zhang, T. J. R. Hughes, Isogeometric fluid–structure interaction analysis with applications to arterial blood flow, Computational Mechanics 38 (4–5) (2006) 310–322. doi:10.1007/s00466-006-0084-3.DOI
  111. [150]T. Martin, E. Cohen, M. Kirby, Volumetric parameterization and trivariate B-spline fitting using harmonic functions, in: Proceedings of the 2008 ACM symposium on Solid and Physical Modeling, 2008, pp. 269–280.
  112. [151]F. Massarwi, G. Elber, A B-spline based framework for volumetric object modeling, Computer-Aided Design 78 (2016) 36–47.
  113. [152]W. Yu, K. Zhang, S. Wan, X. Li, Optimizing polycube domain construction for hexahedral remeshing, Computer-Aided Design 46 (2014) 58–68.
  114. [153]Y. Yu, J. G. Liu, Y. J. Zhang, Hexdom: Polycube-based hexahedral-dominant mesh generation, in: Mesh generation and adaptation: Cutting-edge techniques, Springer, 2022, pp. 137–155.
  115. [154]Y. Yu, X. Wei, A. Li, J. G. Liu, J. He, Y. J. Zhang, HexGen and hex2Spline: Polycube-based hexahedral mesh generation and spline modeling for isogeometric analysis applications in LS-DYNA, in: Geometric Challenges in Isogeometric Analysis, Springer, 2022, pp. 333–363.
  116. [155]Y. Yu, Y. Fang, H. Tong, Y. J. Zhang, DL-Polycube: Deep learning enhanced generalized polycube method for high-quality hexahedral mesh generation and volumetric spline construction, Engineering with Computers 41 (3) (2025) 2069–2092.
  117. [156]N. Kowalski, F. Ledoux, M. L. Staten, S. J. Owen, Fun sheet matching: Towards automatic block decomposition for hexahedral meshes, Engineering with Computers 28 (3) (2012) 241–253.
  118. [157]X. Fang, W. Xu, H. Bao, J. Huang, All-hex meshing using closed-form induced polycube, ACM Transactions on Graphics (TOG) 35 (4) (2016) 1–9.
  119. [158]J. Gregson, A. Sheffer, E. Zhang, All-hex mesh generation via volumetric polycube deformation, in: Computer Graphics Forum, Vol. 30, Wiley Online Library, 2011, pp. 1407–1416.
  120. [159]K. Hu, Y. J. Zhang, Centroidal voronoi tessellation based polycube construction for adaptive all-hexahedral mesh generation, Computer Methods in Applied Mechanics and Engineering 305 (2016) 405–421.
  121. [160]L. Piegl, W. Tiller, The NURBS Book, Monographs in Visual Communications, Springer, Berlin, Heidelberg, 1995. doi:10.1007/978-3-642-97385-7. URL http://link.springer.com/10.1007/978-3-642-97385-7DOI
  122. [161]D. P. Kingma, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980.arxiv.org/abs/1412.6980
  123. [162]W. Boehm, H. Prautzsch, The insertion algorithm, Computer-Aided Design 17 (2) (1985) 58–59.
  124. [163]W. Wang, B. Jüttler, D. Zheng, Y. Liu, Computation of rotation minimizing frames, ACM Transactions on Graphics 27 (1) (2008) 1–18. doi:10.1145/1330511.1330513.DOI
  125. [164]H. Xu, N. Cossett, B. Chen, Knowledge and heuristic-based modeling of laser-scanned trees, ACM Trans. Graph. 26 (4) (2007) 19–es. doi:10.1145/1289603.1289610. URL https://doi.org/10.1145/1289603.1289610
  126. [165]B. M. van Marrewijk, T. van Daalen, K. Smoleňová, B. Xin, G. Polder, G. Kootstra, TomatoWUR: An annotated dataset of tomato plants to quantitatively evaluate segmentation, skeletonisation, and plant-trait extraction algorithms for 3D plant phenotyping, Data in Brief 61 (2025) 111852.
  127. [166]H. Dobbs, O. Batchelor, R. Green, J. Atlas, Smart-Tree: Neural medial axis approximation of point clouds for 3D tree skeletonization, in: Iberian Conference on Pattern Recognition and Image Analysis, Springer, 2023, pp. 351–362.

Paper details

vlnc:W_d75btacv
Version
v1 · latest
Published
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents