Three-dimensional leaf dynamic reconstruction method fusing branch structure identification and random disturbance

Abstract

The invention discloses a three-dimensional leaf dynamic reconstruction method fusing branch structure recognition and random disturbance, and belongs to the technical field of tree three-dimensional reconstruction, forest digital and visual simulation and computer programs in forestry scientific research. The method comprises the following steps: automatically judging a leaf growth region by identifying spatial geometric features (thickness of local branches), simulating leaf spatial distribution in a natural state in combination with a multi-angle disturbance strategy, and realizing design of different leaf profiles and regulation and control of a leaf area index (LAI) by adjusting leaf morphological parameters and quantity control. And automatic transition from a three-dimensional branch structure to a complete tree model is realized.

Description

Technical Field

[0001] The present invention belongs to the field of three-dimensional tree reconstruction, forest digitization and visualization simulation, and computer program technology in forestry science research, and relates to a three-dimensional leaf dynamic reconstruction method that integrates branch structure recognition and random disturbance, and relates to leaf model reconstruction technology, tree model library construction technology, tree model retrieval, and sample plot visualization simulation technology. Background Art

[0002] Efficient and accurate forest reconstruction is crucial for forest management simulation, forest ecological process simulation (such as photosynthetic radiation transfer simulation), and the development of forest digital twin systems. With the advancement of computer graphics and visualization simulation technologies, forest structure analysis, interactive management, and growth simulation are gradually being digitally implemented in virtual forest environments. This not only reduces the cost of trial and error in real life but also enhances intuitive understanding of forest structure and ecological mechanisms.

[0003] Currently, the 3D reconstruction technology for tree branches is relatively mature and has been widely used in many fields, including scientific research, ecological monitoring, film and television animation, and games. However, compared with branch modeling, modeling technology that adaptively constructs and restores the natural distribution of leaves based on real structural features still faces many challenges. Existing mainstream leaf modeling methods, such as SpeedTree, 0nyxTree and other regular growth modeling software, have low modeling efficiency and tend to have a uniform leaf density distribution, making it difficult to restore the heterogeneous structure of real forests; modeling methods based on L-systems mostly remain at the theoretical level and are difficult to connect with branch models obtained from real scans and reconstruction; and point cloud attachment or grid scattering methods generally lack spatial logic and attachment structure basis, and only achieve surface-level random attachment. Summary of the Invention

[0004] In response to the problems in the prior art, the present invention provides a three-dimensional leaf dynamic reconstruction method that integrates branch structure recognition and random disturbance.

[0005] A three-dimensional leaf dynamic reconstruction method that integrates branch structure recognition and random perturbation includes the following steps: based on a real tree trunk model, the leaf growth area is automatically determined by identifying its spatial geometric characteristics (local branch thickness), and a multi-angle perturbation strategy is combined to simulate the spatial distribution of leaves in a natural state. By adjusting the leaf morphological parameters and quantity control, different leaf shape designs and leaf area index (LAI) regulation are achieved, realizing an automated transition from a three-dimensional branch structure to a complete tree model.

[0006] The advantages of this invention include achieving precise positioning and naturalistic layout of leaf models in three-dimensional tree modeling. This method offers advantages such as high automation, strong adaptability, high fidelity, and rapid reconstruction speed, with accuracy fully meeting the requirements of grassroots forestry applications. Furthermore, this method can verify leaf modeling results in plot-level stand modeling, demonstrating its versatility and potential for widespread application.

[0007] The three-dimensional leaf dynamic reconstruction method proposed in the present invention integrates branch structure recognition and random perturbation. It can automatically determine the leaf growth area by accurately identifying the spatial geometric characteristics of tree branches (such as the local branch thickness), and simulate the spatial distribution of leaves in natural conditions by combining multi-angle perturbation strategies. Specifically, the present invention improves the authenticity, scientificity and reconstruction efficiency of leaf three-dimensional reconstruction. In the leaf parameter control part, the leaf number parameter, morphological parameter, and leaf tilt and displacement parameter can be independently adjusted to achieve high-fidelity reconstruction of leaves of different tree species under various leaf area indices (LAIs), realizing the automated transition from three-dimensional branch structure to complete tree model, significantly improving the naturalness and realism of tree modeling. In addition, the use of complete tree models after leaf reconstruction to carry out plot-level forest stand three-dimensional reconstruction and photosynthetic radiation simulation further demonstrated the realism, practicality and scalability of the leaves reconstructed by this method. This method not only improves the precise positioning and natural layout of the leaf model construction position, but also significantly improves the reconstruction speed while ensuring high simulation. Its technical effect fully meets the accuracy and speed requirements of three-dimensional tree modeling in grassroots forestry applications, and has good versatility and application promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. As shown in the figure:

[0009] Figure 1 Technical roadmap for the construction and verification of a three-dimensional leaf dynamic reconstruction algorithm integrating structure recognition and random perturbations.

[0010] Figure 2 Technology roadmap for adaptive leaf model reconstruction.

[0011] Figure 3(a) Rectangular leaf structure diagram of the leaf model.

[0012] Figure 3(b) is the rectangular leaf structure of the leaf model, which is the compressed form of Figure 3(a).

[0013] Figure 3(c) Elliptical leaf structure diagram of the leaf model.

[0014] Figure 3(d) is the structure of the elliptical leaf of the leaf model, which is the compressed form of Figure 3(c).

[0015] Figure 4 Visual display of some tree species.

[0016] Figure 5 Digital sample site construction technology roadmap.

[0017] Figure 6 The green coverage area of ​​the experimental area is the location map of 144 small sample plots (20m×20m).

[0018] Figure 7(a) shows the tree point cloud image after the branches and leaves are separated, showing the details of the Qinghai spruce leaf reconstruction.

[0019] Figure 7(b) shows the branch model of the Qinghai spruce after reconstruction of the leaves.

[0020] Figure 7(c) shows the complete tree model (colored) after reconstructing the leaves of Picea crassifolia.

[0021] Figure 8 Leaf reconstruction visualization interface diagram.

[0022] Figure 9 Complete Qinghai spruce model diagram (partial example).

[0023] Figure 10 A large sample site (240m×240m) 3D image (based on UE5) was constructed.

[0024] Figure 11 (a) is a map of one of the reconstructed digitized small sample plots.

[0025] Figure 11 (b) is the second map of the reconstructed digitized small sample plot.

[0026] Figure 11 (c) Three maps of the reconstructed digitized small sample plot.

[0027] Figure 11 (d) is a graph of photosynthetically active radiation in the small sample plot.

[0028] Figure 11 (e) is the second graph of photosynthetically active radiation of the small sample plot.

[0029] Figure 11 (f) Three graphs of photosynthetically active radiation of the small sample plot. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] Example 1: Figure 1 、 Figure 2 , Figure 3(a), Figure 3(b), Figure 3(c), Figure 3(d), Figure 4 、 Figure 5 、 Figure 6 , Figure 7(a), Figure 7(b), Figure 7(c), Figure 8 、 Figure 9 、 Figure 10 、 Figure 11 (a) Figure 11 (a) Figure 11 (b) Figure 11 (c) Figure 11 (d) Figure 11 (e) and Figure 11 Figure (f) shows an Adaptive Leaf Synthesis Algorithm (ALSA) method for 3D dynamic leaf reconstruction that combines branch structure recognition with random perturbations. This method addresses the difficulty of achieving natural leaf attachment in current leaf reconstruction algorithms after processing measured tree trunk data (such as point clouds or 3D meshes). This method effectively enhances the realism of leaf distribution, improves algorithmic efficiency and automation during the modeling process, and addresses the shortcomings of existing technologies in leaf morphology restoration and structural logic modeling. It provides key technical support for achieving high-precision 3D reconstruction of digital plots, forest stands, and even regional-scale forests.

[0032] First, a novel adaptive leaf synthesis algorithm (ALSA) that integrates branch structure recognition and random perturbations is used to reconstruct leaves from the branch model. Then, a digital plot is constructed based on the complete reconstructed tree model and a newly constructed attribute similarity matching algorithm, thereby further extending the reconstructed complete tree model to the digital plot reconstruction and verifying the leaf modeling effect.

[0033] A 3D leaf dynamic reconstruction method that integrates branch structure recognition and random perturbation, such as Figure 1As shown, the Adaptive Leaf Synthesis Algorithm (ALSA) leaf reconstruction algorithm was applied to the tree trunk model, a tree model library was established, and attribute matching and model screening were performed. (a) The Euclidean distance method was used to define the similarity function (NEDS), (b) the tree model library was matched with the single tree attribute library, and (c) the optimal matching model was screened. Using digitized DEM data, tree models, and ground material data, Unreal Engine blueprint encoding, terrain reconstruction, and tree generation were performed. A digital sample plot was established and the leaf reconstruction algorithm was tested.

[0034] Specifically, the original tree trunk obj model is first imported into the newly constructed Adaptive Leaf Synthesis Algorithm (ALSA). Adaptive leaves are added to the trunk model. The parameters involved are determined based on the tree species and leaf area index being constructed. The reconstructed complete tree model is then imported into the tree model library for future use. Next, attribute matching and model screening are performed. This step uses the Euclidean distance method to define a similarity function (NEDS) to further match the tree model library with the individual tree attribute library. The optimal matching model is selected based on the matching results, and the selected tree model is finally exported. Combining digital DEM data acquired by airborne radar and ground texture data captured by cameras, Unreal Engine Blueprint coding is used to digitally reconstruct the terrain and generate trees by reading the tree models at the corresponding locations on the digital terrain, ultimately completing the construction of a digital plot. Finally, the leaf reconstruction algorithm is tested by observing the reconstruction results of the digital plot and performing photosynthetic radiation simulation on leaves.

[0035] A 3D leaf dynamic reconstruction method that integrates branch structure recognition and random perturbation. The operation steps are as follows:

[0036] 1. Construct a 3D leaf dynamic reconstruction algorithm that integrates branch structure recognition and random perturbations

[0037] Build an adaptive leaf synthesis algorithm (ALSA), using the tree trunk model as the input model. The technical roadmap is as follows: Figure 2 As shown in the figure, the tree trunk model is imported into the Adaptive Leaf Synthesis Algori thm (ALSA) leaf reconstruction algorithm. The neighborhood search of the model vertex and the thickness threshold judgment are carried out through the KD tree spatial data structure to clearly reconstruct the branches. The leaf model construction (rectangular and elliptical leaves) is based on parameter control to carry out leaf layout. Random offset + Z-axis rotation + adaptive tilt are adopted to make the leaves attached and the model colored and visualized. After batch processing and model output, the tree model library (attributes, models) is obtained.

[0038] Specifically, the tree branch model is imported into the Adaptive Leaf Synthesis Algorithm (ALSA) leaf reconstruction algorithm. The algorithm first uses a KD tree spatial data structure to perform a neighborhood search. Then, a thickness threshold is applied to the branches of the branch model, and a neighborhood search of the model's vertices is performed to determine the branches to be reconstructed and their specific locations. The algorithm then constructs parameter-controlled leaf models (rectangular and elliptical leaves), primarily defining the leaf's morphological and structural parameters and quantity. Leaf layout is then performed, primarily through random offsets, Z-axis rotations, and adaptive tilting to achieve a more randomized and naturalistic layout. After determining the leaf reconstruction location, leaves are attached. Finally, the model is colored and visualized by defining leaf and branch colors and designing a visualization interface. The model can be freely rotated for easy viewing. The entire process enables batch processing of model leaf addition and model output. Finally, the model is imported into the tree model library (attributes, model). Attributes include parameters such as the tree model's diameter at breast height, height, crown width, crown area, and crown volume.

[0039] (1) Leaf shape design and parameter control

[0040] The present invention provides two morphological design schemes for leaf model construction: one is a rectangular leaf modeling method, which is constructed by constructing four vertices and splicing them from two triangular facets. It has the advantages of simple structure and high computational efficiency, and is suitable for coniferous tree species or rapid modeling scenarios with low requirements for leaf morphological accuracy. The other is an approximately elliptical leaf modeling method, which uses fan-shaped triangular facets to construct multiple boundary vertices distributed around a center point to simulate the natural contours of real broad-leaved tree leaves. The number of segments can be adjusted as needed to control accuracy. It is suitable for ecological simulation and visualization reconstruction tasks that require high fidelity. As shown in Figures 3(a), 3(b), 3(c), and 3(d). The leaf shape in Figure 3(a) is a long strip (rectangular), which is composed of two triangular facets. This leaf shape has a simple structure and can simulate most lanceolate leaves and some broad-leaved tree leaves. Figure 3(b) is the result of stretching the leaf in Figure 3(a). The leaf shape in Figure 3(c) is an elliptical shape, which is composed of several fan-shaped facets, and the edge of each facet has a certain curvature. This configuration can simulate most broad-leaved tree leaves. Figure 3(d) is the result of stretching Figure 3(c).

[0041] (2) Blade attachment and layout design

[0042] First, a neighborhood search is performed on the vertices of the branch model to estimate the local thickness characteristics of each vertex. By setting a thickness threshold, the vertex area representing the twigs is identified and screened as the potential attachment point of the leaves. On this basis, the leaves are automatically generated and randomly placed, including: a small spatial offset at the attachment point, a random angle rotation around the Z axis or other axes, and adaptive tilt angle control combined with the growth direction. This process effectively simulates the natural growth distribution state of real leaves. Finally, the generated leaf model is integrated with the original trunk model to construct a complete three-dimensional tree structure ( Figure 4 The reconstruction effects of some tree species are displayed), and the attribute information of each tree (such as tree height, diameter at breast height, crown width, crown area and crown volume) is stored in the tree model database together with its three-dimensional model, forming a standardized callable model library.

[0043] (3) Visual window design

[0044] After completing the reconstruction of the tree leaf model, color the model, and then use Python's built-in visualization library (0pen3D) to write a visualization window to view the modeling results, so as to facilitate viewing the modeling effect.

[0045] 2. Digital plot construction

[0046] To further verify the reconstruction and display effect of the leaf model reconstructed by the leaf reconstruction algorithm in the sample plot-level forest stand conditions, a multi-attribute similarity matching algorithm was constructed. The Normalized Euclidean Distance Similarity (NEDS) function was used to calculate the attribute difference values ​​(DBH, tree height, canopy characteristics) between each candidate model and the target tree. The model with the smallest NEDS was selected for model matching and replacement to ensure that the model and the corresponding tree geometry were highly similar. The matching results were used for the subsequent automatic construction of the digital forest stand. The similarity function calculation formula is as follows:

[0047]

[0048] Where: H d , W d , D d , A d , V d are the height, crown width, diameter at breast height, crown area and crown volume of the target tree; H m , W m , D m , A m , V mare the corresponding values ​​of a tree model in the model library; W1~w5 are weight coefficients (w1=0.4, w2=0.2, w3=0.1, w4=0.2, w5=0.1).

[0049] Import the DEM data into the visualization rendering engine to generate the sample plot terrain and apply material mapping to the terrain. Then use blueprint programming to read the attribute data of each tree in the entire sample plot and the matched 3D tree model library data, and automatically generate the 3D model of the tree at the corresponding coordinate position to realize the construction and display of the digital sample plot. The process is as follows: Figure 5 As shown, the single tree attribute data and the tree models selected from the tree model library are read through attributes and programmed in UE5 blueprint. The sample plots are then digitized through model, position, and attribute matching. The airborne radar point cloud data is extracted through the forest stand DEM and ground texture and then referenced with the actual forest stand to digitize the sample plots.

[0050] Specifically, we first extracted a forest stand DEM based on airborne radar point cloud data. The DEM file was converted into a .png format recognizable by UE5 and imported into UE5. A ground texture material was then added to the terrain to create a ground texture map. UE5's Blueprint programming capabilities were then used to read individual tree attribute data. Tree models selected from the tree model library were then read and matched to their model, location, and attributes. This allowed us to generate 3D tree models at the corresponding locations, identical to those in the real world. These steps enabled the construction of a digital plot and provided a real-world forest stand reference.

[0051] Example 2: Figure 1 、 Figure 2 , Figure 3(a), Figure 3(b), Figure 3(c), Figure 3(d), Figure 4 、 Figure 5 、 Figure 6 , Figure 7(a), Figure 7(b), Figure 7(c), Figure 8 、 Figure 9 、 Figure 10 、 Figure 11 (a) Figure 11 (a) Figure 11 (b) Figure 11 (c) Figure 11 (d) Figure 11 (e) and Figure 11 (f) shows an Adaptive Leaf Synthesis Algorithm (ALSA), a three-dimensional leaf dynamic reconstruction method that integrates branch structure recognition and random perturbation, and an example of digital Qinghai spruce leaf reconstruction and sample site reconstruction in Qilianshan National Forest Park, Qinghai Province.

[0052] 1. Data acquisition of branch and trunk models

[0053] The Qinghai spruce pure forest large plot (240m×240m, including 144 small plots, Figure 6 This example collects and fuses airborne and ground-based handheld LiDAR data, generating high-quality point clouds after resampling, denoising, and normalization. Individual tree segmentation is used to extract point cloud data for each tree, and attributes such as its diameter at breast height, height, canopy morphology, and spatial position are obtained. Non-defective point clouds are then selected for branch and leaf separation, and 3D reconstruction of the point cloud into branches and trunks is completed.

[0054] 2. Leaf reconstruction

[0055] The reconstructed branch model is imported into the leaf 3D reconstruction algorithm (ALSA) proposed in this invention to perform leaf modeling. First, the parameter threshold is set to clearly define the leaves and reconstruct the branches. Then, the parameters (rectangular leaves, length, width, and number of leaves) are set according to the actual Qinghai spruce leaf morphology. Then, the leaf model is set to rotate around the Z axis (angle range 355° to 360°) and then tilted by about 60°. A small random displacement is superimposed, so that the generated Qinghai spruce leaves have the characteristics of natural random distribution ( Figure 7c ), based on the visualization function of the algorithm, you can freely rotate to view the quality of the generated model ( Figure 8 All reconstructed trunk and leaf data are combined to form a complete 3D tree model ( Figure 9 ), and store it in the tree model library, while recording the attribute parameters of each model.

[0056] Verification of the effects of digital forest stand reconstruction and leaf reconstruction

[0057] Based on the extracted single tree attribute data, the NEDS similarity function is used to automatically select the most matching 3D tree model from the model library to achieve accurate model replacement. The DEM data obtained by the airborne lidar is combined with the shp files of 144 20m×20m small sample plots to cut and reconstruct the terrain. The terrain and textures are imported into UE5, and the tree models are batch loaded and light rendered through the blueprint to complete the digital modeling of the large-scale forest stand. Figure 10 Within the small plot, a three-dimensional radiation transfer model was further used to render the reconstruction results and simulate the absorbed photosynthetic active radiation ( Figure 11 (a) Figure 11 (b) Figure 11 (c) Figure 11 (d) Figure 11 (e) and Figure 11(f)). The digital plot construction process described above further validated the expanded application capabilities of the adaptive foliage reconstruction algorithm of the present invention at the forest stand scale. The foliage models generated by this algorithm exhibit realistic and natural spatial distribution and good visualization effects. Furthermore, the three-dimensional radiation transfer simulation results also indirectly reflect the potential application value of this algorithm in areas such as ecological process modeling and structure-radiation interaction research. It has broad prospects for promotion and application in multiple fields, including digital twin forests, ecological simulation, and forestry visualization.

[0058] like Figure 11 (a) Figure 11 (b) Figure 11 (c) Figure 11 (d) Figure 11 (e) and Figure 11 As shown in (f), the yellow arrow indicates the incident direction of light. The difference in leaf color of different trees reflects the spatial heterogeneity of light in the sample plot, which indirectly confirms the naturalness and accuracy of leaf reconstruction, reflecting good visualization effects and potential for expanded applications.

[0059] Judging from the results, the three-dimensional leaf dynamic reconstruction method that integrates branch structure recognition and random perturbations proposed in the present invention is effective. It not only improves the realism of the leaf attachment position and the overall simulation of the model, but also realizes the natural difference in tree morphology in plot modeling. This method supports efficient, fast, and accurate automated modeling and visualization of some coniferous and broad-leaved tree species and structural conditions, and can effectively improve the detail expression and immersiveness of forest structure simulation. Further digital plot construction effectively tested the display effect and practical application significance of the leaf reconstruction algorithm in large-scale modeling, showing good scalability and engineering promotion potential.

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A three-dimensional leaf dynamic reconstruction method integrating branch structure recognition and random perturbation, characterized in that: The method includes the following steps: based on the real tree branch model, the leaf growth area is automatically determined by identifying its spatial geometric characteristics (local branch thickness), and the multi-angle perturbation strategy is combined to simulate the spatial distribution of leaves in the natural state. By adjusting the leaf morphological parameters and quantity control, different leaf shape designs and leaf area index (LAI) regulation are achieved, realizing the automatic transition from three-dimensional branch structure to complete tree model.

2. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 1 is characterized in that: It also contains the following steps: The Adaptive Leaf Synthesis Algorithm (ALSA), which combines branch structure recognition with random perturbations, was used to reconstruct leaves from the branch model. A digital plot was constructed based on the complete reconstructed tree model and the constructed attribute similarity matching algorithm. This was used to further extend the reconstructed complete tree model to the digital plot reconstruction and verify the leaf modeling effect.

3. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 1 is characterized in that: The Adaptive Leaf Synthesis Algorithm (ALSA) was used to reconstruct tree trunk models. A tree model library was established, and attribute matching and model screening were performed. (a) The Euclidean distance definition similarity function (NEDS) was used, (b) the tree model library was matched with the single tree attribute library, and (c) the optimal matching model was screened. Using digitized DEM data, tree models, and ground material data, Unreal Engine blueprint encoding, terrain reconstruction, and tree generation were performed. A digital sample plot was established and the leaf reconstruction algorithm was tested.

4. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 1 is characterized in that: First, the original tree trunk model is imported into the newly constructed Adaptive Leaf Synthesis Algorithm (ALSA), and leaves are adaptively added to the trunk model. The parameters involved in this part need to be determined according to the constructed tree species and leaf area index. The reconstructed complete tree model is imported into the tree model library for standby use. Then, attribute matching and model screening are performed. This step uses the Euclidean distance method to define the similarity function (NEDS) to further match the tree model library with the individual tree attribute library. The optimal matching model is screened based on the matching results. Finally, the screened tree model is exported and combined with the digital DEM data obtained by the airborne radar and the ground material data obtained by the camera. The Unreal Engine Blueprint code is used to realize digital terrain reconstruction and read the tree model to the corresponding position of the digital terrain to realize tree generation, thereby finally realizing the construction of the digital sample plot. Finally, the leaf reconstruction algorithm is tested by observing the reconstruction effect of the digital sample plot and conducting photosynthetic radiation simulation on the leaves.

5. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 3 is characterized in that: An Adaptive Leaf Synthesis Algorithm (ALSA) was constructed. The tree trunk model was used as the input model and imported into the Adaptive Leaf Synthesis Algorithm (ALSA). The neighborhood search of the model vertices and the thickness threshold judgment were carried out through the KD tree spatial data structure to clearly reconstruct the branches. The leaf layout was carried out based on the parameter-controlled leaf model construction (rectangular and elliptical leaves). Random offset + Z-axis rotation + adaptive tilt were adopted to enable leaf attachment and model coloring and visualization. After batch processing and model output, a tree model library (attributes, models) was obtained.

6. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 1 is characterized in that: The tree branch model is imported into the Adaptive Leaf Synthesis Algorithm (ALSA) leaf reconstruction algorithm. The algorithm first uses a KD tree spatial data structure to perform a neighborhood search. Then, a thickness threshold is applied to the branches of the branch model and a neighborhood search of the model vertices is performed to determine the reconstructed branches and their specific reconstruction locations. The algorithm then constructs a parameter-controlled leaf model (rectangular and elliptical leaves). This step primarily defines the leaf morphological and structural parameters and quantity parameters. Leaf layout is then performed, primarily through random offsets, Z-axis rotations, and adaptive tilting to make the leaf layout more random and natural. After determining the leaf reconstruction location, leaves are attached. Finally, the model is colored and visualized by defining leaf and branch colors and designing a visualization interface. The model angle can be freely rotated in the visualization interface, enabling batch processing of model leaf additions and model output. Finally, the model is imported into the tree model library (attributes, models). Attributes include parameters such as the tree model's diameter at breast height, height, crown width, crown area, and crown volume.

7. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 3 is characterized in that: The tree models selected from the tree model library and the single tree attribute library are read through attributes and programmed with UE5 blueprints. The digitized plots are then made through model, position, and attribute matching. The airborne radar point cloud data is extracted through forest stand DEM and ground texture, and then referenced with the actual forest stand for digitized plots.

8. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 7 is characterized in that: First, forest stand DEM extraction is realized based on airborne radar point cloud data. The DEM file is converted into a .png format file recognizable by UE and imported into UE5. The ground texture material is imported into the terrain to realize ground texture mapping. The UE5 blueprint programming function is used to read the attribute data of single tree, and the tree models selected by the tree model library are read to match the model, position and attributes, so as to generate a three-dimensional tree model at the corresponding position that is the same as the corresponding position in the real world.

9. The method for 3D leaf dynamic reconstruction integrating branch structure recognition and random perturbation according to claim 4, characterized in that: The matching results are used for the subsequent automatic construction of digital forest stands, and the similarity function calculation formula is as follows: Where: H d , W d , D d , A d , V d are the height, crown width, diameter at breast height, crown area and crown volume of the target tree; H m , W m , D m , A m , V m are the corresponding values ​​of a tree model in the model library; w1~w5 are weight coefficients (w1=0.4, w2=0.2, w3=0.1, w4=0.2, w5=0.1).

Images