Corn population canopy semantic three-dimensional reconstruction method and system

Abstract

The invention provides a corn population canopy semantic three-dimensional reconstruction method and system. The method comprises the following steps: determining a plant azimuth angle and a plant height corresponding to a corn plant at each growth position in a to-be-reconstructed corn population canopy point cloud; according to the growth period information, the plant azimuth angle and the plant height, obtaining a plurality of target node unit three-dimensional grid models corresponding to each corn plant in a preset corn three-dimensional node unit template library; constructing an initial plant three-dimensional model according to the plurality of target section unit three-dimensional grid models; and based on an improved Chamfer distance formula, sequentially carrying out iterative optimization on the initial plant three-dimensional model and the target node unit three-dimensional network model to obtain a target plant three-dimensional model, and constructing a corn population canopy semantic three-dimensional reconstruction model according to the target plant three-dimensional model. According to the method, the corn population canopy three-dimensional model of which the whole corn population canopy contains semantic information and complements deficient organs at the middle and lower parts of the canopy can be obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method and system for semantic three-dimensional reconstruction of a maize population canopy. Background Art

[0002] The crop population structure is an organizational system that performs crop production functions and has become a key element in the fields of crop cultivation, crop phenomics, and plant functional-structural models. It is of crucial significance for promoting crop plant type breeding work, optimizing crop planting density to achieve high yields, and improving cultivation management measures such as crop photosynthesis efficiency.

[0003] Current three-dimensional reconstruction methods for maize populations have high requirements for the quality of the input point cloud data. However, due to the serious cross-occlusion phenomenon within the maize population, existing data acquisition means are difficult to obtain point cloud data that meets the requirements of three-dimensional reconstruction. Especially in high-density maize populations, it poses greater challenges to the three-dimensional reconstruction work.

[0004] Therefore, there is an urgent need for a method and system for semantic three-dimensional reconstruction of a maize population canopy to solve the above problems. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and system for semantic three-dimensional reconstruction of a maize population canopy.

[0006] The present invention provides a method for semantic three-dimensional reconstruction of a maize population canopy, including: Determining the plant azimuth angle and plant height corresponding to each maize plant at each growth position in the point cloud of the maize population canopy to be reconstructed; According to the growth stage information, the plant azimuth angle, and the plant height, obtaining a plurality of target joint unit three-dimensional grid models corresponding to each maize plant in a preset maize three-dimensional joint unit template library, where the growth stage information is the current growth stage of the maize plants in the point cloud of the maize population canopy to be reconstructed; the preset maize three-dimensional joint unit template library is composed of joint unit three-dimensional grid models corresponding to different maize joint units, and the joint unit three-dimensional grid model is composed of maize organ three-dimensional grids with semantic information; Constructing an initial plant three-dimensional model corresponding to each maize plant in the point cloud of the maize population canopy to be reconstructed according to the plurality of target joint unit three-dimensional grid models; Based on the improved Chamfer distance formula, the initial three-dimensional plant model and the target three-dimensional network model of the joint unit are iteratively optimized in sequence to obtain the target three-dimensional plant model, and a three-dimensional semantic reconstruction model of the maize population canopy corresponding to the point cloud of the maize population canopy to be reconstructed is constructed according to the target three-dimensional plant model; wherein, the improved Chamfer distance formula is constructed based on the similarity distance between the initial three-dimensional plant model and the three-dimensional point cloud of the plant before reconstruction, and the three-dimensional point cloud of the plant before reconstruction is the point cloud of the maize plant corresponding to the initial three-dimensional plant model in the point cloud of the maize population canopy to be reconstructed.

[0007] According to a method for three-dimensional semantic reconstruction of a maize population canopy provided by the present invention, the determination of the plant azimuth angle and plant height corresponding to the maize plants at each growth position in the point cloud of the maize population canopy to be reconstructed includes: Obtain the seedling maize azimuth angle and the coordinates of the center point of the seedling maize corresponding to each maize seedling plant point cloud in the maize population seedling data, wherein the maize population seedling data is point cloud data or image data of maize plants in the seedling stage, and the maize population seedling data and the point cloud of the maize population canopy to be reconstructed belong to the same maize plant area; Based on direct linear transformation, perform temporal alignment of each maize seedling plant point cloud in the maize population seedling data with each maize plant point cloud in the point cloud of the maize population canopy to be reconstructed; According to the temporal alignment result, take the seedling maize azimuth angle of each maize seedling plant point cloud in the maize population seedling data as the plant azimuth angle of the corresponding maize plant point cloud in the point cloud of the maize population canopy to be reconstructed; take the coordinates of the center point of the seedling maize of each maize seedling plant point cloud in the maize population seedling data as the center coordinates of the plant point cloud of the corresponding maize plant point cloud in the point cloud of the maize population canopy to be reconstructed; According to the plant azimuth angle and the center coordinates of the plant point cloud, determine the cropping area of each maize plant point cloud in the point cloud of the maize population canopy to be reconstructed, and obtain the plant height according to the size information of the cropping area.

[0008] According to a method for three-dimensional semantic reconstruction of a maize population canopy provided by the present invention, the obtaining of a plurality of target three-dimensional network models of joint units corresponding to each maize plant in a preset three-dimensional maize joint unit template library according to the growth period information, the plant azimuth angle, and the plant height includes: According to the plant height and the growth period information, determine the number of maize joint units and the growth height of the maize joint units required for each maize plant; According to the plant azimuth angle, determine the azimuth angle of the joint unit leaves; Based on the growth period information, the number of maize node units, the growth height of the maize node units, and the azimuth angle of the node unit leaves, match the node unit three-dimensional grid models in the preset maize three-dimensional node unit template library to obtain a plurality of the target node unit three-dimensional grid models corresponding to each of the maize plants.

[0009] According to a method for semantic three-dimensional reconstruction of a maize population canopy provided by the present invention, constructing an initial plant three-dimensional model corresponding to each of the maize plants in the point cloud of the maize population canopy to be reconstructed based on the plurality of target node unit three-dimensional grid models includes: Based on the growth height of the maize node units corresponding to each of the target node unit three-dimensional grid models of the maize plant, translate and align each of the target node unit three-dimensional grid models according to the height position, and rotate each of the target node unit three-dimensional grid models by the azimuth angle based on a preset node unit azimuth angle, so as to splice the target node unit three-dimensional grid models after translation alignment and azimuth angle rotation to obtain the initial plant three-dimensional model.

[0010] According to a method for semantic three-dimensional reconstruction of a maize population canopy provided by the present invention, iteratively optimizing the initial plant three-dimensional model based on an improved Chamfer distance formula includes: Based on a preset plant point cloud rotation range and a preset plant point cloud rotation step size, rotate the current initial plant three-dimensional model around the central axis of the maize plant; Based on the improved Chamfer distance formula, calculate the first Chamfer distance between the initial plant three-dimensional model after each rotation and the pre-reconstruction plant three-dimensional point cloud; Determine the rotated initial plant three-dimensional model corresponding to the smallest first Chamfer distance as the optimized target plant three-dimensional model.

[0011] According to a method for semantic three-dimensional reconstruction of a maize population canopy provided by the present invention, the method further includes: Based on a preset node unit point cloud rotation range and a preset node unit point cloud rotation step size, rotate the current target node unit three-dimensional grid model in the optimized initial plant three-dimensional model around the node central axis; Based on the improved Chamfer distance formula, calculate the second Chamfer distance between the target node unit three-dimensional grid model after each rotation and the pre-reconstruction node unit three-dimensional point cloud, where the pre-reconstruction node unit three-dimensional point cloud is the node unit point cloud corresponding to the target node unit three-dimensional grid model in the point cloud of the maize population canopy to be reconstructed; Determine the rotated target joint unit three-dimensional grid model corresponding to the smallest of the second Chamfer distances as the target joint unit three-dimensional grid model after joint unit angle optimization; Determine the inter-joint connection positions between the target joint unit three-dimensional grid models after optimizing each joint unit angle, and perform interpolation processing on the inter-joint connection positions; Perform splicing processing based on the target joint unit three-dimensional grid model after interpolation processing to construct the target plant three-dimensional models corresponding to each corn plant in the to-be-reconstructed corn population canopy point cloud.

[0012] According to a semantic three-dimensional reconstruction method for corn population canopy provided by the present invention, the improved Chamfer distance formula is: ; Wherein, represents the three-dimensional point cloud of the plant before reconstruction or the three-dimensional point cloud of the joint unit before reconstruction, represents the current initial plant three-dimensional model or the current target joint unit three-dimensional grid model; is the vertex in, is the point in, represents the Chamfer distance.

[0013] The present invention also provides a semantic three-dimensional reconstruction system for corn population canopy, including: A processing module for determining the plant azimuth angle and plant height corresponding to each corn plant at each growth position in the to-be-reconstructed corn population canopy point cloud; A joint unit matching module for obtaining a plurality of target joint unit three-dimensional grid models corresponding to each corn plant in a preset corn three-dimensional joint unit template library according to the growth stage information, the plant azimuth angle, and the plant height, wherein the growth stage information is the current growth stage of the corn plants in the to-be-reconstructed corn population canopy point cloud; the preset corn three-dimensional joint unit template library is composed of joint unit three-dimensional grid models corresponding to different corn joint units, and the joint unit three-dimensional grid model is composed of corn organ three-dimensional grids with semantic information; A plant three-dimensional model construction module for constructing an initial plant three-dimensional model corresponding to each corn plant in the to-be-reconstructed corn population canopy point cloud according to the plurality of target joint unit three-dimensional grid models; The semantic three-dimensional model construction module is used to iteratively optimize the initial plant three-dimensional model and the target node unit three-dimensional network model in sequence based on the improved Chamfer distance formula to obtain the target plant three-dimensional model, and construct the corn population canopy semantic three-dimensional reconstruction model corresponding to the point cloud of the corn population canopy to be reconstructed according to the target plant three-dimensional model; wherein, the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction, and the three-dimensional point cloud of the plant before reconstruction is the corn plant point cloud corresponding to the initial plant three-dimensional model in the point cloud of the corn population canopy to be reconstructed.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the corn population canopy semantic three-dimensional reconstruction method as described in any one of the above.

[0015] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the corn population canopy semantic three-dimensional reconstruction method as described in any one of the above.

[0016] The corn population canopy semantic three-dimensional reconstruction method and system provided by the present invention select a plurality of target node unit three-dimensional grid models with semantic information corresponding to each corn plant in the point cloud of the corn population canopy to be reconstructed through the growth position, azimuth angle, and height of each corn plant in the point cloud of the corn population canopy to be reconstructed. Then, using these selected target node unit three-dimensional grid models, a three-dimensional model of each corn plant is constructed, so as to obtain a three-dimensional model of the corn population canopy that contains semantic information and complements the missing organs in the middle and lower parts of the canopy. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of the corn population canopy semantic three-dimensional reconstruction method provided by the present invention; Figure 2 It is a schematic structural diagram of the corn population canopy semantic three-dimensional reconstruction system provided by the present invention; Figure 3 It is a schematic structural diagram of the electronic device provided by the present invention. Detailed Embodiments

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] The acquisition and reconstruction of existing three-dimensional data of maize population canopies mainly adopt the following several methods: I. In-situ field acquisition method based on a three-dimensional digitizer: This method uses a digitizer to directly obtain the three-dimensional coordinate point set of the maize population in the field, achieving precise three-dimensional data acquisition and model construction. Although its reconstruction accuracy is high and the data is rich in semantic information, with less dependence on subsequent processing algorithms, due to the need to obtain the feature points of all organs of the plants, the working efficiency is extremely low, and it is difficult to apply to the construction of three-dimensional models of large-scale populations.

[0021] II. Modeling method based on statistical models or optimization calculations: First, construct a three-dimensional model of a single maize plant, and then generate a three-dimensional model of the population canopy by replication and translation. However, this method is highly mechanical, lacks a sense of reality, and cannot accurately reflect the spatial heterogeneity in the canopy. In addition, for the method using statistical models, for example, constructing a three-dimensional model of a maize population based on the t-distribution function, although it can combine measured data to generate the main plant type parameters of each plant in the population, it still needs to rely on the population structure information extracted by manual interaction or images to generate a geometric model. And the method based on optimization calculations maximizes light energy interception by adjusting the official azimuth angle of the intelligent regulator to achieve three-dimensional model construction, but these methods are more about three-dimensional simulation of the canopy rather than true three-dimensional reconstruction.

[0022] III. Reconstruction method based on three-dimensional point clouds: Use equipment carried by drones, backpack radars, ground-based radars, or orbital phenotyping platforms to obtain three-dimensional point cloud data of maize population canopies at the plot scale. However, due to the problem of cross-occlusion, there are a large number of missing data in the obtained point cloud data, and there is a lack of semantic information such as plants and organs. Although semantic three-dimensional reconstruction of maize populations can be achieved through methods such as point cloud segmentation of population-plant and plant-organ, and grid generation, these methods have extremely high requirements for the integrity and data accuracy of the point cloud. Due to the missing and low resolution of the maize population point cloud, the feasibility of this method is limited.

[0023] Aiming at the problems existing in the above-mentioned existing technologies, the present invention takes the three-dimensional point cloud data of maize population canopies obtained by an orbital phenotyping platform in the field as the input, and through the analysis of time-series data and iterative optimization calculation methods, reconstructs a three-dimensional model of maize population canopies containing semantic information, providing a core algorithm for maize population canopy phenotyping analysis and functional-structure model construction.

[0024] Figure 1 Schematic flow chart of the maize population canopy semantic 3D reconstruction method provided by the present invention, as Figure 1 shown, the present invention provides a maize population canopy semantic 3D reconstruction method, including: Step 101, determine the plant azimuth angle and plant height corresponding to the maize plants at each growth position in the point cloud of the maize population canopy to be reconstructed.

[0025] In the present invention, first, it is necessary to identify and determine the specific growth position of each maize plant in the population from the acquired point cloud data of the maize population canopy to be reconstructed. The point cloud of the maize population canopy to be reconstructed is a three-dimensional data set composed of countless points, and these points represent the surface morphology of the maize population canopy. By analyzing and processing these point cloud data, the coordinate positions of each maize plant in three-dimensional space can be located.

[0026] Next, for each maize plant whose growth position has been determined, it is necessary to further calculate its plant azimuth angle. The plant azimuth angle refers to the deflection angle of the maize plant relative to a reference direction (such as the due north direction) on the horizontal plane. This angle can determine the orientation of the maize plant in the population, thereby obtaining the mutual occlusion relationship, light distribution, ventilation conditions, etc. among the plants.

[0027] Furthermore, measure the height of each maize plant. The plant height is the vertical distance from the ground to the top of the maize plant (usually the ear or the highest point of the leaves). By measuring the plant height, the growth status of the maize population, the competition relationship among the plants, and the overall structural characteristics of the population can be obtained.

[0028] Step 102, according to the growth stage information, the plant azimuth angle, and the plant height, obtain a plurality of target joint unit three-dimensional grid models corresponding to each of the maize plants in the preset maize three-dimensional joint unit template library, wherein the growth stage information is the growth stage at which the maize plants in the point cloud of the maize population canopy to be reconstructed are currently in; the preset maize three-dimensional joint unit template library is composed of joint unit three-dimensional grid models corresponding to different maize joint units, and the joint unit three-dimensional grid model is composed of maize organ three-dimensional grids with semantic information.

[0029] In the present invention, after determining the growth position, azimuth angle, and height of each maize plant in the point cloud of the maize population canopy to be reconstructed, it is also necessary to obtain the current growth stage of the maize population in the point cloud of the maize population canopy to be reconstructed, that is, the growth stage information. The growth stage information describes which stage of its life cycle the maize plant is currently in, such as the seedling stage, jointing stage, tasseling stage, and filling stage, etc. Maize plants in different growth stages have different morphologies, structures, and organ development degrees.

[0030] In order to more accurately reconstruct the three-dimensional model of the corn population canopy, the present invention pre-establishes a three-dimensional corn node unit template library. This three-dimensional corn node unit template library is composed of three-dimensional grid models corresponding to different corn node units (such as leaf nodes, ear nodes, etc.). Each three-dimensional grid model of the node unit is modeled according to the real morphology and structure of the corn organs and has semantic information. Through the semantic information, it can be determined which corn organ (such as leaf, stem, ear, etc.) each three-dimensional grid of the corn organ represents.

[0031] Furthermore, after obtaining the growth period information, azimuth angle, and height of each corn plant, corresponding multiple target node unit three-dimensional grid models can be searched for and obtained in the preset three-dimensional corn node unit template library according to this information. Specifically, select the node unit model suitable for the current growth stage according to the growth period information of the corn; then, adjust the orientation of the model according to the azimuth angle of the plant so that it is consistent with the orientation of the real plant; finally, adjust the size of the model according to the height of the plant so that it matches the height of the real plant. In this way, a set of accurate and realistic three-dimensional grid models can be obtained for each corn plant. These models combined together can constitute the three-dimensional reconstruction model of the entire corn population canopy. This model not only has a high degree of realism and accuracy but also contains rich semantic information, providing strong support for subsequent analysis and applications.

[0032] Step 103: Construct an initial plant three-dimensional model corresponding to each corn plant in the point cloud of the corn population canopy to be reconstructed according to the multiple target node unit three-dimensional grid models.

[0033] In the present invention, through the above embodiments, corresponding multiple target node unit three-dimensional grid models are found for each corn plant in the preset three-dimensional corn node unit template library. These target node unit three-dimensional grid models represent the three-dimensional morphology of each organ (such as leaves, stems, ears, etc.) of the corn plant at different growth stages, different azimuth angles, and heights.

[0034] Furthermore, use these target node unit three-dimensional grid models to construct the initial three-dimensional model of each corn plant. Specifically, combine and splice the selected three-dimensional grid models of the node units according to the real plant structure and growth law. For example, install the leaf model on the stem model according to the actual growth position and angle, and place the ear model at the top of the plant, etc. In this way, a complete initial three-dimensional model with real morphology and structure can be constructed for each corn plant.

[0035] After the initial 3D models are constructed for each maize plant in the maize population canopy point cloud to be reconstructed, these models are combined together to form an overall semantic 3D reconstruction model of the maize population canopy. This model not only contains the 3D shape and structural information of the maize population canopy, but also has rich semantic information, that is, it can directly determine which maize organ each part in the model represents, as well as their relative positions and relationships.

[0036] Step 104: Use the improved Chamfer distance formula to iteratively optimize the initial plant 3D model and the target node unit 3D network model in sequence to obtain the target plant 3D model, and construct the semantic 3D reconstruction model of the maize population canopy corresponding to the maize population canopy point cloud to be reconstructed based on the target plant 3D model; wherein, the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant 3D model and the 3D point cloud of the plant before reconstruction, and the 3D point cloud of the plant before reconstruction is the point cloud of the maize plant corresponding to the initial plant 3D model in the maize population canopy point cloud to be reconstructed.

[0037] In the present invention, the initial plant 3D model directly constructed through the above embodiments realizes the restoration of each plant in the population according to plant height and growth position, but the positions of the whole plant and each node unit still need to be further optimized and adjusted according to the point cloud to improve the reconstruction accuracy.

[0038] Therefore, based on the preset rotation range and rotation step of the plant point cloud, the present invention rotates the current initial plant 3D model around the central axis of the maize plant on the basis of obtaining the initial plant 3D model. After each rotation, based on the improved Chamfer distance formula, calculate the improved Chamfer distance between the current initial plant 3D model and the 3D point cloud of the plant before reconstruction, and select the rotation angle corresponding to the minimum distance as the final rotation angle of the current initial plant 3D model, so that after the current initial plant 3D model is rotated by this final rotation angle, an optimized initial plant 3D model is obtained.

[0039] Further, based on the preset node unit point cloud rotation range and the preset node unit point cloud rotation step size, rotate the current target node unit three-dimensional grid model in the optimized initial plant three-dimensional model around the node central axis. After each rotation of the node unit three-dimensional grid model, use the improved Chamfer distance formula to calculate the Chamfer distance between the currently rotated target node unit three-dimensional grid model and the node unit three-dimensional point cloud before reconstruction, and then determine the rotated node unit model corresponding to the minimum Chamfer distance as the target node unit three-dimensional grid model after node unit angle optimization. Finally, after determining that all node unit three-dimensional grid models are optimized, obtain the target plant three-dimensional model, and then complete the construction process of the corn population canopy semantic three-dimensional reconstruction model through the target plant three-dimensional model.

[0040] The corn population canopy semantic three-dimensional reconstruction method provided by the present invention selects a plurality of corresponding target node unit three-dimensional grid models with semantic information from a pre-corn three-dimensional node unit template library according to the growth positions, azimuth angles, and heights of each corn plant in the point cloud of the corn population canopy to be reconstructed, and then uses these selected target node unit three-dimensional grid models to construct a three-dimensional model of each corn plant, so as to obtain a corn population canopy three-dimensional model that includes semantic information and complements the missing organs in the middle and lower parts of the canopy of the entire corn population.

[0041] Based on the above embodiments, the determination of the plant azimuth angle and plant height corresponding to each corn plant at each growth position in the point cloud of the corn population canopy to be reconstructed includes: Obtain the seedling-stage corn azimuth angle and the seedling-stage corn center point coordinates corresponding to each corn seedling plant point cloud in the corn population seedling-stage data, where the corn population seedling-stage data is point cloud data or image data of corn plants in the seedling stage, and the corn population seedling-stage data and the point cloud of the corn population canopy to be reconstructed belong to the same corn plant area; Based on direct linear transformation, perform temporal alignment on each corn seedling plant point cloud in the corn population seedling-stage data and each corn plant point cloud in the point cloud of the corn population canopy to be reconstructed; According to the temporal alignment result, use the seedling-stage corn azimuth angle of each corn seedling plant point cloud in the corn population seedling-stage data as the plant azimuth angle of the corresponding corn plant point cloud in the point cloud of the corn population canopy to be reconstructed; use the seedling-stage corn center point coordinates of each corn seedling plant point cloud in the corn population seedling-stage data as the plant point cloud center coordinates of the corresponding corn plant point cloud in the point cloud of the corn population canopy to be reconstructed; According to the plant azimuth angle and the plant point cloud center coordinates, determine the clipping area of each corn plant point cloud in the point cloud of the corn population canopy to be reconstructed, and obtain the plant height according to the size information of the clipping area.

[0042] Since the maize population is small at the seedling stage and there is no plant crossing yet, the present invention uses the seedling stage data to segment and locate each plant within the population at other developmental stages of maize, and constructs a digital map of the maize seedling population. Specifically, the present invention uses devices such as unmanned aerial vehicles or orbital phenotyping platforms to obtain maize seedling population images or point cloud data for plant segmentation and location. For the top image data of the maize seedling population, first, the image sequence is stitched to form an overall image of the maize seedling population, and the image is globally transformed into the world coordinate system using the feature points in the image; then, networks such as PlantU-net can be used to achieve plant segmentation of the overall image of the maize seedling population. For the point cloud data of the maize seedling population, algorithms such as GDQuickshift++ can be used to segment individual plants from the point cloud data of the seedling population. In one embodiment, based on the image or point cloud segmentation results, the plant segmentation accuracy can be improved through manual interaction.

[0043] In the present invention, for each maize seedling image or point cloud obtained, the coordinates of its center on the X-axis and Y-axis in the two-dimensional plane are extracted, that is, the coordinates of the center point of the maize seedling, and the principal component analysis (Principal Component Analysis, abbreviated as PCA) is used to process the single-plant data extracted, calculate the principal direction of each seedling plant, and project it into the two-dimensional plane. By calculating the angle between the principal direction and the Y-axis (the maize planting row direction), the azimuth angle of each seedling plant is determined, that is, the azimuth angle of the maize seedling.

[0044] In the present invention, the maize population seedling stage data contains the three-dimensional morphological information of maize plants at the seedling stage, and this data belongs to the same maize plant area as the maize population canopy point cloud to be reconstructed later. That is to say, the maize population seedling stage data and the maize population canopy point cloud to be reconstructed represent the plants in the same maize field, only at different growth stages.

[0045] Then, using the Direct Linear Transformation (DLT) method, the point clouds of each maize seedling plant in the maize population seedling stage data are temporally aligned with the point clouds of each maize plant in the maize population canopy point cloud to be reconstructed. The purpose of temporal alignment is to unify the point cloud data at different growth stages into the same spatio-temporal reference framework so that the changes of plants at different growth stages can be compared and analyzed. Through the DLT method, a direct linear relationship between the planar coordinates of key points in the image / point cloud and the spatial coordinates of points in the corresponding three-dimensional point cloud is established, so as to obtain the corresponding relationship between the key points in the seedling stage point cloud and the canopy point cloud, and achieve the temporal alignment of the point cloud data.

[0046] After the temporal alignment is completed, the present invention uses the azimuth angle of each maize seedling plant point cloud in the maize population seedling stage data as the plant azimuth angle of the corresponding maize plant point cloud in the maize population canopy point cloud to be reconstructed. Similarly, the central point coordinates of each maize seedling plant point cloud in the seedling stage point cloud are used as the plant point cloud central coordinates of the corresponding maize plant point cloud in the maize population canopy point cloud to be reconstructed, so as to obtain the azimuth angle and central coordinate information of each plant point cloud in the maize population canopy point cloud to be reconstructed.

[0047] Finally, the present invention determines the cropping area of each maize plant point cloud in the maize population canopy point cloud to be reconstructed according to the plant azimuth angle and the plant point cloud central coordinates. The size and shape of the cropping area can be set according to actual needs, but a rectangular area is determined based on the central point and azimuth angle of the plant. Then, the canopy point cloud is cropped, and only the point cloud data within this area is retained, and it is considered that these data represent the approximate growth range of each maize plant in the population. Finally, according to the size information of each plant point cloud data obtained by cropping, the plant height of each plant is calculated, that is, the height information of each plant in the population is obtained.

[0048] Based on the above embodiments, the obtaining of a plurality of target joint unit three-dimensional grid models corresponding to each of the maize plants in the preset maize three-dimensional joint unit template library according to the growth period information, the plant azimuth angle, and the plant height includes: Determining the number of maize joint units and the growth height of the maize joint units required by each of the maize plants according to the plant height and the growth period information; Determining the joint unit leaf azimuth angle according to the plant azimuth angle; Matching the joint unit three-dimensional grid models in the preset maize three-dimensional joint unit template library based on the growth period information, the number of maize joint units, the growth height of the maize joint units, and the joint unit leaf azimuth angle to obtain a plurality of the target joint unit three-dimensional grid models corresponding to each of the maize plants.

[0049] In the present invention, by analyzing the height of the maize plant and its growth period, it can be inferred the number of joint units that the plant should currently contain (i.e., the number of joints on the maize stem, and usually one leaf grows on each joint) and the growth height of each joint unit (i.e., the vertical distance from each joint to the ground or the growth height relative to the previous joint). The plant azimuth angle describes the orientation of the plant on a two-dimensional plane (or a certain plane projection in three-dimensional space). Based on this azimuth angle, the orientation of the leaves on each joint unit, that is, the leaf azimuth angle, can be determined.

[0050] In the preset 3D maize node unit template library, according to the number of node units, growth height, leaf azimuth angle, and growth period information determined above, find the most matching 3D mesh model of the node unit. In the present invention, this preset 3D maize node unit template library contains 3D mesh models of different varieties, different growth periods, and different node unit serial numbers. Each 3D mesh model of the node unit contains organ meshes with semantic information such as leaves, internodes, leaf sheaths (some node units also contain tassels or ears). During the matching process, the keywords of each 3D mesh model of the node unit include variety, acquisition time, acquisition location, growth period, node unit serial number, leaf length, leaf width, leaf included angle, growth height of the node unit, internode length, and leaf sheath length, etc. In the present invention, the most similar 3D mesh model of the node unit in the preset 3D maize node unit template library can be called by constructing a node unit similarity function based on the above keyword information.

[0051] Through the above matching process, multiple corresponding target 3D mesh models of the node unit are obtained for each maize plant. These target 3D mesh models of the node unit will be translated according to the growth height of the node unit and rotated according to the azimuth angle of the node unit to simulate the growth state of the real plant. These target 3D mesh models of the node unit will be used to construct the initial 3D model of the maize plant, and further to construct the 3D models of each plant within the population, and finally complete the construction of the initial 3D model of the population.

[0052] Based on the above embodiments, constructing the initial plant 3D model corresponding to each maize plant in the point cloud of the canopy of the maize population to be reconstructed according to the multiple target 3D mesh models of the node unit includes: Based on the growth height of the maize node unit corresponding to each target 3D mesh model of the maize plant, translate and align each target 3D mesh model of the node unit according to the height position, and rotate each target 3D mesh model of the node unit according to the preset node unit azimuth angle, so as to splice the target 3D mesh models of the node unit after translation alignment and azimuth angle rotation to obtain the initial plant 3D model.

[0053] In the present invention, each target 3D mesh model of the node unit represents a real node unit and has a specific growth height (i.e., the vertical position of the node unit on the plant). In order to simulate the growth state of the real plant, it is necessary to translate each target 3D mesh model of the node unit in the vertical direction according to these growth heights, so that they are arranged in the real growth order and alignment method to ensure that in the constructed 3D model, the vertical position of the node unit is consistent with the real plant, providing a basis for subsequent model splicing.

[0054] Furthermore, in addition to the vertical position, the leaves on the node unit also have a specific orientation, namely the azimuth angle. This azimuth angle describes the rotation angle of the leaf on the horizontal plane. To simulate the orientation of real leaves, the present invention needs to rotate each target node unit three-dimensional mesh model on the horizontal plane according to the preset azimuth angle of the node unit, so as to ensure that in the constructed three-dimensional model, the orientation of the leaves is consistent with that of the real plant, and the fidelity and accuracy of the model are increased.

[0055] After the translation and rotation operations, each target node unit three-dimensional mesh model has been arranged in the real growth order and orientation. Then, these models need to be spliced, that is, combined into a complete three-dimensional model of the plant. The splicing process usually involves aligning and fusing the connecting parts of adjacent node units to ensure the integrity and coherence of the model. Through the operations of translation alignment, azimuth angle rotation and splicing, an initial three-dimensional model of the plant is obtained. This initial three-dimensional model of the plant simulates the growth state and structural characteristics of the real corn plant and can be used for subsequent applications such as the construction of a three-dimensional model of the population, growth simulation and yield prediction.

[0056] Based on the above embodiments, the initial three-dimensional model of the plant is iteratively optimized based on an improved Chamfer distance formula, including: Based on a preset plant point cloud rotation range and a preset plant point cloud rotation step size, the current initial three-dimensional model of the plant is rotated around the central axis of the corn plant; Based on the improved Chamfer distance formula, the first Chamfer distance between the initial three-dimensional model of the plant after each rotation and the three-dimensional point cloud of the plant before reconstruction is calculated; The rotated initial three-dimensional model of the plant corresponding to the smallest first Chamfer distance is determined as the optimized initial three-dimensional model of the plant.

[0057] In the present invention, the initial three-dimensional model of the plant directly constructed through the above embodiments realizes the reduction of each plant in the population according to the plant height and growth position, but the position of the whole plant and each node unit still needs to be further optimized and adjusted according to the point cloud to improve the reconstruction accuracy.

[0058] Therefore, on the basis of obtaining the initial three-dimensional model of the plant, the present invention rotates the current initial three-dimensional model of the plant around the central axis of the corn plant based on a preset plant point cloud rotation range and a preset plant point cloud rotation step size. The preset plant point cloud rotation range is a preset angle interval. In the present invention, the preset plant point cloud rotation range is and , indicating that the model can be rotated within this range; the preset plant point cloud rotation step size determines the increment of each rotation, that is, at what small angle step the model iterates within the rotation range.

[0059] The specific iterative optimization process is as follows: 1. Calculate the overlapping part of the initial plant three-dimensional model and the constrained point cloud (i.e., the three-dimensional point cloud of the plant before reconstruction) in the Z-axis direction of the three-dimensional coordinate system to participate in the iterative calculation.

[0060] 2. Set the rotation range and rotation step of the plant, so that the current initial plant three-dimensional model rotates iteratively around the plant central axis within the rotation range with the rotation step as the step. After each rotation, calculate the improved Chamfer distance between the current initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction, that is, the first Chamfer distance. Select the rotation angle corresponding to the smallest first Chamfer distance as the final rotation angle of the current initial plant three-dimensional model, so that after the current initial plant three-dimensional model is rotated by this final rotation angle, an optimized initial plant three-dimensional model is obtained.

[0061] In the present invention, the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction. After each rotation, this formula is used to calculate the first Chamfer distance between the current rotated initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction. This distance reflects the similarity between the model and the point cloud. The smaller the distance, the higher the similarity. In all rotation iterations, record the Chamfer distance after each rotation, and find the smallest one among them. Then, determine the rotated model corresponding to this smallest distance as the optimized initial plant three-dimensional model.

[0062] Based on the above embodiments, the method further includes: Rotate the current target node unit three-dimensional grid model in the optimized initial plant three-dimensional model around the node central axis based on the preset node unit point cloud rotation range and preset node unit point cloud rotation step; Based on the improved Chamfer distance formula, calculate the second Chamfer distance between the target node unit three-dimensional grid model after each rotation and the three-dimensional point cloud of the node unit before reconstruction, where the three-dimensional point cloud of the node unit before reconstruction is the node unit point cloud corresponding to the target node unit three-dimensional grid model in the point cloud of the maize population canopy to be reconstructed; Determine the target node unit three-dimensional grid model after rotation corresponding to the smallest second Chamfer distance as the target node unit three-dimensional grid model with optimized node unit angle; Determine the internode connection positions between the target node unit three-dimensional grid models with optimized node unit angles for each of them, and perform interpolation processing on the internode connection positions; Perform splicing processing on the target node unit three-dimensional grid model after interpolation processing to construct the target plant three-dimensional model corresponding to each corn plant in the corn population canopy point cloud to be reconstructed.

[0063] In the present invention, based on a preset node unit point cloud rotation range and a preset node unit point cloud rotation step size, rotate the current target node unit three-dimensional grid model in the optimized initial plant three-dimensional model around the node central axis. It should be noted that in the present invention, on the basis of the plant iterative optimization in the above embodiment, each node unit in the plant is iteratively optimized. At the same time, to ensure that the plant iterative optimization result is not overturned and to ensure the growth rules of the node units on the plant, the range of azimuth angle iterative optimization of each node unit is set, that is, the preset node unit point cloud rotation range is , where generally , indicating that the node unit can be rotated and adjusted within this relatively small range. The preset node unit point cloud rotation step size determines the increment of each rotation, that is, at what small angular step the node unit iterates within the rotation range.

[0064] Similarly, after each rotation of the node unit three-dimensional grid model, the improved Chamfer distance formula is used to calculate the second Chamfer distance between the currently rotated target node unit three-dimensional grid model and the pre-reconstruction node unit three-dimensional point cloud. This distance reflects the similarity between the node unit model and the point cloud. The smaller the distance, the higher the similarity. By calculating the second Chamfer distance, the difference between the node unit model and the point cloud is quantified, providing a basis for subsequent node unit selection.

[0065] Furthermore, the rotated target node unit three-dimensional grid model corresponding to the smallest second Chamfer distance is determined as the target node unit three-dimensional grid model after node unit angle optimization. In the present invention, during all rotation iterations, the second Chamfer distance after each rotation is recorded, and the smallest one among them is found. Then, the rotated node unit model corresponding to this smallest distance is determined as the target node unit three-dimensional grid model after node unit angle optimization.

[0066] On the basis of the above embodiment, the improved Chamfer distance formula is: ; where represents the pre-reconstruction plant three-dimensional point cloud or the pre-reconstruction node unit three-dimensional point cloud, represents the current initial plant three-dimensional model or the current target node unit three-dimensional grid model; is the vertex in , is the point in . It represents the Chamfer distance, indicating that the higher the similarity between the currently reconstructed plant mesh and the corresponding point cloud, the smaller the reconstruction error.

[0067] Since each plant is assembled in node units, after the rotation iteration of the node units, there is a certain separation between the three-dimensional node unit models, and node unit splicing and mesh fusion processing are required to further enhance the realism of the plant geometric model.

[0068] Specifically, first calculate the internode positions in each three-dimensional node unit, that is, the connection part between two adjacent node units. Interpolation is performed using the internode points so that all internodes on the plant approximate to form a smooth and natural curve or straight line, which reflects the actual growth form of the plant internodes as much as possible, thereby ensuring that the overall form of the plant is more natural and smooth. Then, revise the diameters of each internode in the order from bottom to top to ensure that the thickness change of the plant conforms to the actual growth situation.

[0069] Furthermore, according to the original vertex positions and new vertex positions of the internodes, other organs (such as leaves, ears, etc.) on the current node unit are translated to ensure that the overall node unit does not separate and deform.

[0070] In the present invention, by performing interpolation processing on the internode connection positions, a series of transition points can be generated, which fill the gaps between adjacent node units, making the internode part look more continuous and realistic. After completing the interpolation processing of the internode connection positions, splicing processing is performed according to the three-dimensional mesh models of these processed node units. The splicing processing is to connect adjacent node unit models according to the transition points generated by interpolation to form a more complete and realistic plant model.

[0071] In the present invention, the growth positions and directions of each plant in the maize population canopy are determined using seedling stage data, combined with data alignment and plant height extraction in subsequent key growth stages, to realize the construction of the initial three-dimensional model of the maize population canopy. And, by constructing an improved Chamfer distance as the loss function, iterative optimization of each plant within the population is carried out successively from the plant scale and the node unit scale, so that the reconstructed three-dimensional model of the population canopy gradually approaches the obtained population three-dimensional point cloud, thereby realizing the three-dimensional reconstruction of the maize population canopy.

[0072] The semantic three-dimensional reconstruction method of the maize population canopy provided by the present invention is applicable to the three-dimensional reconstruction of the canopy of maize populations at different growth stages obtained by platforms such as field track-type phenotyping platforms and unmanned aerial vehicles. The reconstructed three-dimensional model of the population canopy has a high consistency with the input point cloud in the overall contour, and a large number of missing organs in the middle and lower parts of the canopy can be complemented. In addition, the semantic information of each plant and each organ can be accurately retrieved within the reconstructed three-dimensional model of the population canopy, which can meet the requirements of phenotype analysis, construction of maize functional-structure models, and realistic rendering.

[0073] The maize population canopy semantic three-dimensional reconstruction system provided by the present invention will be described below. The maize population canopy semantic three-dimensional reconstruction system described below can be correspondingly referred to the maize population canopy semantic three-dimensional reconstruction method described above.

[0074] Figure 2 It is a schematic structural diagram of the maize population canopy semantic three-dimensional reconstruction system provided by the present invention. As Figure 2 shown, the present invention provides a maize population canopy semantic three-dimensional reconstruction system, including a processing module 201, a node unit matching module 202, a plant three-dimensional model construction module 203, and a semantic three-dimensional model construction module 204. Among them, the processing module 201 is used to determine the plant azimuth angle and plant height corresponding to the maize plants at each growth position in the point cloud of the maize population canopy to be reconstructed; the node unit matching module 202 is used to obtain a plurality of target node unit three-dimensional grid models corresponding to each of the maize plants in a preset maize three-dimensional node unit template library according to the growth stage information, the plant azimuth angle, and the plant height, where the growth stage information is the current growth stage of the maize plants in the point cloud of the maize population canopy to be reconstructed; the preset maize three-dimensional node unit template library is composed of node unit three-dimensional grid models corresponding to different maize node units, and the node unit three-dimensional grid model is composed of maize organ three-dimensional grids with semantic information; the plant three-dimensional model construction module 203 is used to construct an initial plant three-dimensional model corresponding to each of the maize plants in the point cloud of the maize population canopy to be reconstructed according to the plurality of target node unit three-dimensional grid models; the semantic three-dimensional model construction module 204 is used to iteratively optimize the initial plant three-dimensional model and the target node unit three-dimensional network model in turn based on an improved Chamfer distance formula to obtain a target plant three-dimensional model, and construct a maize population canopy semantic three-dimensional reconstruction model corresponding to the point cloud of the maize population canopy to be reconstructed according to the target plant three-dimensional model; among them, the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction, and the three-dimensional point cloud of the plant before reconstruction is the point cloud of the maize plant corresponding to the initial plant three-dimensional model in the point cloud of the maize population canopy to be reconstructed.

[0075] The maize population canopy semantic three-dimensional reconstruction system provided by the present invention selects a plurality of target node unit three-dimensional grid models with semantic information corresponding to each maize plant in the point cloud of the maize population canopy to be reconstructed through the growth position, azimuth angle, and height of each maize plant, and then uses these selected target node unit three-dimensional grid models to construct a three-dimensional model of each maize plant, so as to obtain a maize population canopy three-dimensional model that contains semantic information and complements the missing organs in the middle and lower parts of the canopy for the entire maize population canopy.

[0076] The system provided by the embodiments of the present invention is used to execute the above method embodiments. For the specific process and detailed content, please refer to the above embodiments and will not be elaborated here.

[0077] Figure 3 It is a schematic structural diagram of the electronic device provided by the present invention. As Figure 3 shown, the electronic device may include: a processor (Processor) 301, a communication interface (Communications Interface) 302, a memory (Memory) 303, and a communication bus 304. Among them, the processor 301, the communication interface 302, and the memory 303 complete mutual communication through the communication bus 304. The processor 301 can call the logical instructions in the memory 303 to execute the method for three-dimensional semantic reconstruction of the corn population canopy, and the method includes: determining the plant azimuth angle and plant height corresponding to the corn plants at each growth position in the point cloud of the corn population canopy to be reconstructed; obtaining a plurality of target joint unit three-dimensional grid models corresponding to each of the corn plants in a preset corn three-dimensional joint unit template library according to the growth stage information, the plant azimuth angle, and the plant height, where the growth stage information is the current growth stage of the corn plants in the point cloud of the corn population canopy to be reconstructed; the preset corn three-dimensional joint unit template library is composed of three-dimensional grid models of different corn joint units, and the three-dimensional grid model of the joint unit is composed of three-dimensional grids of corn organs with semantic information; constructing an initial plant three-dimensional model corresponding to each of the corn plants in the point cloud of the corn population canopy to be reconstructed according to the plurality of target joint unit three-dimensional grid models; based on an improved Chamfer distance formula, iteratively optimizing the initial plant three-dimensional model and the target joint unit three-dimensional network model in sequence to obtain a target plant three-dimensional model, and constructing a three-dimensional semantic reconstruction model of the corn population canopy corresponding to the point cloud of the corn population canopy to be reconstructed according to the target plant three-dimensional model; where the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction, and the three-dimensional point cloud of the plant before reconstruction is the point cloud of the corn plant corresponding to the initial plant three-dimensional model in the point cloud of the corn population canopy to be reconstructed.

[0078] In addition, when the logical instructions in the above-mentioned memory 303 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0079] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the maize population canopy semantic three-dimensional reconstruction method provided by the above-mentioned various methods. The method includes: determining the plant azimuth angle and plant height corresponding to the maize plants at each growth position in the maize population canopy point cloud to be reconstructed; according to the growth stage information, the plant azimuth angle, and the plant height, obtaining a plurality of target joint unit three-dimensional grid models corresponding to each of the maize plants in a preset maize three-dimensional joint unit template library, where the growth stage information is the growth stage currently in which the maize plants in the maize population canopy point cloud to be reconstructed are located; the preset maize three-dimensional joint unit template library is composed of three-dimensional grid models of different maize joint units, and the three-dimensional grid model of the joint unit is composed of three-dimensional grids of maize organs with semantic information; according to the plurality of target joint unit three-dimensional grid models, constructing an initial plant three-dimensional model corresponding to each of the maize plants in the maize population canopy point cloud to be reconstructed; based on an improved Chamfer distance formula, iteratively optimizing the initial plant three-dimensional model and the target joint unit three-dimensional network model in sequence to obtain a target plant three-dimensional model, and constructing a maize population canopy semantic three-dimensional reconstruction model corresponding to the maize population canopy point cloud to be reconstructed according to the target plant three-dimensional model; where the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction, and the three-dimensional point cloud of the plant before reconstruction is the point cloud of the maize plant corresponding to the initial plant three-dimensional model in the maize population canopy point cloud to be reconstructed.

[0080] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to implement the maize population canopy semantic three-dimensional reconstruction method provided in the above embodiments. The method includes: determining the plant azimuth angle and plant height corresponding to the maize plants at each growth position in the maize population canopy point cloud to be reconstructed; obtaining a plurality of target joint unit three-dimensional grid models corresponding to each of the maize plants in a preset maize three-dimensional joint unit template library according to the growth stage information, the plant azimuth angle, and the plant height, where the growth stage information is the growth stage at which the maize plants in the maize population canopy point cloud to be reconstructed are currently in; the preset maize three-dimensional joint unit template library is composed of three-dimensional grid models of different maize joint units, and the three-dimensional grid model of the joint unit is composed of three-dimensional grids of maize organs with semantic information; constructing an initial plant three-dimensional model corresponding to each of the maize plants in the maize population canopy point cloud to be reconstructed according to the plurality of target joint unit three-dimensional grid models; based on an improved Chamfer distance formula, iteratively optimizing the initial plant three-dimensional model and the target joint unit three-dimensional network model in sequence to obtain a target plant three-dimensional model, and constructing a maize population canopy semantic three-dimensional reconstruction model corresponding to the maize population canopy point cloud to be reconstructed according to the target plant three-dimensional model; where the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction, and the three-dimensional point cloud of the plant before reconstruction is the point cloud of the maize plant corresponding to the initial plant three-dimensional model in the maize population canopy point cloud to be reconstructed.

[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0082] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for three-dimensional semantic reconstruction of a maize population canopy, characterized in that Including: Determine the plant azimuth angle and plant height corresponding to each corn plant at each growth position in the point cloud of the corn population canopy to be reconstructed; According to the growth stage information, the plant azimuth angle, and the plant height, obtain a plurality of target joint unit three-dimensional grid models corresponding to each of the corn plants in a preset corn three-dimensional joint unit template library, where the growth stage information is the growth stage at which the corn plants in the point cloud of the corn population canopy to be reconstructed are currently in; the preset corn three-dimensional joint unit template library is composed of three-dimensional grid models of joint units corresponding to different corn joint units, and the three-dimensional grid model of the joint unit is composed of three-dimensional grids of corn organs with semantic information; According to the plurality of target joint unit three-dimensional grid models, construct an initial plant three-dimensional model corresponding to each of the corn plants in the point cloud of the corn population canopy to be reconstructed; Based on an improved Chamfer distance formula, iteratively optimize the initial plant three-dimensional model and the target joint unit three-dimensional network model in sequence to obtain a target plant three-dimensional model, and construct a corn population canopy semantic three-dimensional reconstruction model corresponding to the point cloud of the corn population canopy to be reconstructed according to the target plant three-dimensional model; where the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the three-dimensional point cloud of the plant before reconstruction, and the three-dimensional point cloud of the plant before reconstruction is the point cloud of the corn plant corresponding to the initial plant three-dimensional model in the point cloud of the corn population canopy to be reconstructed.

2. The three-dimensional semantic reconstruction method of the maize population canopy according to claim 1, characterized in that The determination of the plant azimuth angle and plant height corresponding to each corn plant at each growth position in the point cloud of the corn population canopy to be reconstructed includes: Obtain the seedling-stage corn azimuth angle and the seedling-stage corn center point coordinates corresponding to each seedling-stage corn plant point cloud in the corn population seedling-stage data, where the corn population seedling-stage data is point cloud data or image data of corn plants in the seedling stage, and the corn population seedling-stage data and the point cloud of the corn population canopy to be reconstructed belong to the same corn plant area; Based on direct linear transformation, perform temporal alignment between each seedling-stage corn plant point cloud in the corn population seedling-stage data and each corn plant point cloud in the point cloud of the corn population canopy to be reconstructed; According to the temporal alignment result, use the seedling-stage corn azimuth angle of each seedling-stage corn plant point cloud in the corn population seedling-stage data as the plant azimuth angle of the corresponding corn plant point cloud in the point cloud of the corn population canopy to be reconstructed; use the seedling-stage corn center point coordinates of each seedling-stage corn plant point cloud in the corn population seedling-stage data as the plant point cloud center coordinates of the corresponding corn plant point cloud in the point cloud of the corn population canopy to be reconstructed; According to the plant azimuth angle and the plant point cloud center coordinates, determine the cropping area of each corn plant point cloud in the point cloud of the corn population canopy to be reconstructed, and obtain the plant height according to the size information of the cropping area.

3. The method for three-dimensional semantic reconstruction of the maize population canopy according to claim 1, wherein The obtaining of a plurality of target joint unit three-dimensional grid models corresponding to each of the corn plants in a preset corn three-dimensional joint unit template library according to the growth stage information, the plant azimuth angle, and the plant height includes: Based on the plant height and the growth stage information, determine the number of maize node units and the growth height of maize node units required for each of the maize plants; Based on the plant azimuth angle, determine the leaf azimuth angle of the node unit; Based on the growth stage information, the number of maize node units, the growth height of maize node units, and the leaf azimuth angle of the node unit, match the node unit three-dimensional grid model in the preset maize three-dimensional node unit template library to obtain a plurality of target node unit three-dimensional grid models corresponding to each of the maize plants.

4. The method for three-dimensional semantic reconstruction of the maize population canopy according to claim 3, wherein The constructing the initial plant three-dimensional model corresponding to each of the maize plants in the maize population canopy point cloud to be reconstructed according to the plurality of target node unit three-dimensional grid models includes: Based on the growth height of the maize node units corresponding to each of the target node unit three-dimensional grid models of the maize plant, translate and align each of the target node unit three-dimensional grid models according to the height position, and rotate each of the target node unit three-dimensional grid models by the azimuth angle based on the preset node unit azimuth angle, so as to splice the target node unit three-dimensional grid models after translation alignment and azimuth angle rotation to obtain the initial plant three-dimensional model.

5. The maize population canopy semantic three-dimensional reconstruction method according to claim 4, characterized in that, Based on the improved Chamfer distance formula, perform iterative optimization on the initial plant three-dimensional model, including: Based on the preset plant point cloud rotation range and the preset plant point cloud rotation step, rotate the current initial plant three-dimensional model around the central axis of the maize plant; Based on the improved Chamfer distance formula, calculate the first Chamfer distance between the initial plant three-dimensional model after each rotation and the pre-reconstruction plant three-dimensional point cloud; Determine the initial plant three-dimensional model after rotation corresponding to the minimum first Chamfer distance as the optimized initial plant three-dimensional model.

6. The method for three-dimensional semantic reconstruction of the maize population canopy according to claim 5, wherein The method further includes: Based on the preset node unit point cloud rotation range and the preset node unit point cloud rotation step, rotate the current target node unit three-dimensional grid model in the optimized initial plant three-dimensional model around the node central axis; Based on the improved Chamfer distance formula, calculate the second Chamfer distance between the target node unit three-dimensional grid model after each rotation and the pre-reconstruction node unit three-dimensional point cloud, where the pre-reconstruction node unit three-dimensional point cloud is the node unit point cloud corresponding to the target node unit three-dimensional grid model in the maize population canopy point cloud to be reconstructed; Determine the target node unit three-dimensional grid model after rotation corresponding to the minimum second Chamfer distance as the target node unit three-dimensional grid model optimized by the node unit angle; Determine the internode connection positions between the target node unit three-dimensional grid models optimized by each of the node unit angles, and perform interpolation processing on the internode connection positions; Perform splicing processing according to the target node unit three-dimensional grid models after interpolation processing to construct the target plant three-dimensional model corresponding to each of the maize plants in the maize population canopy point cloud to be reconstructed.

7. The method for three-dimensional semantic reconstruction of the maize population canopy according to claim 6, wherein The improved Chamfer distance formula is: ； Among them, represents the three-dimensional point cloud of the plant before reconstruction or the three-dimensional point cloud of the unit segment before reconstruction, represents the current three-dimensional model of the initial plant or the current three-dimensional mesh model of the target unit segment; is a vertex in is a point in represents the Chamfer distance.

8. A three-dimensional semantic reconstruction system for the canopy of a corn population, characterized in that, Including: A processing module, configured to determine the plant azimuth angle and plant height corresponding to each maize plant at each growth position in the maize population canopy point cloud to be reconstructed; A node unit matching module, configured to obtain a plurality of target node unit three-dimensional grid models corresponding to each of the maize plants in a preset maize three-dimensional node unit template library according to the growth stage information, the plant azimuth angle, and the plant height, wherein the growth stage information is the growth stage at which the maize plants in the maize population canopy point cloud to be reconstructed are currently in; the preset maize three-dimensional node unit template library is composed of node unit three-dimensional grid models corresponding to different maize node units, and the node unit three-dimensional grid models are composed of maize organ three-dimensional grids with semantic information; A plant three-dimensional model construction module, configured to construct an initial plant three-dimensional model corresponding to each of the maize plants in the maize population canopy point cloud to be reconstructed according to the plurality of target node unit three-dimensional grid models; A semantic three-dimensional model construction module, configured to iteratively optimize the initial plant three-dimensional model and the target node unit three-dimensional network model in sequence based on an improved Chamfer distance formula to obtain a target plant three-dimensional model, and construct a maize population canopy semantic three-dimensional reconstruction model corresponding to the maize population canopy point cloud to be reconstructed according to the target plant three-dimensional model; wherein the improved Chamfer distance formula is constructed based on the similarity distance between the initial plant three-dimensional model and the plant three-dimensional point cloud before reconstruction, and the plant three-dimensional point cloud before reconstruction is the maize plant point cloud corresponding to the initial plant three-dimensional model in the maize population canopy point cloud to be reconstructed.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the maize population canopy semantic three-dimensional reconstruction method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the maize population canopy semantic three-dimensional reconstruction method according to any one of claims 1 to 7.

Images