Vegetation growth evaluation method and system based on laser point cloud recognition

Abstract

The invention discloses a vegetation growth evaluation method and system based on laser point cloud recognition. The method comprises the following operation steps: collecting two-period point cloud data of an interval time period in a target area; the two-period point cloud data comprises first-period point cloud data and second-period point cloud data; determining a shielding candidate region for a comprehensive difference value between the point cloud densities in the point cloud data in the two periods, and analyzing spatial statistical characteristics of the shielding candidate region; dividing vegetation category point cloud to analyze a growth parameter set for the sheltered candidate region by using spatial statistical characteristics; plant growth simulation is carried out on the growth parameter set, and the shielding result of the vegetation point cloud after growth simulation is judged. And screening the sheltering candidate region through the point cloud to obtain a vegetation point cloud, and reflecting the sheltering relationship between the crown and the shrub by using the vegetation point cloud so as to determine whether sheltering exists between the crown and the shrub.

Description

Technical Field

[0001] This invention relates to the field of power transmission line operation and maintenance, and in particular to a vegetation growth assessment method and system based on laser point cloud recognition. Background Technology

[0002] In the intricate network of power grids, the safe and stable operation of overhead transmission lines is crucial. Vegetation beneath and within these lines is often a dynamically changing potential source of risk. Assessing vegetation growth and species, especially managing the "line-tree relationship," is the first line of defense against power grid accidents. When vegetation grows too rapidly and approaches transmission lines, it can cause electrical conductivity faults or damage trees, leading to wildfires. Accurate identification of vegetation, especially fast-growing tree species and flammable trees, is a vital indicator for vegetation growth assessment. Traditional vegetation growth assessment methods often rely on manual surveys or image processing techniques based on optical remote sensing data. However, these methods are often affected by external factors such as weather, light variations, and viewing angles in complex terrain or areas with high-density vegetation, resulting in insufficient accuracy and reliability. Assessing shading effects is particularly challenging in environments with multiple vegetation types, such as areas where trees and shrubs are intertwined.

[0003] To overcome the shortcomings of traditional methods, laser point cloud technology is widely used in vegetation growth assessment. Laser scanning technology uses laser sensors to emit laser beams from different angles and receive reflected signals, accurately measuring the three-dimensional spatial coordinates of an object's surface to construct high-precision point cloud data. The density and accuracy of this point cloud data can reflect the three-dimensional morphology of vegetation and can be compared at different time periods, allowing for real-time monitoring of vegetation growth.

[0004] However, studies have found tree shading during vegetation growth monitoring and assessment, particularly in environments where trees and shrubs coexist. Tree vegetation often obscures parts of the shrub vegetation because trees generally have a distinct upright trunk, typically exceeding 6 meters in height, and can be further categorized into small, medium, large, and towering trees, such as camphor trees and ginkgo. Shrubs, on the other hand, lack a prominent trunk, are low-growing and clump-forming, and are generally less than 6 meters tall. For example, in temperate coniferous and broad-leaved mixed forests, the trees in Northeast China's forests can be clearly divided into two categories: conifers and broad-leaved trees. The most common trees include conifers and Scots pine, while the most common shrubs include hazelnut and eleutherococcus senticosus. This tree-shrub shading significantly impacts the accurate assessment of tree growth and plant-environment interactions.

[0005] Traditional assessment methods are difficult to effectively quantify this shading effect and often fail to accurately distinguish the shading relationships between different vegetation types. Summary of the Invention

[0006] The purpose of this invention is to provide a vegetation growth assessment method and system based on laser point cloud recognition, which solves the above-mentioned technical problems pointed out in the prior art.

[0007] This invention provides a vegetation growth assessment method based on laser point cloud recognition, comprising the following steps:

[0008] Two periods of point cloud data are collected from the target area at an interval; the two periods of point cloud data include point cloud data of the first period and point cloud data of the second period.

[0009] Based on the comprehensive difference value between the point cloud density in the two periods of point cloud data, candidate occlusion regions are determined, and spatial statistical characteristics of the candidate occlusion regions are analyzed. The spatial statistical characteristics are used to classify the candidate occlusion regions into vegetation categories and point cloud analysis growth parameter sets. Plant growth simulation is performed on the growth parameter sets to determine the occlusion result of the vegetation point cloud after growth simulation.

[0010] This invention also provides a vegetation growth assessment system based on laser point cloud recognition, comprising: a data acquisition module; and an analysis module.

[0011] The acquisition module is used to acquire point cloud data of the target area at two time intervals; the point cloud data of the two time intervals includes point cloud data of the first time interval and point cloud data of the second time interval.

[0012] The analysis module is used to determine occlusion candidate regions by analyzing the comprehensive difference value between point cloud densities in the point cloud data of the two periods, and to analyze the spatial statistical characteristics of the occlusion candidate regions; to classify the occlusion candidate regions into vegetation categories using the spatial statistical characteristics; to perform plant growth simulation on the growth parameter set, and to determine the occlusion result of the vegetation point cloud after growth simulation.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0014] Analysis of the above-mentioned vegetation growth assessment method and system based on laser point cloud recognition provided by this invention reveals that, in practical applications, point cloud data from two periods are allocated to three-dimensional grids of different scales. Each scale grid represents a different resolution, capable of capturing local changes within different ranges. Density statistics are performed on voxels within each grid to obtain point cloud density differences at different time periods. These differences reflect the point cloud changes in local areas between different time points. The scheme optimizes sensitivity to changes at each scale by assigning weight coefficients to grids of different scales, comprehensively considering both the global view at a coarse scale and local changes at a fine scale. The weighted comprehensive difference value accurately expresses the changes in local areas within two time periods, and based on this, it determines whether there are occlusion candidate areas. Further, possible occlusion areas are precisely identified, and these areas are further converted into binary data to simplify subsequent analysis. Then, geometric features are extracted for each occlusion candidate area, the area and boundary descriptor are calculated, and the convex hull algorithm is used to determine the region's outline. By calculating the point cloud density and spatial statistical characteristics of the region, it is possible to further accurately locate and analyze occlusion areas of different forms, especially in identifying geometric features to more accurately determine the occlusion relationship between trees and shrubs.

[0015] Furthermore, the above technical solution utilizes spatial statistical features such as regional point cloud density and boundary descriptors to calculate occlusion candidate regions, forming a three-dimensional circular buffer region. Then, by expanding the buffer region, the integrity of data coverage is ensured, thereby effectively handling the complex occlusion relationship between trees and shrubs and ensuring the accuracy of subsequent analysis. On this basis, the solution extracts multi-dimensional features such as laser echo intensity and color attributes, and combines them with a point cloud segmentation model pre-trained by a deep learning network to classify point clouds into vegetation and non-vegetation categories.

[0016] Furthermore, the aforementioned technical solution further divides spatial scale intervals and analyzes the vegetation point cloud in each interval, extracting curvature, color ratio, and texture features to provide rich spatial information on vegetation morphology, structure, and growth patterns. The solution calculates the height growth rate of the vegetation point cloud by comparing point cloud data from different time periods and uses principal component analysis (PCA) to determine the canopy expansion direction to accurately reflect vegetation growth dynamics, enabling a comprehensive prediction of vegetation growth rate and expansion direction. Overall, this solution provides a precise framework for vegetation growth trend analysis.

[0017] Furthermore, this scheme utilizes L-system rules and the sun-facing mechanism to construct a three-dimensional geometric model of plant branches and leaves by simulating the plant growth process. By collecting point cloud data of the target area and combining multi-dimensional feature extraction and error correction, it can effectively supplement the shaded vegetation data and generate a high-precision point cloud model, thus more realistically reflecting the distribution and growth of plants in the environment. By optimizing the sampling density, it ensures that key parts (such as forks and leaf edges) are densely sampled, making the point cloud data more accurate in detail. After adjusting the difference between the sampling density and the point cloud density, it ensures seamless connection between the supplemented point cloud and the surrounding point cloud, ultimately generating a complete and accurate three-dimensional vegetation model. In the subsequent point cloud segmentation stage, by accurately segmenting the vegetation point cloud and the non-vegetation point cloud, efficient vegetation area identification can be achieved. In particular, for the analysis of tall shrubs, it helps to assess the shading impact of trees on buildings. The calculation of area overlap can quantify the shading effect of the canopy and the bottom shrubs. Attached Figure Description

[0018] Figure 1 This is a flowchart of the main process of a vegetation growth assessment method based on laser point cloud recognition, as described in Example 1.

[0019] Figure 2 This is a flowchart illustrating the spatial characteristics of a vegetation growth assessment method based on laser point cloud recognition, as described in Example 1.

[0020] Figure 3 This is a flowchart of the growth parameter set for a vegetation growth assessment method based on laser point cloud recognition, as described in Example 1.

[0021] Figure 4 This is a flowchart illustrating the mutual occlusion of independent point cloud clusters in a vegetation growth assessment method based on laser point cloud recognition, as described in Example 1.

[0022] Figure 5 This is a schematic diagram of the surrounding point cloud of a vegetation growth assessment method based on laser point cloud recognition, as described in Example 1.

[0023] Figure 6 This is a schematic diagram of the overlap of a vegetation growth assessment method based on laser point cloud recognition in Example 1.

[0024] Figure 7 This is a flowchart of a vegetation growth assessment system based on laser point cloud recognition, as described in Example 2.

[0025] Labels: Acquisition Module 10; Analysis Module 20. Detailed Implementation

[0026] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0028] Example 1

[0029] like Figure 1 As shown in the figure, this embodiment of the invention provides a vegetation growth assessment method based on laser point cloud recognition, including the following steps:

[0030] S10: Collect point cloud data of the target area at two time intervals (e.g., a two-day time interval); the two time interval point cloud data includes point cloud data of the first time period and point cloud data of the second time period.

[0031] It should be noted that laser scanning equipment is deployed in the target area to be collected. The target area includes its geographical location and environmental characteristics (such as hills, buildings, vegetation cover, etc.). The first laser scan is carried out according to the predetermined plan, which is the first period of point cloud data collection. The sensor position, attitude, and time information are recorded, and the density and accuracy of the acquired point cloud data are ensured to meet the requirements. In the predetermined second period, laser scanning is also carried out according to the planned plan, which means ensuring that the parameters and coverage of the point cloud data in the second period are consistent with or comparable to those in the first period (that is, the point cloud coordinates are the same in the two periods).

[0032] S20: Determine occlusion candidate regions based on the comprehensive difference value between point cloud densities in the two periods of point cloud data, and analyze the spatial statistical characteristics of the occlusion candidate regions; use the spatial statistical characteristics to classify the occlusion candidate regions into vegetation categories and point cloud analysis growth parameter sets; perform plant growth simulation on the growth parameter sets to determine the occlusion result of the vegetation point cloud after growth simulation.

[0033] Specifically, such as Figure 2 As shown, in step S20, the comprehensive difference value between the point cloud density in the point cloud data of the two periods is used to determine the occlusion candidate region, and the spatial statistical characteristics of the occlusion candidate region are analyzed; the spatial statistical characteristics are used to classify the occlusion candidate region into vegetation categories; the growth parameter set of the point cloud is analyzed; plant growth simulation is performed on the growth parameter set to determine the occlusion result of the vegetation point cloud after growth simulation. The specific operation steps are as follows:

[0034] S21: Set multi-scale grids for the point cloud data of the two periods respectively, and use the multi-scale grids to divide the point cloud data of the two periods into grids to form a three-dimensional grid layer of multiple scales.

[0035] For each scale of the 3D mesh layer (i.e., a set of voxels corresponding to each resolution (each scale is equivalent to the corresponding resolution), that is, each scale of the 3D mesh layer has a set of voxels), all voxels are traversed.

[0036] Obtain all point clouds in the first-period point cloud data and the second-period point cloud data in the 3D mesh layer corresponding to the current scale;

[0037] Count all points in the first-period point cloud data and the second-period point cloud data in each voxel to obtain the first-period point cloud density and the second-period point cloud density respectively.

[0038] It should be noted that the above multi-scale settings are based on the resolution parameters of the point cloud data collected in the two periods. That is, each resolution represents a grid layer, and each grid layer represents a scale information. The purpose is to capture the local density change information at different scales (i.e., different resolutions).

[0039] When processing tasks such as laser point clouds and 3D modeling, continuous 3D space is often divided into many small units, namely voxels, for the convenience of calculation and analysis. Information such as the number of points contained in each voxel and the average intensity can be statistically analyzed, which helps to extract local features or perform data downsampling and fast retrieval. In point cloud processing, voxels are used to divide point cloud data into many small regions, so that the data in each voxel can be independently statistically analyzed.

[0040] S22: Calculate the density difference value (i.e., the difference, reflecting the change in point cloud density in the local area between the two time periods) for the point cloud density in the first period and the point cloud density in the second period for each voxel.

[0041] Assign weight coefficients to each scale grid; sum the density difference value of each voxel with the weight coefficient of each scale grid to obtain the comprehensive difference value;

[0042] ;

[0043] in: Indicates the overall difference value;

[0044] It is the total number of voxels;

[0045] It is the first Weighting coefficients for each scale of the grid;

[0046] This represents the density difference between the point cloud density in the first period and the point cloud density in the second period for each voxel;

[0047] It is the first Density of point cloud in the first period of individual units;

[0048] It is the first Density of point cloud in the second period in individual units;

[0049] It should be noted that the weighting coefficients assigned to each grid scale are based on the resolution characteristics of each grid layer. Coarser scales have a better global view (i.e., larger scales, such as 1); finer scales can capture local fine changes (i.e., smaller scales, such as 0.1); at the same time, the weighting factor for each scale is determined by taking into account the local changes in spatial location (e.g., the smoothness of the surrounding density).

[0050] Meanwhile, the comprehensive density difference value can reflect the changes in point cloud density in local areas at different time stages. The above steps, by assigning weights to grids of different scales, are used to balance the sensitivity to global and local changes at different scales, making the results more accurate and targeted. Grids of different scales capture different change information; coarser scales can provide a global view, while finer scales help to capture detailed changes. The above weighted synthesis can better combine the advantages of each scale, improve the accuracy of the analysis, and directly reflect the changes in point cloud at different times, which is crucial for detecting potential occlusion candidate regions.

[0051] S23: Preset difference density threshold; determine whether the overall difference value in each voxel is greater than the difference density threshold;

[0052] If so, voxels that meet the conditions are determined to form potential occlusion candidate regions; each occlusion candidate region is binarized and used as an occlusion candidate region.

[0053] It should be noted that the above steps, by judging the difference density threshold, can accurately identify potential occlusion candidate regions in the point cloud data; converting potential occlusion candidate regions into binary data facilitates subsequent processing and analysis, ensuring that the localization and subsequent repair of occlusion candidate regions are simpler and more efficient; the above judgment by using the difference density threshold (set based on experience) can accurately distinguish occlusion candidate regions and avoid interference from noisy point clouds; and the binarization method in the above steps simplifies complex data into an easy-to-process form, providing convenience for subsequent analysis.

[0054] S24: Calculate the number of pixels for each occlusion candidate region, and use the result as the region area;

[0055] Determine the coordinates of voxels in each occlusion candidate region, find the maximum and minimum values ​​of the voxel coordinates in each occlusion candidate region, and obtain the bounding box of the occlusion candidate region (i.e., the bounding box can be represented by (x...)). min , y min , z min ) and (x max, y max , z max) (Representation); The convex hull algorithm is used to construct the convex hull for each voxel, and the minimum convex polyhedron of all voxels is determined to form the outline of the occlusion candidate region (i.e., the outline contains information such as the vertex set of the convex hull and the area of ​​the occlusion candidate region); The bounding box and the outline are combined to form a boundary descriptor;

[0056] The number of point clouds in each occlusion candidate region is calculated and used as the region point cloud density.

[0057] The region area, boundary descriptor, and region point cloud density are combined to form a spatial statistical feature.

[0058] It should be noted that the above steps, by calculating geometric features such as area, can provide detailed spatial information for each occlusion candidate region; the bounding box and convex hull in the above steps determine the bounding box and outline of each occlusion candidate region (i.e., can describe the shape regularity or complexity of the occlusion candidate region), which helps to make the shape geometrically accurate and can provide accurate geometric information for subsequent analysis or repair; the above steps, through the extraction of geometric features, can comprehensively understand the shape and other features of the occlusion candidate region; the bounding box and convex hull can provide effective spatial constraints and structural descriptions for subsequent calculations, and the convex hull can especially accurately describe the outline of the object;

[0059] The above steps, by calculating the area and bounding box of each occlusion candidate region, can accurately depict the shape and structure of each region. Trees and shrubs typically have different growth forms; trees are usually taller and have larger crowns, while shrubs are relatively shorter and denser. Through these geometric features, based on the spatial distribution characteristics of each occlusion candidate region, it can be determined whether it is a tree or a shrub, or an occlusion relationship between the two. For example, the bounding box of a tree is usually larger, while the geometric features of a shrub are more compact. Comparing these geometric features helps to identify which region is a tree and which region is a shrub, thereby distinguishing the occlusion effect of trees on shrubs.

[0060] S25: Expand the occlusion candidate region using the spatial statistical features to obtain a three-dimensional circular buffer region; analyze the circular buffer region to obtain multi-dimensional features; input the multi-dimensional features into a pre-constructed point cloud segmentation model to classify vegetation category point clouds; divide the circular buffer region into spatial scale intervals and analyze the growth parameter set of the vegetation category point clouds in each spatial scale interval; use the growth parameter set to perform plant growth simulation to update the point cloud segmentation model and obtain a new vegetation point cloud; perform clustering calculation on the vegetation point cloud to determine the occlusion result.

[0061] It should be noted that research has found that point cloud data is obtained from LiDAR scanning technology, and the density and distribution of point clouds may be uneven due to differences in the shape and size of occlusions and the data acquisition angle. Therefore, relying solely on the original occlusion candidate regions (such as bounding boxes or outlines) may not be sufficient to cover the target area, especially when the boundaries are complex or irregular. The above implementation method, by expanding the buffer radius to form a three-dimensional circular buffer region, can ensure uniform coverage of point cloud data within the target area, effectively avoiding analysis errors caused by insufficient data or uneven distribution. At the same time, trees and shrubs may form complex occlusion relationships during their growth due to factors such as spatial competition. Especially when large-scale trees occlude shrubs, the canopy or branches of the trees may cover part or all of the shrub's area. By expanding the buffer region, this occlusion effect can be better simulated, and a larger spatial range can be provided for subsequent analysis. This helps to identify the occlusion relationship between shrubs and trees and can also quantify the impact of occlusion effects on plant growth and environmental changes.

[0062] Specifically, such as Figure 3 As shown, in step S25, the occlusion candidate region is expanded using the spatial statistical features to obtain a three-dimensional circular buffer region; the circular buffer region is analyzed to obtain multi-dimensional features; the multi-dimensional features are input into a pre-constructed point cloud segmentation model to classify vegetation category point clouds; the circular buffer region is divided into spatial scale intervals, and the growth parameter set of the vegetation category point cloud in each spatial scale interval is analyzed; the growth parameter set is used to perform plant growth simulation to update the point cloud segmentation model and obtain the vegetation point cloud again; the vegetation point cloud is clustered to calculate the height and determine the occlusion result. The specific operation steps are as follows:

[0063] S251: The occlusion candidate region is analyzed and calculated using the spatial statistical features (i.e., using the regional point cloud density and boundary descriptor in the spatial statistical features, where the boundary descriptor obtains the size of the region through its outline; the buffer radius of the occlusion candidate region is calculated in detail using the size of the region obtained from the outline and the regional point cloud density (i.e., the size of the region obtained from the outline reflects the regional point cloud density of the canopy (or shrubs) (i.e., reflecting the density and structural complexity of the vegetation), and a basic radius is introduced (i.e., even if a region is small and sparse, this basic analysis range will be guaranteed) to generate a spherical buffer region with a radius (i.e., buffer radius) of approximately [value missing] for the occlusion candidate region, as shown in the formula: ;in This is the final buffer radius; Base radius; Size weight; The dimensions of the region are obtained from the outer contour. Density weights; The buffer radius is obtained by calculating the point cloud density of the region.

[0064] The buffer radius is expanded (i.e., the buffer radius can be expanded using a bounding box to form a circular or spherical region) to form a three-dimensional circular buffer region;

[0065] It should be noted that the buffer radius obtained by calculating the density and outline of the point cloud, and then expanding it with a bounding box, forms a three-dimensional circular buffer region. By utilizing the bounding box and the expansion of the buffer radius, a region containing the point cloud data is established, ensuring that the point cloud data for subsequent analysis can cover the target area. Constructing this buffer region ensures effective processing of the point cloud data within the target area and provides a clear regional division for obtaining information such as laser echo intensity values ​​and color attributes. The aforementioned circular buffer region refers to a dynamically calculated buffer radius centered on the spatial geometric center or sphere of the occlusion candidate region. A three-dimensional spherical spatial region is generated with radius . The purpose is to extend the analysis scope from the original candidate region, which may be incomplete or irregular, to a regular, continuous space that better covers the potential influence range (such as canopy projection, root competition zone).

[0066] Due to the shape differences among different plant types (such as trees and shrubs), the geometric features of their shading areas can be very complex. For trees and shrubs, especially when considering shading relationships, their spatial distribution is often non-uniform and variable. By expanding the buffer area and forming a regular three-dimensional circular buffer area, the complex shading morphology can be simplified, making subsequent analyses more consistent and comparable. This expansion ensures the integrity of data coverage even in irregularly shaped areas, especially in the case of trees shading shrubs, ensuring that complete point cloud data of the shaded shrubs can be effectively obtained. At the same time, by expanding the buffer area, this shading effect can be better simulated, and a larger spatial range can be provided for subsequent analysis. This not only helps to identify the shading relationship between shrubs and trees, but also enables the quantification of the impact of shading effects on plant growth and environmental changes.

[0067] S252: Determine the three-dimensional coordinates of the point cloud in the occlusion candidate region, and determine whether the three-dimensional coordinates of all point clouds are located in the circular buffer region;

[0068] If so, the buffer point cloud density is calculated for the point cloud located in the circular buffer region;

[0069] The laser point cloud device acquires the laser echo intensity value (i.e., the echo intensity reflects the material and reflection properties of the target surface, and can be used as a reference indicator to distinguish between vegetation and non-vegetation).

[0070] Multispectral images are acquired from the target area; the circular buffer area is mapped to the multispectral image to obtain pixel coordinates, and the color attributes of the corresponding pixels (i.e., the colors of the red, green, and blue components, which helps to identify vegetation, buildings, ground, etc.) are extracted.

[0071] The buffer point cloud density, laser echo intensity value, and color attribute are combined to form a multidimensional feature of the circular buffer region;

[0072] It should be noted that in the above steps, the spatial statistical features are used to expand the occlusion candidate region to obtain a three-dimensional circular buffer region. The spatial statistical features are a regional range feature, thus confirming the circular buffer region. Subsequently, step S252, "combining the buffer point cloud density, laser echo intensity value, and color attribute to form a multi-dimensional feature of the circular buffer region," involves analyzing the circular buffer region based on features such as color to obtain multi-dimensional features (i.e., a type of morphological color feature). In specific execution, by determining whether the three-dimensional coordinates of the point cloud are located within the buffer region, valid point cloud data is filtered out, ensuring that only point clouds located within the target region are used to further remove irrelevant or unnecessary data. This allows for focused analysis of point cloud information related to the target region, improving the accuracy and efficiency of the analysis. Furthermore, combining feature data such as laser echo intensity and color attributes (i.e., multispectral images) helps to more accurately analyze the vegetation information of the target region.

[0073] The canopy of a tree may block part of a shrub, causing changes in the light conditions for the shrub's leaves. As the sun's position changes, the area blocked by the tree also changes, thus affecting the amount of light received by the shrub's leaves. When the weather is sunny and the sun is shining directly, the shrub's leaves may appear a brighter green, but in shady areas or during periods of insufficient sunlight, the leaves may turn yellow or yellowish. This may be due to reduced chlorophyll synthesis, which affects photosynthesis, or the plant entering a dormant state, reducing the energy required for photosynthesis.

[0074] The reflectivity of shrub leaves is also closely related to their physiological state. Generally, when chlorophyll is abundant, leaves will exhibit strong green light reflectance (i.e., lower red and blue light reflectance and higher green light reflectance). However, if the chlorophyll concentration decreases, the reflectance spectrum of the leaves will change, and they may exhibit higher red light reflectance, making the leaves appear yellow or brown, especially during periods of strong light.

[0075] Studies have found that the lower angle of sunlight in the morning and evening results in a greater difference in the angle of sunlight received by the leaves. This may cause some shrub leaves to receive more red light and reflect less green light due to the angle, thus they may appear more yellow. At noon, the angle of sunlight is more vertical, and the depth of light penetration into the vegetation varies, which also affects the efficiency of photosynthesis and thus changes the color reflection characteristics of the leaves.

[0076] S253: Utilize deep learning networks to pre-build point cloud segmentation models for the framework (i.e., pre-trained classification tasks for extracting vegetation and non-vegetation point clouds, which is common knowledge and will not be elaborated further).

[0077] The multidimensional features of the circular buffer region are input into the point cloud segmentation model, and the category label of each point cloud in the circular buffer region is output; the category label is vegetation category point cloud and non-vegetation category point cloud;

[0078] It should be noted that the point cloud segmentation model pre-trained by deep learning can accurately classify vegetated and non-vegetated point clouds. By utilizing the efficiency of deep learning models, it can identify vegetated and non-vegetated areas from a large amount of point cloud data and automatically classify point clouds into vegetated and non-vegetated categories, providing a clear vegetated area division for subsequent growth trend analysis.

[0079] S254: Divide the circular buffer area into spatial scale intervals (e.g., if the height is 10m, each 2m can be a scale to form an interval to determine the vegetation type point cloud situation between each interval).

[0080] The number of vegetation type point clouds in each spatial scale interval of the circular buffer area is counted, and the proportion of vegetation type point clouds in each spatial scale interval is calculated (that is, the proportion of vegetation type point clouds to the total number of point clouds in that spatial scale interval, which is used to describe the vegetation height distribution).

[0081] Planar fitting is performed on the vegetation category point cloud for each spatial scale interval, and the curvature of each vegetation category point cloud (i.e., reflecting the local shape details of the vegetation surface) is calculated.

[0082] The color ratio (i.e., the proportion of red, green, and blue components in each spatial scale interval, reflecting whether the vegetation is more green or the tree trunk is more green) is calculated using the color attribute in the multidimensional features; the laser echo intensity value in the multidimensional features and the color ratio are used as texture features (i.e., describing the roughness or structural complexity of the vegetation canopy).

[0083] The proportion of the vegetation category point cloud, the curvature of the vegetation category point cloud, and the texture features are weighted and integrated to obtain a multi-scale feature vector (that is, the multi-scale feature vector contains information such as the proportion, curvature, and texture of each vegetation category point cloud, which can be used to analyze and predict the growth trend of the vegetation category point cloud in the subsequent process).

[0084] It should be noted that by dividing the space into different spatial scale intervals (e.g., by height stratification), vegetation at different height intervals can be analyzed; the proportion, curvature, color proportion, and texture features of vegetation category point clouds can be statistically analyzed to provide detailed information for further analysis of vegetation geometry, color distribution, structural complexity, etc.; the multi-scale analysis method described above is used to identify and analyze different growth characteristics of vegetation at different scales, which helps to describe the spatial distribution and morphological structure of vegetation, and express the growth patterns and environmental adaptability of vegetation;

[0085] In the above steps, curvature and color ratio provide richer spatial information for texture features. By performing planar fitting on the vegetation point cloud of each interval and calculating the curvature, the local shape details of the vegetation surface can be reflected. For example, the canopy structure of shrubs and trees, the curvature of branches and leaves, etc., are very evident in the curvature calculation. Color ratio (especially the RGB components) reflects the distribution of color on the vegetation surface and can distinguish the color differences between the green leaves and the trunk of the plant. For example, in a higher spatial scale interval, there may be more green vegetation (representing the leaves of trees), while in a lower interval, there may be more brown or gray trunks. Therefore, curvature and color ratio can help further distinguish different types of vegetation and their surface features.

[0086] S255: Compare the height change of vegetation category point clouds in the same spatial scale interval (i.e., the same height) between the point cloud data of the first period and the point cloud data of the second period with the time interval (i.e., the two periods of point cloud data were collected at an interval of time, such as 2 days, and the height change is obtained by comparing the height of vegetation category point clouds in the first period and the second period of point cloud data (i.e., the height at the two-day interval)); calculate the height growth rate of vegetation category point clouds by using the height change and time interval.

[0087] The vegetation category point cloud is projected onto a horizontal plane, and a two-dimensional data matrix is ​​constructed from the vegetation category point cloud projected onto the horizontal plane.

[0088] Principal component analysis (PCA) is used to perform PCA on the two-dimensional data matrix to obtain eigenvectors of the eigenvalues; the eigenvector with the largest eigenvalue is selected as the direction of maximum dispersion of the eigenvectors, which is then used as the canopy expansion direction.

[0089] The height growth rate and canopy expansion direction are combined to form growth trend parameters (that is, the canopy growth trend, including the canopy growth rate, height, direction, etc.).

[0090] A growth parameter set is formed using the aforementioned growth trend parameters and multi-scale feature vectors;

[0091] It should be noted that by comparing point cloud data at different time points and analyzing the height changes of vegetation category point clouds and the canopy expansion direction extracted by principal component analysis (PCA), the growth trend of vegetation can be predicted. This method, by combining time series data, captures the dynamic changes in vegetation growth and calculates the rate of vegetation growth and the direction of expansion. It can provide a time-dimensional analysis for monitoring vegetation growth and thus predict the future growth trend of vegetation. In particular, by comparing vegetation changes at different time points, it can express its growth rate and expansion direction, thereby helping decision-makers to plan for vegetation protection, management and restoration.

[0092] S256: Use the growth parameter set to simulate plant growth and construct a branch and leaf geometric model; supplement the branch and leaf geometric model with point clouds; update the point cloud segmentation model with the supplemented point clouds, and re-analyze the multidimensional features of the circular buffer area to obtain a vegetation point cloud; cluster the vegetation point cloud to obtain independent point cloud clusters; calculate the height coordinates of each independent point cloud cluster, and determine the occlusion result based on the height coordinates.

[0093] It should be noted that the above steps, by using a set of growth parameters to simulate plant growth, can accurately simulate the dynamic growth process of plants; this step uses L-system rules to generate the geometric structure of plant branches and leaves, reflecting the expansion of plants over time and their heliotropic mechanisms.

[0094] Based on the geometric model of the plants, supplementary point clouds are generated and the original data is updated. By adding more point cloud data, the spatial information of the plants is enriched, making the model more complete and detailed. The supplementary point cloud ensures that key plant parts are not missed during the simulation, improving the model's accuracy and detail representation. The above-mentioned method of updating the point cloud segmentation model by supplementing point cloud data can more accurately separate vegetation point clouds from non-vegetation point clouds. The multidimensional features of the circular buffer area are re-analyzed to obtain more refined vegetation point clouds. The vegetation point clouds are clustered to obtain independent point cloud clusters, which helps to identify different types or different heights of vegetation groups. Finally, by calculating the height coordinates of each point cloud cluster, it is determined whether plants at different heights are mutually occluding.

[0095] Specifically, such as Figure 4 As shown, in step S256, a branch and leaf geometric model is constructed using the growth parameter set to simulate plant growth; a supplementary point cloud is generated for the branch and leaf geometric model; the point cloud segmentation model is updated using the supplementary point cloud, and the multidimensional features of the circular buffer region are re-analyzed to obtain a vegetation point cloud; the vegetation point cloud is clustered to obtain independent point cloud clusters; the height coordinates of each independent point cloud cluster are calculated, and the occlusion result is determined based on the height coordinates. The specific operation steps are as follows:

[0096] S2561: Use the growth parameter set to set L system rules (i.e. define recursively extended production rules according to plant growth laws, such as A → AB or other extended rules that conform to biological characteristics, and introduce the growth parameter set into the rules).

[0097] The L-system rules are introduced with a sun-facing mechanism (i.e., setting nodes to preferentially grow towards the light source) to simulate plant growth over a continuous time period, generating branch and leaf geometry (as time changes, plant growth is simulated according to the L-system rules and the introduced sun-facing mechanism, generating complete branch and leaf network data, including information on the position, direction, and length (i.e., height or spread growth) of each component).

[0098] Collect the surrounding point cloud of the target area (i.e., the point cloud around the target area, which supplements the point cloud in the target area); extract multi-dimensional features from the surrounding point cloud (i.e., the extraction step is explained in step S252 and will not be repeated here);

[0099] The multidimensional features of the surrounding point cloud are matched with the multidimensional features of the target region (i.e., the circular buffer region) to calculate the error (i.e., the error is calculated by matching using Euclidean distance). The error of these multidimensional features is used to correct the growth parameter set. The corrected growth parameter set is then used to construct a branch and leaf geometric model based on the branch and leaf geometry (i.e., the branch and leaf geometric model reflects the distribution and coverage of the canopy branches and leaves after the above growth simulation; this model is not an algorithm model, but a three-dimensional point cloud model of the branches and leaves). The simulation is considered a differentiable function, with the growth parameter set as input and the simulated point cloud features as output. Then, by comparing the error between the simulated features and the real features, the input parameters are adjusted in reverse to make the simulation result infinitely close to the real world, such as... Figure 5 (as shown)

[0100] It should be noted that the point cloud data of the surrounding area of ​​the target region was collected because the density of the shrubs in the occluded area was unknown, as were the steepness or flatness of their branches. Therefore, the state of the occluded area was inferred by analyzing the surrounding unoccluded area (which is known, i.e., the point cloud data of the surrounding area of ​​the target region). The surrounding point cloud clearly outlines the contours of the unoccluded parts of the vegetation. The simulated growth must start from this contour and maintain geometric continuity, so as to compare the features of the simulated supplementary point cloud with the features of the surrounding point cloud.

[0101] The above steps, using the Lindenmayer system rules, can accurately simulate the recursive growth process of plants and the expansion of their branches and leaves. The Lindenmayer system, based on biological characteristics, defines how different parts of a plant grow over time, ensuring the naturalness and realism of the growth pattern. Defining the Lindenmayer system rules and the set of growth parameters is the foundation for building realistic plant models and supports the dynamic evolution of plants in three-dimensional space. It simulates the growth behavior of plants in natural environments, especially their tendency to grow towards light sources. By introducing a heliotropic mechanism, it simulates how plants optimize their growth direction and distribution according to environmental light sources. This mechanism makes the branch and leaf structure of the plant model more consistent with the growth patterns of real plants, thus making the final model more biologically accurate.

[0102] Based on L-system rules and the sun-facing mechanism, the geometric structure of plant branches and leaves is dynamically generated, including information such as the position, direction, and length of each point. This step provides detailed plant structure data for point cloud generation, forming a plant-based geometric model. Collecting point cloud data of the target area can accurately understand the spatial position of plants in the environment, providing basic data for subsequent modeling and correction. The above steps, by acquiring point clouds, are used to understand the distribution of plants in three-dimensional space in greater detail, thereby helping to analyze and correct errors in plant growth simulation.

[0103] S2562: Set the sampling density for the branch and leaf geometric model (i.e., ensure that the sampling point density is high in key parts (such as forks, leaf edges), randomly sample the points with the highest sampling point density in the branch and leaf geometric model according to the sampling density, and determine the current sampling coordinates;

[0104] Add multidimensional features of the point cloud to the sampled coordinates to serve as a supplementary point cloud;

[0105] The point cloud density is calculated for both the surrounding point cloud and the supplementary point cloud, and the point cloud density difference is further calculated (i.e., for the same vegetation, the shaded and unshaded parts should have the same or similar physical and statistical characteristics; therefore, if the characteristics of the supplementary point cloud are highly consistent with the characteristics of the surrounding point cloud, we consider the simulation to be credible; the lower the error threshold, the more consistent the characteristics).

[0106] The sampling density is adjusted based on the point cloud density difference to obtain a new sampling density (i.e., the density of low-density areas is improved by increasing sampling points; and areas with excessively high density are randomly removed or resampled to achieve density balance).

[0107] The branch and leaf geometry model is resampled using the new sampling density, and a new point cloud density difference is calculated.

[0108] A preset error threshold is established; it is then determined whether the new point cloud density difference is less than the error threshold.

[0109] If so, it is determined that the new supplementary point cloud is seamlessly connected with the branch and leaf geometric model to obtain the final supplementary point cloud (that is, the fundamental purpose of the supplementary point cloud is to repair the data loss caused by occlusion and construct a high-precision digital model that can reflect the true and complete three-dimensional morphology of the vegetation (i.e., the updated point cloud segmentation model); the supplementary point cloud scientifically "repairs" this data gap by simulating growth and generating the most likely morphology of the occluded part).

[0110] It should be noted that the above steps, by setting different sampling density strategies, ensure that key parts of the plant (such as forks, leaf edges, etc.) can be sampled more densely, especially when performing detailed analysis of plant structure; uniform sampling density can enhance the detailed representation of the microstructure of branches and leaves while ensuring computational efficiency.

[0111] The above steps calculate the point cloud density differences and adjust the sampling density to avoid overly dense or sparse point cloud regions, thereby optimizing the distribution of point cloud data, ensuring balanced point cloud density, and preventing errors from accumulating in areas with excessively high or low density. This results in a more accurate and representative model.

[0112] The above process ensures that the supplemented point cloud is seamlessly connected with the original geometric model by judging whether the new point cloud density difference is less than a preset threshold, thus avoiding obvious seams or inconsistencies. Achieving seamless point cloud synthesis is the key to generating a high-quality 3D model and improving the accuracy of subsequent simulations.

[0113] Based on the set sampling density, the optimized branch and leaf geometric model (vector model) is sampled in three-dimensional space to generate an initial set of sampling points containing only geometric positions (x, y, z). Each point in the initial sampling point set is assigned multi-dimensional features learned from the surrounding point cloud to generate a preliminary supplementary point cloud with real physical properties. The preliminary supplementary point cloud is density-matched with the surrounding point cloud, and the sampling strategy is iteratively adjusted to ensure that the density distribution of the two is consistent, and finally the final supplementary point cloud is generated.

[0114] S2563: The point cloud segmentation model is corrected using the final supplementary point cloud to obtain an updated point cloud segmentation model;

[0115] The multidimensional features of the circular buffer region are re-input into the updated point cloud segmentation model for point cloud segmentation to obtain the final vegetation point cloud and non-vegetation point cloud.

[0116] The coordinate positions of the vegetation point cloud are determined; the coordinate positions of the lowest vegetation point cloud are filtered, and the minimum height threshold (usually 0.5-5.0 meters, the minimum height threshold is for shrubs that are blocked by trees, shrubs are usually below 5 meters, mostly between 0.5-2 meters) and the maximum height limit threshold (usually above 5 meters, trees with obviously tall canopies) are preset.

[0117] Determine whether the coordinate position (i.e., height position) of each vegetation point cloud is greater than or equal to the minimum height threshold of tall shrubs and simultaneously less than or equal to the maximum height limit threshold;

[0118] If so, then set a clustering search radius for the selected vegetation point clouds; cluster each selected vegetation point cloud according to the clustering search radius to obtain an independent point cloud cluster (that is, tall shrubs, each tall shrub has different heights, and the height of the tall shrubs is used to determine whether they block the surrounding buildings).

[0119] It should be noted that the above steps involve supplementing point cloud data to correct the segmentation model, which is used to accurately extract vegetated and non-vegetated areas, ensuring the integrity and accuracy of the model. The correction of the point cloud segmentation model allows the final point cloud to more accurately represent different types of land features (such as trees, shrubs, etc.), providing data support for subsequent analysis. In the above steps, by setting a height threshold for tall shrubs, high-rise vegetation that may be blocked by buildings is effectively screened out. Cluster analysis can help to accurately identify these vegetation types and further analyze their shading effects. The analysis of tall shrubs is very important for urban greening, landscape design, and urban planning, as it can help understand the impact of plants on the surrounding environment, especially the shading of light and views.

[0120] S2564: Determine the coordinates of the vertex point clouds within each independent point cloud cluster (i.e., tall shrubs);

[0121] The height of the independent point cloud cluster is calculated based on the coordinate position of the vertex point cloud in the independent point cloud cluster.

[0122] The heights of each independent point cloud cluster are sorted, and the size percentiles are calculated using the heights of different independent point cloud clusters (i.e., the 3rd, 5th (median), 5.5th, and 7th percentile heights are calculated, the height values ​​are sorted, and the percentiles are calculated by position); the size percentile of the independent point cloud cluster with the lowest height among the size percentiles is taken as the lower percentile, and all other size percentiles are taken as the upper percentiles.

[0123] Calculate the area of ​​each independent point cloud cluster at each height (i.e., convert all independent point cloud clusters into a two-dimensional coordinate system, extract the coordinates of all pixels in the point cloud, take the pixel with the smallest x-coordinate (i.e., the leftmost pixel on the x-axis; if there are multiple, take the pixel with the smallest y-coordinate) as the vertex of the convex hull, calculate the angle between this convex hull vertex and all pixels, sort them in descending order of angle, calculate the angle difference between the first two angles, set a preset angle threshold, if the angle difference is greater than the angle threshold, then include the pixels corresponding to the two angles into the convex hull set, and continue to calculate the angle difference for the second and third angles in descending order until all pixels of the convex hull set are obtained, and connect the pixels of all convex hull sets in pixel order to form a convex hull polygon, which is the projected area of ​​the independent point cloud cluster).

[0124] Calculate the area of ​​independent point cloud clusters at each height. Calculate the area overlap using the area of ​​the lower percentile (i.e., the area of ​​independent point cloud clusters at the lower percentile) and the area of ​​each of the remaining upper-level percentiles (i.e., the area of ​​independent point cloud clusters at the upper percentile). (Note: The calculated area overlap is not based on the areas of two adjacent height percentiles, but rather on the area of ​​the lower percentile (i.e., the 3rd percentile height). The lower percentile (i.e., independent point cloud clusters at the 3rd percentile height) might be bottom-level shrubs, so the calculation uses the area of ​​the lower percentile and all other height percentiles; that is, the bottommost shrubs. The overlap is calculated with all other tree canopies at each height to determine the extent of shading of the lowest-lying shrubs by the canopy at each height (area overlap, i.e., point cloud cluster area). This is then compared with the independent point cloud clusters at each higher percentile. Due to the growth structure of trees, the canopy of a tree grows around the trunk, and the canopy layer and the underlying shrubs are dispersed. Not the entire canopy of a tree can shade a single shrub; perhaps only one location of the canopy actually shades the underlying shrubs. However, as sunlight changes, the canopies of other trees may gradually shade the underlying shrubs, such as... Figure 6 As shown; to make the calculations in this scheme more realistic, the area overlap is calculated based on the vertical illumination of the trees by sunlight, reducing the error in calculating the area overlap that may occur when sunlight shines from different directions (due to the rotation of the sun, the sunlight illuminating the canopy will result in different overlaps of shadow areas on the underlying shrubs); first, the area of ​​the independent point cloud clusters at the percentile of the height dimension is subtracted from the area of ​​the independent point cloud clusters at the lower percentile to obtain the intersection area, and then divided by the percentage is the area overlap).

[0125] A preset overlap threshold is set; it is then determined whether the area overlap (i.e., the overlap of the point cloud cluster area) is greater than the overlap threshold.

[0126] If so, then it is determined that the areas of two independent point cloud clusters at the height percentiles are mutually occluded;

[0127] It should be noted that calculating the area overlap of independent point cloud clusters for each height percentile can effectively assess the relative position of tree canopies and shrubs, identify shading candidate areas, and help express the impact of vegetation on buildings, especially in scenarios that consider the shading of trees and shrubs; the calculation and analysis of area overlap can help understand the growth status of different plants, especially the shading of low-canopy trees by tall canopies.

[0128] The above steps, by determining whether the area overlap exceeds a set threshold, can accurately identify and quantify the plant shading effect within different height ranges, and then analyze its shading impact on low-level environmental elements. This can help determine the impact of different plants on light and field of vision, thereby enabling reasonable planning.

[0129] Example 2

[0130] like Figure 7 As shown, the present invention provides a vegetation growth assessment system based on laser point cloud recognition, comprising: a data acquisition module 10; and an analysis module 20.

[0131] The acquisition module 10 is used to acquire point cloud data of the target area at two time intervals (e.g., a two-day time interval); the point cloud data of the two time periods includes point cloud data of the first time period and point cloud data of the second time period.

[0132] The analysis module 20 is used to determine occlusion candidate regions by the comprehensive difference value between the point cloud densities in the point cloud data of the two periods, analyze the spatial statistical characteristics of the occlusion candidate regions, classify the occlusion candidate regions into vegetation categories using the spatial statistical characteristics, analyze the growth parameter set of the point cloud, perform plant growth simulation on the growth parameter set, and determine the occlusion result of the vegetation point cloud after the growth simulation.

[0133] The vegetation growth assessment method and system based on laser point cloud recognition provided by this invention, in practical applications, distributes point cloud data from two periods into three-dimensional grids of different scales. Each scale grid represents a different resolution, capable of capturing local changes within different ranges. Density statistics are performed on voxels within each grid to obtain point cloud density differences at different time periods. These differences reflect the point cloud changes in local areas between different time points. The scheme optimizes sensitivity to changes at each scale by assigning weight coefficients to grids of different scales, comprehensively considering both the global field of view at a coarse scale and local changes at a fine scale. The weighted comprehensive difference value accurately expresses the changes in local areas within two time periods, and based on this, determines whether there are occlusion candidate areas.

[0134] Further refine the identification of potential occlusion regions, convert these regions into binary data to simplify subsequent analysis; then extract geometric features for each occlusion candidate region, calculate area, AAA, and boundary descriptors, and use the convex hull algorithm to determine the region's outline. By calculating the point cloud density and spatial statistical characteristics of the region, it is possible to further accurately locate and analyze occlusion regions of different shapes, especially in identifying the occlusion relationship between trees and shrubs in terms of geometric features.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vegetation growth assessment method based on laser point cloud recognition, characterized in that, The following steps are included: Two periods of point cloud data are collected from the target area at an interval; the two periods of point cloud data include point cloud data of the first period and point cloud data of the second period. Based on the comprehensive difference value between the point cloud density in the point cloud data of the two periods, occlusion candidate regions are determined, and the spatial statistical characteristics of the occlusion candidate regions are analyzed. The spatial statistical features are used to classify the occlusion candidate region into vegetation categories and a point cloud analysis growth parameter set; plant growth simulation is performed on the growth parameter set to determine the occlusion result of the vegetation point cloud after growth simulation.

2. The vegetation growth assessment method based on laser point cloud recognition according to claim 1, characterized in that, Based on the comprehensive difference in point cloud density between the two periods of point cloud data, candidate occlusion regions are determined. The specific operation steps are as follows: The point cloud data of the two periods are respectively set into voxels with multi-scale grids; the point cloud density of the first period and the point cloud density of the second period are calculated according to the number of points of each voxel in the point cloud data of the two periods; the comprehensive difference value is calculated using the point cloud density of the first period and the point cloud density of the second period; voxels are selected as occlusion candidate regions according to the comprehensive difference value of each voxel and the preset difference density threshold.

3. The vegetation growth assessment method based on laser point cloud recognition according to claim 2, characterized in that, The spatial statistical characteristics of the occlusion candidate regions are analyzed, and the specific operation steps are as follows: The number of pixels in each occlusion candidate region is calculated as the region area; the coordinates of voxels in each occlusion candidate region are determined, and the maximum and minimum values ​​of the voxel coordinates in each occlusion candidate region are found to obtain the bounding box of the occlusion candidate region; the bounding box is combined with the outline to form a boundary descriptor. The number of point clouds in each occlusion candidate region is calculated and used as the region point cloud density. The region area boundary descriptor and the region point cloud density are combined as spatial statistical features.

4. The vegetation growth assessment method based on laser point cloud recognition according to claim 3, characterized in that, The spatial statistical features are used to classify the occlusion candidate region into vegetation categories, and a point cloud growth parameter set is analyzed. Plant growth simulation is performed on the growth parameter set to determine the occlusion result of the vegetation point cloud after growth simulation. The specific operation steps are as follows: The occlusion candidate region is expanded using the spatial statistical features to obtain a three-dimensional circular buffer region; the circular buffer region is analyzed to obtain multi-dimensional features. The multidimensional features are input into a pre-constructed point cloud segmentation model to classify vegetation category point clouds; the circular buffer region is divided into spatial scale intervals, and the growth parameter set of the vegetation category point cloud in each spatial scale interval is analyzed; the point cloud segmentation model is updated using the growth parameter set to simulate plant growth and obtain vegetation point clouds again; the vegetation point clouds are clustered to calculate height and determine occlusion results.

5. The vegetation growth assessment method based on laser point cloud recognition according to claim 4, characterized in that, The occlusion candidate region is expanded using the spatial statistical features to obtain a three-dimensional circular buffer region; the circular buffer region is then analyzed to obtain multi-dimensional features. The specific operation steps are as follows: The spatial statistical features are used to calculate the occlusion candidate region to obtain the buffer radius; the buffer radius is then expanded to form a three-dimensional circular buffer region. The three-dimensional coordinates of the point cloud in the occlusion candidate region are determined, and it is determined whether the three-dimensional coordinates of all point clouds are located in the circular buffer region. If so, the buffer point cloud density is calculated for the point cloud located in the circular buffer region; Acquire laser echo intensity values ​​from the laser point cloud device; A multispectral image is acquired from the target area; the circular buffer area is mapped to the multispectral image to obtain the pixel coordinates, and the color attributes of the corresponding pixels are extracted. The buffer point cloud density, laser echo intensity value, and color attribute are combined to form a multidimensional feature of the circular buffer region.

6. The vegetation growth assessment method based on laser point cloud recognition according to claim 5, characterized in that, The multidimensional features are input into a pre-constructed point cloud segmentation model to divide the point cloud into vegetation categories; the circular buffer region is divided into spatial scale intervals, and the growth parameter set of the vegetation category point cloud in each spatial scale interval is analyzed. The specific operation steps are as follows: A point cloud segmentation model is pre-constructed using a deep learning network as the framework; the multidimensional features of the circular buffer region are input into the point cloud segmentation model, and the vegetation category point cloud in the circular buffer region is output; the circular buffer region is divided into spatial scale intervals. Multi-scale feature vectors are extracted from the spatial scale range using the point cloud of the vegetation categories. Compare the height growth rate of vegetation category point clouds in the same spatial scale interval between the point cloud data of the first period and the point cloud data of the second period; project the vegetation category point clouds onto a horizontal plane, and use the vegetation category point clouds projected onto the horizontal plane to obtain the canopy expansion direction; The height growth rate and canopy expansion direction are combined to form growth trend parameters; A growth parameter set is formed using the growth trend parameters and multi-scale feature vectors.

7. The vegetation growth assessment method based on laser point cloud recognition according to claim 6, characterized in that, The point cloud segmentation model is updated using the growth parameter set to simulate plant growth, and the vegetation point cloud is obtained again. The vegetation point cloud is then clustered to calculate its height and determine the occlusion results. The specific steps are as follows: A branch and leaf geometric model is constructed using the growth parameter set to simulate plant growth; a supplementary point cloud is generated for the branch and leaf geometric model; the point cloud segmentation model is updated using the supplementary point cloud, and the multidimensional features of the circular buffer region are re-analyzed to obtain a vegetation point cloud; the vegetation point cloud is clustered to obtain independent point cloud clusters; the height coordinates of each independent point cloud cluster are calculated, and the occlusion result is determined based on the height coordinates.

8. The vegetation growth assessment method based on laser point cloud recognition according to claim 7, characterized in that, A branch and leaf geometric model is constructed using the aforementioned growth parameter set to simulate plant growth; the point cloud is then supplemented into the branch and leaf geometric model. The specific operation steps are as follows: The growth parameter set is used to set L-system rules to introduce a sun-seeking mechanism; Collect the surrounding point cloud of the target area; extract multidimensional features from the surrounding point cloud; The multidimensional features of the surrounding point cloud are matched with the multidimensional features of the target region to calculate the error; the error of the multidimensional features is used to correct the growth parameter set; and the corrected growth parameter set is used to construct a branch and leaf geometric model with the branch and leaf geometric structure. A sampling density is set for the branch and leaf geometric model. Based on the sampling density, the points with the highest sampling density in the branch and leaf geometric model are randomly sampled, and the current sampling coordinates are determined. Multidimensional features of the point cloud are added to the sampling coordinates as supplementary point cloud. The point cloud density is calculated for the surrounding point cloud and the supplementary point cloud respectively, and the point cloud density difference is further calculated; the sampling density is adjusted according to the point cloud density difference to obtain a new sampling density; The branch and leaf geometric model is resampled using the new sampling density, and a new point cloud density difference is calculated; it is then determined whether the new point cloud density difference is less than a preset error threshold. If so, the new supplementary point cloud is determined to be seamlessly integrated with the branch and leaf geometric model to obtain the final supplementary point cloud.

9. The vegetation growth assessment method based on laser point cloud recognition according to claim 8, characterized in that, The point cloud segmentation model is updated using supplementary point clouds, and the multidimensional features of the circular buffer region are re-analyzed to obtain vegetation point clouds. These vegetation point clouds are then clustered to obtain independent point cloud clusters. The height coordinates of each independent point cloud cluster are calculated, and the occlusion result is determined based on these height coordinates. The specific steps are as follows: The point cloud segmentation model is corrected and updated using the final supplementary point cloud. The multidimensional features of the circular buffer region are re-input into the updated point cloud segmentation model for point cloud segmentation to obtain the final vegetation point cloud. A clustering search radius is set. Each selected vegetation point cloud is clustered according to the clustering search radius to obtain an independent point cloud cluster. For each independent point cloud cluster, determine the coordinate positions of the vertex point clouds within the cluster, and use these coordinate positions as the coordinate positions of the independent point cloud cluster. Sort all independent point cloud clusters by height and calculate the size percentiles. The lowest height of the independent point cloud cluster in the size percentiles is taken as the lower percentile, and all other size percentiles are taken as the upper percentile. Calculate the area of ​​independent point cloud clusters at each height; calculate the area overlap using the area of ​​the lower percentile and the area of ​​each of the remaining upper-level size percentiles; determine, based on the area overlap, that the areas of independent point cloud clusters at two height size percentiles are mutually occluding.

10. A vegetation growth assessment system based on laser point cloud recognition, characterized in that, include: Data acquisition module; Analysis module; The acquisition module is used to acquire point cloud data of the target area at two time intervals; the point cloud data of the two time intervals includes point cloud data of the first time interval and point cloud data of the second time interval. The analysis module is used to determine occlusion candidate regions by the comprehensive difference value between the point cloud density in the point cloud data of the two periods, and to analyze the spatial statistical characteristics of the occlusion candidate regions. The spatial statistical features are used to classify the occlusion candidate region into vegetation categories and a point cloud analysis growth parameter set; plant growth simulation is performed on the growth parameter set to determine the occlusion result of the vegetation point cloud after growth simulation.

Images