A single tree detection method, system and computer device

Abstract

The present application relates to a kind of single wood detection method, system and computer equipment, the method comprises: obtaining the point cloud data of target forest area, according to the point cloud data, constructs the digital surface model of target forest area;Based on the elevation information of the digital surface model, using the preset multi-order partial derivative calculation mode, the gradient feature information of target forest area is obtained;According to the gradient feature information, over-segmentation is carried out to crown patch, obtains several over-segmented crown patches;Based on the preset multi-feature similarity weight calculation method, the clustering analysis is carried out to the several over-segmented crown patches, the crown patch belonging to the same single wood is merged, and the single wood detection result is obtained.The method provided in the present application can accurately obtain the crown boundary and tree top point, and the over-segmented crown patches are merged according to the similarity weight value of multi-feature, effectively avoid segmentation error, and improve the single wood detection precision.

Description

Technical Field

[0001] The present invention relates to the technical field of forest resource surveys, and in particular to a method, system and computer equipment for detecting individual trees. Background Technology

[0002] Forests, as a vital component of terrestrial ecosystems, play a crucial role in maintaining ecosystem balance, protecting biodiversity, and mitigating global climate change. Individual trees, as the smallest unit of a forest, require efficient and accurate extraction of their structural parameters. These parameters are essential for precise carbon storage assessment and ecological environment modeling research, and are crucial for forest resource surveys, improving forest resource quality and efficiency, and maintaining ecosystem functions. Given the fragmented distribution and wide coverage of forest resources in my country, traditional manual field survey methods are time-consuming, labor-intensive, costly, and heavily influenced by subjective factors, such as significant deviations in crown width and tree height measurements, making them unsuitable for the need for accurate and efficient continuous large-scale forest resource surveys. New remote sensing technologies, especially UAV remote sensing technology, have established a multi-platform, multi-angle, and multi-mode three-dimensional observation system, offering advantages such as large-scale, spatially continuous, and dynamic monitoring, effectively solving the problem of researchers being unable to conduct in-depth operations in complex terrain conditions. In recent decades, researchers have developed numerous individual tree segmentation algorithms. Among them, digital surface models / canopy height models are the main data source for individual tree segmentation.

[0003] However, existing tree segmentation algorithms based on Digital Surface Model (DSM) / Canopy Height Models all exhibit oversegmentation and undersegmentation to varying degrees, resulting in significant uncertainty in the segmentation results. To address this issue, researchers have attempted to combine algorithms prone to oversegmentation and undersegmentation to fully leverage their respective advantages. For example, watershed segmentation algorithms combined with point cloud spatial information (Ncut algorithm), mean-shift algorithms combined with tree vertex detection strategies (Ncut algorithm), and marker-controlled watershed algorithms combined with hyperspectral information (Ncut algorithm). These methods can reduce the number of nodes and time complexity in forest environments with minimal topographic relief and low canopy closure, improving segmentation accuracy and reducing oversegmentation and undersegmentation to some extent. However, for high-canopy-closure stands with varying canopy shapes, watershed segmentation algorithms still struggle to completely avoid undersegmentation, making it difficult to maximize the effectiveness of different algorithm combination strategies. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a single tree detection method, system and computer equipment that can accurately obtain the tree crown boundary and tree apex, and merge over-segmented tree crown patches according to the similarity weight values ​​of multiple features, effectively avoiding segmentation errors and improving the accuracy of single tree detection.

[0005] Firstly, this application provides a method for detecting individual trees, including:

[0006] Acquire point cloud data of the target forest area, and construct a digital surface model of the target forest area based on the point cloud data;

[0007] Based on the elevation information of the digital surface model, the gradient feature information of the target forest area is obtained by using a preset multi-order partial derivative calculation method.

[0008] Based on the gradient feature information, the canopy patches are oversegmented to obtain several oversegmented canopy patches;

[0009] Based on a preset multi-feature similarity weight calculation method, cluster analysis is performed on the several over-segmented canopy patches, and canopy patches belonging to the same single tree are merged to obtain the single tree detection results.

[0010] Secondly, this application provides a single-tree detection system, comprising:

[0011] The data acquisition module is used to acquire point cloud data of the target forest area and construct a digital surface model of the target forest area based on the point cloud data.

[0012] The multi-order gradient calculation module is used to obtain the gradient feature information of the target forest area based on the elevation information of the digital surface model and by using a preset multi-order partial derivative calculation method.

[0013] The oversegmentation module is used to oversegment the canopy patches according to the gradient feature information to obtain several oversegmented canopy patches;

[0014] The single tree detection module is used to perform cluster analysis on the several over-segmented canopy patches based on a preset multi-feature similarity weight calculation method, merge canopy patches belonging to the same single tree, and obtain the single tree detection result.

[0015] Thirdly, this application provides a computer device including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the single-tree detection method as described in any of the above.

[0016] The tree detection method, system, and computer equipment provided in this application utilize UAV remote sensing technology to acquire point cloud data of a target forest area, construct a digital surface model based on the point cloud data, and acquire the elevation information of the digital surface model. Using the elevation information, a multi-order partial derivative calculation method is employed to obtain the gradient feature information of the target forest area. Based on the gradient feature information, the area is over-segmented to obtain several over-segmented canopy patches. Then, based on a multi-feature similarity weight calculation method, the similarity weights between adjacent canopy patches are compared. Based on the similarity weight values, it is determined whether the canopy patches originate from the same tree, thereby clustering the canopy patches, merging canopy patches belonging to the same tree, and finally outputting the tree detection result. Compared to existing technologies, this application directly uses the elevation information of digital surface models and employs multi-order partial derivative calculations to identify gradient features of target forest areas, thereby determining canopy patches. By integrating tree growth characteristics and differences in canopy structure, and based on the similarity weights of spectral, texture, and color space information, over-segmented canopy patches are merged to complete individual tree detection. This enables the extraction of canopy parameters for individual trees of different species, providing technical support for detailed forest resource surveys and accurate carbon storage estimation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the steps of a single-tree detection method provided in Embodiment 1 of this application;

[0019] Figure 2 A schematic diagram of a single-tree detection method provided in Embodiment 1 of this application;

[0020] Figure 3 A flowchart illustrating the steps for obtaining gradient feature information of a target forest region, provided in Embodiment 1 of this application;

[0021] Figure 4 A schematic diagram of an oversegmentation method provided in Embodiment 1 of this application;

[0022] Figure 5 A flowchart illustrating the steps of single-tree segmentation based on multi-feature similarity provided in Embodiment 1 of this application;

[0023] Figure 6 This is a schematic diagram of the structure of a single-tree detection system provided in Embodiment 2 of this application;

[0024] Figure 7 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the protection scope of this application.

[0026] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0027] In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0028] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0029] Existing tree segmentation algorithms based on Digital Surface Models (DSMs) / Canopy Height Models all exhibit oversegmentation and undersegmentation to varying degrees, resulting in significant uncertainty in the segmentation results. Improvements to existing technologies, such as the Ncut algorithm combining watershed segmentation with point cloud spatial information, the Ncut algorithm combining mean-shift algorithm with tree vertex detection strategy, and the Ncut algorithm combining marker-controlled watershed segmentation with hyperspectral information, can reduce the number of nodes and time complexity in forest environments with relatively flat terrain and low canopy closure, thus improving segmentation accuracy and reducing oversegmentation and undersegmentation to some extent. However, for high-canopy-closure stands with varying canopy shapes, watershed segmentation algorithms still struggle to completely avoid undersegmentation, making it difficult to maximize the effectiveness of different algorithm combination strategies.

[0030] To address this, this application provides a method, system, and computer device for detecting individual trees. It directly uses the elevation information of a digital surface model, calculates the gradient feature information of the forest area based on multi-order partial derivatives, and performs over-segmentation of the forest area into canopy patches. Then, based on the calculation method of multi-feature similarity weights, it performs cluster analysis on the canopy patches by integrating spectral features, texture features, and color space features, thereby realizing individual tree detection, effectively avoiding segmentation errors, and improving the accuracy of individual tree detection.

[0031] The term "single tree" as used in this application refers to the basic unit in forest resource surveys, namely, a single tree in a forest ecosystem. As the smallest unit of a forest, the identification and segmentation of a single tree is fundamental to the accurate extraction of forest parameters (such as tree height, crown width, and crown area).

[0032] Example 1

[0033] Please see Figure 1 and Figure 2 , Figure 1 This is a flowchart illustrating the steps of a single-tree detection method provided in Embodiment 1 of this application. Figure 2 This is a schematic diagram of a single-tree detection method provided in Embodiment 1 of this application.

[0034] The single-tree detection method provided in this application includes:

[0035] S101, acquire point cloud data of the target forest area, and construct a digital surface model of the target forest area based on the point cloud data;

[0036] S102, Based on the elevation information of the digital surface model, the gradient feature information of the target forest area is obtained by using a preset multi-order partial derivative calculation method;

[0037] S103, based on the gradient feature information, the canopy patches are over-segmented to obtain several over-segmented canopy patches;

[0038] S104. Based on the preset multi-feature similarity weight calculation method, cluster analysis is performed on the several over-segmented canopy patches, and canopy patches belonging to the same single tree are merged to obtain the single tree detection result.

[0039] The single-tree detection method provided in this application uses UAV remote sensing technology to acquire point cloud data of the target forest area, constructs a digital surface model based on the point cloud data, and acquires the elevation information of the digital surface model. Using the elevation information, a multi-order partial derivative calculation method is used to obtain the gradient feature information of the target forest area. Based on the gradient feature information, the area is over-segmented to obtain several over-segmented canopy patches. Then, based on the calculation method of multi-feature similarity weight, the similarity weight between adjacent canopy patches is compared. Based on the similarity weight value, it is determined whether the canopy patches come from the same single tree, thereby clustering the canopy patches, merging canopy patches belonging to the same single tree, and finally outputting the single-tree detection result. Compared to existing technologies, this application directly uses the elevation information of digital surface models and employs multi-order partial derivative calculations to identify gradient features of target forest areas, thereby determining canopy patches. By integrating tree growth characteristics and differences in canopy structure, and based on the similarity weights of spectral, texture, and color space information, over-segmented canopy patches are merged to complete individual tree detection. This enables the extraction of canopy parameters for individual trees of different species, providing technical support for detailed forest resource surveys and accurate carbon storage estimation.

[0040] For step S101, point cloud data of the target forest area is obtained, and a digital surface model of the target forest area is constructed based on the point cloud data.

[0041] In one embodiment, step S101 includes:

[0042] The S10 acquires point cloud data of the target forest area based on multi-angle measurement by UAV and LiDAR scanning.

[0043] In one embodiment, a target forest area is established through field surveys to obtain the location information of trees within the target forest area. For example, the location information of trees can be obtained using a Zenith 45 Global Navigation Satellite System (GNSS) receiver. The point cloud data is a set of three-dimensional spatial data acquired by remote sensing equipment, consisting of a large number of points, each containing spatial coordinate information. In one embodiment, point cloud data of the target forest area is obtained through LiDAR or UAV multi-angle measurement data. UAV multi-angle measurement refers to using a UAV equipped with a multi-angle photogrammetry (UMP) system to collect image data of the target forest area from different angles, thereby generating point cloud data. LiDAR scanning can directly obtain elevation point cloud data, and LiDAR scanning is not affected by lighting conditions and can adapt to complex terrain.

[0044] S11, perform data preprocessing on the point cloud data.

[0045] The initial point cloud data acquired by remote sensing equipment contains noise. Data preprocessing is performed on the initial point cloud data to eliminate the noise and thus eliminate data errors.

[0046] The data preprocessing steps may include filtering and denoising. Filtering algorithms are applied to remove outliers and low-quality data, while retaining relevant point cloud data. Smoothing processes are used to reduce random noise and ensure the continuity and consistency of the point cloud data.

[0047] In this embodiment, the acquired point cloud data is preprocessed to improve the quality of the point cloud data and increase the accuracy of single tree detection.

[0048] S12, based on the preprocessed point cloud data, construct a digital surface model of the target forest area.

[0049] Based on the point cloud data, a digital surface model of the target forest area is constructed. The digital surface model stores elevation data in raster form, with each pixel representing the canopy surface elevation.

[0050] In this embodiment, after constructing a digital surface model based on point cloud data, elevation information is directly obtained to extract gradient feature information, instead of using a CHM model obtained by normalizing a digital terrain model (DTM) for single tree detection. Directly using the elevation information data of the model can reduce the influence of complex terrain (such as hillsides and valleys) on the shape of the tree crown and the position of the tree apex, thereby improving the reliability of segmentation.

[0051] For step S102, based on the elevation information of the digital surface model, the gradient feature information of the target forest area is obtained by using a preset multi-order partial derivative calculation method.

[0052] The digital surface model can be defined as follows: x and y represent the horizontal coordinates in the x and y directions, respectively, and z represents the elevation value.

[0053] In this embodiment, the continuous elevation distribution of the canopy surface of the target forest area is directly reflected based on the digital surface model, thereby reflecting the canopy structure.

[0054] Please see Figure 3 , Figure 3 A flowchart illustrating the steps for obtaining gradient feature information of a target forest region, provided in Embodiment 1 of this application. In one embodiment, step S102 includes:

[0055] S201, Based on the digital surface model, the first-order partial derivative of the digital surface model is calculated to obtain the first-order gradient information of the digital surface model; wherein, the first-order gradient information includes the first-order gradient vector and the magnitude of the first-order gradient vector.

[0056] In one embodiment, step S201 includes:

[0057] Calculate the digital surface model in x direction and y The gradient vector in the direction is used to obtain the first-order gradient vector:

[0058]

[0059] in, and Digital surface models exist x direction and y First-order partial derivative in the direction, This is the first-order gradient vector of the digital surface model.

[0060] Calculate the magnitude of the first-order gradient vector of the digital surface model based on the first-order gradient vector:

[0061]

[0062] in, The magnitude of the first-order gradient vector.

[0063] In this embodiment, the first-order gradient vector represents the rate of change of elevation in the gradient direction, and the magnitude of the first-order gradient vector represents the magnitude of the rate of change. The rate of change of tree crown elevation can be obtained based on the first-order gradient vector and the magnitude of the first-order gradient vector, which is used to identify the tree crown boundary.

[0064] S202, based on the calculation result of the first-order partial derivative of the digital surface model, the second-order partial derivative of the digital surface model is calculated to obtain the second-order gradient information of the digital surface model; wherein, the second-order gradient information includes the second-order gradient vector and the magnitude of the second-order gradient vector.

[0065] In one embodiment, step S202 includes:

[0066] Based on the first-order gradient vector, calculate the second-order gradient vector of the digital surface model:

[0067]

[0068] in, and Digital surface models in x direction and y The second partial derivative in the direction, is the second-order gradient vector of the digital surface model.

[0069] In this embodiment, after calculating the first-order partial derivative of the digital surface model, the first-order derivative of the digital surface model is obtained. Then, partial derivatives are calculated on the first derivative of the digital surface model to calculate the second derivative of the digital surface model.

[0070] Calculate the magnitude of the second-order gradient vector of the digital surface model based on the second-order gradient vector:

[0071]

[0072] in, The magnitude of the second-order gradient vector.

[0073] In this embodiment, the second-order partial derivatives of the digital surface model are calculated to obtain second-order gradient information. The magnitude of the second-order gradient vector is used to determine the concavity and convexity of the tree canopy surface. Convex points correspond to tree apexes, and concave points correspond to tree canopy boundaries. Based on the second-order gradient information, the extreme values ​​of the tree canopy surface can be distinguished, thereby further identifying tree apexes and tree canopy boundaries.

[0074] S203, based on the first-order gradient information and the second-order gradient information, the gradient feature information of the target forest area is obtained; wherein, the gradient feature information includes the rate of change of tree canopy elevation and the concavity and convexity of the tree canopy surface.

[0075] The gradient feature information includes first-order gradient information and second-order gradient information. It integrates the rate of change of tree canopy elevation and the concavity and convexity of the tree canopy surface, and uses the edge detection operator of first-order derivative and second-order derivative to perform feature recognition, providing conditions for over-segmentation of tree canopy patches.

[0076] Please see Figure 4 , Figure 4 This is a schematic diagram of an oversegmentation method provided in Embodiment 1 of this application. In step S103, the canopy patches are oversegmented based on the gradient feature information to obtain several oversegmented canopy patches.

[0077] In one embodiment, step S103 includes:

[0078] S301, Based on the gradient feature information, obtain the canopy boundary and several tree vertices in the target forest area.

[0079] S302, based on the canopy boundary and several tree vertices, the target forest area is over-segmented to obtain several over-segmented canopy patches.

[0080] This application uses an oversegmentation method to divide the target forest area, dividing the canopy area into several canopy patches. The number of canopy patches can be much greater than the actual number of individual trees, thereby subdividing the canopy area.

[0081] The canopy patches are canopy components obtained through oversegmentation. In the subsequent step of clustering the canopy patches to form individual trees, each individual tree may include several canopy patches.

[0082] In one embodiment, the canopy boundary is obtained based on the first-order gradient information in the gradient feature information. The canopy boundary can be the boundary between different tree canopies, or the boundary between a tree and the ground or other non-tree areas. The location of trees is identified based on the canopy boundary. Then, the concavity and convexity features of the canopy surface are obtained based on the second-order gradient information in the gradient feature information, thereby identifying the tree apex. The canopy boundary is further distinguished based on the second-order gradient information, thereby improving the recognition accuracy.

[0083] In this embodiment, gradient feature information of the forest canopy region is obtained by using multi-order partial derivatives of elevation information. The canopy region is over-segmented, dividing it into several canopy patches. By over-segmenting to the maximum extent, the problem of under-segmentation caused by canopy overlap in forests with high canopy density is solved, thereby improving detection accuracy.

[0084] For step S104, based on the preset multi-feature similarity weight calculation method, cluster analysis is performed on the several over-segmented canopy patches, and canopy patches belonging to the same single tree are merged to obtain the single tree detection result.

[0085] In this embodiment, a multi-feature similarity weight calculation method is used for the several over-segmented canopy patches. Based on the similarity magnitude, it is determined whether the canopy patches belong to the same single tree. Canopy patches belonging to the same single tree are clustered and merged to obtain the single tree detection result.

[0086] Please see Figure 5 , Figure 5 A flowchart illustrating the steps of single-tree segmentation based on multi-feature similarity provided in Embodiment 1 of this application. In one embodiment, step S104 includes:

[0087] S401, obtain the feature information of the several over-segmented canopy patches; wherein, the feature information includes spectral features, texture features and color space features.

[0088] Because different trees have different growth characteristics and crown structures, they will present different characteristic information. The characteristic information of the crown of the same tree is more similar to that of other different trees. Therefore, the characteristic information of the crown patch can be used to determine whether the crown patch belongs to the same tree.

[0089] In one embodiment, the feature information includes spectral features, texture features, and color space features. By calculating the similarity of feature information between canopy patches, the over-segmented canopy patches are merged, thereby achieving individual tree segmentation and extraction of individual tree parameters.

[0090] Among them, the spectral features are the reflectance attributes of the canopy patches in optical images, used to distinguish vegetation types; the texture features are the surface structure information of the canopy patches, such as mean, entropy, contrast, etc., reflecting the roughness or uniformity of the canopy; the color space features are the attributes of the canopy patches in the color model, such as hue, saturation, etc., used to enhance feature distinguishability.

[0091] S402, Based on the feature information of the canopy patches, a weighted undirected graph is constructed; wherein, the vertices of the weighted undirected graph are the center points of the oversegmented canopy patches, and the similarity between the oversegmented canopy patches is used as the similarity weight value of the connected edges.

[0092] In one embodiment, a weighted undirected graph is constructed. The weighted undirected graph, used to quantify the similarity between canopy patches, consists of several vertices. V and several edges E The system is composed of a weighted undirected graph, with the center point of each canopy patch as a vertex. Adjacent canopy patches are connected by edges, and the similarity between the canopy patches is used as the weight of the connecting edges. Based on the constructed weighted undirected graph, the relationships between adjacent canopy patches are obtained to aid in the similarity analysis of canopy patches.

[0093] In one embodiment, the similarity of the spectral features, texture features, and color space features of the canopy patches is used as the weight value of the connected edges.

[0094] S403, based on a preset similarity weight function, calculates the similarity weight value between adjacent canopy patches.

[0095] In one embodiment, step S403 includes:

[0096] S4031, respectively obtain the feature variables of several spectral features, several texture features and several color space features corresponding to each pixel of the several over-segmented canopy patches, and perform normalization calculation to obtain the average value of several spectral features, several texture features and several color space feature variables corresponding to each canopy patch; wherein, the spectral features, the texture features and the color space features each include several feature variables.

[0097] The spectral features, texture features, and color space features may each include several feature variables. For example, in one embodiment, six spectral feature variables (e.g., overgreen vegetation index, green leaf algorithm, normalized differential vegetation index, etc.), 25 texture feature variables (e.g., mean, entropy, contrast, second moment of angle, homogeneity, etc.), and four color space feature variables (e.g., hue, saturation, etc.) are selected. Based on the parameter values ​​of the feature variables corresponding to each feature information of each pixel in each canopy patch, the average value of the feature variables of the several oversegmented canopy patches is calculated by normalization.

[0098] In this embodiment, by obtaining the parameter values ​​of several feature variables corresponding to each feature information in each pixel of the canopy patch, and performing normalization calculation on the feature variables, the average value of each feature variable of the canopy patch is obtained.

[0099] S4032, according to the preset weight calculation formula, calculate the weight values ​​of the several feature variables between adjacent canopy patches respectively.

[0100] In one embodiment, the average value of the same feature variable among adjacent canopy patches is obtained, and a weighting formula is used to calculate the weight value of the feature variable of adjacent canopy patches. For example, in one embodiment, the weighting formula is:

[0101]

[0102] in, Canopy patches i and canopy patches j The weight value, and Canopy patchesi and canopy patches j The average value of a certain characteristic variable in the spectral information.

[0103] S4033, select the feature variable with the largest weight value among the spectral features, texture features and color space features as input variables and input them into the preset similarity weight function to obtain the similarity weight value between adjacent canopy patches.

[0104] In this embodiment, based on the average value of each feature variable in each feature information between adjacent canopy patches, the corresponding weight value is calculated respectively. The feature variable with the highest weight value in each feature information is selected as the input variable and input into the similarity weight function to calculate the similarity weight value of adjacent canopy patches, thereby clustering the canopy patches.

[0105] In one embodiment, the similarity weight function is:

[0106]

[0107] in, Adjacent patches i and j The similarity weight between them and These are the average spectral characteristics of canopy patches i and j, respectively. and Canopy patches i and canopy patches j The average value of the texture features, and Canopy patches i and canopy patches j The average value of the color space characteristics.

[0108] In this embodiment, the logarithmic function ( As the basic function of the similarity weight function, the normalization method is used to... x The range is fixed between 0 and 1 to increase the growth rate of the similarity weight function. For the logarithmic function, its magnitude increases with... x The decrease in similarity weights leads to a rapid increase in similarity weights between canopy patches, thereby increasing the weights between adjacent canopy patches of the same tree.

[0109] S404, perform cluster analysis on the canopy patches based on the weight values ​​between the canopy patches, merge canopy patches with similarity weight values ​​higher than the preset value into canopy patches of the same single tree, and obtain the single tree detection result.

[0110] In this embodiment, a similarity weight calculation method is used to perform cluster analysis on several over-segmented canopy patches based on similarity weight values. Canopy patches with high similarity weight values ​​are determined to be canopy regions of the same single tree. Canopy patches of the same single tree are then clustered and merged to complete single tree detection.

[0111] In one embodiment, the feature variable with the largest weight value is input into the weighted undirected graph to generate a weight symmetric matrix. and diagonal matrix D (The elements on the diagonal are the sum of the similarity weights of canopy patch i with other canopy patches). The higher the weight value between canopy patches, the more likely they belong to the same tree, and vice versa.

[0112] In one embodiment, the Rayleigh entropy algorithm is used to obtain the system's feature values ​​and feature vectors. Based on the feature vectors, the canopy patches are segmented, thereby merging canopy patches with higher weight values ​​into the same single tree.

[0113] In one embodiment, after clustering, the individual tree detection results are output, which may include parameters such as the individual tree position (e.g., tree vertex coordinates), crown width, crown area, and crown perimeter.

[0114] Using the location data of each tree from the field survey as static reference data for tree vertex identification, however, factors such as competition among trees, terrain, and weather conditions during data acquisition may cause horizontal deviations between the tree vertex and tree base locations. Therefore, this study overlays the tree vertices detected by the individual tree segmentation algorithm with the reference canopy boundary to determine the number of Truepositive (TP), False Positive (FP), and False Negative (FN) detections. If only one tree vertex detected by the algorithm exists within the reference canopy boundary, it is TP (number of correctly detected trees). When multiple tree vertices detected by the algorithm exist within the reference canopy boundary, the point closest to the individual tree location is selected as TP, and the rest are FP (number of falsely detected trees). The case where no tree vertices are detected by the algorithm within the reference canopy boundary is called FN (number of undetected trees).

[0115] In one embodiment, recall (r), precision (p), overall precision (F_score, f), and coefficient of determination (R) are used. 2The accuracy of tree vertex identification and canopy parameter extraction is evaluated by metrics such as root mean square error (RMSE) and relative root mean square error (rRMSE).

[0116] The single-tree detection method provided in this application acquires point cloud data of a target forest area, constructs a digital surface model based on the preprocessed point cloud data, calculates the first and second derivatives of the digital surface model using a multi-order partial derivative calculation method to obtain first-order gradient features, reflecting the rate of change of canopy elevation, and second-order gradient features, used to determine the concavity and convexity of the canopy surface. Based on the gradient feature information, the method analyzes and identifies canopy structural features, and identifies tree apex and canopy boundary. Based on elevation information and gradient features, the forest canopy area is over-segmented to obtain several over-segmented canopy patches. Based on the high self-similarity features of canopy patches of the same single tree in spectral, texture, and color space features, a similarity weight calculation formula is used to calculate the similarity weight between canopy patches. Canopy patches with high weights are clustered and merged to finally obtain the single-tree detection result.

[0117] The method provided in this application obtains the gradient feature information of the tree canopy based on digital surface model data, thereby over-segmenting the canopy region. This avoids the problems of canopy shape distortion and tree apex displacement caused by normalization in complex terrain, significantly improving the adaptability to complex terrain. Furthermore, by comprehensively considering the growth characteristics of trees and differences in canopy structure, and based on the similarity weights of spectral, texture, and color space information, the over-segmented canopy patches are merged to complete individual tree detection. This enables the extraction of individual tree canopy parameters for different tree species, providing technical support for detailed forest resource surveys and accurate carbon storage estimation.

[0118] Example 2

[0119] Secondly, please refer to Figure 6 , Figure 6 This is a schematic diagram of a single-tree detection system provided in Embodiment 2 of this application.

[0120] The single-tree detection system provided in this application includes:

[0121] Data acquisition module 11 is used to acquire point cloud data of the target forest area and construct a digital surface model of the target forest area based on the point cloud data;

[0122] The multi-order gradient calculation module 12 is used to obtain the gradient feature information of the target forest area based on the elevation information of the digital surface model and by using a preset multi-order partial derivative calculation method.

[0123] The oversegmentation module 13 is used to oversegment the canopy patches according to the gradient feature information to obtain several oversegmented canopy patches;

[0124] The single tree detection module 14 is used to perform cluster analysis on the several over-segmented canopy patches based on a preset multi-feature similarity weight calculation method, merge canopy patches belonging to the same single tree, and obtain the single tree detection result.

[0125] It should be noted that the single-wood detection system provided in the above embodiments is only illustrated by the division of the above functional modules when executing the single-wood detection method. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the equipment can be divided into different functional modules to complete all or part of the functions described above. The single-wood detection system provided in the above embodiments is used to execute the single-wood detection method described in the above embodiments. Its operation method and principle are the same as the single-wood detection method described above. That is, the single-wood detection system and the single-wood detection method provided in the above embodiments belong to the same concept. The implementation process is detailed in the above method embodiments and will not be repeated here.

[0126] Example 3

[0127] Thirdly, this embodiment provides a computer device. Please refer to [link / reference needed]. Figure 7 , Figure 7 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of this application. Figure 7 As shown, the computer device 21 includes: a processor 210, a memory 211, and a computer program 212 stored in the memory 211 and executable on the processor 210, such as a single-tree detection program; the processor 210 executes the computer program 212 to implement the methods described in the above embodiments.

[0128] The processor 210 may include one or more processing cores. The processor 210 connects to various parts within the computer device 21 using various interfaces and lines. It executes various functions of the computer device 21 and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 211, and by accessing data in the memory 211. Optionally, the processor 210 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 210 may integrate one or more of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for the touch screen; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip, rather than integrated into the processor 210.

[0129] The memory 211 may include random access memory (RAM) or read-only memory. Optionally, the memory 211 may include a non-transitory computer-readable storage medium. The memory 211 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 211 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch instructions), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data involved in the various method embodiments described above, etc. Optionally, the memory 211 may also be at least one storage device located remotely from the aforementioned processor 210.

[0130] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0131] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for detecting individual wood species, characterized in that, include: Acquire point cloud data of the target forest area, and construct a digital surface model of the target forest area based on the point cloud data; Based on the elevation information of the digital surface model, gradient feature information of the target forest region is obtained using a preset multi-order partial derivative calculation method, including: Based on the digital surface model, the first-order partial derivatives of the digital surface model are calculated to obtain the first-order gradient information of the digital surface model; wherein, the first-order gradient information includes the first-order gradient vector and the magnitude of the first-order gradient vector; Based on the first-order partial derivative calculation results of the digital surface model, the second-order partial derivative calculation is performed on the digital surface model to obtain the second-order gradient information of the digital surface model; wherein, the second-order gradient information includes the second-order gradient vector and the magnitude of the second-order gradient vector; Based on the first-order gradient information and the second-order gradient information, gradient feature information of the target forest area is obtained; wherein, the gradient feature information includes the rate of change of tree canopy elevation and the surface roughness of the tree canopy. Based on the gradient feature information, the canopy patches are oversegmented to obtain several oversegmented canopy patches; Based on a preset multi-feature similarity weight calculation method, cluster analysis is performed on the several over-segmented canopy patches, merging canopy patches belonging to the same single tree to obtain single tree detection results, including: Obtain feature information of the several over-segmented canopy patches; wherein, the feature information includes spectral features, texture features, and color space features; Based on the feature information of canopy patches, a weighted undirected graph is constructed; wherein the vertices of the weighted undirected graph are the center points of the oversegmented canopy patches, and the similarity between the oversegmented canopy patches is used as the similarity weight value of the connected edges. Based on a preset similarity weight function, the similarity weight values ​​between adjacent canopy patches are calculated; Cluster analysis is performed on the canopy patches based on the weight values ​​between them. Canopy patches with similarity weight values ​​higher than a preset value are merged into canopy patches of the same tree to obtain the tree detection results. The step of calculating the similarity weight value between adjacent canopy patches based on a preset similarity weight function includes: Each pixel of the several oversegmented canopy patches is obtained with several spectral features, several texture features, and several color space features as feature variables, and normalization calculations are performed to obtain the average values ​​of the several spectral features, several texture features, and several color space features of each canopy patch; wherein, the spectral features, the texture features, and the color space features each include several feature variables. According to the preset weight calculation formula, the weight values ​​of the several feature variables between adjacent canopy patches are calculated respectively; The feature variable with the largest weight value among the spectral features, texture features, and color space features is selected as the input variable and input into a preset similarity weight function to obtain the similarity weight value between adjacent canopy patches.

2. The method for detecting individual wood grains according to claim 1, characterized in that, The step of calculating the first-order partial derivative of the digital surface model to obtain the first-order gradient information of the digital surface model includes: Calculate the digital surface model in x direction and y The gradient vector in the direction is used to obtain the first-order gradient vector: in, and Digital surface models exist x direction and y First-order partial derivative in the direction, The first-order gradient vector of the digital surface model; Calculate the magnitude of the first-order gradient vector of the digital surface model based on the first-order gradient vector: in, The magnitude of the first-order gradient vector; The second-order gradient information of the digital surface model is obtained by calculating the second-order partial derivative of the digital surface model based on the first-order partial derivative calculation result, including: Based on the first-order gradient vector, calculate the second-order gradient vector of the digital surface model: in, and Digital surface models in x direction and y The second partial derivative in the direction, This represents the second-order gradient vector of the digital surface model; Calculate the magnitude of the second-order gradient vector of the digital surface model based on the second-order gradient vector: in, The magnitude of the second-order gradient vector.

3. The method for detecting individual wood grains according to claim 1, characterized in that, The process involves oversegmenting the canopy patches based on the gradient feature information to obtain several oversegmented canopy patches, including: Based on the gradient feature information, obtain the canopy boundary and several tree vertices in the target forest area; Based on the canopy boundary and several tree vertices, the target forest area is over-segmented to obtain several over-segmented canopy patches.

4. The method for detecting individual wood grains according to claim 1, characterized in that, The similarity weight function is: in, Adjacent patches i and j The similarity weight between them and Canopy patches i and canopy patches j The average value of the spectral characteristics, and Canopy patches i and canopy patches j The average value of the texture features, and Canopy patches i and canopy patches j The average value of the color space characteristics.

5. The method for detecting individual wood grains according to claim 1, characterized in that, The process of acquiring point cloud data of the target forest area and constructing a digital surface model of the target forest area based on the point cloud data includes: Point cloud data of the target forest area is acquired based on multi-angle measurement by UAV and LiDAR scanning. Perform data preprocessing on the point cloud data; A digital surface model of the target forest area is constructed based on the preprocessed point cloud data.

6. A single-tree detection system, characterized in that, include: The data acquisition module is used to acquire point cloud data of the target forest area and construct a digital surface model of the target forest area based on the point cloud data. The multi-order gradient calculation module is used to obtain the gradient feature information of the target forest region based on the elevation information of the digital surface model and using a preset multi-order partial derivative calculation method, including: Based on the digital surface model, the first-order partial derivatives of the digital surface model are calculated to obtain the first-order gradient information of the digital surface model; wherein, the first-order gradient information includes the first-order gradient vector and the magnitude of the first-order gradient vector; Based on the first-order partial derivative calculation results of the digital surface model, the second-order partial derivative calculation is performed on the digital surface model to obtain the second-order gradient information of the digital surface model; wherein, the second-order gradient information includes the second-order gradient vector and the magnitude of the second-order gradient vector; Based on the first-order gradient information and the second-order gradient information, gradient feature information of the target forest area is obtained; wherein, the gradient feature information includes the rate of change of tree canopy elevation and the surface roughness of the tree canopy. The oversegmentation module is used to oversegment the canopy patches according to the gradient feature information to obtain several oversegmented canopy patches; The single-tree detection module is used to perform cluster analysis on the several over-segmented canopy patches based on a preset multi-feature similarity weight calculation method, merge canopy patches belonging to the same single tree, and obtain single-tree detection results, including: Obtain feature information of the several over-segmented canopy patches; wherein, the feature information includes spectral features, texture features, and color space features; Based on the feature information of canopy patches, a weighted undirected graph is constructed; wherein the vertices of the weighted undirected graph are the center points of the oversegmented canopy patches, and the similarity between the oversegmented canopy patches is used as the similarity weight value of the connected edges. Based on a preset similarity weight function, the similarity weight values ​​between adjacent canopy patches are calculated; Cluster analysis is performed on the canopy patches based on the weight values ​​between them. Canopy patches with similarity weight values ​​higher than a preset value are merged into canopy patches of the same tree to obtain the tree detection results. The step of calculating the similarity weight value between adjacent canopy patches based on a preset similarity weight function includes: Each pixel of the several oversegmented canopy patches is obtained with several spectral features, several texture features, and several color space features as feature variables, and normalization calculations are performed to obtain the average values ​​of the several spectral features, several texture features, and several color space features of each canopy patch; wherein, the spectral features, the texture features, and the color space features each include several feature variables. According to the preset weight calculation formula, the weight values ​​of the several feature variables between adjacent canopy patches are calculated respectively; The feature variable with the largest weight value among the spectral features, texture features, and color space features is selected as the input variable and input into a preset similarity weight function to obtain the similarity weight value between adjacent canopy patches.

7. A computer device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the single-tree detection method as described in any one of claims 1 to 5.

Images