Tree identification method and system based on ground-air fusion three-dimensional laser point cloud

Abstract

The invention provides a tree identification method and system based on ground-air fusion three-dimensional laser point cloud, and belongs to the field of environment remote sensing data identification, and the method comprises the steps: obtaining laser radar point cloud data in a set forest region through employing an unmanned plane, and collecting the point cloud data in the set forest region through employing a laser scanning device; preprocessing the two groups of data to obtain preprocessed laser radar point clouds, and performing ground-air data seamless fusion on the preprocessed laser radar point clouds to obtain fused data; generating a canopy height model based on a watershed segmentation algorithm, and performing individual tree segmentation on the fused point cloud data by using the model; extracting a vector boundary after single tree segmentation, distributing a tree ID for each tree, and extracting a single tree point cloud according to the tree ID; the single tree point cloud is further processed to construct a multi-angle point cloud sample data set; and training a deep learning network model by using the multi-angle point cloud sample data set, and detecting the tree species to be detected by using the trained deep learning network model.

Description

Technical Field

[0001] The present invention belongs to the field of environmental remote sensing data recognition, and in particular relates to a tree recognition method and system based on ground-air fusion three-dimensional laser point cloud. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Forests are the main body of terrestrial ecosystems, the core carrier of terrestrial biodiversity, and the most important cornerstone of national ecological security. However, with the intensification of human activities and the impact of global climate change, forest species diversity is facing the severe challenge of continuous reduction. At the same time, the existing biodiversity monitoring network is still imperfect, so rapid and accurate monitoring of large-scale forest resources is particularly important. Tree species surveys are an important part of forest resource surveys. At present, traditional tree species surveys in my country usually require the establishment of sample plots, which have problems such as high intensity of field work, high cost, and long cycle. With the development of a series of emerging technologies such as remote sensing technology and drone technology, the application of LiDAR and drone aerial surveys in the forestry field is becoming more and more extensive, opening up new ways to improve the efficiency and accuracy of forest resource surveys.

[0004] As an advanced active remote sensing technology, LiDAR accurately measures the distance of target objects by emitting laser beams, thereby obtaining rich spatial information. Compared with traditional optical passive remote sensing, LiDAR data can provide three-dimensional information of target objects and accurately extract vertical information of forest stands. It has unparalleled advantages in forestry applications over traditional optical remote sensing. Ground-based, backpack and handheld LiDAR usually operate on the ground under the forest and can obtain the structural parameters of single trees in the forest stand. Among them, handheld LiDAR can obtain data in a mobile manner, with flexible operation methods and higher data quality. In recent years, remote sensing sensors have developed towards diversification and lightweight, and the advancement of drone technology has brought new opportunities to the remote sensing field. Compared with other remote sensing platforms, drones have low operating costs, flexible take-off and landing, low flight altitude, and are easier to obtain high-precision remote sensing data, making up for the lack of spatial resolution of manned or satellite-borne images.

[0005] Although LiDAR data can obtain spatial structural information of trees, the laser light source is in the understory when ground-based LiDAR data is collected. Therefore, the obstruction of leaves, branches, etc. and the limitation of the ground three-dimensional laser scanning angle make ground-based LiDAR unfavorable for the extraction of information on the top layer of the tree canopy, especially in forest stands with high canopy density. This will lead to the loss of spatial structural data on the top layer of the tree canopy, which has an adverse effect on the extraction of tree parameters. The use of photography technology can obtain forest canopy information, but it is easy to have problems such as not being able to capture the front and back landscapes under the canopy.

[0006] Traditional tree species identification can be divided into manual identification and machine learning identification. Manual identification is mainly based on visual interpretation of external morphological features such as leaf shape and bark texture of trees. It has high recognition accuracy, but high cost and low efficiency. Machine learning-based identification establishes a model of tree species and feature information by mining the hidden rules of macroscopic features. The evaluation results are objective and the recognition speed is fast, but the model is greatly affected by parameter settings and feature selection. It is difficult to effectively extract key features manually based on traditional machine learning algorithms. Deep learning networks can automatically and intelligently mine the connections between deep features in the data, and continuously adjust and optimize according to samples, freeing up manpower, improving both the recognition accuracy and the universality of the solution. They are widely used in single tree segmentation and tree species identification research.

[0007] In terms of constructing point cloud datasets, traditional point cloud processing methods mostly rely on direct analysis of three-dimensional models. However, in practical applications, the processing of three-dimensional data is highly complex and consumes huge computing resources.

[0008] With the development of technology, there are technical solutions in the existing technology that use LiDAR data and images collected by drones for fusion processing and use deep learning networks for recognition, such as: Li Yiying and others used ground-air fused 3D laser point cloud data to extract structural features and input them into the deep learning model to identify and classify tree species through feature integration and further obtain crown width changes. Zhu Ruoning and others proposed an identification method based on the ITPNet network model, using hyperspectral images and lidar data to meet the needs of single tree identification. Li Dan and others used ground-air fused radar point cloud data to perform instance segmentation and biomass inversion through a 3D segmentation framework.

[0009] However, the above technical solutions still have the problem that tree identification based directly on point clouds is inefficient, and the labeling process of three-dimensional samples is relatively difficult. In addition, it is usually necessary to fuse point clouds with other sensor data for multimodal data, which significantly increases the amount of data and further increases the complexity of processing. The overlapping and crossing of trees in the point cloud leads to unsatisfactory individual differentiation. At the same time, in the process of semantic recognition of land objects, simply relying on the two-dimensional or multi-angle feature information of point clouds fails to fully explore and utilize the deep-level features of point cloud data, thus affecting the recognition accuracy and effect. Summary of the invention

[0010] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a tree recognition method based on ground-air fusion three-dimensional laser point cloud, which can realize the semantic (qualitative) intelligent recognition of tree species at a single tree scale and the precise extraction of structural parameters (quantitative) by using the ground-air fusion three-dimensional laser point cloud.

[0011] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: In the first aspect, a tree recognition method based on ground-air fusion three-dimensional laser point cloud is disclosed, comprising: Use drones to obtain lidar point cloud data within a set forest area, and use laser scanning equipment to collect point cloud data within a set forest area; The two sets of data are preprocessed to obtain a preprocessed LiDAR point cloud, and the preprocessed LiDAR point cloud is seamlessly fused with ground and air data to obtain fused data; The canopy height model is generated based on the watershed segmentation algorithm, and the model is used to segment individual trees from the fused point cloud data. Extract the vector boundaries after the individual tree segmentation, assign a tree ID to each tree, and extract the individual tree point cloud based on the tree ID; The single tree point cloud is further processed to construct a multi-angle point cloud sample dataset; The deep learning network model is trained using a multi-angle point cloud sample dataset, and the trained deep learning network model is used to detect the tree species to be tested.

[0012] As a further technical solution, the two sets of data are preprocessed, including denoising, normalizing, removing ground points, and cropping the point cloud to obtain a preprocessed lidar point cloud.

[0013] As a further technical solution, an iterative closest point algorithm is used to seamlessly fuse ground and air data for the preprocessed lidar point cloud. This algorithm seamlessly fuses ground and air data based on a position and feature matching method.

[0014] As a further technical solution, when performing single tree segmentation, a dynamic window of variable size is used to search for the local maximum value as the crown vertex on the canopy height model, and the crown vertex is used as a marker. The marker-controlled watershed algorithm is used to outline the crown boundary, and the tree height, breast diameter and crown width parameters are further extracted.

[0015] As a further technical solution, CHM canopy height model = DSM digital surface model - DEM digital elevation model.

[0016] As a further technical solution, the extracted single tree point cloud is first used to determine the vertical axis of the trunk, and then the trunk detection is completed through a multi-layer merging method based on hierarchical clustering; Secondly, all-round observations around the vertical axis of the tree trunk are used to form point cloud images at multiple angles, which are combined with ground-based survey data to construct a multi-angle observation dataset of single standing trees.

[0017] As a further technical solution, on the basis of hierarchical clustering, the tree trunk detection is completed by multi-layer merging method. The specific process is as follows: The tree point cloud P is divided into n layers from bottom to top according to a certain distance, and the trunk position is extracted by layer-by-layer point aggregation from bottom to top; After obtaining the point clouds Pi of each layer, the DBSCAN algorithm is used to cluster them layer by layer to obtain a set, and the clustering results of each layer are merged.

[0018] As a further technical solution, the clustering results of each layer are merged, specifically including: 1) Calculate the centroid and area information of the convex hull formed by the clustering point sets of each layer; 2) Taking any cluster unit in the first layer of clustering set obtained by hierarchical clustering as each growth starting unit, starting from this unit, calculate the distance between the convex hull centroid of each cluster unit and the convex hull centroid of the previous layer, as well as the convex hull area formed by the projection of each cluster on the horizontal plane; 3) Determine whether the two clusters belong to the same feature based on the distance and area of ​​the two stratified point clusters; 3) Repeat the above steps until all layers are traversed, that is, i=n, ​​and obtain the trunk point cloud of a single tree.

[0019] As a further technical solution, when stratifying, the number of point clouds of each trunk segment should be greater than T minPts , the segment distance should satisfy:

[0020] Where: T den is the average point cloud density at the trunk; r is the average breast diameter; T minPts Determined by the point cloud density of the tree trunk.

[0021] As a further technical solution, the deep learning network model YOLO v11-cls,YOLO v11-cls adopts the improved Backbone and Neck architecture, and introduces C3k2 and C2PSA components; The C3k2 component is used for multi-scale feature fusion and can effectively capture the subtle features of trees.

[0022] In the second aspect, a tree recognition system based on ground-air fusion three-dimensional laser point cloud is disclosed, including: The point cloud data acquisition module is configured to: use a drone to acquire laser radar point cloud data within a set forest area, and use a laser scanning device to collect point cloud data within the set forest area; The fusion module is configured to: pre-process the two sets of data to obtain a pre-processed lidar point cloud, and seamlessly fuse the ground and air data of the pre-processed lidar point cloud to obtain fused data; The single tree segmentation module is configured to: generate a canopy height model based on a watershed segmentation algorithm, and use the model to perform single tree segmentation on the fused point cloud data; The dataset construction module is configured to: extract the vector boundary after the single tree segmentation, assign a tree ID to each tree, and extract the single tree point cloud based on the tree ID; The single tree point cloud is further processed to construct a multi-angle point cloud sample dataset; The tree species detection module is configured to: use a multi-angle point cloud sample data set to train a deep learning network model, and use the trained deep learning network model to detect the tree species to be detected.

[0023] One or more of the above technical solutions have the following beneficial effects: The technical solution of the present invention adopts ground-based lidar point cloud data and drone images, generates three-dimensional point cloud information of tree canopies through high-overlap drone images, and aligns the point cloud data generated by the ground-based lidar point cloud data and drone images for fusion. The fused data not only contains the tree spatial structure information and trunk structure information of the ground-based lidar point cloud data, but also the drone point cloud makes it have more complete canopy spatial structure information and canopy color information, and then single tree segmentation is performed based on the fused point cloud data.

[0024] The technical solution of the present invention integrates new technologies such as lidar scanning, multi-scale object segmentation, and intelligent semantic recognition, improves the problems of appearance defects, structural dislocation, etc. in the three-dimensional collection of forest elements, improves the accuracy of point cloud single tree segmentation and tree species identification, and realizes the integrated collection of key semantics such as appearance, structure, and attributes of forest elements.

[0025] The technical solution of the present invention proposes a dimensionality reduction scheme based on single tree point cloud data, which aims to form multi-angle point cloud images through all-round observation around the vertical axis of the trunk, and generate a high-quality single-tree scale tree species semantic data set. This method not only retains the structural characteristics of trees, but also enhances the diversity of data sets, thereby providing rich training samples for subsequent deep learning models. Compared with the existing technology, a significant advantage of this method is that it can accurately identify individual trees, avoiding interference from other plants and environmental factors in large forests, thereby improving the accuracy and efficiency of tree species identification, and has broad application prospects.

[0026] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0028] Figure 1 is a flow chart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of 12-angle point cloud image effects formed by all-round observation of a single tree around the vertical axis of the trunk according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the network structure of the ResNet152+SPP pooling deep learning model described in the tree species semantic recognition in an embodiment of the present invention; Figure 4 This is a schematic diagram of the network structure of the YOLO v11 deep learning classification model described in the tree species semantic recognition in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0030] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0031] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0032] Embodiment 1 This embodiment discloses a technical method for realizing intelligent semantic recognition of tree species at a single tree scale based on ground-based and air-based fusion three-dimensional laser point cloud, including: Step 1: Data acquisition: First, use drones to obtain lidar point cloud data in the study area, and then use handheld laser scanning equipment to collect point cloud data in the study area. Sample trees are obtained through ground surveys, which will be combined with point cloud data to form a single standing tree sample image dataset, which can be used to train tree species recognition models and evaluate model accuracy and the accuracy of structural parameter extraction.

[0033] Step 2: Data preprocessing and data fusion: First, the point cloud data of the research area collected by the UAV is synthesized and spliced ​​to obtain a set of air-based point cloud data; the point cloud data of the research area collected by the handheld laser scanning equipment is solved to obtain a set of ground-based point cloud data; Secondly, the two sets of data are preprocessed by denoising, normalizing, removing ground points, and cropping the point cloud to obtain the preprocessed lidar point cloud; The iterative closest point (ICP) algorithm is used for the preprocessed lidar point cloud to seamlessly fuse ground and air data based on position and feature matching methods.

[0034] 2-1) Denoising: Common noise includes high-order gross errors and low-order gross errors. High-order gross errors are usually caused by the airborne LiDAR system being affected by low-flying objects during data collection, and the signals reflected from these objects are mistakenly recorded as reflected signals of the measured target. Low-order gross errors are extremely low points caused by multipath errors or errors in the laser rangefinder during the measurement process. By selecting appropriate parameters, noise points can be removed to improve data quality.

[0035] Assuming the standard deviation multiple is meanK, the algorithm will search for the specified number of neighboring points for each point, calculate the average distance D from the point to the neighboring points, calculate the median meanD and standard deviation S of these average distances, and if D is greater than the maximum distance MaxD (MaxD=meanD+meanK*S), it is considered a noise point and will be removed. Specific parameter settings: number of neighborhood points = 10; standard deviation S = 5.

[0036] 2-2) Normalization: The process of normalizing the air-based point cloud and the ground-based point cloud according to the ground points involves identifying the ground points, calculating their heights, and subtracting the ground height from the height of each point in the point cloud to achieve height standardization. This operation eliminates height differences, enhances data consistency, and facilitates subsequent analysis.

[0037] Regarding ground point recognition: 1) Sort the elevation coordinates of the point cloud and select the lowest 10% of points as candidate ground points. 2) Screen the points below the threshold through eight iterations to identify the ground points.

[0038] 2-3) Ground point classification (progressive encryption triangulation filtering algorithm): 1) Selection of initial seed points: In point cloud data containing buildings, the maximum building size is taken as the grid size to grid the point cloud data, and the lowest point in the grid is taken as the starting seed point.

[0039] 2) Construct triangulated network. Use the starting seed point to construct the initial triangulated network.

[0040] 3) Iterative encryption process. Traverse all points to be classified, query the triangles where the horizontal plane projection of each point falls, calculate the distance d from the point to the triangle and the maximum angle between the point and the three vertices of the triangle and the plane where the triangle is located, as shown in the figure below, and compare them with the iterative distance and iterative angle respectively. If it is less than the corresponding threshold, the point is determined to be a ground point and added to the triangulation network. Repeat this process until all ground points are classified. Finally, extract the point cloud data except the ground points to achieve the purpose of removing the ground points.

[0041] Specific parameter settings: maximum building size = 20m; maximum terrain slope = 88°; iteration angle = 8°; iteration distance = 1.4m. Compared with the traditional algorithm, the ground seed points of this algorithm are obtained through morphological opening operations; the simulated ground points along the buffer zone participate in the construction of the original digital terrain model based on the progressive triangulated irregular network (TIN) to improve the quality of TIN and avoid the formation of inappropriate triangles; downward encryption is performed before upward encryption, which improves the ability of encryption to handle slope changes.

[0042] 2-4) Iterative Closest Point (ICP) algorithm: 1) Sample the point clouds P to be registered from the two sets of data, the air-based point cloud and the ground-based point cloud, respectively, and then find the point set corresponding to the target point cloud Q, so that the Euclidean distance between each corresponding point is the smallest, and finally obtain two new point sets, that is, remove the points without corresponding points and the wrong points.

[0043] 2) Based on the coordinate information of the point set found in the first step, calculate the centroid of the corresponding point set to provide a reference point for subsequent calculations; 3) Obtain the center of gravity, and based on the center of gravity, calculate the rigid transformation (rotation matrix R and translation vector T) between the source point cloud and the target point cloud by minimizing the error function E(R,T) to minimize its value;

[0044] Where: P i and Q i are the coordinates of the corresponding points in the source point cloud P and the target point cloud Q, respectively, and n is the number of paired points.

[0045] 4) For the points in the source point cloud P, the rotation matrix R and translation vector T calculated in the third step are rigidly transformed. The new point set obtained after the transformation is named P′; the exchanged point set P′ is the result of the current iteration and should be closer to the target point cloud Q.

[0046] 5) From the new point set P′ and all the points in the target point cloud Q, find the average distance between the corresponding points, which is recorded as ;

[0047] Where: P i and Q i are the coordinates of the corresponding points in the source point cloud P and the target point cloud Q, respectively, and n is the number of paired points.

[0048] 6) According to the above formula, the average distance is obtained , and determine whether the iteration is over. If If it is greater than the preset threshold, then continue to return to the second step and calculate until the condition is met; if If it is smaller than the preset threshold or meets the set number of iterations k, the algorithm is judged to have converged and the iteration is terminated.

[0049] The advantages of the ICP algorithm are that it is simple, efficient, easy to implement, and can accurately align two sets of 3D point cloud data. Through an iterative optimization process, ICP can effectively find the rigid transformation between point clouds and is widely used in 3D reconstruction, robot navigation, and object recognition. Its main advantage is that it has low requirements for initial guesses and can handle complex point cloud registration tasks, especially for accurate local alignment and small-scale point cloud matching.

[0050] Step 3: Single tree segmentation: The fused point cloud data is used to generate a canopy height model (CHM) based on the watershed segmentation algorithm, which reflects the height changes of the tree canopy. A dynamic window of variable size is used to search for the local maximum value as the crown vertex on the canopy height model, and the crown vertex is used as a marker and input into the watershed algorithm. The watershed algorithm controls the segmentation of the crown area based on these markers, and judges the boundary according to the height changes between different crown areas in the CHM and the position of the "ridge" formed during the watershed algorithm to simulate the rising flood. By extracting the boundary of each crown, the precise segmentation of a single tree is achieved. Finally, based on the segmentation results, key parameters such as tree height, breast diameter and crown width are further extracted. The advantages of this method are that it has high segmentation accuracy and strong automatic processing capabilities, so as to accurately extract various forestry parameters and reduce the problems of over-segmentation and under-segmentation in traditional methods.

[0051] About generating the canopy height model: The Digital Surface Model (DSM) is a ground elevation model that includes the heights of buildings, bridges, and trees on the ground. The DSM is generated by interpolation calculation based on the point cloud data (including ground points and non-ground points). The specific parameters are set as follows: Xsize=2, Ysize=2, weight=2, buffer size=5, interpolation method=inverse distance weighted interpolation, radius search=variable radius, distance=5, and number of field points=12.

[0052] Digital Elevation Model (DEM) is a digital simulation of the ground terrain (i.e., a digital expression of the terrain surface morphology) through limited terrain elevation data. It is a physical ground model that represents the ground elevation in the form of a set of ordered numerical arrays. Interpolation calculations are performed based on the ground point data in the point cloud to generate a continuous DEM surface. The specific parameters are set as follows: Xsize=2, Ysize=2, weight=2, buffer size=5, interpolation method=inverse distance weighted interpolation, radius search=variable radius, distance=5, number of field points=12.

[0053] CHM (Canopy Height Model) = DSM (Digital Surface Model) - DEM (Digital Elevation Model) Step 4: Construction of a multi-angle semantic dataset of tree species at the scale of a single tree: Extract the vector boundaries after the segmentation of a single tree, assign a tree ID to each tree (vector outline), and extract the single tree point cloud based on the tree ID. The extracted single tree point cloud is first used to determine the vertical axis of the trunk. In order to effectively distinguish the trunk point cloud, the trunk detection is completed through a multi-layer merging method based on hierarchical clustering. Secondly, a full-scale observation around the vertical axis of the trunk is formed to form a point cloud image of 12 angles, which is combined with the tree species name in the ground survey data to construct a multi-angle observation single standing tree image dataset.

[0054] About hierarchical clustering and multi-layer merging detection tree trunks: The tree point cloud P is divided into n layers from bottom to top according to a certain distance, and the trunk position is extracted by layer-by-layer point clustering from bottom to top. When stratifying, in order to effectively merge and detect the trunks later, the number of point clouds in each trunk segment should be greater than T minPts , the segment distance should satisfy:

[0055] Where: T den is the average point cloud density at the trunk; r is the average breast diameter; T minPts It is the minimum point cloud density for trunk clustering, which is determined by the overall point cloud density of the trunks. The default value is 5.

[0056]

[0057] Where: Z j Indicates the elevation value of the jth point.

[0058] After obtaining the point clouds Pi of each layer, the DBSCAN algorithm is used to cluster them layer by layer to obtain a set, and the clustering results of each layer are merged.

[0059] 1) Calculate the centroid and area information of the convex hull formed by the clustering point sets of each layer.

[0060] The convex hull of the point set is calculated using the Graham scanning method. Given the convex hull point Ok (k=1,2,…,n) detected by the point set, the coordinates of the convex hull centroid T can be expressed as: , Among them, S is the convex hull area, and the calculation formula is S= .

[0061] 2) Taking any cluster unit in the first layer of clustering set obtained by hierarchical clustering as each growth starting unit, starting from this unit, calculate the distance D between the convex hull centroid of each cluster unit and the convex hull centroid of the previous layer, as well as the convex hull area S formed by the projection of each cluster on the horizontal plane.

[0062] The calculation formula of the convex hull centroid distance D between two hierarchical point clusters is:

[0063] 3) Determine whether the two clusters belong to the same feature based on the distance and area of ​​the two layered point clusters. Repeat the above steps until all layers are traversed, that is, i=n, ​​and obtain the trunk point cloud of a single tree.

[0064] Step 5: Semantic identification and optimization of tree species: Use the multi-angle semantic dataset of single-tree-scale tree species to train, optimize, and compare deep learning algorithms (YOLO v11 and ResNet152 + SPP pooling), and finally select the more accurate YOLO v11 for tree species identification.

[0065] YOLO v11-cls: In the tree species recognition task, by adopting a series of improved technical architectures, especially the optimized Backbone and Neck architectures, the model's ability to extract tree features has been significantly improved, thereby effectively improving the accuracy of tree species detection and its performance in complex environments. It is particularly noteworthy that the introduction of two innovative components, C3k2 and C2PSA, has greatly enhanced the model's ability to capture tree detail features, enabling it to perform well in a variety of practical applications.

[0066] First, the C3k2 component can deeply mine and fuse image information from different scales through multi-scale feature fusion. In the tree species identification task, the various components of the tree, such as leaves, trunks, and crowns, may present different spatial scale characteristics. By strengthening the extraction of these multi-scale features, the C3k2 component enables the model to more accurately capture subtle tree features, such as leaf shape, trunk diameter changes, crown density and morphology, etc. Especially in complex forest environments, the differences between trees are often subtle, and C3k2 can effectively improve the sensitivity to these details, thereby improving recognition accuracy.

[0067] Secondly, the C2PSA component further optimizes the feature weighting process in the Neck part by introducing an attention mechanism, thereby enhancing the model's ability to focus on important areas. C2PSA performs adaptive weighting in the spatial and channel dimensions, and can focus on the most distinguishing features in the image, especially when the trees are occluded by the background or other trees, and can effectively distinguish the unique visual features of the tree species. In this way, the model can improve its ability to recognize key tree structures (such as trunks, branches, etc.) when dealing with complex backgrounds and interference information, thereby further improving classification accuracy.

[0068] In addition, YOLO v11-cls not only significantly reduces the consumption of computing resources by reducing the number of parameters by 22%, but also improves the processing speed, which enables the model to maintain efficient operation in real-time applications.

[0069] ResNet152 + SPP pooling: In tree species recognition, the deep residual network structure of ResNet152 can deeply explore the multi-level features of trees and provide richer feature representations. This structure is particularly suitable for processing tree images of different varieties and ages, helping to identify diverse tree species characteristics. Combined with spatial pyramid pooling (SPP), the model can flexibly process input images of various sizes without fixing the input size, avoiding information loss caused by adjusting the image size. By performing feature pooling at multiple scales, ResNet152 can capture the diversity of trees, including the characteristics of leaves, crowns, and branches, thereby improving the accuracy and robustness of tree species recognition.

[0070] In the specific implementation process, it is also necessary to consider technical means such as data enhancement, learning rate adjustment, batch normalization, etc. to further improve the performance and stability of the model. In addition, appropriate evaluation indicators (such as accuracy, recall rate, F1 value, etc.) can also be used to evaluate the performance of the model in remote sensing semantic recognition tasks. The specific indicators are as follows: Accuracy: Accuracy = (TP + TN) / (TP + TN + FP + FN) Recall: Recall = TP / (TP + FN) Precision: Precision = TP / (TP + FP) F1 Score: F1 Score = 2 * (Precision * Recall) / (Precision +Recall) In the formula, TP represents true positive, that is, the number of samples correctly predicted as positive by the model; TN represents true negative, that is, the number of samples correctly predicted as negative by the model; FP represents false positive, that is, the number of samples incorrectly predicted as positive by the model; FN represents false negative, that is, the number of samples incorrectly predicted as negative by the model.

[0071] In a more detailed implementation example, the tree recognition method based on ground-air fusion 3D laser point cloud includes: S1. Data acquisition: The present invention first uses the DJI M300 drone with the L1 Zenmuse camera to obtain the laser radar point cloud data in the study area, and then uses the China Measurement RS10 series handheld laser scanning equipment to collect the point cloud data of the study area. The tree species, tree height, breast diameter, crown width, longitude and latitude coordinates and other data of more than 2,000 trees were surveyed on the ground.

[0072] S2. Data preprocessing and data fusion: The UAV flight data is imported into DJI Zhitu to complete data synthesis and image stitching; the handheld scanning data is solved by CoPre2 software; secondly, the two sets of data are preprocessed in LiDAR360 software, and the point cloud is denoised, normalized, ground points are removed, and cropped. The preprocessed lidar point cloud uses the iterative closest point (ICP) algorithm to seamlessly fuse ground and air data based on the position and feature matching method.

[0073] S3. Single tree segmentation: The fused point cloud data is used to generate a canopy height model (CHM) based on the watershed segmentation algorithm. A dynamic window of variable size is used to search for the local maximum value as the crown vertex on the canopy height model. The crown vertex is used as a marker to outline the crown boundary using the marker-controlled watershed algorithm, and further extract parameters such as tree height, breast diameter and crown width.

[0074] S4. Construction of multi-angle semantic dataset of tree species at single tree scale: Extract vector boundaries after single tree segmentation, assign tree ID to each tree (vector outline), and extract single tree point cloud based on tree ID. The extracted single tree point cloud is used to determine the vertical axis of the trunk. In order to effectively distinguish the trunk point cloud, the trunk detection is completed through a multi-layer merging method based on hierarchical clustering. Secondly, all-round observation around the vertical axis of the trunk forms point cloud images at 12 angles, which are combined with ground survey data to construct a multi-angle observation single standing tree image dataset.

[0075] S5. Semantic recognition and optimization of tree species: Use multi-angle point cloud sample data sets to train, optimize, and compare the YOLOv11 model and ResNet152 model + SPP pooling.

[0076] YOLO v11-cls training: The network is built using the Pytorch library in Python, with 80% of the data used as the training set and 20% of the data used as the test set. Use the YOLO11-cls model configuration file and load the pre-trained weights; pass the defined training parameters into the model to start training.

[0077] ResNet152 model + SPP pooling: This network is built using the Pytorch library in Python, with 80% of the data used as the training set and 20% of the data used as the test set. A spatial pyramid pooling layer is added after the last convolutional layer of ResNet152, and pooling parameters of different scales (such as 1x1, 2x2, 3x3, etc.) are configured to obtain multi-level features.

[0078] The example of the present invention finally chooses to use the YOLO11-cls model, and the tree species intelligent identification evaluation indicators are: precision=0.87, recall=0.82, F1-score=0.83, accuracy=0.91.

[0079] In response to the problem of "easy obstruction of the ground foundation and poor penetration of the air foundation" in forest scanning, the present invention first collects three-dimensional laser point cloud data of the study area through low-altitude aerial photography of unmanned aerial vehicles and handheld laser scanning equipment, and realizes seamless integration of ground and air dual platform data through GPS positioning and feature matching, thereby improving the phenotype and structural integrity of forest point cloud elements. In view of the limitations of traditional artificial tree species surveys and machine learning algorithm identification, a single standing tree image data set with multi-angle observation is constructed based on single tree point clouds and survey data, and high-precision automatic recognition of tree species semantics at a single tree scale is achieved through training, optimization, and comparison of multiple artificial intelligence algorithms. The method proposed in the present invention can realize the construction of key parameters such as the position, attributes, and structure of surface forest elements, and provide technical support for application needs such as refined forest resource management, biodiversity investigation and protection, and urban livable environment regulation.

[0080] When constructing a data set in the technical solution of the present invention, all-round observation is performed around the vertical axis of the tree trunk to form point cloud images at 12 angles, and the three-dimensional point cloud data is reduced to two-dimensional image data to simplify the analysis process and improve the training efficiency of the model, which has important practical significance.

[0081] Embodiment 2 The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0082] Embodiment 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0083] A computer-readable storage medium stores a computer program, which executes the steps of the above method when executed by a processor.

[0084] Embodiment 4 The purpose of this embodiment is to provide a tree recognition system based on ground-air fusion three-dimensional laser point cloud, including: The point cloud data acquisition module is configured to: use a drone to acquire laser radar point cloud data within a set forest area, and use a laser scanning device to collect point cloud data within the set forest area; The fusion module is configured to: pre-process the two sets of data to obtain a pre-processed lidar point cloud, and seamlessly fuse the ground and air data of the pre-processed lidar point cloud to obtain fused data; The single tree segmentation module is configured to: generate a canopy height model based on a watershed segmentation algorithm, and use the model to perform single tree segmentation on the fused point cloud data; The dataset construction module is configured to: extract the vector boundary after the single tree segmentation, assign a tree ID to each tree, and extract the single tree point cloud based on the tree ID; The single tree point cloud is further processed to construct a multi-angle point cloud sample dataset; The tree species detection module is configured to: use a multi-angle point cloud sample data set to train a deep learning network model, and use the trained deep learning network model to detect the tree species to be detected.

[0085] Embodiment 5 The purpose of this embodiment is to provide a computer program product containing instructions, which, when running on a computer, enables the computer to execute the methods and functions involved in any of the above embodiments. The steps involved in the apparatus of the above embodiment correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0086] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0087] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A tree recognition method based on ground-air fusion three-dimensional laser point cloud, characterized by: include: Use drones to obtain lidar point cloud data within the set forest area, and use laser scanning equipment to collect point cloud data within the set forest area; The two sets of data are preprocessed to obtain a preprocessed LiDAR point cloud, and the preprocessed LiDAR point cloud is seamlessly fused with ground and air data to obtain fused data; The canopy height model is generated based on the watershed segmentation algorithm, and the model is used to segment individual trees from the fused point cloud data. Extract the vector boundaries after the individual tree segmentation, assign a tree ID to each tree, and extract the individual tree point cloud based on the tree ID; The single tree point cloud is further processed to construct a multi-angle point cloud sample dataset; The deep learning network model is trained using a multi-angle point cloud sample dataset, and the trained deep learning network model is used to detect the tree species to be tested.

2. The tree recognition method based on ground-air fusion three-dimensional laser point cloud as claimed in claim 1 is characterized in that: When seamlessly fusing ground and air data for the preprocessed lidar point cloud, an iterative closest point algorithm is used, which performs seamless fusion of ground and air data based on position and feature matching methods.

3. The tree recognition method based on ground-air fusion three-dimensional laser point cloud as claimed in claim 1 is characterized in that: When performing single tree segmentation, a dynamic window of variable size is used to search for the local maximum value as the crown vertex on the crown height model. The crown vertex is used as a marker, and the marker-controlled watershed algorithm is used to outline the crown boundary, and further extract the tree height, breast diameter and crown width parameters.

4. The tree recognition method based on ground-air fusion three-dimensional laser point cloud as claimed in claim 1 is characterized in that: The extracted single tree point cloud is first used to determine the vertical axis of the tree trunk, and then the tree trunk detection is completed through the multi-layer merging method based on the hierarchical clustering. Secondly, all-round observations around the vertical axis of the tree trunk are used to form point cloud images at multiple angles, which are combined with ground-based survey data to construct a multi-angle observation dataset of single standing trees.

5. The tree recognition method based on ground-air fusion three-dimensional laser point cloud as claimed in claim 1, characterized in that: On the basis of hierarchical clustering, tree trunk detection is completed through a multi-layer merging method. The specific process is as follows: The tree point cloud P is divided into n layers from bottom to top according to a certain distance, and the trunk position is extracted by layer-by-layer point aggregation from bottom to top; After obtaining the point clouds Pi of each layer, the DBSCAN algorithm is used to cluster them layer by layer to obtain a set, and the clustering results of each layer are merged.

6. The tree recognition method based on ground-air fusion three-dimensional laser point cloud as claimed in claim 1, characterized in that: The clustering results of each layer are merged, including: 1) Calculate the centroid and area information of the convex hull formed by the clustering point sets of each layer; 2) Taking any cluster unit in the first layer of clustering set obtained by hierarchical clustering as each growth starting unit, starting from this unit, calculate the distance between the convex hull centroid of each cluster unit and the convex hull centroid of the previous layer, as well as the convex hull area formed by the projection of each cluster on the horizontal plane; 3) Determine whether the two clusters belong to the same feature based on the distance and area of ​​the two stratified point clusters; 3) Repeat the above steps until all layers are traversed, that is, i=n, ​​and obtain the trunk point cloud of a single tree.

7. A tree recognition system based on ground-air fusion three-dimensional laser point cloud, characterized by: include: The point cloud data acquisition module is configured to: use a drone to acquire laser radar point cloud data within a set forest area, and use a laser scanning device to collect point cloud data within the set forest area; The fusion module is configured to: pre-process the two sets of data to obtain a pre-processed lidar point cloud, and seamlessly fuse the ground and air data of the pre-processed lidar point cloud to obtain fused data; The single tree segmentation module is configured to: generate a canopy height model based on a watershed segmentation algorithm, and use the model to perform single tree segmentation on the fused point cloud data; The dataset construction module is configured to: extract the vector boundary after the single tree segmentation, assign a tree ID to each tree, and extract the single tree point cloud based on the tree ID; The single tree point cloud is further processed to construct a multi-angle point cloud sample dataset; The tree species detection module is configured to: use a multi-angle point cloud sample data set to train a deep learning network model, and use the trained deep learning network model to detect the tree species to be detected.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are performed.

Images