Ancient tree unmanned aerial vehicle inspection path planning method and system based on absolute safety constraint and multi-objective self-adaptation

Abstract

This invention proposes a method and system for UAV inspection path planning of ancient trees based on absolute safety constraints and multi-objective adaptation. The method includes: analyzing vulnerable structural information, disease information, and environmental information of ancient trees based on point cloud data and spectral images to assess vulnerability levels and obtain the vulnerability level of each point in the point cloud; calculating the safety distance increment of each point based on its vulnerability level to construct a non-uniform safety shell model; constructing an adaptive multi-objective cost function based on the vulnerability cost of the ancient trees and the inspection coverage quality cost in the outer region of the safety shell according to the non-uniform safety shell model; performing improved A* algorithm path planning based on the adaptive multi-objective cost function to obtain the optimal path; and verifying the optimal path. This invention satisfies absolute safety constraints while minimizing costs, thereby improving the absolute safety rate, inspection efficiency, and path quality of UAV inspection path planning.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of UAV inspection and digital cultural relic protection technology, and in particular relates to a method and system for UAV inspection path planning of ancient trees based on absolute safety constraints and multi-objective adaptive design. Background Technology

[0002] Currently, the application of drones in the protection of ancient and famous trees is mainly focused on the execution of inspection tasks. However, drone path planning, as the core link in drone inspection of ancient trees, faces many technical bottlenecks. Most existing path planning methods are based on simple geometric models or uniform obstacle avoidance rules, such as treating ancient trees as spheres or cylinders for uniform avoidance. This ignores the morphological differences and varying vulnerabilities of different parts of the tree, resulting in insufficient avoidance of vulnerable areas (such as trunk cracks and canopy branches) during path planning, posing potential collision risks. Furthermore, existing methods often employ single-objective optimization, such as pursuing only the shortest path or shortest time, neglecting the specific characteristics of ancient tree inspection tasks. They fail to organically integrate the needs of ancient tree protection, inspection quality requirements, and flight efficiency, leading to the need for manual intervention and adjustments in practical applications, reducing inspection efficiency and safety. In addition, existing algorithms are not adaptable to complex terrain and dense forest areas, easily getting trapped in local optima and unable to generate high-quality inspection paths while ensuring safety. Especially in areas with dense ancient trees, the efficiency and quality of path planning often drop significantly.

[0003] In existing technologies, some studies have attempted to introduce multi-objective optimization to solve path planning problems, such as combining path length, flight altitude, and energy consumption for optimization. However, most of these methods use linear combinations with fixed weights, failing to dynamically adjust the weights of each objective based on the specific characteristics of the ancient tree (such as tree species, age, shape, and vulnerability level). This results in planned paths that cannot adaptively optimize for the protection needs of different ancient trees. Furthermore, existing research lacks precise modeling of the three-dimensional morphology of ancient trees, using only simplified geometric models for obstacle avoidance. This fails to accurately reflect the true shape and distribution of vulnerable parts of the ancient tree, significantly reducing obstacle avoidance effectiveness. More importantly, existing technologies do not prioritize "absolute safety," treating safety as an optimizable objective, which is unacceptable in the high-risk scenario of ancient tree protection. Due to the unique nature of ancient tree protection, any potential collision risk can lead to irreversible damage; therefore, absolute safety must be ensured in path planning, rather than merely pursuing a certain optimization metric. These shortcomings of existing technologies pose challenges to the application of drones in ancient tree inspection, including high safety risks, low inspection quality, and unsatisfactory efficiency, severely restricting the widespread application of drones in the field of ancient tree protection. Summary of the Invention

[0004] This invention proposes a method and system for planning the path of UAV inspection of ancient trees based on absolute safety constraints and multi-objective adaptation. Under the premise of satisfying absolute safety constraints, it minimizes the vulnerability cost and coverage quality cost, thereby improving the absolute safety rate of UAV inspection path planning, inspection efficiency, and path quality.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows: A method for UAV inspection path planning of ancient trees based on absolute safety constraints and multi-objective adaptive approach includes: S1. Based on the ancient tree point cloud data and spectral images, analyze the information on the vulnerable structure, disease, and environment of the ancient trees, conduct a vulnerability level assessment, and obtain the vulnerability level of each point in the ancient tree point cloud. S2. Calculate the safety distance increment of each point based on the vulnerability level of each point in the ancient tree point cloud, and construct a non-uniform safe shell model. S3. Based on the non-uniform containment model, construct an adaptive multi-objective cost function in the outer region of the containment based on the vulnerability cost of ancient trees and the inspection coverage quality cost. S4. Based on the adaptive multi-objective cost function, perform improved A* algorithm path planning to obtain the optimal path; S5. Verify the optimal path.

[0006] Furthermore, step S1 specifically includes: S11. Perform 3D scanning of ancient trees using lidar and multispectral cameras to obtain 3D point cloud data of ancient trees, including the root system area, as well as multispectral image data of ancient trees. S12. Preprocess the 3D point cloud data of ancient trees, segment non-ground points, and obtain the point cloud of ancient trees. S13. Calculate the vulnerability level of ancient trees: ; Where V represents the vulnerability level; The structural vulnerability is represented by analyzing ancient tree point cloud and multispectral image data to identify vulnerable structural information of ancient trees, including cracks, tilting, and bark damage, and then calculating the vulnerability. It indicates the vulnerability to diseases. By analyzing the point cloud and multispectral image data of ancient trees, it identifies disease information of ancient trees, including lesions and insect infestation, and calculates the vulnerability. The environmental vulnerability is represented by analyzing the point cloud data of ancient trees and the three-dimensional point cloud data of the root system area to obtain environmental information including the area of ​​soil loosening and the area affected by wind, and then calculating it. These are the weighting coefficients; S14. Map the vulnerability levels onto the ancient tree point cloud to obtain the vulnerability level of each point.

[0007] Furthermore, the preprocessing in step S12 includes: 3D point cloud data filtering is achieved using a Gaussian filter. Use the RANSAC algorithm to fit the ground plane; Calculate the vertical distance between the point cloud and the ground plane, and classify points whose distance is greater than a threshold as non-ground points, thereby dividing the point cloud into ground points and non-ground points; The region growing algorithm is used to segment non-ground points to obtain ancient tree point clouds.

[0008] Furthermore, step S2 specifically includes: S21. Construct a spherical coordinate system with the center of the ancient tree cloud as the origin, and calculate the angle and distance of each point relative to the origin; the angle of each point relative to the origin includes the azimuth and elevation angles; S22. Calculate the safety distance increment for each point based on the difference between the vulnerability level of each point in the ancient tree point cloud and the lowest vulnerability level, combined with the safety distance coefficient. S23. Calculate the safe distance of each point in the ancient tree point cloud, and construct a non-uniform safe shell model of the ancient tree based on the safe distance; the safe distance of each point is the sum of the distance of each point relative to the origin and the safe distance increment of each point.

[0009] Furthermore, the non-uniform containment model is represented as: ; in Let i be the point on the containment model corresponding to point i in the ancient tree point cloud. Let i be the safe distance. Let i be the azimuth angle of point i. Let be the elevation angle of point i; N is the total number of points in the ancient tree point cloud.

[0010] Furthermore, step S3 specifically includes: S31. Define the feasible space for path planning as the region outside the non-uniform containment model; S32. Define an adaptive multi-objective cost function for planning paths in the feasible space. The adaptive multi-objective cost function includes the ancient tree vulnerability cost and the inspection coverage quality cost, which are weighted and summed according to their respective adaptive weights. S33. Determine the point in the ancient tree point cloud that is closest to a certain path point in the planned path. Use the ratio of the vulnerability level of the point in the ancient tree point cloud to the closest distance as the ancient tree vulnerability cost of the path point. The sum of the ancient tree vulnerability costs of all path points is the ancient tree vulnerability cost of the path. S34. Calculate the coverage area ratio of the planned route to the key parts of the ancient trees, and use this to calculate the inspection coverage quality cost. S35. Based on historical data, calculate the mean of the vulnerability cost of ancient trees and the mean of the inspection and coverage quality cost, and use these to calculate the adaptive weights of the vulnerability cost of ancient trees and the inspection and coverage quality cost, respectively.

[0011] Furthermore, step S35, based on historical data, includes the average values ​​of the vulnerability cost of ancient trees and the average values ​​of the inspection coverage quality cost, which include: Data on the vulnerability cost of ancient trees and the quality cost of inspection coverage obtained from existing inspection records of the same ancient tree; Mean data on the vulnerability cost and inspection coverage quality cost of ancient trees of the same species and similar age; Statistical baseline data on the vulnerability cost of ancient trees and the quality cost of inspection and coverage of all ancient trees; Weights are assigned to the data at the three levels mentioned above, with the weights summing to 1. Then, a weighted sum is performed to obtain the mean of the vulnerability cost of ancient trees and the mean of the inspection coverage quality cost. After each complete inspection task is completed, the data at the three levels mentioned above are automatically updated, and the average value of the vulnerability cost of ancient trees and the average value of the inspection coverage quality cost are recalculated.

[0012] Furthermore, step S4 specifically includes: S41. Initialization: Set the starting point F and ending point G of the UAV, as well as the OPEN list for storing nodes to be expanded and the CLOSED list for storing expanded nodes; the nodes are discrete grid points in the A* algorithm search space and are candidate positions in the path planning process. S42. Set the total cost function: ; in, Let n be the total cost of node n. This is the actual cost from the starting point F to node n, which is the sum of the physical lengths of the paths already traversed. The estimated cost from node n to the endpoint G is calculated using Euclidean distance. The adaptive adjustment coefficient is dynamically adjusted according to the protection level of the ancient tree. The adaptive cost of node n is calculated based on the adaptive multi-objective cost function; S43. Choosing a heuristic function: ; Let be the Euclidean distance from node n to the endpoint G. These are adaptive coefficients, predefined based on the complexity of the environment surrounding the ancient tree; S44, Node Expansion: Select from the OPEN list For the smallest node n, move node n from the OPEN list to the CLOSED list; generate the neighbor nodes of node n, and calculate the total cost for each neighbor node of node n according to the total cost function; if a neighbor node is not in the OPEN list, add it to the OPEN list, otherwise check whether the neighbor node needs to be updated. S45. Continue to perform the node expansion until the endpoint G is added to the CLOSED list or the OPEN list is empty, then terminate. S46. Backtrack from the endpoint G to the starting point F to obtain the optimal path.

[0013] Furthermore, step S5 includes: S51. Verify whether the path meets the safety constraints: Check whether all points on the optimal path are outside the safe shell; S52. Verify path quality: Calculate the adaptive multi-objective cost of the optimal path based on the adaptive multi-objective cost function to confirm whether it is the optimal path.

[0014] In another aspect, this invention proposes a path planning system for ancient tree UAV inspection based on absolute safety constraints and multi-objective adaptation, comprising: Vulnerability Level Module: Based on ancient tree point cloud data and spectral images, analyze the vulnerable structure information, disease information, and environmental information of ancient trees to conduct vulnerability level assessment and obtain the vulnerability level of each point in the ancient tree point cloud; Containment Module: Calculates the safety distance increment for each point based on its vulnerability level in the ancient tree point cloud, and constructs a non-uniform containment model; Adaptive multi-objective cost function module: Based on the non-uniform containment model, an adaptive multi-objective cost function is constructed in the outer region of the containment based on the vulnerability cost of ancient trees and the inspection coverage quality cost; Path planning module: Based on the adaptive multi-objective cost function, it performs improved A* algorithm path planning to obtain the optimal path; Verification module: Verifies the optimal path.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Absolute safety constraint: This invention constructs a non-uniform three-dimensional no-fly safety shell model that closely fits the shape of ancient trees. As an insurmountable rigid spatial constraint for path planning, it eliminates the possibility of physical collisions from the root and ensures the absolute safety of ancient and famous trees.

[0016] 2. Adaptive Multi-Objective Optimization: This invention establishes an adaptive multi-objective evaluation model that integrates the vulnerability cost of ancient trees and the quality cost of inspection and coverage. It can dynamically adjust the weight of each objective according to the specific situation of the ancient tree (tree species, tree age, tree shape, vulnerability level) to achieve adaptive optimization of path planning.

[0017] 3. High-precision modeling: This invention uses high-precision 3D point cloud reconstruction technology to accurately obtain the geometric shape of ancient trees, and constructs a non-uniform safety shell based on the protection vulnerability level of each part of the ancient tree. Compared with the simplified geometric model of the existing technology, it greatly improves obstacle avoidance accuracy.

[0018] 4. Highly efficient inspection: Under the premise of ensuring absolute safety, this invention optimizes the inspection path, which can improve the inspection efficiency by more than 30% and the path quality (including path length, flight stability and coverage integrity) by more than 35%, providing an efficient, safe and accurate inspection technology solution for the protection of ancient and famous trees.

[0019] 5. Wide applicability: This invention is not only suitable for the protection of ancient and famous trees, but can also be applied to the protection of cultural relics and buildings, the inspection of precision equipment, and other fields, and has a wide range of application prospects. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the system structure of Embodiment 3 of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0022] To make the purpose and features of this invention patent more apparent and understandable, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0023] Example 1: The ancient tree UAV inspection path planning method based on absolute safety constraints and multi-objective adaptive method applied in Example 1 is as follows: Figure 1 As shown, it includes: Step 1: Acquisition of high-precision 3D point cloud of ancient trees and assessment of their vulnerability level.

[0024] 1.1 Three-dimensional scanning of ancient trees was carried out using lidar and multispectral cameras to obtain three-dimensional point cloud data of ancient trees, including the root system area, as well as multispectral image data of ancient trees. This embodiment uses a drone equipped with a high-precision LiDAR and a multispectral camera to perform 3D scanning of ancient trees, obtaining high-precision 3D point cloud data and multispectral image data of the ancient trees.

[0025] 1.2 Preprocessing of point cloud data, including point cloud filtering, ground point separation, and point cloud segmentation, is performed to obtain an accurate geometric model of the ancient tree.

[0026] Specifically, point cloud filtering is achieved using a Gaussian filter to remove noise points; ground point separation uses the RANSAC algorithm to fit the ground plane, calculates the vertical distance between the point cloud and the ground plane, and classifies points with a distance greater than a threshold as non-ground points, thus dividing the point cloud into ground points and non-ground points; point cloud segmentation is achieved using a region growing algorithm, which segments the point cloud based on the proximity and normal vector similarity of the point cloud to obtain the ancient tree point cloud.

[0027] Ancient tree dotted cloud is represented by P: ; in Point i represents the ancient tree point cloud data. Let be the three-dimensional coordinates of point i, and N be the total number of points in the ancient tree point cloud.

[0028] 1.3 Based on the point cloud of ancient trees, calculate the vulnerability level of each part of the ancient tree.

[0029] Vulnerability level assessment is based on the following factors: ; Where V represents the vulnerability level, The vulnerability weighting coefficient, for example, can be obtained through historical data statistics and expert evaluation, or it can be defined based on historical experience. . Indicates structural fragility. This indicates the vulnerability of the disease. This indicates environmental vulnerability.

[0030] The method for calculating structural vulnerability is as follows: ; in, For example, in this embodiment, a deep learning semantic segmentation network (such as U-Net) is used to identify crack regions in the ancient tree point cloud to determine the crack length. After mapping the identification results to the ancient tree point cloud, the point cloud at the crack is subjected to skeleton extraction processing to obtain the centerline trajectory of the crack. The actual length of the crack is obtained by calculating and summing the Euclidean distances between adjacent points on the centerline. For branched cracks, the length of the main trunk and the length of each branch can be calculated separately, and the maximum value is selected as the final crack length according to the protection criteria.

[0031] To determine the total diameter of the ancient tree, this embodiment, for example, processes the point cloud of the ancient tree into layers along the height direction, focusing on three key layers: the root layer (0.3 meters high), the diameter at breast height (DBH) layer (e.g., 1.3 meters high), and the middle layer (the highest point of the tree). Each layer of the point cloud is projected onto a horizontal plane, and a computational geometry algorithm (such as the Welzl algorithm) is used to calculate the minimum circumscribed circle diameter containing all projected points. According to industry standards for ancient tree conservation, the DBH layer data accounts for 50% of the weight, the root layer for 30%, and the middle layer for 20%, and the total diameter of the ancient tree is obtained through a weighted average. For irregular trunks, the measurement value at the location of the maximum diameter can be additionally considered.

[0032] To determine the tilt height, for example, in this embodiment, the ground plane is first accurately fitted using the RANSAC algorithm to determine the ground reference height. Then, the centroid coordinates of the ancient tree point cloud are calculated; these centroid coordinates are the weighted average of all point positions. The bottom center coordinates are also calculated, using the centroids of the point cloud within a 0.5-meter height range above the ground as the bottom center coordinates. Next, the horizontal distance between the projection of the centroid onto the horizontal plane and the bottom center is calculated. Combined with the total height of the ancient tree, the tilt angle is calculated using trigonometric functions. The tilt height is equal to the total height of the ancient tree multiplied by the sine of the tilt angle. When the tilt angle is less than 5 degrees, the ancient tree is considered essentially vertical, and the tilt height is automatically set to zero.

[0033] For the total height of the ancient tree, as exemplified in this embodiment, attitude correction is performed on the point cloud of the ancient tree to ensure that the Z-axis is consistent with the direction of gravity. The ground plane is fitted using the RANSAC algorithm to accurately determine the ground height reference. The maximum value of the point cloud in the Z-axis direction is calculated, outliers (such as noise caused by birds or swaying branches) are automatically eliminated, and the true highest point is determined using a statistical method such as the 95th percentile. The total height of the ancient tree is equal to the difference between the Z-coordinate of the highest point and the ground height.

[0034] To calculate the area of ​​bark damage, this embodiment, for example, requires fusing the multispectral image data of the ancient tree. Utilizing the differences in reflectance characteristics of the bark across different wavelengths, a deep learning classification model is used to automatically identify damaged bark areas, including peeling, rotting, and insect infestation. The identification results are accurately mapped to the ancient tree point cloud, forming a subset of the damaged area point cloud. High-precision surface reconstruction is then performed on this subset of the damaged area point cloud to generate a triangular mesh model. By calculating the sum of the areas of all triangular facets, the total area of ​​bark damage is obtained.

[0035] To determine the total surface area of ​​the ancient tree, this embodiment employs a Poisson surface reconstruction algorithm to convert the discrete point cloud data of the ancient tree into a continuous triangular mesh surface. The algorithm first estimates the normal vector of each point, then solves the Poisson equation to generate an indicator function, and finally extracts the triangular mesh through isosurfaces. Each triangular facet in the mesh is traversed, its area is calculated, and then the areas of all faces are summed to obtain the total surface area of ​​the ancient tree.

[0036] This is a structural vulnerability weighting coefficient. It is obtained through historical data statistics and expert evaluation, and is exemplified in this embodiment. .

[0037] The method for calculating disease vulnerability is as follows: ; in, For example, this embodiment uses multispectral image data of the ancient tree to obtain spectral images of the tree surface, focusing on the analysis of visible and near-infrared data. The lesion area is automatically identified by utilizing the reflectance differences between lesion areas and healthy tissue in different wavelength bands, combined with machine learning classification algorithms (such as support vector machines). The identification results are mapped to the corresponding ancient tree point cloud, and surface reconstruction is performed on the lesion area point cloud to generate a triangular mesh model. The total area of ​​the mesh is then calculated as the lesion area.

[0038] To determine the length of insect damage, this embodiment, for example, combines point clouds of ancient trees to identify insect-damaged holes and channels. The identification method involves using a deep learning object detection algorithm (such as YOLO) to locate the insect-damaged entrance, and then tracing the extension path of the insect-damaged channel inside the trunk using ray casting. The skeleton of the insect-damaged channel is extracted from the point cloud, and the centerline length is calculated to obtain the actual length of the insect damage.

[0039] This is a disease vulnerability weighting coefficient; it can be obtained through expert assessment and statistical analysis of historical disease data. For example, in this embodiment... .

[0040] The method for calculating environmental vulnerability is as follows: ; in, For example, in this embodiment, the 3D point cloud data of the soil surface can be obtained from the 3D point cloud data of the ancient tree acquired through the original scan. The loosened areas are automatically identified by utilizing the differences between the loosened and stable areas in terms of surface roughness, elevation changes, etc., combined with a machine learning classification algorithm. The point cloud of the loosened areas is then reconstructed to generate a triangular mesh model, and the total area of ​​the mesh is calculated as the loosened soil area.

[0041] The total area of ​​the ancient tree root system region is exemplified in this embodiment. The three-dimensional point cloud data of the ancient tree root system region can be obtained from the original scanned three-dimensional point cloud data of the ancient tree. The total area of ​​the ancient tree root system region can be obtained by calculating the convex hull area of ​​the root system projection range.

[0042] For the wind-affected area, this embodiment integrates meteorological station data, topographic data, and ancient tree point cloud data for comprehensive analysis. First, historical wind direction and speed data are acquired, and the wind field distribution is calculated using a topographic elevation model. Then, the influence of ancient tree morphology on the wind field is analyzed, and the wind concentration area is determined through computational fluid dynamics simulation. The wind-affected area is projected onto the surface of the ancient trees, and the actual affected surface area is determined by combining this with the distribution of the ancient trees' mechanical strength.

[0043] The total area surrounding the ancient tree is represented by a cube space containing the ancient tree and its surrounding environment, constructed with the center of the ancient tree point cloud as the origin. The total area surrounding the ancient tree is obtained by calculating the projected area of ​​the smallest bounding cube of the point cloud onto the XY plane.

[0044] The environmental vulnerability weighting coefficient can be obtained through meteorological data statistics and topographic analysis. For example, in this embodiment... .

[0045] 1.4 Map the vulnerability levels onto the ancient tree point cloud to obtain the vulnerability level of each point. .

[0046] The mapping process involves associating point cloud data with vulnerability levels based on spatial location and calculating the vulnerability using the nearest neighbor interpolation method. The value range is [0,1], where 1 represents the highest vulnerability.

[0047] Step 2: Construct a three-dimensional no-fly non-uniform containment model.

[0048] 2.1 Construct a spherical coordinate system with the center of the ancient tree point cloud as the origin, and calculate the angle and distance of each point relative to the center: ; in, The azimuth angle is calculated from the point cloud coordinates. The elevation angle is calculated using point cloud coordinates. The distance from point i to the center of the point cloud is calculated using the point cloud coordinates.

[0049] 2.2 Based on the vulnerability level of point clouds Calculate the safety distance increment for each point: ; in, The safety distance coefficient ranges from 0.5 to 2.0 and is dynamically adjusted according to the protection level of the ancient tree. In this embodiment, the value is exemplarily set as follows: =1.5, The minimum vulnerability level is obtained through vulnerability level statistics from point cloud data. In this embodiment, the statistics are as follows: .

[0050] 2.3 Construct a non-uniform containment model, where the safety distance at each point is: ; in, Let i be the safe distance at point i.

[0051] 2.4 The containment model is represented as: ; in, Let i be the point on the containment model corresponding to point i in the ancient tree point cloud.

[0052] The containment model defines the minimum safe distance that drones must maintain, serving as an absolute safety constraint to ensure that no collision between drones and ancient trees occurs under any circumstances.

[0053] Step 3: Construct an adaptive multi-objective cost function.

[0054] 3.1 Define the feasible space for path planning as the region outside the containment structure, i.e., satisfying: ; indicates that for the planned path path, each point The distance to set S is greater than 0.

[0055] in, Represents the path points in the path. The shortest distance to the containment model S is obtained by calculating the Euclidean distance using point cloud data.

[0056] 3.2 Define the adaptive multi-objective cost function, including: The cost of ancient tree vulnerability : Related to the distance between the path and the vulnerable part; Inspection coverage quality cost : Related to the degree of coverage of different parts of the ancient tree by the path.

[0057] The adaptive multi-objective cost function is: ; in, Represents the adaptive multi-objective cost; As an adaptive weight, it can be dynamically adjusted according to the specific conditions of the ancient tree.

[0058] 3.3 Calculation of the vulnerability cost of ancient trees: ; in, This represents the number of points in the current temporary path, which is dynamically determined by the path planning algorithm. For point The vulnerability level at a location is calculated by selecting a path point. Recent ancient tree cloud points Vulnerability level As its vulnerability level; path point To the nearest point of ancient tree cloud The distance is calculated by the point. To the nearest ancient tree dotted with clouds The distance is obtained.

[0059] The purpose of calculating the vulnerability cost of ancient trees is that, although the drone's flight path is strictly outside the containment shell (not inside the ancient tree point cloud), the spatial location of the path points and their relative distance to the vulnerable areas of the ancient tree directly determines the potential risk level of the flight operation to the ancient tree. The vulnerability level calculation is not based on the attributes of the path points themselves, but rather quantifies the impact of the flight position on the vulnerable areas of the ancient tree. The goal of this cost is to keep the drone as far away as possible from the vulnerable areas of the ancient tree.

[0060] 3.4 Calculation of the quality cost of inspection coverage: ; Where k is the number of key parts of the ancient tree. For example, if key parts include the trunk, crown, and root system, then k=3. For key parts The coverage level is obtained by calculating the proportion of the coverage area of ​​the key parts by the path, and the value range is [0,1], where 1 represents complete coverage.

[0061] For example, in this embodiment The calculation method is as follows: (1) Key parts division: The ancient tree is divided into three key parts: trunk, crown and root system. The point cloud of each part is accurately separated by point cloud segmentation algorithm (deep learning); (2) Determining the observation range: For each path point on the planned path, the visible range is calculated based on the UAV camera parameters (field of view, resolution) to form a frustum model; (3) Visible point identification: Project the point cloud of each key part onto the UAV camera coordinate system to determine which points are located within the view frustum and are unobstructed. Occlusion can be detected by ray projection method. (4) Coverage area calculation: for each key area The observed point cloud is reconstructed using surface reconstruction to generate a triangular mesh, and the area of ​​this mesh is calculated to obtain the observed area. Simultaneously calculate the total surface area of ​​this critical component. ; (5) Determining the coverage ratio: Coverage level equal Divide by .

[0062] 3.5 Calculation of Adaptive Weights: ; in, This is the average cost of the vulnerability of ancient trees, updated through historical data statistics. For example, in this embodiment... ; The average cost of inspection coverage quality is updated using historical data statistics. For example, in this embodiment... .

[0063] The historical data statistical update method includes: 1. Three-level data and weights: Based on the data of the vulnerability cost and inspection coverage quality cost of the ancient tree obtained from the existing inspection records of the same ancient tree, for example, the weight is assigned as 0.6; The average data of the vulnerability cost of ancient trees and the quality cost of inspection and coverage for ancient trees of the same species and similar age are used as an example, with a weight of 0.3. The statistical baseline data of the vulnerability cost of ancient trees and the quality cost of inspection and coverage of all ancient trees in historical data are assigned a weight of 0.1 for example. The weights of the three factors add up to 1; The weighted summation of the data from the three levels above yields the mean of the vulnerability cost of ancient trees and the mean of the quality cost of inspection and coverage.

[0064] 2. Update mechanism: Regular updates: After each complete inspection task, the data records at the three levels mentioned above are automatically updated. The latest data is then obtained by weighted summation of the data at these three levels.

[0065] 3. Initialization Settings: When a new ancient tree is used for the first time, the average value of the vulnerability cost and the inspection coverage quality cost of similar ancient trees (same species, similar age) is used as the initial value (e.g., As the number of inspections increases, the parameters will be gradually adjusted to be personalized.

[0066] Step 4: Improved A* algorithm path planning.

[0067] 4.1 Initialization: Starting point F: The initial position of the drone, obtained through GPS positioning; Destination G: The target location of the drone, determined based on inspection requirements; OPEN list: Stores nodes to be expanded; initialized as an empty list. CLOSED list: Stores expanded nodes, initialized as an empty list.

[0068] 4.2 Cost Function Calculation: ; in, This is the actual cost from the starting point F to node n, which is the sum of the physical lengths of the paths already traversed. The estimated cost from node n to the endpoint G is calculated using Euclidean distance. This is an adaptive adjustment coefficient, with a value range of [0.1, 0.9]. It is dynamically adjusted according to the protection level of the ancient tree. In this example, it is set to [value missing]. ; The cumulative cost of node n is calculated as follows: In the improved A* algorithm, node n refers to a discrete grid point in the A* algorithm's search space, representing a candidate location considered by the algorithm during path planning. The entire search space is divided into a three-dimensional grid, with the center point of each grid forming a node. Node n represents a candidate location currently being evaluated by the algorithm.

[0069] During the execution of the A* algorithm, when the algorithm expands from the starting point to node n, the current temporary path contains all visited nodes from the starting point F to node n. The vulnerability cost of all points on the current temporary path is calculated using the formula in step 3.3, resulting in... The total inspection coverage quality cost of all points on the current temporary path is calculated using formula 3.4, resulting in... Then, the adaptive multi-objective cost of the current temporary path is calculated using the formula in step 3.2. The result calculated at this time As the cumulative cost of node n .

[0070] 4.3 Choosing a heuristic function: choose The calculation of is used as the heuristic function for the improved A* algorithm in this embodiment.

[0071] ; in, Let be the Euclidean distance from node n to the endpoint G, calculated using coordinates. This is an adaptive coefficient, with a value range of [0.5, 1.5]. It is predefined based on the complexity of the environment surrounding the ancient tree. For example, in this embodiment, the value is [value missing]. .

[0072] 4.4 Node Expansion: Select from the OPEN list The smallest node n is moved from the OPEN list to the CLOSED list to generate the neighbor nodes of n, including the six directions of up, down, left, right, front, and back; For each neighbor node m, calculate If m is not in the OPEN list, add it to the OPEN list; if m is already in the OPEN list, check if it needs to be updated.

[0073] The node n initially selected in the above node expansion process is definitely the starting point F, that is, the expansion continues from the starting point F until the termination condition is met.

[0074] 4.5 Termination Conditions: The termination condition is met when the endpoint G is added to the CLOSED list or the OPEN list is empty.

[0075] 4.6 Path backtracking: The path obtained by tracing back from the endpoint G to the starting point F is the optimal path.

[0076] Step 5: Path verification.

[0077] Verify that the path meets safety constraints: Check whether all points on the optimal path are outside the safe shell, i.e. ; Represents a point on the optimal path; Verify path quality: Calculate the adaptive multi-objective cost of the optimal path based on the aforementioned adaptive multi-objective cost function. The goal is to confirm whether this is the optimal path. The confirmation objective is to find another optimal path through further iterations or multiple iterations, with the path that minimizes the adaptive multi-objective cost being verified as the optimal path.

[0078] The core innovations of this embodiment include: First, a non-uniform three-dimensional no-fly safety shell model was constructed that closely matches the shape of the ancient tree, eliminating the possibility of physical collisions from the root and ensuring the absolute safety of the ancient and famous trees.

[0079] Second, an adaptive multi-objective evaluation model was established that integrates the vulnerability cost of ancient trees and the quality cost of inspection and coverage. This model can dynamically adjust the weight of each objective according to the specific situation of the ancient trees, thereby achieving adaptive optimization of path planning.

[0080] Example 2: This embodiment is a practical application example of the method described in Embodiment 1.

[0081] Take, for example, a thousand-year-old ginkgo tree in a forest park. The tree is about 1,000 years old, about 25 meters tall, and has a trunk diameter of about 3 meters. The tree has multiple cracks in its trunk and diseases in the crown, and the soil in the root area is loose.

[0082] (1) Use a drone equipped with a high-precision LiDAR and a multispectral camera to perform three-dimensional scanning of ancient trees and obtain high-precision three-dimensional point cloud data and multispectral image data.

[0083] (2) Preprocess the point cloud data to obtain the accurate geometric model of the ancient tree point cloud.

[0084] (3) Based on the ancient tree point cloud model, the vulnerability level of each part was calculated, and the vulnerability level of the trunk crack area was 0.85, the vulnerability level of the crown disease area was 0.75, and the vulnerability level of the root soil loosening area was 0.65.

[0085] (4) Construct a non-uniform safety shell model. The safety shell increases the safety distance by 0.5 meters in the trunk crack area, 0.3 meters in the crown disease area, and 0.2 meters in the root soil loose area.

[0086] (5) In the flyable space outside the containment structure, establish an adaptive multi-objective cost function and dynamically adjust the adaptive weights according to the specific situation of the ancient tree: .

[0087] (6) The improved A* algorithm is used to optimize the path and obtain the optimal inspection path.

[0088] (7) Verify the path to ensure that all points on the path are outside the safe zone and that the path quality is optimal.

[0089] Example 3: This embodiment proposes a path planning system for ancient tree UAV inspection based on absolute safety constraints and multi-objective adaptive design, such as... Figure 2 As shown, it includes: Vulnerability Level Module: Based on ancient tree point cloud data and spectral images, analyze the vulnerable structure information, disease information, and environmental information of ancient trees to conduct vulnerability level assessment and obtain the vulnerability level of each point in the ancient tree point cloud; Containment Module: Calculates the safety distance increment for each point based on its vulnerability level in the ancient tree point cloud, and constructs a non-uniform containment model; Adaptive multi-objective cost function module: Based on the non-uniform containment model, an adaptive multi-objective cost function is constructed in the outer region of the containment based on the vulnerability cost of ancient trees and the inspection coverage quality cost; Path planning module: Based on the adaptive multi-objective cost function, it performs improved A* algorithm path planning to obtain the optimal path; Verification module: Verifies the optimal path.

[0090] The vulnerability level module includes: Three-dimensional scanning of ancient trees was carried out using lidar and multispectral cameras to obtain three-dimensional point cloud data of ancient trees, including the root system area, as well as multispectral image data of ancient trees. Preprocess the 3D point cloud data of ancient trees to segment non-ground points and obtain the point cloud of ancient trees; Calculate the vulnerability level of ancient trees: ; Where V represents the vulnerability level; The structural vulnerability is represented by analyzing ancient tree point cloud and multispectral image data to identify vulnerable structural information of ancient trees, including cracks, tilting, and bark damage, and then calculating the vulnerability. It indicates the vulnerability to diseases. By analyzing the point cloud and multispectral image data of ancient trees, it identifies disease information of ancient trees, including lesions and insect infestation, and calculates the vulnerability. The environmental vulnerability is represented by analyzing the point cloud data of ancient trees and the three-dimensional point cloud data of the root system area to obtain environmental information including the area of ​​soil loosening and the area affected by wind, and then calculating it. These are the weighting coefficients; The vulnerability levels are mapped onto the ancient tree point cloud to obtain the vulnerability level of each point.

[0091] The preprocessing includes: 3D point cloud data filtering is achieved using a Gaussian filter. Use the RANSAC algorithm to fit the ground plane; Calculate the vertical distance between the point cloud and the ground plane, and classify points whose distance is greater than a threshold as non-ground points, thereby dividing the point cloud into ground points and non-ground points; The region growing algorithm is used to segment non-ground points to obtain ancient tree point clouds.

[0092] The containment module includes: A spherical coordinate system is constructed with the center of the ancient tree cloud as the origin, and the angle and distance of each point relative to the origin are calculated; the angle of each point relative to the origin includes the azimuth and elevation angles. The safe distance increment for each point is calculated based on the difference between the vulnerability level of each point in the ancient tree point cloud and the lowest vulnerability level therein, combined with the safe distance coefficient. Calculate the safe distance of each point in the ancient tree point cloud, and construct a non-uniform safe shell model of the ancient tree based on the safe distance; the safe distance of each point is the sum of the distance of each point relative to the origin and the safe distance increment of each point.

[0093] The non-uniform containment model is represented as follows: ; in Let i be the point on the containment model corresponding to point i in the ancient tree point cloud. Let i be the safe distance. Let i be the azimuth angle of point i. Let be the elevation angle of point i; N is the total number of points in the ancient tree point cloud.

[0094] The adaptive multi-objective cost function module includes: The feasible space for path planning is defined as the region outside the non-uniform containment model; An adaptive multi-objective cost function for planning paths in the feasible space is defined. The adaptive multi-objective cost function includes the ancient tree vulnerability cost and the inspection coverage quality cost, which are weighted and summed according to their respective adaptive weights. Identify the point in the ancient tree point cloud that is closest to a certain path point in the planned path. Use the ratio of the vulnerability level of the point in the ancient tree point cloud to the closest distance as the ancient tree vulnerability cost of that path point. The sum of the ancient tree vulnerability costs of all path points is the ancient tree vulnerability cost of the path. Calculate the coverage area ratio of the planned route to key parts of ancient trees, and use this to calculate the cost of inspection coverage quality. Based on historical data, the mean values ​​of ancient tree vulnerability cost and inspection coverage quality cost are statistically analyzed, and adaptive weights for ancient tree vulnerability cost and inspection coverage quality cost are calculated accordingly.

[0095] The average values ​​of the vulnerability cost of ancient trees and the average values ​​of the inspection and coverage quality cost based on historical data include: Data on the vulnerability cost of ancient trees and the quality cost of inspection coverage obtained from existing inspection records of the same ancient tree; Mean data on the vulnerability cost and inspection coverage quality cost of ancient trees of the same species and similar age; Statistical baseline data on the vulnerability cost of ancient trees and the quality cost of inspection and coverage of all ancient trees; Weights are assigned to the data at the three levels mentioned above, with the weights summing to 1. Then, a weighted sum is performed to obtain the mean of the vulnerability cost of ancient trees and the mean of the inspection coverage quality cost. After each complete inspection task is completed, the data at the three levels mentioned above are automatically updated, and the average value of the vulnerability cost of ancient trees and the average value of the inspection coverage quality cost are recalculated.

[0096] The route planning module includes: Initialization: Set the starting point F and ending point G of the UAV, and store the OPEN list of nodes to be expanded and the CLOSED list of expanded nodes; the nodes are discrete grid points in the search space of the A* algorithm, which are candidate positions in the path planning process; Set the total cost function: ; in, Let n be the total cost of node n. This is the actual cost from the starting point F to node n, which is the sum of the physical lengths of the paths already traversed. The estimated cost from node n to the endpoint G is calculated using Euclidean distance. The adaptive adjustment coefficient is dynamically adjusted according to the protection level of the ancient tree. The adaptive cost of node n is calculated based on the adaptive multi-objective cost function; Choosing a heuristic function: ; Let be the Euclidean distance from node n to the endpoint G. These are adaptive coefficients, predefined based on the complexity of the environment surrounding the ancient tree; Node expansion: Select from the OPEN list For the smallest node n, move node n from the OPEN list to the CLOSED list; generate the neighbor nodes of node n, and calculate the total cost for each neighbor node of node n according to the total cost function; if a neighbor node is not in the OPEN list, add it to the OPEN list, otherwise check whether the neighbor node needs to be updated. Continue performing the node expansion until the endpoint G is added to the CLOSED list or the OPEN list is empty, at which point the process terminates. By tracing back from the endpoint G to the starting point F, the optimal path is obtained.

[0097] The verification module includes: Verify that the path meets the safety constraints: check whether all points on the optimal path are outside the safe shell; Verify path quality: Calculate the adaptive multi-objective cost of the optimal path based on the adaptive multi-objective cost function to confirm whether it is the optimal path.

[0098] The ancient tree UAV inspection path planning system based on absolute safety constraints and multi-objective adaptation proposed in this embodiment can achieve the ancient tree UAV inspection path planning method based on absolute safety constraints and multi-objective adaptation described in Embodiment 1, and has the same technical effect.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for planning a patrol route of an unmanned aerial vehicle (UAV) for an ancient tree based on absolute safety constraints and multi-objective self-adaptation, characterized in that, include: S1. Based on the ancient tree point cloud data and spectral images, analyze the information on the vulnerable structure, disease, and environment of the ancient trees, conduct a vulnerability level assessment, and obtain the vulnerability level of each point in the ancient tree point cloud. S2. Calculate the safety distance increment for each point based on its vulnerability level in the ancient tree point cloud, and construct a non-uniform safe shell model; including: S21. Construct a spherical coordinate system with the center of the ancient tree cloud as the origin, and calculate the angle and distance of each point relative to the origin; the angle of each point relative to the origin includes the azimuth and elevation angles; S22. Calculate the safety distance increment for each point based on the difference between the vulnerability level of each point in the ancient tree point cloud and the lowest vulnerability level, combined with the safety distance coefficient. S23, calculate the safety distance of each point in the ancient tree point cloud, and construct a non-uniform safety shell model of the ancient tree according to the safety distance; the safety distance of each point is the sum of the distance of each point relative to the origin and the safety distance increment of each point; the non-uniform safety shell model is expressed as: ; in Let i be the point on the containment model corresponding to point i in the ancient tree point cloud. Let i be the safe distance. Let i be the azimuth angle of point i. Let be the elevation angle of point i; N is the total number of points in the ancient tree point cloud; S3. Based on the non-uniform containment model, construct an adaptive multi-objective cost function in the outer region of the containment based on the vulnerability cost of ancient trees and the inspection coverage quality cost. S4. Based on the adaptive multi-objective cost function, perform improved A* algorithm path planning to obtain the optimal path; S5. Verify the optimal path.

2. The ancient tree UAV inspection path planning method based on absolute safety constraints and multi-objective adaptive method according to claim 1, characterized in that, Step S1 specifically includes: S11. Perform 3D scanning of ancient trees using lidar and multispectral cameras to obtain 3D point cloud data of ancient trees, including the root system area, as well as multispectral image data of ancient trees. S12. Preprocess the 3D point cloud data of ancient trees, segment non-ground points, and obtain the point cloud of ancient trees. S13. Calculate the vulnerability level of ancient trees: ; Where V represents the vulnerability level; The structural vulnerability is represented by analyzing ancient tree point cloud and multispectral image data to identify vulnerable structural information of ancient trees, including cracks, tilting, and bark damage, and then calculating the vulnerability. It indicates the vulnerability to diseases. By analyzing the point cloud and multispectral image data of ancient trees, it identifies disease information of ancient trees, including lesions and insect infestation, and calculates the vulnerability. The environmental vulnerability is represented by analyzing point cloud data of ancient trees and three-dimensional point cloud data of the root system area to obtain environmental information including the area of ​​soil loosening and the area affected by wind, and then calculating it. These are the weighting coefficients; S14. Map the vulnerability levels onto the ancient tree point cloud to obtain the vulnerability level of each point.

3. The ancient tree UAV inspection path planning method based on absolute safety constraints and multi-objective adaptive method according to claim 2, characterized in that, The preprocessing in step S12 includes: 3D point cloud data filtering is achieved using a Gaussian filter. Use the RANSAC algorithm to fit the ground plane; Calculate the vertical distance between the point cloud and the ground plane, and classify points whose distance is greater than a threshold as non-ground points, thereby dividing the point cloud into ground points and non-ground points; The region growing algorithm is used to segment non-ground points to obtain ancient tree point clouds.

4. The method for UAV inspection path planning of ancient trees based on absolute safety constraints and multi-objective adaptive methods according to claim 1, characterized in that, Step S3 specifically includes: S31. Define the feasible space for path planning as the region outside the non-uniform containment model; S32. Define an adaptive multi-objective cost function for planning paths in the feasible space. The adaptive multi-objective cost function includes the ancient tree vulnerability cost and the inspection coverage quality cost, which are weighted and summed according to their respective adaptive weights. S33. Determine the point in the ancient tree point cloud that is closest to a certain path point in the planned path. Use the ratio of the vulnerability level of the point in the ancient tree point cloud to the closest distance as the ancient tree vulnerability cost of the path point. The sum of the ancient tree vulnerability costs of all path points is the ancient tree vulnerability cost of the path. S34. Calculate the coverage area ratio of the planned route to the key parts of the ancient trees, and use this to calculate the inspection coverage quality cost. S35. Based on historical data, calculate the mean of the vulnerability cost of ancient trees and the mean of the inspection and coverage quality cost, and use these to calculate the adaptive weights of the vulnerability cost of ancient trees and the inspection and coverage quality cost, respectively.

5. The ancient tree UAV inspection path planning method based on absolute safety constraints and multi-objective adaptive method according to claim 4, characterized in that, Step S35, based on historical data, includes the average values ​​of the vulnerability cost of ancient trees and the average values ​​of the inspection coverage quality cost, which include: Data on the vulnerability cost of ancient trees and the quality cost of inspection coverage obtained from existing inspection records of the same ancient tree; Average data on the vulnerability cost and inspection coverage quality cost of ancient trees of the same species and similar age; Statistical baseline data on the vulnerability cost of ancient trees and the quality cost of inspection and coverage of all ancient trees; Weights are assigned to the data at the three levels mentioned above, with the weights summing to 1. Then, a weighted sum is performed to obtain the mean of the vulnerability cost of ancient trees and the mean of the inspection coverage quality cost. After each complete inspection task is completed, the data at the three levels mentioned above are automatically updated, and the average value of the vulnerability cost of ancient trees and the average value of the inspection coverage quality cost are recalculated.

6. The method for UAV inspection path planning of ancient trees based on absolute safety constraints and multi-objective adaptive design according to claim 1, characterized in that, Step S4 specifically includes: S41. Initialization: Set the starting point F and ending point G of the UAV, as well as the OPEN list for storing nodes to be expanded and the CLOSED list for storing expanded nodes; the nodes are discrete grid points in the A* algorithm search space and are candidate positions in the path planning process. S42. Set the total cost function: ; in, Let n be the total cost of node n. This is the actual cost from the starting point F to node n, which is the sum of the physical lengths of the paths already traversed. The estimated cost from node n to the endpoint G is calculated using Euclidean distance. The adaptive adjustment coefficient is dynamically adjusted according to the protection level of the ancient tree. The adaptive cost of node n is calculated based on the adaptive multi-objective cost function; S43. Choosing a heuristic function: ; Let be the Euclidean distance from node n to the endpoint G. These are adaptive coefficients, predefined based on the complexity of the environment surrounding the ancient tree; S44, Node Expansion: Select from the OPEN list For the smallest node n, move node n from the OPEN list to the CLOSED list; generate the neighbor nodes of node n, and calculate the total cost for each neighbor node of node n according to the total cost function; if a neighbor node is not in the OPEN list, add it to the OPEN list, otherwise check whether the neighbor node needs to be updated. S45. Continue to perform the node expansion until the endpoint G is added to the CLOSED list or the OPEN list is empty, then terminate. S46. Backtrack from the endpoint G to the starting point F to obtain the optimal path.

7. The method for UAV inspection path planning of ancient trees based on absolute safety constraints and multi-objective adaptive methods according to claim 1, characterized in that, Step S5 includes: S51. Verify whether the path meets the safety constraints: Check whether all points on the optimal path are outside the safe shell; S52. Verify path quality: Calculate the adaptive multi-objective cost of the optimal path based on the adaptive multi-objective cost function to confirm whether it is the optimal path.

8. A path planning system for unmanned aerial vehicle (UAV) inspection of ancient trees based on absolute safety constraints and multi-objective adaptive design, characterized in that, include: Vulnerability Level Module: Based on ancient tree point cloud data and spectral images, analyze the vulnerable structure information, disease information, and environmental information of ancient trees to conduct vulnerability level assessment and obtain the vulnerability level of each point in the ancient tree point cloud; The safe shell module calculates the safety distance increment for each point in the ancient tree point cloud based on its vulnerability level, and constructs a non-uniform safe shell model. The module includes: constructing a spherical coordinate system with the center of the ancient tree point cloud as the origin; calculating the angle and distance of each point relative to the origin; the angle of each point relative to the origin includes azimuth and elevation; calculating the safety distance increment for each point based on the difference between its vulnerability level and the lowest vulnerability level, combined with a safety distance coefficient; calculating the safety distance for each point in the ancient tree point cloud; and constructing a non-uniform safe shell model for the ancient tree based on these safety distances; the safety distance for each point is the sum of its distance relative to the origin and the safety distance increment for each point; the non-uniform safe shell model is represented as follows: ;in Let i be the point on the containment model corresponding to point i in the ancient tree point cloud. Let i be the safe distance. Let i be the azimuth angle of point i. Let be the elevation angle of point i; N is the total number of points in the ancient tree point cloud; Adaptive multi-objective cost function module: Based on the non-uniform containment model, an adaptive multi-objective cost function is constructed in the outer region of the containment based on the vulnerability cost of ancient trees and the inspection coverage quality cost; Path planning module: Based on the adaptive multi-objective cost function, it performs improved A* algorithm path planning to obtain the optimal path; Verification module: Verifies the optimal path.

Images