Tree health state assessment method based on proxy model

Abstract

The invention discloses a tree health state assessment method based on a proxy model, which comprises the following steps: establishing a three-dimensional finite element model of trees by collecting geometrical shape data and material attribute data of different types of trees, and performing finite element analysis. And according to the analysis result and the geometric data, constructing an adaptive agent model by using a neural network, training the adaptive agent model, and generating a prediction model of the tree health state. Characteristic data such as stress, strain and damage factor distribution of a tree to be evaluated are collected in real time through a sensor and an unmanned aerial vehicle and input into the trained adaptive agent model for dynamic analysis, and data optimization model parameters are fed back in real time. And according to an evaluation result, windproof reinforcement countermeasures for trees with different health state levels are automatically generated, and the scientificity and the accuracy of the reinforcement measures are improved. According to the method, the accuracy of tree health state evaluation is improved, real-time monitoring and dynamic early warning are realized, timely identification and feedback can be realized at the initial stage of tree damage, a corresponding windproof reinforcement countermeasure is given, and delayed processing is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of tree assessment, and particularly relates to a method for assessing the health status of trees based on a surrogate model. Background Art

[0002] Trees play a crucial role in urban greening, ecological maintenance, and environmental beautification, and have significant ecological benefits and social values. However, with the frequent occurrence of global climate change and extreme weather events, trees in urban and natural environments are facing increasingly severe threats to their safety and health. Factors such as wind disasters, pests and diseases, and aging often lead to structural damage to trees, affecting their stability and even posing potential safety hazards to surrounding people and facilities. In the face of these problems, traditional visual inspections and regular patrols often struggle to meet the requirements for efficient and real-time assessment of the health status of trees. Therefore, there is an urgent need for an intelligent and automated health monitoring system to accurately assess and predict the health status of trees.

[0003] In existing tree health monitoring technologies, the finite element model is widely used in the structural stress analysis of trees. It can simulate the stress distribution and deformation of trees under different load conditions, providing data support for health assessment. However, the finite element model is computationally complex and difficult to dynamically respond to real-time changes in the environment. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for assessing the health status of trees based on a surrogate model to overcome the defects of the above-mentioned existing technologies.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] The present invention provides a method for assessing the health status of trees based on a surrogate model, including the following steps:

[0007] Collect geometric shape data and material property data of different types of trees. The geometric shape data includes the dimensions, lengths, diameters, bifurcation angles, and bifurcation numbers of tree trunks and branches, and the material property data includes density, elastic modulus, and shear modulus;

[0008] Establish three-dimensional finite element models of different types of trees based on the geometric shape data and material property data of different types of trees, and perform finite element analysis according to the three-dimensional finite element models of the trees;

[0009] Construct an adaptive surrogate model of the tree through a neural network according to the geometric shape data and finite element analysis results of different types of trees, and train the adaptive surrogate model;

[0010] Collect real-time characteristic data of the trees to be evaluated using sensors and drones, and input the real-time characteristic data into the trained adaptive proxy model for analysis. The adaptive proxy model adopts a feedback optimization mechanism, and the actual data obtained from each evaluation is fed back to the adaptive proxy model for parameter update.

[0011] Based on the analysis results of the adaptive proxy model, evaluate the health status of the trees to be evaluated and generate countermeasures for wind prevention and reinforcement of the trees.

[0012] Furthermore, the process of establishing the three-dimensional finite element model of the trees includes:

[0013] Collect geometric shape data of the trees;

[0014] Obtain material property data of different parts of the trees, and the different parts of the trees include tree trunks, branches, and bifurcation points;

[0015] Establish a three-dimensional finite element model of the trees in the finite element analysis software ABAQUS according to the collected geometric shape data and material property data;

[0016] Perform mesh division on the finite element model, and apply fine meshes in the bifurcation structure, tree trunk nodes, and high-stress regions of the trees.

[0017] Furthermore, the finite element analysis according to the three-dimensional finite element model of the trees specifically includes:

[0018] Set fixed constraints in the root region of the three-dimensional finite element model of the trees to simulate the stability of the trees rooted in the soil;

[0019] Apply static loads and dynamic loads, where the static loads include the self-weight load of the trees, and the dynamic loads include wind loads;

[0020] Gradually apply the wind load, observe the stress response of the trees under different load levels, and capture the stress and strain changes of the trees under different load levels;

[0021] According to the stress and strain changes of the trees under various natural conditions, select a damage model based on the damage mechanics theory and calculate the distribution of damage factors of the trees.

[0022] Furthermore, the finite element analysis results include the stress, strain, and damage factor distributions under different load conditions.

[0023] Furthermore, according to the geometric shape data of different types of trees and the finite element analysis results, construct an adaptive proxy model of the trees through a neural network and train the adaptive proxy model, specifically including:

[0024] According to the geometric shape data of different types of trees and the corresponding finite element analysis results, feature extraction, fusion, and mapping are performed to construct an input feature dataset, where the input feature dataset includes input feature vectors of each type of tree.

[0025] Perform a health status score on different types of trees through expert experience and knowledge.

[0026] Input the input feature dataset into a neural network model. With the health status score as the target, through the multi-layer non-linear transformation of the neural network, learn the relationship between the tree health status and damage evolution, and perform model training to obtain a trained adaptive surrogate model.

[0027] Furthermore, the step of performing feature extraction, fusion, and mapping according to the geometric shape data of different types of trees and the corresponding finite element analysis results to construct an input feature dataset specifically includes:

[0028] Perform parametric processing on the geometric shape data of the tree to obtain the geometric features of each node of the tree.

[0029] Extract the stress, strain, and damage factor distributions of each part of the tree under different load conditions from the finite element analysis results, and the stress, strain, and damage factor distributions are presented in the form of a three-dimensional grid.

[0030] Map the stress, strain, and damage factor distributions of each part of the tree under different load conditions to the corresponding nodes of the tree.

[0031] Perform weighted fusion on the geometric features of the mapped nodes and the stress, strain, and damage factor distributions under different load conditions to obtain the input feature vector of the tree.

[0032] Perform the same processing on different types of trees to obtain the input feature vectors of each type of tree.

[0033] Obtain the input feature dataset through the input feature vectors of each type of tree.

[0034] Furthermore, the loss function of the adaptive surrogate model is:

[0035]

[0036] where L is the loss function of the adaptive surrogate model, N is the number of input feature vectors in the input feature dataset, y i is the true health status score obtained through expert experience and knowledge, is the predicted health status score output by the adaptive surrogate model.

[0037] Further, the real-time feature data of the trees to be evaluated collected by the sensors and the UAVs specifically includes:

[0038] Stress and strain data of the trees to be evaluated are collected in real time through sensors installed at different parts of the trees to be evaluated;

[0039] Surface images and multi-spectral images of the trees are obtained through the high-definition cameras of the UAVs, and the distribution of damage factors of the trees to be evaluated is obtained through the obtained surface images and multi-spectral images of the trees to be evaluated;

[0040] The UAV is used to inspect the trees to be evaluated, and the geometric features of the trees to be evaluated are collected, including the shape, size, and bifurcation angle of the trees;

[0041] The stress, strain data, damage factor distribution, and geometric features of the trees to be evaluated obtained are subjected to feature extraction, mapping, and fusion to obtain the real-time feature data of the trees to be evaluated.

[0042] Further, based on the analysis results of the adaptive agent model, the health status of the trees to be evaluated is evaluated, and countermeasures for wind prevention and reinforcement of the trees are generated, specifically including:

[0043] According to the health status score of the trees to be evaluated output by the adaptive agent model, the health status level of the trees to be evaluated is determined. The health status level includes three levels: slight hazard, moderate hazard, and severe hazard;

[0044] Corresponding countermeasures for wind prevention and reinforcement of the trees are generated according to the health status level.

[0045] Further, the generation of corresponding countermeasures for wind prevention and reinforcement of the trees according to the health status level specifically includes:

[0046] For trees with a slight hazard level, basic maintenance measures are taken, including pruning overcrowded or overweight branches, repairing slightly damaged branches, and filling and compacting the loosened root areas;

[0047] For trees with a moderate hazard level, reinforcement measures are taken, including installing support structures or flexible cables in the high-stress areas of the trees, repairing the trunks and branches with cracks or rot, filling the cracks and performing waterproof treatment, and reinforcing the pillars or ground anchors in the root areas. The high-stress areas of the trees include bifurcation nodes or trunk cracks;

[0048] For trees with a severe hazard level, comprehensive reinforcement and repair measures are taken, including installing support structures or reinforcement frames at key parts of the trees, repairing severe cracks and rotten parts of the trunks, strengthening root support and performing disease treatment. For trees with a continuously deteriorating health status and poor repair effects, removal treatment is carried out. The key parts of the trees include trunks, main branches, and bifurcation points.

[0049] Compared with the prior art, the present invention has the following advantages:

[0050] (1) Through the deep coupling of finite element stress analysis and an adaptive surrogate model (neural network), the present invention not only simulates the surface damage of trees, but also accurately quantifies the stress distribution, damage factors, and deformation conditions, deeply reflecting the structural changes of trees under natural conditions and external forces. This coupling method ensures the accurate characterization of the internal stress state of trees, surpassing the limitations of traditional visual inspections and making the evaluation results scientifically based and highly reliable.

[0051] (2) Through the data linkage of sensors and drones, the present invention breaks the limitations of static inspections and realizes real-time monitoring and dynamic early warning. Compared with traditional regular inspections, the present invention can identify risks at the early stage of damage changes and quickly give feedback, avoiding delayed diagnosis. This efficient dynamic response ability is particularly important before emergencies such as wind disasters and can greatly reduce environmental risks.

[0052] (3) Based on the real-time health assessment results, the present invention automatically generates tree reinforcement plans. Different from simple manual judgments in the past, the present invention fully considers the current stress distribution, health status, and environmental impacts of trees and scientifically determines specific measures such as pruning and bracing. This targeted plan significantly improves the effectiveness of preventive measures and the scientific nature of decision-making, enhancing the reliability of disaster protection.

[0053] (4) The adaptive surrogate model of the present invention adopts a feedback optimization mechanism, feeding the actual data obtained from each assessment back to the surrogate model to enhance the self-learning ability of the model, making the present invention more and more accurate in long-term use. This data-driven optimization mechanism can automatically adapt to the tree health assessment needs under different regions and climate conditions, ensuring its high efficiency in diverse environments.

[0054] (5) By integrating multi-dimensional data such as stress and strain data collected by sensors, surface images obtained by drones, and multi-spectral images, the present invention performs feature extraction and fusion. This comprehensive data fusion method not only improves the comprehensiveness of tree health assessment, but also can comprehensively capture different damage types and development trends of trees under the action of various environmental factors, making the assessment results more accurate and diverse and avoiding the limitations of a single data source.

[0055] (6) By automatically generating windproof reinforcement countermeasures for trees based on the health assessment results, the present invention optimizes the reinforcement decision-making process. Different from traditional manual assessment methods, the present invention, based on the tree health assessment and combined with the structural analysis results of trees, intelligently generates specific reinforcement measures, avoiding the subjectivity of human judgment, improving the scientific nature and accuracy of reinforcement countermeasures, and ensuring the reinforcement effect and the long-term stability of trees. Description of the Drawings

[0056] Figure 1 It is a flowchart of the method according to an embodiment of the present invention;

[0057] Figure 2 It is a flowchart of the stress analysis of the finite element model according to an embodiment of the present invention;

[0058] Figure 3 It is a diagram of the integration of a single surrogate model and a category according to an embodiment of the present invention. Detailed Embodiments

[0059] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, rather than all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0060] Embodiment 1:

[0061] Figure 1 Exemplarily, a method for evaluating the health status of trees based on a surrogate model according to an embodiment of the present invention is shown, and the method includes:

[0062] Collect geometric shape data and material property data of different types of trees. The geometric shape data includes the dimensions, lengths, diameters, bifurcation angles, and bifurcation numbers of tree trunks and branches. The material property data includes density, elastic modulus, and shear modulus;

[0063] Based on the geometric shape data and material property data of different types of trees, establish three-dimensional finite element models of different types of trees, and perform finite element analysis according to the three-dimensional finite element models of the trees;

[0064] According to the geometric shape data and finite element analysis results of different types of trees, construct an adaptability surrogate model of the trees through a neural network, and train the adaptability surrogate model;

[0065] Use sensors and drones to collect real-time feature data of the trees to be evaluated, and input the real-time feature data into the trained adaptability surrogate model for analysis. The adaptability surrogate model adopts a feedback optimization mechanism, and the actual data obtained each time is fed back to the adaptability surrogate model for parameter update.

[0066] Based on the analysis results of the adaptability surrogate model, evaluate the health status of the trees to be evaluated and generate tree windproof reinforcement countermeasures.

[0067] For the three-dimensional finite element model of the trees, the establishment process includes:

[0068] Collect geometric shape data of trees;

[0069] Obtain material property data of different parts of the tree, where the different parts of the tree include the trunk, branches, and bifurcation points;

[0070] Establish a three-dimensional finite element model of the tree in the finite element analysis software ABAQUS according to the collected geometric shape data and material property data;

[0071] Perform mesh division on the finite element model, and apply fine meshes in the bifurcation structure, trunk nodes, and high-stress regions of the tree.

[0072] The finite element analysis based on the three-dimensional finite element model of the tree specifically includes:

[0073] Set fixed constraints in the root region of the three-dimensional finite element model of the tree to simulate the stability of the tree rooted in the soil;

[0074] Apply static loads and dynamic loads, where the static loads include the self-weight load of the tree, and the dynamic loads include wind loads;

[0075] Gradually apply the wind load, observe the stress response of the tree under different load levels, and capture the stress and strain changes of the tree under different load levels;

[0076] According to the stress and strain changes of the tree under various natural conditions, select a damage model based on the damage mechanics theory, and calculate the distribution of the damage factor of the tree.

[0077] The finite element analysis results include the stress, strain, and damage factor distributions under different load conditions.

[0078] The construction and training of the adaptability proxy model of the tree through a neural network according to the geometric shape data of different types of trees and the finite element analysis results specifically include:

[0079] According to the geometric shape data of different types of trees and the corresponding finite element analysis results, perform feature extraction, fusion, and mapping to construct an input feature data set, where the input feature data set includes the input feature vectors of each type of tree;

[0080] Score the health status of different types of trees through expert experience knowledge;

[0081] Input the input feature data set into the neural network model, with the health status score as the target, and through the multi-layer non-linear transformation of the neural network, learn the relationship between the tree health status and damage evolution, and perform model training to obtain the trained adaptability proxy model.

[0082] According to the geometric shape data of different types of trees and the corresponding finite element analysis results, feature extraction, fusion, and mapping are performed to construct an input feature dataset, which specifically includes:

[0083] Perform parametric processing on the geometric shape data of the trees to obtain the geometric features of each node of the trees;

[0084] Extract the stress, strain, and damage factor distributions of each part of the trees under different load conditions from the finite element analysis results, and the stress, strain, and damage factor distributions are presented in the form of a three-dimensional grid;

[0085] Map the stress, strain, and damage factor distributions of each part of the trees under different load conditions to the corresponding nodes of the trees;

[0086] Perform weighted fusion on the geometric features of the mapped nodes and the stress, strain, and damage factor distributions under different load conditions to obtain the input feature vectors of the trees;

[0087] Perform the same processing on different types of trees to obtain the input feature vectors of each type of tree;

[0088] Obtain the input feature dataset through the input feature vectors of each type of tree.

[0089] The loss function of the adaptive surrogate model is:

[0090]

[0091] where L is the loss function of the adaptive surrogate model, N is the number of input feature vectors in the input feature dataset, and y i is the true health status score obtained through expert experience knowledge, is the predicted health status score output by the adaptive surrogate model.

[0092] The use of sensors and drones to collect real-time feature data of the trees to be evaluated specifically includes:

[0093] Through sensors installed in different parts of the trees to be evaluated, collect the stress and strain data of the trees to be evaluated in real time, specifically including:

[0094] Stress sensors: Installed at key parts of the trees (such as the trunk, branches, and roots, etc.), used to monitor in real time the stress conditions generated by the trees under the action of external environmental factors (such as wind speed, temperature, humidity, etc.) and their own loads (such as tree weight, wind force, snow accumulation, etc.). By detecting the bending or compression of the trees, it can reflect the bearing capacity and stress distribution of the trees under different environmental conditions, and timely identify the risks of possible structural damage or excessive load.

[0095] Strain sensors: Installed at different parts of the tree to measure the minute deformation of the tree under external actions. Through strain data, the local deformation conditions of the tree can be reflected, such as the stretching or compression of the tree trunk, the bending of branches, etc. These data can be used to analyze the health status of the tree, identify possible pest infestations, damages, cracks or rotting areas, and then predict the stability of the tree under different environmental conditions.

[0096] Obtain the surface images and multi-spectral images of the tree through the high-definition camera of the drone. Obtain the distribution of damage factors of the tree to be evaluated through the obtained surface images and multi-spectral images of the tree to be evaluated, specifically including:

[0097] Obtain the surface images and multi-spectral images of the tree through the high-definition camera of the drone. Obtain the distribution of damage factors of the tree to be evaluated through the obtained surface images and multi-spectral images of the tree to be evaluated, specifically including the detection of cracks and damages on the tree surface. The high-resolution images taken by the high-definition camera can effectively capture the minute cracks, scars, pest infestations or other obvious damage areas on the tree surface. Through image processing techniques, such as image enhancement and edge detection, the system can accurately identify the cracks appearing on the bark surface, and by calculating the length, depth and distribution of the cracks, reflect the damage situation on the tree surface. This can provide detailed information about the damage on the tree surface and help further evaluate the health status of the tree.

[0098] In addition, through multi-spectral image analysis, different spectral information of the tree surface and the surrounding environment can be obtained. Especially the reflection spectra in the visible and infrared bands can effectively evaluate the health status of the tree. For example, infrared imaging can help identify problems such as water loss, withering or pest infestations in the tree. The ultraviolet and near-infrared spectral images can help reveal the possible hidden disease signs on the tree surface. This information can help accurately evaluate the nutritional status of the tree and the degree of pest infestations on the tree surface, providing reference data for tree management.

[0099] The rotting and aging areas of the tree can also be analyzed through multi-spectral images. The rotting or aging areas of the tree usually have different spectra from the healthy areas. Image analysis techniques can identify the changes in these areas, and then help locate the rotting or aging problems of the tree. This can provide a scientific basis for subsequent repair or replacement, thus improving the accuracy and efficiency of tree health management.

[0100] By analyzing the foliage density and growth condition of trees, the health condition of trees can be further understood. Multispectral imagery can provide the health condition of tree leaves and the distribution of foliage. Especially in the visible and near-infrared bands, this data can reveal the state of trees at different growth stages. The distribution and density of tree foliage directly affect its adaptability to the external environment. Especially in an environment with strong winds, monitoring the health condition of foliage helps to evaluate the wind resistance of trees.

[0101] Finally, through high-definition cameras and infrared imaging technology, it is also possible to monitor in real time whether there is water accumulation on the tree surface, especially in areas prone to water accumulation such as the bottom of the tree trunk and the joints of branches. Water accumulation may cause corrosion on the tree surface and the growth of mold, which in turn accelerates the aging or decay of the tree. Through the collection and analysis of these image data, the system can timely detect these potential risks and provide comprehensive support for the health management of trees.

[0102] Through the above technical means, the real-time monitoring of high-definition cameras and multispectral imagery can comprehensively capture the surface damage of trees, accurately identify problems such as cracks, decay, pests, and water accumulation, and thus provide accurate data support for tree health assessment. Combining stress sensor and environmental monitoring data, a health assessment report of the tree can be comprehensively generated, providing a scientific basis for subsequent reinforcement, repair, and management measures.

[0103] Using drones to conduct inspections on the trees to be evaluated, and collect the geometric features of the trees to be evaluated, including the shape, size, and bifurcation angle of the trees, specifically including:

[0104] Obtain the three-dimensional spatial information of the trees through the lidar (LiDAR) system and high-definition imaging equipment carried by the drones. The lidar can accurately measure the height, crown width, and diameter of the tree trunk by emitting laser beams and receiving reflected signals. Through the lidar scan data, the three-dimensional model of the tree can be accurately obtained, including the specific geometric shapes of the tree trunk, main branches, and secondary branches, providing accurate data support for the morphological characteristics of the tree.

[0105] The morphological characteristics of trees mainly include the diameter of the tree trunk, the width of the tree crown, and the distribution of branches. The drone obtains images of the tree from different angles through a high-definition camera. Combining the lidar data, it can accurately determine the thickness of the tree trunk and the distribution of foliage. Through the analysis of multi-angle images, the system can generate a two-dimensional projection of the tree and calculate the three-dimensional structure of the tree, including the diameter of the tree trunk, the radius of the tree crown, and the length of the branches.

[0106] The dimensional characteristics of trees can be accurately measured through lidar or computer vision technology. The lidar system obtains the geometric dimensions of different parts of the tree through high-precision distance measurement, especially the basal diameter of the trunk and the widest part of the crown. Using image processing algorithms, such as stereovision technology, the dimensional characteristics of the tree can be further accurately extracted, including the depth of the trunk, the distribution and extension angles of the branches.

[0107] The measurement of the bifurcation angle relies on high-definition camera equipment and computer vision technology. The system can automatically identify the starting position of the branch and the bifurcation angle by collecting images of the tree trunk and the branch junction and combining image recognition algorithms. Specifically, the system analyzes the surface image of the tree to determine the angle between the main trunk and the branch, and further estimates the growth trend of the tree. Through accurate angle measurement, the system can quantify the bifurcation angle of the tree, which is of great significance for evaluating the stability and wind resistance of the tree. Especially in strong wind weather, changes in the bifurcation angle may directly affect the overall structure of the tree.

[0108] Extract, map, and fuse the stress, strain data, damage factor distribution, and geometric features of the tree to be evaluated to obtain the real-time feature data of the tree to be evaluated.

[0109] Based on the analysis results of the adaptive agent model, evaluate the health status of the tree to be evaluated and generate countermeasures for windproof reinforcement of the tree, specifically including:

[0110] According to the health status score of the tree to be evaluated output by the adaptive agent model, classify the health status level of the tree to be evaluated. The health status level includes three levels: slight hazard, moderate hazard, and severe hazard;

[0111] Generate corresponding countermeasures for windproof reinforcement of the tree according to the health status level.

[0112] The generation of corresponding countermeasures for windproof reinforcement of the tree according to the health status level specifically includes:

[0113] For trees with a slight hazard level, take basic maintenance measures, including pruning overcrowded or overweight branches, repairing slightly damaged branches, filling and compacting the loose root area;

[0114] For trees with a moderate hazard level, take reinforcement measures, including installing support structures or flexible cables in the high-stress areas of the tree, repairing the trunk and branches with cracks or decay, filling the cracks and performing waterproof treatment, and strengthening the pillars or ground anchors in the root area. The high-stress areas of the tree include bifurcation nodes or trunk cracks;

[0115] For trees with a severe hazard level, comprehensive reinforcement and repair measures are taken, including installing support structures or reinforcement frames at key parts of the trees, repairing severe cracks and rotten parts of the tree trunks, strengthening root support and conducting disease treatment. For trees with continuously deteriorating health conditions and poor repair effects, removal treatment is carried out. The key parts of the trees include the tree trunks, main branches, and bifurcation points.

[0116] Example 2:

[0117] The parts not mentioned in this example are the same as those in Example 1.

[0118] This example presents a tree health status assessment system based on a surrogate model. By combining a neural network surrogate model with finite element analysis, it realizes the intelligent prediction and damage assessment of the health status of different tree species under various external stresses. The construction and training of the surrogate model are based on a large number of tree images, mesh models, and point cloud data, simulating the damage distribution, damage factors, and deformation conditions of trees under stress, and achieving efficient and low-power health status monitoring.

[0119] The tree health assessment system of this example contains surrogate models of various single trees. The models learn the damage characteristics and stress responses of trees through neural networks, especially refining the characteristics of different types of trees such as pine trees and trees with few leaves, so that the system can accurately capture the disease areas and structural characteristics of trees during the assessment process. In addition, the system can deploy sensors and drones to collect data such as the stress and deformation of trees in real time, and combine the prediction ability of the surrogate model to provide guarantee for the dynamic monitoring of the tree health status. Compared with traditional methods, this system not only improves the real-time performance and accuracy of monitoring, but also can realize comprehensive health risk identification, providing a scientific basis for preventing tree lodging under extreme weather such as wind disasters.

[0120] Finally, based on the analysis results of the assessment system, this example can generate specific suggestions for wind disaster prevention measures, including reasonable pruning and support plans, etc., to deal with different tree health conditions and reduce the risk of damage to trees by disastrous weather. The introduction of this system will effectively improve the scientific nature and convenience of tree health monitoring, and has important significance for ecological protection, public safety, and greening management.

[0121] This example provides a tree health status assessment system based on a surrogate model, which can achieve the following functions:

[0122] 1. For different types of trees (such as pine trees, trees with few leaves, etc.), based on the finite element model and the surrogate model, calculate the damage distribution, damage factors, and deformation conditions of the trees under stress, providing a scientific basis for tree health status assessment.

[0123] 2. Based on the characteristics of tree bifurcation structure, disease areas, etc., establish surrogate models suitable for different tree species and health states, simulate the damage forms of single trees through neural networks, and accurately predict the health changes under various environmental stresses.

[0124] 3. Through real-time data collection by sensors and drones, monitor the stress state and health condition of trees, combine the surrogate model to dynamically analyze the degree of tree damage, identify potential risks and give early warnings.

[0125] 4. According to the real-time monitoring results and health assessment data, provide specific windproof reinforcement suggestions for trees in different environments, including methods such as reasonable pruning and bracing, to help prevent the risk of trees falling and being damaged in bad weather.

[0126] 5. The health assessment report generated by the system includes damage distribution, health prediction and preventive suggestions, which can provide data support for ecological protection and urban greening management, and help decision-makers achieve efficient tree health management.

[0127] The process of the tree health assessment system based on the surrogate model includes four key steps: establishing a tree stress model through finite element analysis, constructing surrogate models of different types of tree damage, real-time monitoring and evaluating the health state, and formulating windproof reinforcement countermeasures according to the evaluation results. The specific explanations and processes are as follows:

[0128] (1) Stress analysis of the tree finite element model:

[0129] As Figure 2 shown, in order to accurately evaluate the stress distribution and damage of trees under different load conditions, it is first necessary to collect data on the structural and material properties of trees, including the geometric shapes, densities, elastic moduli, shear moduli, etc. of the tree trunks and branches. These data are used to construct a three-dimensional finite element model of the tree in ABAQUS, and fine mesh division is ensured to improve the calculation accuracy of the model, especially at key positions such as the tree trunk and bifurcation structure, to better simulate the stress concentration effect.

[0130] After the model is constructed, boundary conditions and external loads are applied. Fixed constraints are set in the root area of the tree to simulate the stability of the tree rooted in the soil. At the same time, static and dynamic loads are applied, such as the self-weight of the tree and wind loads. The wind loads can be set according to different wind speeds and applied in a step-by-step manner to facilitate observing the stress response of the tree under different load levels. This process captures the stress and strain changes of the tree under various natural conditions through multi-scenario and multi-time-step dynamic analysis to ensure that the simulation process is closer to reality.

[0131] Next, based on the damage mechanics theory, the distribution of the damage factor of the tree is calculated. A suitable damage model is selected and combined with the stress-strain data to generate a damage factor contour map to identify potential vulnerable areas of the tree, such as internal damage points that may be caused by insect infestation, diseases, etc. In addition, through the calculation of stress and damage distribution, the overall deformation of the tree can also be deduced, and a deformation distribution contour map is generated to visually display the deformation degree of the tree under extreme conditions and evaluate whether there is a risk of structural instability or fracture.

[0132] Finally, the results of these finite element analyses, such as the stress concentration positions at key nodes, strain distribution characteristics, and damage factor distribution, will serve as the input data basis for the subsequent training and prediction of the neural network model. The finite element results can provide a high-precision stress state of the tree under complex loading conditions, ensuring that the neural network model can maintain physical consistency and prediction accuracy in the surrogate analysis.

[0133] (2) Establish a surrogate model for tree damage:

[0134] Since there are significant differences in the structure and mechanical properties of various types of trees, the design of the model needs to fully consider the characteristics of tree types (such as pine trees, common trees, few-leaf trees, and flat trees, etc.). For example, the conical structure of pine trees is prone to stress concentration under wind load, while the crown shapes of common trees and few-leaf trees significantly affect their stress distribution characteristics. Therefore, when establishing the surrogate model, it is necessary to conduct detailed parameterization of the geometric and mechanical properties of the tree, including the bifurcation angle, bifurcation number, and branch position, etc., to better capture the stress transfer path and damage distribution law at the bifurcation nodes. This parameterization not only improves the simulation accuracy of the model for the structural response of the tree but also provides a basis for the integration of subsequent finite element analysis results.

[0135] In the surrogate model, the results of the finite element analysis are organically integrated into the original feature framework to form a more comprehensive input feature space. Key data such as the stress distribution, strain field, and damage factor obtained from the finite element calculation further enrich the input features of the model through feature extraction and mapping methods. For example, in the stress concentration area at the bifurcation node, the finite element results can provide a high-precision mechanical property description for the surrogate model, helping the model to focus on the stress transfer path and damage evolution of these key parts. For wind load and dynamic response, the finite element analysis provides the stress and deformation distribution that changes with time, and these time-series features are encoded as dynamic inputs for predicting the damage evolution of the tree at different time steps.

[0136] The refined modeling of the disease area also benefits from the support of the finite element analysis results. Through finite element analysis, the mechanical property changes in the areas of insect damage, breakage, and decay are quantified, and these change characteristics are directly introduced into the training data of the surrogate model, significantly enhancing the model's sensitivity to the characteristics of local stiffness reduction and damage accumulation. Through this process, the surrogate model is more physically consistent and accurate in predicting disease damage.

[0137] As Figure 3 shown, to cover the complex scenarios of different types of trees and various damage forms, the model adopts a multi-class integration method. The original feature fusion, hierarchical prediction, and weighted output framework form a highly collaborative integration mechanism with the incorporation of the finite element results. In the feature fusion stage, the geometric parameters, stress distribution, and dynamic responses in the finite element results, together with the original bifurcation structure features and disease area features, construct a unified integrated feature space. This fusion not only enhances the model's perception of the overall structure and local characteristics of the tree but also improves the physical meaning of the input features. In the hierarchical prediction stage, according to the tree structure complexity and disease distribution characteristics, the key area features revealed by the finite element results are gradually introduced to improve the accuracy of the model in predicting local damage. In the weighted output stage, the finite element results provide a basis for the applicability evaluation of different single surrogate models, ensuring that the prediction results of the integrated model are more robust and reliable.

[0138] By deeply integrating the finite element analysis results with the original features of the surrogate model, the method proposed in this paper not only inherits the physical authenticity of the finite element analysis but also achieves efficient prediction under complex structures and diverse damage scenarios through the neural network model. This collaborative mechanism enables the model to demonstrate higher applicability and accuracy in the assessment of tree health status, providing a unified and flexible solution for the damage prediction of various types of trees.

[0139] (3) Real-time monitoring and assessment of health status:

[0140] In this health status monitoring and assessment system, the perception system and the decision-making system work together to achieve real-time monitoring and risk assessment of the tree health status. The perception system continuously obtains physical data such as the stress, acceleration, temperature, and humidity of the tree, as well as the three-dimensional structure and multi-spectral images of the tree through sensors and drone inspections. The sensors are installed at key parts of the tree to capture the stress distribution and minute deformations of the tree under external forces such as wind loads and gravity. The drone takes pictures of the overall structure and multi-spectral images of the tree from the outside, and analyzes the surface diseases, breakages, and decays of the tree through high-precision images.

[0141] The decision-making system inputs this real-time data into the proxy model for analysis, identifies the highly damaged areas of the trees, predicts the potential damage risks through stress and deformation data, and provides data support for subsequent risk prediction. And a health status report is generated based on the output of the proxy model to help managers understand the current health status of the trees. The system will also generate risk warnings according to the changes in the health status, such as prompting potential structural damage risks when the wind is strong or the temperature changes suddenly. If the stress level of a certain part of the tree is detected to continue to increase or the deformation is abnormal, the system will automatically trigger the warning mechanism to prompt managers to take appropriate protective measures.

[0142] Finally, the health status assessment results will be updated regularly to achieve dynamic monitoring of the tree health status. Through the continuous input of data from sensors and drones, the proxy model can adjust the health assessment of the trees in real time under various climate and environmental conditions to ensure that managers always have the latest health status of the trees.

[0143] (4) Wind-proof reinforcement countermeasures for trees based on health assessment

[0144] Based on the analysis of the tree health assessment results, the wind-proof reinforcement measures are divided into three levels: minor hazards, moderate hazards, and severe hazards. Each level corresponds to different treatment methods to achieve the best protection effect. In addition, the effect monitoring part ensures the long-term effectiveness of the reinforcement measures and provides a basis for dynamic adjustment.

[0145] First, for minor hazards, such as excessive wind resistance in the tree crown, aging of local branches, or slight loosening of the roots, the protection measures focus on the basic maintenance of the trees. Specific operations include pruning overly dense or heavy branches to reduce wind resistance and maintain the balance of the trees; for aging or slightly diseased branches, pruning is carried out to avoid breaking under strong winds; for areas with slightly loosened roots, filling and compacting are carried out to enhance the supporting ability of the roots. The treatment methods for minor hazards are relatively simple, but they can effectively improve the wind resistance of the trees in daily maintenance.

[0146] For moderate hazards, such as stress concentration at the bifurcation nodes or branches, small cracks in the tree trunk, and insufficient root support, the protection measures begin to introduce support structures and repair means. Install support structures or flexible cables at the bifurcation nodes or high-stress parts to disperse the stress and reduce the risk of fracture in strong winds; fill and treat the cracks or slightly decayed areas on the tree trunk to repair the damaged parts and prevent further deterioration; for the situation of insufficient root support, add ground anchors or struts in the root area to enhance stability. These reinforcement measures for moderate hazards are highly targeted and can effectively prevent further structural damage.

[0147] For severe hazards, such as large - scale structural damage, severely loose roots or root diseases, and continuous deterioration of the health status, comprehensive reinforcement or removal measures need to be taken. For obvious cracks and decay on the trunk or main branches, filling and waterproofing treatments are adopted, and multiple layers of support structures are arranged around to maximize stability. If the roots are severely loose or there are root diseases, the root support is further strengthened and the diseases are treated. In the case of continuous deterioration of the health status and limited repair effect, it is recommended to remove the tree to ensure the safety of the surrounding environment. These measures provide the most comprehensive protection and treatment solutions for high - risk trees.

[0148] After implementing the reinforcement measures, the system will regularly check the reinforcement structure and the tree status to ensure the continuous effectiveness of the reinforcement measures. At the same time, the dynamic assessment and adjustment mechanism can update the reinforcement measures in a timely manner according to the actual monitoring situation to ensure the stability of the tree in a harsh environment. All reinforcement measures will be summarized and archived to form a health and maintenance file of the tree, providing a scientific basis for subsequent maintenance management and further optimization.

[0149] Compared with traditional methods, the tree health status assessment system based on the surrogate model proposed in this embodiment has significant advantages in the following aspects:

[0150] (1) Deep - coupled stress calculation and surrogate modeling to ensure assessment accuracy

[0151] Through the deep coupling of finite - element stress analysis and the surrogate model (neural network), this system not only simulates the surface damage of the tree, but also accurately quantifies the stress distribution, damage factors, and deformation conditions, deeply reflecting the structural changes of the tree under natural conditions and external forces. This coupling method ensures the accurate characterization of the internal stress state of the tree, exceeding the limitations of traditional visual inspection, making the assessment results scientifically based and highly reliable.

[0152] (2) High - adaptability surrogate models for various tree structural characteristics

[0153] The system designs high - adaptability surrogate models according to different tree types (such as pine trees, ordinary trees, trees with few leaves, etc.), enabling it to specifically consider key characteristics such as bifurcation structures and disease areas. This type - based modeling method, combined with the big - data training of neural networks, enables the model to learn and identify fine damage features from images, grids, point clouds, and stress data, providing strong support for health assessment and avoiding the one - size - fits - all assessment errors.

[0154] (3) Efficient real - time monitoring and dynamic risk warning

[0155] Through the data linkage of sensors and drones, the system breaks the limitations of static inspection, enabling real-time monitoring and dynamic early warning. Compared with traditional regular inspections, the system can identify risks at the early stage of damage changes and provide prompt feedback, avoiding delayed diagnosis. This efficient dynamic response ability is particularly important before emergencies such as wind disasters, which can greatly reduce environmental risks.

[0156] (4) Generation of scientific windproof reinforcement plans

[0157] Based on the real-time health assessment results, the system automatically generates tree reinforcement plans. Different from simple manual judgments in the past, this plan fully considers the current stress distribution, health status, and environmental impacts of the trees, and scientifically determines specific measures such as pruning and bracing. This targeted plan significantly improves the effectiveness of preventive measures and the scientific nature of decision-making, enhancing the reliability of disaster protection.

[0158] (5) Feedback-driven model dynamic optimization mechanism

[0159] This system adopts a feedback optimization mechanism, feeding the actual data obtained from each assessment back to the proxy model to enhance the model's self-learning ability, making the system more and more accurate during long-term use. This data-driven optimization mechanism enables the system to automatically adapt to the tree health assessment requirements in different regions and climate conditions, ensuring its high efficiency in diverse environments.

[0160] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, etc., which can store program codes.

[0161] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A tree health status assessment method based on an agent model, characterized in that: The following steps are involved: Collecting geometric data and material property data of different types of trees, wherein the geometric data includes the size, length, diameter, bifurcation angle, and number of bifurcations of the trunk and branches, and the material property data includes density, elastic modulus, and shear modulus; Establish three-dimensional finite element models of different types of trees based on their geometric shape data and material property data, and perform finite element analysis based on the three-dimensional finite element models of the trees; According to the geometric shape data of different types of trees and the results of finite element analysis, an adaptive proxy model of trees is constructed through a neural network, and the adaptive proxy model is trained; Use sensors and drones to collect real-time characteristic data of the trees to be evaluated, and input the real-time characteristic data into the trained adaptive agent model for analysis; Based on the analysis results of the adaptive agent model, the health status of the trees to be evaluated is evaluated and wind-proof reinforcement measures for the trees are generated.

2. A method for evaluating tree health status based on an agent model according to claim 1, characterized in that: The three-dimensional finite element model of the tree is established by: Collect data on the tree's geometry; Obtaining material property data of different parts of a tree, wherein the different parts of the tree include a trunk, branches, and bifurcations; Based on the collected geometric shape data and material property data, a three-dimensional finite element model of the tree is established in the finite element analysis software ABAQUS; The finite element model was meshed, with fine meshes applied in the tree's bifurcations, trunk nodes, and high stress areas.

3. The method for evaluating tree health status based on an agent model according to claim 1, characterized in that: The finite element analysis is performed based on the three-dimensional finite element model of the tree, specifically including: Fixed constraints are set in the root area of ​​the tree's three-dimensional finite element model to simulate the stability of the tree rooted in the soil; Applying static loads and dynamic loads, wherein the static loads include the deadweight loads of trees, and the dynamic loads include wind loads; Gradually load wind loads to observe the stress response of trees under different load levels and capture the stress and strain changes of trees under different load levels; According to the stress and strain changes of trees under various natural conditions and based on the damage mechanics theory, a damage model is selected to calculate the damage factor distribution of trees.

4. The method for evaluating tree health status based on an agent model according to claim 1, characterized in that: The finite element analysis results include stress, strain and damage factor distribution under different loading conditions.

5. The method for evaluating tree health status based on an agent model according to claim 1, characterized in that: The method of constructing an adaptive proxy model of trees by using a neural network based on the geometric shape data of different types of trees and the results of finite element analysis, and training the adaptive proxy model specifically includes: According to the geometric shape data of different types of trees and the corresponding finite element analysis results, feature extraction, fusion and mapping are performed to construct an input feature data set, wherein the input feature data set includes input feature vectors of each type of tree; Scoring the health status of different types of trees based on expert experience; The input feature data set is input into the neural network model. With the health status score as the goal, the relationship between the health status of trees and the evolution of damage is learned through the multi-layer nonlinear transformation of the neural network. The model is trained to obtain the trained adaptive proxy model.

6. A method for evaluating tree health status based on an agent model according to claim 5, characterized in that: The method of extracting, fusing and mapping features based on the geometric shape data of different types of trees and the corresponding finite element analysis results to construct an input feature data set specifically includes: Perform parameterization on the geometric shape data of trees to obtain the geometric features of each node of the tree; Extracting the stress, strain and damage factor distribution of each part of the tree under different load conditions from the finite element analysis results, wherein the stress, strain and damage factor distribution are presented in the form of a three-dimensional grid; The extracted stress, strain and damage factor distribution of each part of the tree under different load conditions are mapped to the corresponding nodes of the tree; The geometric features of each node after mapping are weightedly fused with the distribution of stress, strain and damage factor under different loading conditions to obtain the input feature vector of the tree; Perform the same process on different types of trees to obtain the input feature vectors of each type of tree; The input feature dataset is obtained through the input feature vectors of each type of trees.

7. The method for evaluating tree health status based on an agent model according to claim 5, characterized in that: The loss function of the adaptive proxy model is: Where L is the loss function of the adaptive proxy model, N is the number of input feature vectors in the input feature dataset, and y i Score the real health status obtained through expert experience knowledge, Score the predicted health status output by the adaptive surrogate model.

8. The method for evaluating tree health status based on an agent model according to claim 1, characterized in that: The use of sensors and drones to collect real-time characteristic data of trees to be evaluated specifically includes: By installing sensors at different parts of the trees to be evaluated, the stress and strain data of the trees to be evaluated are collected in real time; The surface image and multispectral image of the tree are obtained by the high-definition camera of the drone, and the damage factor distribution of the tree to be evaluated is obtained by the surface image and multispectral image of the tree to be evaluated; Use drones to inspect trees to be evaluated and collect geometric features of the trees, including their shape, size, and bifurcation angles; The acquired stress, strain data, damage factor distribution and geometric features of the trees to be evaluated are extracted, mapped and fused to obtain real-time feature data of the trees to be evaluated.

9. The method for evaluating tree health status based on an agent model according to claim 1, characterized in that: The analysis results based on the adaptive agent model are used to evaluate the health status of the trees to be evaluated and generate wind-proof reinforcement measures for the trees, including: According to the health status scores of the trees to be assessed output by the adaptive agent model, the health status levels of the trees to be assessed are graded, wherein the health status levels include slight damage, moderate damage and severe damage; Generate corresponding tree wind-proof reinforcement measures based on health status level.

10. A method for evaluating tree health status based on an agent model according to claim 9, characterized in that: The windproof reinforcement measures for trees generated according to the health status level specifically include: For trees with a minor damage rating, basic maintenance measures are taken, including pruning overcrowded or heavy branches, repairing slightly damaged branches, and filling and compacting loose root areas; For trees with a moderate hazard rating, reinforcement measures are taken, including installing support structures or flexible cables in high stress areas of the tree, repairing cracked or rotten trunks and branches, filling cracks and waterproofing them, and reinforcing pillars or anchors in the root area. High stress areas of the tree include bifurcation nodes or trunk cracks. For trees with severe damage levels, comprehensive reinforcement and repair measures are taken, including installing support structures or reinforcement frames at key parts of the trees, repairing serious cracks and rotten parts of the trunks, strengthening root support and treating diseases. For trees whose health status continues to deteriorate and the repair effect is poor, they are removed. The key parts of the trees include the trunk, main branches and bifurcation points.

Images