Construction method and system of forest pest distribution map based on twin simulation

Abstract

The application discloses a kind of based on twin simulation forest tree pest distribution map construction method and system, the present application relates to the technical field of twin simulation, according to the state difference vector of twin simulation information identification and determine forest, on the basis of multiple pest constraint conditions corresponding dynamic infection characteristic field is constructed, dynamic infection characteristic field is mapped to preset twin simulation space, improve the accuracy of twin simulation body. With current time as simulation starting point, corresponding environmental disturbance parameters are loaded in twin simulation body, and the spatial diffusion distribution information of forest tree pest on each topological node in future preset time period and infection gradient evolution surface are output;Mark the three-dimensional topological structure of twin simulation body, and corresponding rendering is carried out in combination with spatial diffusion distribution information and infection gradient evolution surface, to output corresponding rendering result, mapping rendering result to preset three-dimensional space, improve the accuracy of forest tree pest distribution map.

Description

Technical Field

[0001] This invention relates to the technical field of twin simulation, and in particular to a method and system for constructing a distribution map of forest tree diseases and pests based on twin simulation. Background Technology

[0002] Monitoring and controlling forest pests and diseases is a major challenge in the field of ecological protection. With the development of remote sensing technology and the Internet of Things, using digital means to monitor forest conditions has become a mainstream trend. Existing technologies typically use single satellite remote sensing monitoring or manual ground patrols to obtain data, although some technologies attempt to combine satellite and drone data.

[0003] Existing technologies mostly rely on simple geographic coordinate overlay or pixel-level fusion, failing to effectively address the issues of asynchronous temporal scales and spatial inconsistencies in heterogeneous data. This shallow fusion results in the inability to fully extract the complex physiological and ecological characteristics of forest trees, leading to missing or distorted data information input into the simulation space. Due to the lack of cross-modal spatiotemporal alignment of multimodal data and deep fusion based on ecological knowledge graphs in the twin simulation space, existing technologies struggle to output accurate twin simulation information, affecting the precision of the twin simulation and resulting in low accuracy of forest pest and disease distribution maps. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a method and system for constructing a distribution map of forest tree diseases and pests based on twin simulation.

[0005] This invention provides a method for constructing a forest tree disease and pest distribution map based on twin simulation, including: Acquire multimodal sensing data of the target forest area, perform cross-modal spatiotemporal alignment of the multimodal sensing data, and fuse it with twin simulation space, outputting corresponding twin simulation information during the fusion process; Based on the identification of twin simulation information, the state difference vector of forest trees is determined, and the dynamic evolution of the state difference vector is triggered. Thus, a corresponding dynamic infection feature field is constructed on the basis of multiple pest and disease constraints. The dynamic infection feature field is mapped to the preset twin simulation space to obtain the corresponding twin simulation body. Starting from the current moment, the corresponding environmental disturbance parameters are loaded into the twin simulation, and the dynamic infection feature field is forward extrapolated to output the spatial diffusion distribution information of forest pests and diseases on each topological node and the infection gradient evolution surface within a preset time period in the future. The three-dimensional topological structure of the twin simulation object is marked, and the corresponding rendering is performed by combining the spatial diffusion distribution information and the infection gradient evolution surface to output the corresponding rendering result. The rendering result is then mapped to the preset three-dimensional space to construct the corresponding forest pest and disease distribution map.

[0006] This invention provides a system for constructing a forest tree disease and pest distribution map based on twin simulation. This system is applied to the aforementioned method for constructing a forest tree disease and pest distribution map based on twin simulation. The system includes: The simulation information module is used to acquire multimodal sensing data of the target forest area, perform cross-modal spatiotemporal alignment of the multimodal sensing data, and fuse it with the twin simulation space, outputting the corresponding twin simulation information during the fusion process; The twin simulation module is used to determine the state difference vector of forest trees based on the identification of twin simulation information, and trigger the dynamic evolution of the state difference vector. In this way, a corresponding dynamic infection feature field is constructed based on multiple pest and disease constraints. The dynamic infection feature field is then mapped to a preset twin simulation space to obtain the corresponding twin simulation object. The simulation module is used to load the corresponding environmental disturbance parameters into the twin simulation object with the current time as the simulation starting point, and to perform forward simulation of the dynamic infection feature field, thereby outputting the spatial diffusion distribution information of forest pests and diseases on each topological node and the infection gradient evolution surface within a preset time period in the future. The forest pest and disease distribution map module is used to mark the three-dimensional topological structure of the twin simulation object, and combine the spatial diffusion distribution information with the infection gradient evolution surface for corresponding rendering to output the corresponding rendering results. The rendering results are then mapped to a preset three-dimensional space to construct the corresponding forest pest and disease distribution map.

[0007] Compared with the prior art, the beneficial effects of the present invention are: (1) Obtain multimodal perception data of the target forest area, perform cross-modal spatiotemporal alignment of the multimodal perception data, and fuse it with twin simulation space. Output the corresponding twin simulation information during the fusion process. Determine the state difference vector of the forest tree based on the identification of the twin simulation information, and trigger the dynamic evolution of the state difference vector. Construct the corresponding dynamic infection feature field based on multiple pest and disease constraints, map the dynamic infection feature field to the preset twin simulation space, and obtain the corresponding twin simulation body. The twin simulation information is introduced, and the dynamic infection feature field and twin simulation space are further controlled, which improves the accuracy of the twin simulation body.

[0008] (2) Taking the current moment as the simulation starting point, the corresponding environmental disturbance parameters are loaded into the twin simulation body, and the dynamic infection feature field is forward extrapolated to output the spatial diffusion distribution information and infection gradient evolution surface of forest pests and diseases on each topological node within the future preset time period; the three-dimensional topological structure of the twin simulation body is marked, and the spatial diffusion distribution information and infection gradient evolution surface are combined for corresponding rendering to output the corresponding rendering results. The rendering results are mapped to the preset three-dimensional space to construct the corresponding forest pest and disease distribution map, further control the spatial diffusion distribution information and infection gradient evolution surface, fully consider the rendering results and the preset three-dimensional space, and improve the accuracy of the forest pest and disease distribution map. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the method for constructing a forest pest and disease distribution map based on twin simulation in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structural composition of the system for constructing a forest pest and disease distribution map based on twin simulation in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0011] Please see Figures 1 to 2 A method for constructing a forest tree disease and pest distribution map based on twin simulation, applied to a twin simulation scenario; the method for constructing a forest tree disease and pest distribution map based on twin simulation includes: Step S11: Acquire multimodal sensing data of the target forest area, perform cross-modal spatiotemporal alignment on the multimodal sensing data, and fuse it with the twin simulation space. During the fusion process, output the corresponding twin simulation information. Step S12: Determine the state difference vector of the forest trees based on the identification of twin simulation information, and trigger the dynamic evolution of the state difference vector, thereby constructing the corresponding dynamic infection feature field based on multiple pest and disease constraints, and mapping the dynamic infection feature field to the preset twin simulation space to obtain the corresponding twin simulation body. Step S13: Taking the current moment as the simulation starting point, load the corresponding environmental disturbance parameters into the twin simulation and perform forward extrapolation on the dynamic infection feature field, thereby outputting the spatial diffusion distribution information of forest pests and diseases on each topological node and the infection gradient evolution surface within the future preset time period. Step S14: Mark the three-dimensional topology of the twin simulation object, and combine the spatial diffusion distribution information with the infection gradient evolution surface for corresponding rendering to output the corresponding rendering results. Map the rendering results to the preset three-dimensional space to construct the corresponding forest pest and disease distribution map.

[0012] In step S11, the specific steps are as follows: S111: Mark the target forest area, detect the target forest area, and acquire multimodal perception data collected collaboratively by satellite remote sensing, UAV hyperspectral imaging and ground IoT sensor array. Utilize a cross-modal alignment network based on spatiotemporal attention mechanism to extract the low-level semantic features of each submodal data in the multimodal perception data, and perform timestamp synchronization and three-dimensional spatial coordinate system registration to output the registered data combination. S112: The registered data combination is input into the twin simulation space, and multi-factor fusion is performed in the twin simulation space. During the fusion process, the forest ecological knowledge graph is introduced, thereby outputting twin simulation information in different dimensions. The twin simulation information includes forest canopy temperature anomaly index, forest abnormal morphology information, and canopy three-dimensional structure distortion parameters.

[0013] In the embodiments of this application, the target forest area is marked, the target forest area is detected, and multimodal perception data collected collaboratively by satellite remote sensing, UAV hyperspectral imaging and ground IoT sensor array is acquired. A cross-modal alignment network based on spatiotemporal attention mechanism is used to extract the low-level semantic features of each submodal data in the multimodal perception data, and timestamp synchronization and three-dimensional spatial coordinate system registration are performed to output the registered data combination, thus introducing data combination.

[0014] At this point, the predetermined target forest area is geographically marked and its boundaries are delineated to establish the geographical scope and coordinate benchmark for monitoring; satellite remote sensing, UAV hyperspectral imaging, and ground-based IoT sensor arrays are activated to work together to obtain raw multimodal sensing data covering different dimensions and resolutions.

[0015] The system retrieves real-time transit data from archived data or programmed commands from multispectral satellites covering the target forest area (such as Landsat 8 / 9 or Sentinel-2); extracts red, near-infrared, and short-wave infrared images, and records the satellite orbital parameters and metadata at the time of imaging to obtain macroscopic spectral reflectance information of the forest area. Simultaneously, it plans an autonomous drone patrol route covering key patrol grids within the target forest area; carries a hyperspectral imager with a band range of 400nm-1000nm, and performs imaging at a preset altitude, such as 100 meters; it also simultaneously records onboard GPS positioning data, inertial measurement unit (IMU) attitude data (pitch, roll, yaw), and the precise capture time of each frame.

[0016] Activate the wireless sensor network nodes deployed in the understory and canopy of the target forest area; collect "environmental parameters: air temperature, relative humidity, light intensity, soil moisture content" and "biological trapping data, such as phototactic capture counts of specific insect pests"; all sensor nodes generate time-stamped time-series data streams according to the set "sampling frequency: for example, once every 10 minutes".

[0017] A deep neural network based on a spatiotemporal attention mechanism is constructed. This network is configured to include a multi-branch feature extractor and a spatiotemporal attention weighting module. The network is designed to capture the temporal and spatial correlations of different data sources while preserving the unique semantic features of each submodal data, thereby achieving semantic alignment. The network receives satellite imagery, UAV hyperspectral imagery, and ground sensor readings as input. For image data, a convolutional neural network (CNN) branch is used to extract deep semantic features of texture, color, and spectral curves. For temporal sensor data, a long short-term memory network (LSTM) or a Transformer encoder branch is used to extract temporal dependency features of environmental changes.

[0018] A temporal attention module is introduced to calculate the temporal correlation weights of data from different modalities. For asynchronously acquired data, such as the contradiction between long satellite revisit periods and high-frequency sampling by ground sensors, this module learns the implicit temporal correlations between data, interpolates and aligns semantic features along the temporal dimension, and generates feature representations under a unified time axis. Simultaneously, a spatial attention module is introduced to establish a spatial mapping matrix based on high-precision POS data acquired by UAVs. This module learns the spatial adjacency relationships between ground point sensor data and image area pixels, back-projecting the semantic features of the sensor data into the image pixel space to achieve cross-modal spatial feature alignment. Using the alignment results from the constructed network, rigorous data preprocessing operations are performed, including timestamp synchronization and 3D spatial coordinate system registration, to eliminate spatiotemporal deviations in the physical world.

[0019] Establish a globally unified time reference, such as Coordinated Universal Time (UTC); for UAV and satellite imagery, lock the image frames to the global time axis based on the time information in the image metadata; for ground sensor data streams, use linear interpolation or spline interpolation algorithms to calculate the sensor values ​​at the corresponding image imaging time, thereby ensuring the consistency of data from all modalities in the time dimension.

[0020] Aerial triangulation and dense matching are performed on all UAV imagery to generate a digital surface model (DSM) and orthophoto map (DOM) of the target forest area. Next, using ground-based control points and markers with known precise coordinates, the satellite imagery and UAV DOM are geometrically corrected. The latitude and longitude coordinates of the ground sensor nodes in WGS84 format are converted to a Gauss-Kruger projection coordinate system or a local tangent plane coordinate system consistent with the imagery, and the sensor data attributes are mounted to the corresponding three-dimensional spatial points.

[0021] Multimodal feature data that has undergone time synchronization and spatial registration is combined and encapsulated to generate a standardized data combination output. This data combination includes pixel-level spectral data, point cloud environmental data, and temporal attribute data with a unified spatiotemporal reference, forming a digital holographic description of the target forest area.

[0022] Specifically, the target forest species is Masson pine, which is mainly threatened by pests and diseases including pine wilt disease and pine sawyer beetle damage. For the Masson pine forest area, recent Sentinel-2 satellite multispectral images of the region were acquired, with a focus on extracting the reflectance characteristics of Masson pine needles in the "red edge band: 700nm-750nm," because Masson pine infected with pine wilt disease exhibits a unique early anomaly in this band of reflectance.

[0023] Drones equipped with hyperspectral cameras were deployed to conduct low-altitude detailed surveys of suspected Masson pine trees in the target forest area, collecting nanometer-resolution spectral data of the Masson pine canopy. The aim was to capture subtle spectral "red shift" or "blue shift" phenomena caused by "needle discoloration: from green to yellow and then to red." At the same time, the drones recorded high-precision position and attitude information at the moment of shooting.

[0024] Simultaneously, data from the ground-based IoT sensor array in the target forest area was collected, particularly "forest temperature and humidity sensor data: pine wilt disease transmission is highly correlated with temperature" and the counting data from pine sawyer beetle traps. The aforementioned satellite imagery, UAV hyperspectral data, and ground sensor data were input into a cross-modal alignment network based on a spatiotemporal attention mechanism. The network extracted the texture and spectral features of the Masson pine canopy through a CNN branch, while simultaneously using an LSTM branch to analyze the temperature and humidity trends and temporal characteristics of the target forest area over the past two weeks.

[0025] In the temporal dimension, the module focuses on the peak temperatures and humidity before and after the drone's passage, and correlates and weights environmental factors with the spectral state of the trees at that time. In the spatial dimension, based on the drone's high-precision POS data, the module accurately maps sensor data from "specific points on the ground, such as next to a tree where longhorn beetles have been trapped" to the corresponding pixel area of ​​the Masson pine canopy in the drone image, thereby establishing a spatial correspondence between "environment-insect infestation-tree state".

[0026] Assuming the UAV imagery was captured at 10:30:00 Beijing time on August 15, 2023, the system automatically extracts forest temperature and humidity data ("at that moment: or the average of the previous and following 5 minutes") from the ground sensor database, ensuring that the analysis focuses on the environmental state of the Masson pine trees at the same time. Using existing topographic map data of the target forest area, the satellite imagery, UAV orthophotos, and ground sensor coordinates are uniformly converted to the CGCS2000 coordinate system. Using point cloud data generated by the UAV, the position of each Masson pine tree crown on the 2D image is mapped to a 3D geographic space (X, Y, Z coordinates), and the corresponding sensor environmental values ​​and hyperspectral reflectance values ​​are assigned to that 3D position. Step S111 outputs the registered data set for the Masson pine trees in the target forest area. This set not only includes the 3D spatial coordinates and detailed spectral curves of each Masson pine tree crown but also precisely correlates the meteorological environmental data of that crown at a specific time. This provides a precise, synchronous, and spatialized data foundation for subsequent analysis of differences in the health status of Masson pine trees in twin space.

[0027] Furthermore, the registered data combination is input into the twin simulation space, and multi-factor fusion is performed in the twin simulation space. During the fusion process, a forest ecological knowledge graph is introduced, thereby outputting twin simulation information in different dimensions. This twin simulation information includes forest canopy temperature anomaly index, forest abnormal morphology information, and canopy three-dimensional structural distortion parameters, fully taking into account twin simulation information.

[0028] At this point, a digital twin simulation space is established that is consistent with the geometric structure and topological relationship of the physical target forest area; the registered data output in step S111 is combined and mapped into the twin simulation space according to a unified three-dimensional spatial coordinate system (such as CGCS2000).

[0029] A 3D mesh voxel model, scaled proportionally to the target forest area, is constructed in the twin simulation space. Spectral reflectance values ​​from UAV hyperspectral imagery are used as texture attributes and mapped onto the corresponding canopy surface mesh of the 3D model. Temperature and humidity data collected by ground-based IoT sensors and vegetation indices retrieved from satellites are loaded as voxel attributes into the corresponding spatial nodes of the 3D model. At this point, the twin simulation space is no longer an empty shell, but a digital mirror containing real, multi-dimensional state data of "forest trees: Masson pine" in the physical world.

[0030] A pre-constructed forest ecology knowledge graph is deployed in a twin simulation space. This graph stores prior knowledge of forest growth mechanisms, pathological characteristics, and environmental interactions in a graph structure. Entity nodes in the graph are defined, including "forest tree entity" (e.g., Masson pine), "disease and pest entity" (e.g., pine wilt disease, pine sawyer beetle), "environmental factors" (e.g., temperature, humidity), "phenotypic characteristics" (e.g., needle color, canopy morphology, transpiration rate), etc. Semantic relationship edges between entities are also defined. For example, logical chains are established such as "pine wilt disease" – [leading to] – "thin-wall cell damage"; "thin-wall cell damage" – [leading to] – "water transport obstruction"; "water transport obstruction" – [manifesting as] – "increased canopy temperature"; "pine sawyer beetle" – [carrying] – "pine wilt nematode," etc.

[0031] The knowledge graph is transformed into a computer-reasonable rule engine to provide logical constraints and semantic guidance during data fusion, ensuring that the fusion result conforms to biological logic. Multi-factor fusion operations are performed, utilizing the forest ecosystem knowledge graph to semantically associate and reconstruct features from multimodal data loaded into a twin simulation space.

[0032] The input raw data, such as a decrease in the red edge position of the spectrum and an increase in ambient temperature, is fed into the graph inference engine. The engine infers based on the relationship paths in the graph. For example, when it detects a "blue shift in the red edge position of the spectrum" and "ambient temperature suitable for longhorn beetle reproduction, >20℃", it activates the subgraph network in the graph related to "early infection of pine wilt disease". Through graph neural networks (GNNs) or other fusion algorithms, the discrete environmental data, spectral data, and pathological nodes in the graph are fused at the feature level to remove non-pathological interference caused by changes in light intensity or soil drought, and extract pathological feature vectors with high confidence.

[0033] Based on the fused pathological feature vectors, twin simulation information of different dimensions is calculated and output. This information is not merely a data stack, but a quantitative indicator with clear pathological meaning. The physiological and structural deviations of trees under pathological conditions are calculated, outputting a vector set containing the following three core parameters: Tree canopy temperature anomaly index: reflects heat stress changes caused by reduced transpiration due to disease. Tree abnormality morphology information: reflects changes in appearance caused by pathogen invasion, such as needle discoloration and dieback. Canopy three-dimensional structure distortion parameters: reflect changes in canopy volume and structure caused by leaf drop or death.

[0034] Specifically, the "hyperspectral data of Masson pine in the target forest area (showing yellowing needles in some areas)," "ground sensor data (showing forest temperature of 28℃ and humidity of 60%)," and UAV visible light data obtained in step S11 are input into the twin simulation space of the target forest area. In this twin simulation space, each Masson pine tree corresponds to a virtual instance, and its real-time attributes are filled with the above data.

[0035] The system invokes the "Masson pine ecological knowledge graph," which contains specific rules: "Pine wilt nematodes inhibit resin secretion in Masson pine, leading to decreased transpiration and consequently, canopy temperatures higher than surrounding healthy trees and ambient temperatures"; and "Early symptoms of pine wilt disease manifest as needles turning yellowish-brown to reddish-brown, often accompanied by bite marks from the pine sawyer beetle." The fusion module discovered that a Masson pine instance in the twin space exhibited "abnormal spectral characteristics in the red-edge region" and "ambient temperature during the active period of the pine sawyer beetle." Based on the graph logic, the system automatically eliminated the interfering factor of "soil drought," as sensor data showed normal soil moisture, indicating that the anomaly was highly likely caused by "pine wilt disease."

[0036] The system combines thermal infrared data with environmental baseline temperature and uses the "transpiration-thermal anomaly" model in the spectrum to calculate the deviation of the canopy temperature of the Masson pine from the healthy baseline value. For example, if the canopy temperature is calculated to be 2.5℃ higher than that of a healthy Masson pine, the system outputs "Forest canopy temperature anomaly index = 0.85, high risk", indicating that its water transport system has been severely damaged.

[0037] The system analyzes the texture features of hyperspectral data and combines them with the semantic features of "disease discoloration" in the atlas to output "abnormal tree morphology information = needle chlorosis rate 40%, showing reddish-brown mottled appearance, and typical dieback phenomenon," which directly corresponds to the external manifestations of pine wilt disease in its middle and late stages. The system performs Boolean operations and volume difference between the current 3D point cloud model of Masson pine and the historical healthy model; it detects sparse crowns caused by needle shedding and calculates "canopy 3D structural distortion parameters = leaf area index (LAI) decreased by 25%, canopy porosity increased by 15%." Step S112 outputs detailed twin simulation information for the Masson pine in the target forest area in the twin simulation space. This information quantifies the tree's physiological "abnormalities," temperature abnormalities, appearance "discoloration," morphological abnormalities, and structural "shrinkage" and structural distortion.

[0038] In step S12, the specific steps are as follows: S121: The twin simulation information is hierarchically divided. During the division process, the corresponding state recognition is triggered by the recognition mechanism, thereby marking the current physiological state of the trees and the standard state baseline. The corresponding state difference vector is determined based on the comparison between the current physiological state of the trees and the standard state baseline. S122: Obtain multiple pest and disease constraints, including meteorological environmental parameters, tree species resistance index, and forest stand spatial pattern. The dynamic evolution of this state difference vector over a continuous time series is further combined with multiple pest and disease constraints to construct a corresponding dynamic infection feature field, which characterizes the spatiotemporal transmission potential of pests and diseases. The dynamic infection feature field is losslessly mapped to a preset twin simulation space, and a physically consistent forest twin simulation is constructed by combining multiple current images of the target forest area.

[0039] In the embodiments of this application, the twin simulation information is hierarchically divided. During the division process, the corresponding state recognition is triggered by the recognition mechanism, thereby marking the current physiological state of the trees and the standard state baseline. The corresponding state difference vector is determined based on the comparison between the current physiological state of the trees and the standard state baseline, thus introducing the state difference vector.

[0040] At this point, the multi-dimensional twin simulation information output in step S112 is hierarchically classified according to biological attributes and data characteristics. This process deconstructs the high-dimensional and heterogeneous simulation data into a subset of features that can be quantified, facilitating subsequent targeted state identification. Hierarchical definition: A three-level hierarchical system of "physiology-morphology-structure" is established. Physiological function level: mainly includes parameters reflecting internal metabolic state, such as forest canopy temperature anomaly index and water stress index. Apparent morphology level: mainly includes abnormal morphological information of forest trees, such as needle color characteristics and spectral reflectance curve characteristics. Physical structure level: mainly includes canopy three-dimensional structural distortion parameters, such as leaf area index (LAI), canopy closure, and branch biomass loss rate.

[0041] The system automatically reads various parameters from the twin simulation information and, according to predefined mapping rules, categorizes them into the three levels mentioned above. For example, it maps the "canopy temperature anomaly index" to the physiological function level and the "canopy three-dimensional structural distortion parameters" to the physical structure level. For each data subset after level division, a corresponding state recognition mechanism is deployed. This mechanism includes a pre-trained disease recognition model and an expert rule base, used to perform semantic analysis and pattern matching on the features within the level, thereby triggering specific state recognition logic to determine the specific health performance of forest trees in different dimensions.

[0042] At the physiological function level, a threshold-based recognizer is configured; at the morphological level, a deep learning-based spectral classifier is configured; and at the physical structure level, a geometric difference-based structural analyzer is configured. When data is loaded into the corresponding level, the corresponding recognition mechanism is automatically activated. The system compares the input feature vector with the pathological pattern library in the recognition mechanism. For example, at the morphological level, the recognition mechanism analyzes the slope change of the spectral curve in a "specific band, such as the red edge band." If the slope decreases by more than a preset threshold, the "needle discoloration" state recognition is triggered. Based on the output results of the recognition mechanisms at each level, the instances of forest trees in the twin simulation space are labeled with their states. This labeling is not only a description of a single feature but also a comprehensive aggregation of the recognition results at each level, forming a complete qualitative description of the current physiological state of the forest trees.

[0043] Weighted voting or fuzzy logic reasoning is used to integrate the recognition results from the physiological, morphological, and structural levels. A "Current Physiological State" label is generated in the attribute panel of the twin simulation entity. This label can be a discrete state category, such as "Healthy," "Sub-healthy," "Severely Infected," or "Dead," or it can be a vector containing continuous probability values. The identified "specific feature values, such as infection site and infection percentage," are attached to the state label as digital attributes of the forest tree instance.

[0044] Retrieve the standard state baseline of forest trees that is pre-stored in the twin system. This baseline represents the ideal forest tree parameters under normal growth conditions without pest or disease stress. Compare the current physiological state of the forest trees with the standard state baseline in each dimension, calculate the deviation value, and aggregate all the deviation values ​​into a high-dimensional vector, which is the state difference vector.

[0045] Based on information such as tree age and site conditions for "Tree: Masson Pine", the corresponding "Standard Growth Model for Healthy Masson Pine" is retrieved from the knowledge base to obtain its standard parameter range, standard spectral curve, standard crown temperature, and standard LAI for the current season and time period. The residuals between the current state parameters and the standard baseline parameters are calculated; for example, the difference between the current crown temperature and the standard crown temperature, and the ratio of the current LAI to the standard LAI are calculated. The calculated residuals of each dimension are arranged in a preset order to construct a state difference vector Vdiff=[ΔT,ΔS,ΔLAI,...]T. The magnitude of this vector represents the degree to which the tree deviates from a healthy state, and the direction of the vector represents the main pathological factors leading to the deviation.

[0046] Specifically, the system receives twin simulation information about a Masson pine tree in the target forest area from S112. The "abnormal forest canopy temperature index, too high" is classified to the physiological function level; the "needles exhibiting yellowish-brown spectral characteristics" is classified to the appearance morphology level; and the "canopy three-dimensional structure distortion parameters, leaf area index decreased" is classified to the physical structure level.

[0047] At the physiological function level, the recognition mechanism detected that the canopy temperature anomaly index exceeded the threshold of the standard summer transpiration cooling effect of healthy Masson pine, triggering the "water transport obstruction" recognition logic. At the appearance morphology level, the spectral classifier analyzed the hyperspectral data and found that the red light reflectance at 680nm was abnormally high, while the reflectance in the near-infrared band was low. This feature matched the pattern "pine wilt disease causes needle dehydration and discoloration" in the knowledge base, triggering the "needle discoloration, from yellowing to reddening" state recognition. At the physical structure level, the structure analyzer compared the current point cloud with the historical model and detected sparse branches and leaves on the outer edge of the canopy, triggering the "canopy structure defect" recognition.

[0048] The system integrates the identification results from the above three levels: obstructed water transport, reddening of needles, and canopy defects. Based on the pathological development of pine wilt disease, the system marks the current physiological state of the Masson pine instance in the twin simulation space as "mid-stage of pine wilt disease infection, or severe infection".

[0049] The system retrieves the standard baseline of healthy Masson pines of the same age in this forest area, such as: standard canopy temperature 32.5℃, standard red edge slope k, and standard leaf area index (LAI) = 3.5. It compares this baseline with the current state: Physiological dimension: current canopy temperature 35.0℃, difference ΔT = +2.5℃.

[0050] Morphological dimension: Red edge slope decreases by 30%, difference ΔS = -30%. Structural dimension: Current LAI = 2.8, difference ΔLAI = -0.7. Based on the above deviation values, a corresponding state difference vector Vdiff = [2.5, -0.3, -0.2]T is constructed, as shown in the normalized example. This vector quantifies the degree and direction of deviation of this Masson pine from its healthy state, not only confirming the existence of the disease, but also providing precise numerical input for triggering dynamic evolution and constructing the infection feature field in step S12, i.e., serving as the initial condition for disease evolution.

[0051] Furthermore, multiple pest and disease constraints are obtained, including meteorological environmental parameters, tree species resistance indices, and forest stand spatial patterns. The dynamic evolution of this state difference vector over a continuous time series is further combined with these multiple pest and disease constraints to construct a corresponding dynamic infection feature field, which characterizes the spatiotemporal transmission potential of pests and diseases. The dynamic infection feature field is losslessly mapped to a pre-defined twin simulation space, and a physically consistent forest twin simulation body is constructed by combining multiple current images of the target forest area. By introducing twin simulation information, the dynamic infection feature field and twin simulation space are further controlled, improving the accuracy of the twin simulation body.

[0052] At this point, key constraints affecting the occurrence and spread of pests and diseases are extracted and quantified from multidimensional data sources. These factors constitute the environmental boundary and rule parameters of the dynamic evolution model, which determine the evolution trajectory of the state difference vector in space and time.

[0053] Micrometeorological data of the target forest area were collected, including air temperature, relative humidity, precipitation, wind speed, and wind direction. Particular attention was paid to key thresholds affecting pathogen development and vector insect activity, such as the annual average temperature suitable for pine wilt disease and the meteorological conditions during the flight of the pine sawyer beetle. Based on a forest ecological knowledge graph, physiological resistance indicators of different Masson pine trees within the target area were retrieved. These indicators, determined by a combination of tree age, diameter at breast height (DBH), past disease history, and growth vigor, characterize the ability of individual trees to resist pathogen infection. The spatial topology of the forest area was analyzed, and stand density, canopy closure, spatial distribution pattern of trees (e.g., clustered or random distribution), and adjacency distance matrix between individual trees were calculated. These parameters define the physical channels and contact probabilities of disease transmission in space.

[0054] Based on the constraints, a spatiotemporal evolution model based on partial differential equations or cellular automata is constructed. The state difference vector determined in S121 is used as the initial input to drive it to perform dynamic deduction on a continuous time series. The propagation probability and infection intensity at each spatiotemporal node are calculated in combination with the constraints, thereby constructing a dynamic infection feature field.

[0055] A rate-of-change function for the state difference vector over time is defined, with independent variables including the current "infection intensity: vector magnitude" and meteorological environmental parameters; for example, the incremental coefficient of the disease increases exponentially over time when the temperature is within a specific range. Neighborhood rules are defined using the stand spatial pattern, and the resistance matrix for disease propagation is calculated by combining it with the tree species resistance index. The probability of disease spreading from highly infected nodes to less infected nodes is determined by wind direction parameters and adjacency distance; nodes with lower resistance are more easily infected. The temporal and spatial dimensions are discretized, and the infection potential energy of each grid at future moments is calculated. This potential energy field is the dynamic infection characteristic field, a four-dimensional tensor (three-dimensional space + time), whose numerical value characterizes the potential risk and severity of future pest and disease infection at that location.

[0056] The constructed dynamic infection feature field, in the form of a numerical field, is seamlessly injected into the pre-defined twin simulation space through coordinate transformation and interpolation algorithms. This process ensures that each individual tree in the virtual space acquires the corresponding time-varying pathological attributes. Then, using bilinear or trilinear interpolation algorithms, the low-resolution grid data of the dynamic infection feature field is mapped onto a high-precision 3D voxel model in the twin simulation space. The numerical values ​​representing infection intensity in the feature field are attached as dynamic attributes to the corresponding single-tree twin instances. This means that each tree in the twin space not only possesses a geometric appearance but also an "infection heat" attribute that evolves over time. Multiple current images of the target forest area, such as UAV orthophotos or oblique photogrammetry models, are used as texture maps or geometric constraints and deeply fused with the injected dynamic infection feature field to render and generate the final forest twin simulation.

[0057] Using current images, a realistic 3D mesh model of the forest area is generated through photogrammetry, ensuring that the geometric structure of the twin simulation maintains millimeter-level consistency with the real physical world. Real color images are used as the base texture. Based on the mapped dynamic infection feature field values, the color attributes of the texture are dynamically adjusted; for example, for areas showing high infection in the feature field, even if the original image shows green, the system will render corresponding pathological colors, such as yellowish-brown or reddish-brown, according to the pathological evolution patterns. This generates a "physically consistent" twin simulation that possesses both realistic topographical structure and accurately reflects the potential distribution of the disease.

[0058] Specifically, for the Masson pine forest area, the system obtained the current meteorological environmental parameters: the average daily temperature is 28℃, it is in the peak breeding period of pine wilt disease, and the prevailing wind direction is southeast. The tree species resistance index for this area was calculated: based on tree age data, the resistance index for old, over-mature Masson pines was set at 0.3, indicating low resistance, while the index for middle-aged forests was 0.7, indicating moderate resistance. The calculated planting density of Masson pines in the target forest area is 45 trees per acre, with high canopy closure and some areas containing contiguous pure forests, which is conducive to the migration of longhorn beetles.

[0059] The "state difference vector" calculated in S121 was used as the initial infection source. Due to the suitable temperature of 28℃ for the activity of the pine sawyer beetle, the state difference vector increased rapidly over time, indicating a worsening of the disease. The system deduced that the disease mainly spread from southeast to northwest along the southeast wind, and in densely planted areas, older, less resistant Masson pines were rapidly affected. The system generated a dynamic infection characteristic field covering the entire target forest area, with the highest infection potential in the southeast corner, showing a decreasing gradient towards the northwest but localized clustering trend.

[0060] The infection feature field is mapped onto the twin simulation space of the target forest area. At this point, each Masson pine tree crown in the virtual space is assigned an "infection risk value" between 0 and 1. A 3D real-world model is generated by combining the current visible light image of the target forest area taken by a drone. The system corrects the model texture based on the infection risk value. For Masson pines with a "feature field value greater than 0.8: high risk," even if the current photo has not yet fully turned red, the system renders the simulation body as a "typical yellow-red" pathological color according to the disease evolution law. For areas with a value below 0.1, a healthy green is maintained. The constructed forest twin simulation body not only matches the appearance of the real target forest area but also internally pre-sets the logic and visual representation of the disease's spread over time, intuitively demonstrating the potential outbreak path and severity of pine wilt disease in the forest.

[0061] In step S13, the specific steps are as follows: S131: Taking the current moment as the simulation starting point, historical multi-source meteorological time series data and ecological prior knowledge are extracted as condition vectors and input into a pre-trained spatiotemporal conditional generative adversarial network to generate environmental disturbance parameters within the future time window. These environmental disturbance parameters include extreme rainfall, sudden changes in wind direction and speed, and changes in the population density of natural enemies. They are then used as time-varying boundary conditions through the physics engine interface and dynamically loaded into the twin simulation. S132: In the twin simulation, driven by time-varying boundary conditions, a spatiotemporal graph neural network with a hybrid architecture including graph attention mechanism and temporal convolutional network is used to perform forward extrapolation of dynamic infection feature field. The edge weights of the adjacency matrix of graph structure are dynamically updated with environmental disturbance parameters to determine the pest and disease transmission information between each forest topology node based on canopy overlap distance and wind direction propagation resistance. Then, the spatial diffusion distribution information within a future preset time period is output. This spatial diffusion distribution information characterizes the evolution of infection status of each topology node over time. S133: Spatial diffusion distribution information determines corresponding spatial irregularities on the three-dimensional grid of the preset twin simulation space. The spatial irregularities are iterated at multiple levels. During the iteration process, multiple target indicators are combined for multi-level progressive processing to generate the corresponding infection gradient evolution surface. The infection gradient evolution surface represents the infection severity as a continuous nonlinear change in the three-dimensional space.

[0062] In the embodiments of this application, the current moment is taken as the starting point of the simulation. Historical multi-source meteorological time series data and ecological prior knowledge are extracted as condition vectors and input into a pre-trained spatiotemporal conditional generative adversarial network to generate environmental disturbance parameters within the future time window. These environmental disturbance parameters include extreme rainfall, sudden changes in wind direction and speed, and changes in the population density of natural enemies. They are then used as time-varying boundary conditions through the physics engine interface and dynamically loaded into the twin simulation, thus introducing environmental disturbance parameters.

[0063] At this point, the current time t0 is established as the starting moment of the simulation. The system starts the data retrieval and processing module, extracts multi-dimensional time series data and prior rules from the historical database and knowledge base, and encodes them into high-dimensional feature vectors as input conditions for subsequent model generation.

[0064] The simulation starts precisely at time t0, and a snapshot of the current target forest area is read. A preset historical period, such as the past 30 days, is then traced back to extract high-frequency meteorological time-series data for that period, including hourly temperature, humidity, precipitation, dew point temperature, and historical wind direction and speed records. A forest ecology knowledge graph is invoked to extract ecological rules and bioclimatological parameters strongly correlated with pest and disease transmission; for example, the peak emergence period of the pine sawyer beetle, the effective accumulated temperature constant, and the reproductive rate threshold of the pathogen pine wood nematode under specific temperature and humidity conditions are extracted. The numerical time-series data and logical ecological rules are normalized and vectorized, concatenated to form a condition vector Cenv that contains historical trends and ecological constraints. This vector comprehensively characterizes the current environmental state and evolution trend.

[0065] The constructed conditional vector is input into a pre-trained spatiotemporal conditional generative adversarial network (GAN). This network utilizes an adversarial game mechanism between the generator and discriminator to generate a sequence of future environmental perturbation parameters with high randomness and physical plausibility in the spatiotemporal dimension. A Spacetime-GAN model based on the Transformer architecture is adopted, which has the ability to capture long-distance temporal dependencies and spatial local correlations. The conditional vector Cenv is input into the generator network, and a random noise vector is injected to simulate the uncertainty of nature. Based on the conditional constraints, the generator outputs a prediction of environmental evolution within a "preset time period, such as the next 7 days." The discriminator network is responsible for distinguishing the generated environmental sequence from real historical records. It adjusts the generator parameters through backpropagation to ensure that the generated data conforms to statistical distribution patterns and has high continuity. The network finally outputs a parameter sequence containing multiple environmental perturbation factors, specifically including extreme rainfall representing the scouring effect of rainstorms, sudden changes in wind direction and speed representing the migration paths of vector insects, and population density changes of "natural enemies, such as the flower beetle," representing the effectiveness of biological control.

[0066] Establish a communication interface between the generated data and the physical simulation engine, convert the generated environmental disturbance parameters into force field, flow field or property change parameters that the engine can recognize, and apply them dynamically to the twin simulation as time-varying boundary conditions, thereby driving the evolution of the virtual environment.

[0067] Deploy data conversion middleware to map the numerical sequences output by the GAN to API parameters of a physics engine, such as NVIDIA PhysX or a custom fluid dynamics engine; for example, convert predicted wind speed vectors into fluid dynamics wind speed field parameters, and rainfall into emission density parameters for a particle system. These parameters are set as "time-varying boundary conditions" for the simulated world, meaning that in each frame of the simulation, the environment's physical rules, such as gravity, drag, and buoyancy, shift in real time based on the input perturbation parameters. The system dynamically loads these boundary conditions onto the twin simulation of the target forest area according to the time step, transforming the environment of the virtual forest from a static background into a dynamic field with complex dynamic characteristics.

[0068] Specifically, the system uses mid-July as the starting point for simulation. Historical multi-source meteorological time-series data from the target forest area over the past 30 days are extracted: the data shows that temperatures have consistently been above 30℃, with high humidity, and weak rainfall was detected in the preceding period. Combining prior ecological knowledge, the system extracts the rule "The pine sawyer beetle is active in flight under conditions of high temperature and humidity and light evening winds, and heavy rain will inhibit its flight but may promote the spread of pathogens with water flow," constructing a condition vector that includes "high temperature and humidity trends" and "potential monsoon influences."

[0069] The conditional vector is input into a pre-trained spatiotemporal conditional generative adversarial network. Based on the probability distribution of historical typhoon tracks, the network predicts that a severe convective weather event may occur in the region within the next three days. The system outputs specific environmental disturbance parameters: Extreme Rainfall: Short-duration heavy rainfall, reaching 50 mm / h, is predicted in the next 48 hours. Sudden Changes in Wind Direction and Speed: The wind direction is predicted to shift from southeast to northwest in the next 24 hours, with instantaneous wind speeds increasing abruptly from 2 m / s to 8 m / s. Changes in Natural Enemy Population Density: Due to the rainfall, the activity density of parasitic natural enemies, such as the parasitic beetle, is predicted to decrease by 20% in the next 48 hours.

[0070] Through the physics engine interface, the system converts the above parameters into physical rules for the simulated world. In the 24th hour of the simulation, the system applies a strong northwest wind force to the twin simulator, which changes the flight trajectory of the "pine sawyer beetle: carrying pine wood nematodes" in the simulation, causing it to gather in the relatively healthy Masson pine area on the northwest side of the forest, simulating the natural spread of the disease.

[0071] In the 48th hour of the simulation, the system activated the rainstorm particle system; high-density raindrop particles struck the canopy of the Masson pine tree, and the physics engine calculated the collision feedback, causing the simulated longhorn beetle to stop flying and hide in the bark crevices. Simultaneously, the simulation demonstrated the physical process of rainwater washing nematodes carried by insect vectors on the tree trunk to the soil or the roots of nearby healthy trees. The twin simulation entered a realistic simulation state including dynamic environmental disturbances such as "sudden strong winds" and "rainstorm suppression," providing a highly realistic physical boundary for the next step of accurately predicting the spatiotemporal distribution of pests and diseases.

[0072] Furthermore, in the twin simulation, driven by time-varying boundary conditions, a spatiotemporal graph neural network with a hybrid architecture including graph attention mechanism and temporal convolutional network is used to perform forward extrapolation of the dynamic infection feature field. The edge weights of the adjacency matrix of the graph structure are dynamically updated with environmental perturbation parameters to determine the pest and disease transmission information between each forest topology node based on the canopy overlap distance and wind direction propagation resistance. Then, the spatial diffusion distribution information within a future preset time period is output. This spatial diffusion distribution information characterizes the evolution of the infection status of each topology node over time, fully considering the spatial diffusion distribution information.

[0073] At this point, a dedicated spatiotemporal inference engine is deployed within the twin simulation. This engine employs a hybrid architecture that includes graph attention mechanisms and temporal convolutional networks (TCNs) to simultaneously capture the topological propagation characteristics of pests and diseases in space and their cumulative evolutionary trends in time series.

[0074] Each Masson pine tree in the target forest area is represented as a "topological node" in the graph structure; the node feature vector contains the tree's "current infection status, such as infection probability and disease level" and physiological attributes.

[0075] Spatial dimension: A Graph Attention Network (GAT) layer is employed to aggregate information from neighboring nodes, automatically learning the influence weights of different neighboring nodes on the current node's disease propagation through attention coefficients. Simultaneously, temporal dimension: A Temporal Convolutional Network (TCN) layer is used, utilizing causal convolution and dilated convolution to process time-series data, capturing the long-term dependencies and nonlinear trends in infection status over time. Environmental perturbation parameters generated by S131 are used as external control signals, inputting into each layer of the network to dynamically adjust network parameters according to environmental changes, ensuring the inference process conforms to the actual constraints of the physical environment.

[0076] Traditional graph structures typically have fixed adjacency relationships. However, in real ecosystems, the spread efficiency of pests and diseases is greatly affected by meteorological conditions. Therefore, the system utilizes environmental perturbation parameters to reconstruct the adjacency matrix of the graph structure in real time, dynamically adjusting the weights of edges between nodes. Based on environmental parameters such as wind direction, wind speed, and rainfall at the current moment, the propagation resistance coefficient between different nodes is calculated. For example, propagation resistance is lower in the downwind direction and higher in the upwind direction; extreme rainfall may interrupt the propagation path. The calculated propagation resistance coefficient is mapped to the edge weights in the adjacency matrix; the weight value represents the probability or susceptibility of pests and diseases spreading from node i to node j. As the simulation progresses, whenever environmental perturbation parameters, such as changes in wind direction, are updated, the system immediately triggers the reconstruction of the adjacency matrix, ensuring that the propagation topology network always maintains physical consistency with the real-time environmental state.

[0077] In the updated dynamic graph structure, the system comprehensively considers the spatial structure of trees and meteorological driving forces to accurately calculate the disease and pest transmission information between each topological node and quantify the transmission flux. It calculates the three-dimensional Euclidean distance between nodes and, combined with the geometric model of the tree canopy, calculates the "canopy overlap distance." If two Masson pine canopies physically overlap or have very small gaps, their basic transmission probability is set to be higher. A wind direction transmission resistance factor is introduced; the cosine of the angle between the vector from the source node to the target node and the wind direction vector is calculated to correct the basic transmission probability. The "canopy overlap distance, representing the contact probability," and the "wind direction transmission resistance, representing the vector migration efficiency" are weighted and fused to generate the final "disease and pest transmission information." This information determines whether a pathogen or vector insect can successfully migrate from one node to another within the current time step.

[0078] Using a pre-constructed spatiotemporal graph neural network, with a dynamic infection feature field as the initial input and combined with dynamic propagation information, multiple iterations are performed on the time axis to ultimately output spatial diffusion distribution information within a preset future time period. The network starts with the infection feature field at time t0. Within each time step Δt, the GAT layer aggregates the propagation information of neighboring nodes, and the TCN layer updates the node's state vector. The node's infection state changes with the received cumulative propagation flux. When the flux exceeds the resistance threshold, the node's state changes from "healthy" to "susceptible," further evolving into different disease levels. The state of all topological nodes in each future time slice is recorded to generate spatial diffusion distribution information. This information is a spatiotemporal tensor that details the infection probability, infection level, and dynamic evolution of each forest topological node over time.

[0079] Specifically, in the twin simulation of the target forest area, each Masson pine tree is considered a node; node characteristics are initialized with the current pine wilt nematode load and health level. A GAT-TCN hybrid network is constructed; GAT is responsible for monitoring the cross-infection relationships between neighboring pine trees, while TCN is responsible for predicting the temporal trend of disease severity increasing with the number of days. The wind direction changes from southeast to northwest, with a wind speed of 8 m / s.

[0080] In the physics engine, the relationship between wind direction vector and tree position vector is calculated. The system found that the weight of the connection edge between the "sick Masson pine: source of infection" located in "upwind: southeast side" and the "downwind: northwest side" healthy Masson pine was significantly increased, for example, the weight was increased from 0.1 to 0.8, because strong winds will help the pine beetle migrate quickly downwind to the northwest; conversely, the weight of the edge in the upwind direction was significantly reduced.

[0081] The system calculates the canopy overlap distance between topological nodes and finds that tree No. 5 and tree No. 6 in area A are relatively close and their canopies intertwine, indicating a high probability of base contact. Since the wind is blowing from tree No. 5 to tree No. 6, the propagation resistance is extremely low. The system comprehensively determines that within the current time step, the probability of pine wilt nematode spreading from tree No. 5 to tree No. 6 via pine sawyer beetle is extremely high, generating a high-intensity propagation information flow.

[0082] A forward simulation was initiated to model the situation over the next 72 hours. In the 24th hour, driven by a northwesterly wind, severely infected trees on the southeastern edge of Area A rapidly caused a row of Masson pine nodes downwind to change from "healthy" to "latent." In the 48th hour, due to the cumulative effect of the temporal convolutional network, the incubation period of this row of Masson pines ended, and the node status was updated to "early stage of disease," characterized by "yellowing of the canopy." The final output was the spatial diffusion distribution information for the next 72 hours. This information clearly shows that the disease did not spread uniformly, but rather exhibited a significant "band-like diffusion along the northwesterly wind direction" spatiotemporal evolution characteristic, accurately predicting the specific location and time of the next outbreak.

[0083] Therefore, the spatial diffusion distribution information determines the corresponding spatial irregularity information on the three-dimensional grid of the preset twin simulation space. The spatial irregularity information is iterated at multiple levels. During the iteration process, multiple target indicators are combined for multi-level progressive processing, thereby generating the corresponding infection gradient evolution surface. The infection gradient evolution surface represents the infection severity of the infection severity changing nonlinearly in three-dimensional space. The infection gradient evolution surface is introduced.

[0084] At this point, the spatial diffusion distribution information based on "discrete topological nodes: single tree level" output in step S132 is mapped to a high-resolution three-dimensional mesh in the preset twin simulation space. Due to the non-uniformity of the forest distribution, this mapping process will form an initial data field with sparsity and local aggregation characteristics in the mesh, i.e., spatial irregular information.

[0085] The twin simulation space of the target forest area is divided into a three-dimensional voxel grid of uniform size, with each voxel representing a tiny unit in the space, such as 1m×1m×1m. The predicted infection status of each Masson pine node at future time, such as infection probability and pathogen load, is read; based on the three-dimensional geometric coordinates of the tree, these node attributes are projected onto the corresponding voxel grid.

[0086] Because Masson pine trees are scattered or clustered in the forest, a large number of voxel grids do not include the trunk or crown, resulting in a high degree of spatial discontinuity in the data. These non-empty voxel sets containing infection values ​​and their sparse distribution in space constitute "spatial irregular information," which accurately reflects the physical fact that the disease exists only in the tree body and its microenvironment.

[0087] For the given spatial irregularities, a multi-level iterative algorithm framework is constructed. In each iteration, multiple target indicators are introduced as constraints to perform smoothing, interpolation, and diffusion calculations on the spatial data, and to perform multi-level progressive processing, thereby gradually inferring the continuous distribution of the entire field from the discrete node states.

[0088] The core metrics guiding the iterative process are defined, including: Spatial continuity: The infection gradient between neighboring voxels should not undergo abrupt changes, reflecting the natural spread trend of pathogens in air or soil. Directional consistency: In conjunction with wind direction, the rate of change of infection potential energy in the downwind direction should be less than that in the upwind direction. The degree of infection should decrease non-linearly with increasing distance from the canopy center.

[0089] Within each voxel neighborhood containing trees, local interpolation is performed using radial basis functions (RBF) to fill vacant voxels within the canopy, initially forming a single-tree-scale infection field. A Gaussian smoothing kernel or anisotropic diffusion equation is applied, using the initial iteration results as initial values, to moderately diffuse the infection to surrounding voxel-free regions, simulating the activity range of the vector insect, the pine longhorn beetle, flying through the forest. Through a coarse-to-fine grid resolution conversion, the global trend is first quickly obtained on a low-resolution grid, and then details are corrected on a high-resolution grid, gradually optimizing the numerical distribution of the entire field.

[0090] After completing multi-level iteration and progressive processing, the calculated three-dimensional voxel data field is transformed into a visual surface. This surface uses numerical elevation to characterize the severity of infection and can dynamically change with time parameters to form an infection gradient evolution surface.

[0091] The Moving Cube algorithm is used to extract isosurfaces from the 3D mesh that represent "infection severity, such as infection probability," with a specific threshold, such as 0.5, i.e., the severe infection limit. Voxel values ​​are mapped to the geometric properties of the surface; for example, infection severity is mapped to the shade of surface color, with green representing healthy and red representing severe infection, and infection rate is mapped to the normal vector direction or undulation of the surface. Because nonlinear constraints such as wind direction and resistance are incorporated during the iteration process, the final generated surface accurately characterizes the continuous and nonlinear changes in infection severity in 3D space; that is, the surface is steep and has a large gradient at wind convergence points or dense canopy areas, and gentle and has a small gradient in open areas.

[0092] Specifically, in the twin simulation space of the target forest area, the system loads the projection results for the next 72 hours onto a three-dimensional grid. The infection values ​​are mainly concentrated on discrete voxel coordinates, which correspond to the location of the Masson pine canopy; for example, the voxel at coordinates (x1, y1, z1) belongs to Masson pine with the number M05, which has an extremely high infection load; while a large number of air voxels around it have a value of 0. This set of discrete, high-value data points constitutes the spatial irregularity information for pine wilt disease, reflecting the current state of the disease, which exists only inside or on the surface of the tree.

[0093] The system combines vegetation density decay index and starts from the center of the crown of M05 Masson pine tree, interpolating and iterating voxels outward. Due to the infection of branches by pine wilt disease, the iterative algorithm sets the voxel infection value of the upper part of the crown to be higher than that of the lower part, and initially generates a three-dimensional infection field of a diseased Masson pine with "red head and green tail".

[0094] The system, incorporating directional consistency indicators, determined the prevailing wind direction to be northwest and initiated diffusion iterations. For air voxels located "downwind: southeast" of pine tree M05, although there were no trees in the area, the system simulated the migratory habits of the pine sawyer beetle, assigning these voxels a lower "potential infection energy" value. The system adjusted the grid resolution, iterating from a 5m coarse grid to a 0.5m fine grid, precisely depicting the gradient changes in disease penetration from the bark to the xylem.

[0095] The system extracts isosurfaces representing the severity of infection based on the iterated voxel data.

[0096] Surface characteristics: The generated surface is not a simple plane, but a complex fluid-like surface that wraps around the tree canopy and extends downwind.

[0097] Nonlinear characterization: In the center of the crown of M05 Masson pine, the curved surface is dark red and densely wrinkled, indicating an extremely steep infection gradient and severe disease; while at the edge of the crown and in the downwind open area, the curved surface gradually transitions to light green and is smooth, indicating a gentle infection gradient and potential risk of spread.

[0098] The final output infection gradient evolution surface visually demonstrates the vertical distribution of pine wilt disease on each individual Masson pine tree in the target forest area, as well as its horizontal radiation pattern throughout the entire forest stand.

[0099] In step S14, the specific steps are as follows: S141: Based on the identification of the twin simulation object, the corresponding three-dimensional topology is determined. Semantic segmentation and node labeling are performed on the three-dimensional topology of the twin simulation object. Combining spatial diffusion distribution information and infection gradient evolution surface, ray tracing and color mapping are performed using a volumetric rendering mechanism to output the corresponding four-dimensional rendering result, which includes time dimension evolution features. The four-dimensional rendering result is then mapped to a preset three-dimensional space with fused geographic information system coordinates. S142: Virtual camera nodes are deployed in three-dimensional space, and a perspective immersive roaming engine is further introduced. At the same time, a corresponding index structure is built for the time-series data in the four-dimensional rendering results to respond to the time node query command input by the user, and the corresponding pest and disease distribution status is rendered in combination with the perspective immersive roaming engine. S143: Extract infection distribution data from the infection gradient evolution surface and combine it with the distribution status of pests and diseases for multi-factor fusion. In the fusion process, divide the boundaries of high-risk areas, medium-risk areas and low-risk areas, and further generate a semi-transparent superimposed risk thermal early warning layer, thereby constructing a forest pest and disease distribution map that can be dynamically demonstrated.

[0100] In the embodiments of this application, the corresponding three-dimensional topology is determined based on the identification of the twin simulation object. The three-dimensional topology of the twin simulation object is semantically segmented and node-marked. Combined with spatial diffusion distribution information and infection gradient evolution surface, ray tracing and color mapping are performed using a volumetric rendering mechanism to output the corresponding four-dimensional rendering result, which includes time dimension evolution features. The four-dimensional rendering result is then mapped to a preset three-dimensional space with fused geographic information system coordinates, thus introducing it into the three-dimensional space.

[0101] At this point, the three-dimensional mesh model constructed inside the twin simulation is analyzed to identify and determine its topological structure; using semantic segmentation algorithms in computer vision, the model is decomposed into components, and each independent unit is assigned a semantic label with ecological significance.

[0102] The system iterates through the 3D scene map of the twin simulation, extracting all geometric entities constituting the forest area and their connections, clarifying the hierarchical structure of geometric elements such as ground surface, tree trunk, branches, and leaf clusters. A semantic segmentation network based on point clouds or voxels, such as PointNet++ or 3DU-Net, is used to classify the 3D model on a voxel-by-voxel or face-by-face basis. The system divides the pixel units in the model into semantic categories such as "background soil," "healthy Masson pine main body," "sick Masson pine main body," and "forest gaps." Based on semantic segmentation, each segmented Masson pine tree instance is uniquely bound to the topological node in step S132; simultaneously, the canopy region is finely marked, distinguishing between "canopy interior," "canopy edge," and "canopy gaps," providing a geometric mask for subsequent differentiated rendering.

[0103] The marked 3D topology is used as the geometry of the rendering scene. The spatial diffusion distribution information and infection gradient evolution surface generated in step S133 are loaded as the texture field. A volumetric rendering mechanism is used to emit rays in 3D space and perform integral calculations to achieve the mapping from the data field to the visual image. The "numerical attribute: infection rate 0.0-1.0" of the infection gradient evolution surface is used as the volume density or color attribute and sampled into the voxel mesh of the 3D topology. The spatial diffusion distribution information is used as a dynamic source to continuously update the mesh attributes. The ray tracing engine is started, and rays passing through the 3D voxels are emitted from the virtual camera's perspective into the scene. On the path of each ray, the infection value within the voxel is sampled and integrated according to the step size.

[0104] The sampled infection values ​​are mapped to RGBA color values ​​and opacity; for example, an infection value of 0 corresponds to completely transparent green (healthy), 0.5 corresponds to semi-transparent yellow (warning), and 1.0 corresponds to opaque dark red (severe and dead). Simultaneously, the lighting model is adjusted using infection gradients to give the diseased areas a withered, rough, diffuse reflection characteristic. A time variable is introduced into the rendering pipeline to synthesize rendering frames over a continuous time series; the system repeatedly performs ray tracing and color mapping processes according to a preset time step, such as one frame per hour, generating a series of continuous dynamic image sequences, i.e., the four-dimensional rendering result.

[0105] The temporal parameter t of the infection gradient evolution surface generated in S133 is associated with the frame index of the rendering engine; as t increases, the infection value field in the voxel mesh is updated. The color change in each frame of the scene is calculated; for example, simulating the process of disease spreading downwards from the top of the tree canopy, the rendering result shows that the red "infected area" expands frame by frame along the time axis, and the color saturation increases as the disease worsens. The generated image sequence and its corresponding timestamp metadata are packaged and output as a four-dimensional rendering result, 3D space + 1D time. This result not only includes the three-dimensional spatial distribution at each moment, but also embeds the complete evolutionary features of disease occurrence, development, outbreak to decline.

[0106] A coordinate transformation bridge is established between the virtual rendering space and the physical geographic space. The generated 4D rendering results are projected from the virtual simulation coordinate system to the real 3D space of the Geographic Information System (GIS), completing the final mapping from the digital twin to the physical world. A preset GIS coordinate system, such as the CGCS2000 geodetic coordinate system, is read to calculate the affine transformation matrix between the origin of the virtual simulation space and the geographic origin of the real target forest area. Each frame of the 4D rendering result is used as a dynamic texture layer and overlaid onto the GIS 3D digital earth or 3D terrain model. The transformation matrix is ​​used to correct the position, ensuring that the rendered red diseased areas accurately cover the real geographic coordinates of the Masson pine forest. A timeline slider control is built in the GIS platform, allowing users to view the dynamic spread video of pine wilt disease in the target forest area over a future time period on a real map by dragging the slider.

[0107] Specifically, the system loads a twin simulation of the target forest area and identifies 3,500 individual pine tree models growing on the terrain. The system automatically marks the canopy area of ​​each pine tree as an "Object of Interest (ROI)," and marks the shrubs and bare ground beneath the trees as the background. The system binds the ID of each tree (e.g., Tree_M_001 to Tree_M_3500) to the 3D mesh; specifically for high-risk trees predicted in previous simulations, such as Tree_M_0150 located in the northwest downwind direction, the system labels its canopy voxels with a specific "high-risk latent" tag.

[0108] The system loads infection gradient evolution surface data, indicating a high infection gradient at the top of the tree canopy of Tree_M_0150. Light rays are emitted from a virtual bird's-eye view through the canopy of Tree_M_0150; the light samples high infection values ​​at the top of the canopy and low infection values ​​at the bottom. Based on the transfer function, the top of the canopy is rendered as a warning red, indicating severe infection; the middle section is rendered as yellowish-brown, indicating moderate infection and needle discoloration; and the bottom remains dark green, indicating that the tree is not yet dead. Simultaneously, the system renders a light red "infection halo" around the canopy, corresponding to the potential migration path of the pine sawyer beetle in the forest, visually demonstrating the risk of transmission.

[0109] The system generates a 4D rendering of the next 72 hours, as a dynamic video stream. In the first hour's footage, Tree_M_0150 only has a few red spots on the treetop; as the timeline is dragged to the 48th hour, the red spots rapidly spread downwards along the branches, covering half of the canopy, and sporadic yellow spots also begin to appear on the surrounding green trees. This clearly demonstrates the lethal process of pine wilt disease in Masson pine from top to bottom and its spread in the forest.

[0110] The system uses high-precision RTK coordinate data to seamlessly project the rendering results onto the GIS digital elevation model (DEM) of the target forest area. On the GIS screen of the forest fire prevention command center, staff no longer see abstract curves, but a three-dimensional Masson pine forest that is "changing color" and superimposed on a real satellite map.

[0111] When staff click on the red area (Tree_M_0150) on the map, the system immediately pops up a window showing the geographical coordinates of the area, the predicted infection level, and the suggested treatment time, thus achieving the final closed loop from simulation to actual command.

[0112] Furthermore, virtual camera nodes are deployed in three-dimensional space, and an immersive viewpoint roaming engine is introduced. At the same time, a corresponding index structure is constructed for the time-series data in the four-dimensional rendering results to respond to the user's input time node query command. Combined with the rendering of the corresponding pest and disease distribution status by the immersive viewpoint roaming engine, the pest and disease distribution status is fully considered.

[0113] At this point, in the three-dimensional spatial scene that has been mapped to the geographic information system, based on the terrain features, forest stand distribution, and key monitoring areas of the forest area, a series of virtual camera nodes are deployed through a combination of algorithm calculation and manual assistance. These nodes constitute a multi-dimensional perspective acquisition network.

[0114] The system analyzes the digital elevation model (DEM) and canopy height model (CHM) of the target forest area, automatically identifying high-altitude locations with unobstructed views as global overview nodes. Simultaneously, it selects forest paths, valleys, and the areas surrounding key individual trees as near-ground micro-nodes. Optical parameters are configured for each virtual camera node, including field of view (FOV), focal length, depth of field, and orientation vector. For micro-nodes, a large aperture is configured to highlight details of diseased or damaged trees in the foreground. Smooth 3D interpolation paths, such as B-spline curves, are calculated between key nodes to form a pre-defined automatic navigation track, ensuring the continuity and smoothness of perspective switching during the roaming process.

[0115] Based on the deployment of virtual camera nodes, an immersive roaming engine is integrated and initialized. This engine is responsible for handling scene lighting calculations, occlusion culling, and real-time changes in the user's perspective, constructing a virtual observation environment with a sense of spatial depth through the graphics rendering pipeline. The engine assembles 3D terrain, individual pine tree models, time-varying textures from 4D rendering results, and environmental elements such as skyboxes and fog effects into a unified scene graph data structure. The engine employs a physically based rendering (PBR) workflow, calculating the reflection and scattering characteristics of light on pine needles and bark, and combining this with real-time shadow mapping to enhance the realism of the scene. A collision detection module is added to ensure that users do not pass through tree trunks or fall to the ground during roaming. Simultaneously, it supports free switching between first-person (FPV) and third-person (TPV) perspectives, simulating the visual experience of a forest ranger patrolling the forest or a drone hovering low overhead.

[0116] To achieve low-latency retrieval of "massive four-dimensional rendering results: three-dimensional space + time dimension", the system constructs a dedicated index structure for time-series data. This structure uses the time dimension as the primary key or key index item to establish a fast mapping relationship between timestamps and the state of the three-dimensional scene.

[0117] The timeline of the 4D rendering results is analyzed to extract key time nodes, such as the onset, peak, and decline points of disease outbreaks, as well as regular, equally spaced time frames. A spatiotemporal index data structure, such as a time-extended variant of 3DR-Tree, is used to organize the scene state data blocks (including infection levels, color textures, and morphological distortion parameters for each tree) under each time slice. A time index table is created and loaded into memory. This table records the data storage address and compressed file pointer corresponding to each timestamp, allowing the system to locate a specific snapshot of the state without loading the entire time series data.

[0118] The system receives time-node query commands input by the user through the interactive interface, quickly retrieves and loads the corresponding data using the constructed index structure, drives the immersive roaming engine to render the distribution of pests and diseases at that specific moment, and presents it in the virtual camera's view. Simultaneously, it listens for user drag events on the timeline slider or input of specific dates and times, such as "2023-10-01 14:00", converting them into standard timestamp query keys.

[0119] The index structure is used to locate the data block corresponding to the timestamp within milliseconds. If the data is compressed, a streaming decompression algorithm is initiated to load the infection gradient data and texture data into video memory. Based on the retrieved data, the roaming engine updates the attributes of all Masson pine entities in the scene in real time, including updating the crown color and texture to simulate the degree of discoloration, updating the geometric shape to simulate the state of dead branches, and updating environmental effects such as cloud cover and lighting. The virtual camera node captures the updated scene image, which, after post-processing such as color correction and anti-aliasing, is output to the display device, presenting an immersive real-world view of the pest and disease distribution at the time of the user's query.

[0120] Specifically, the system deployed a global overview node at the southeastern ridge of the target forest area's GIS 3D map. This elevated position provides a commanding view of the wind direction across the entire pure pine forest. Simultaneously, a micro-survey node was deployed at a ground level of 1.5 meters above the ground at the predicted "high-risk outbreak center: near Tree_M_0150," simulating the perspective of a forest ranger observing from below the trees. The system pre-programmed a "drone patrol path" connecting these two nodes.

[0121] The immersive walkthrough engine is activated, loading real terrain data and vegetation models of the target forest area. The engine simulates the light and shadow effects of afternoon sunlight penetrating the coniferous forest, the Tyndall effect, and enables physically based material rendering, making the rough texture of the pine bark and the translucency of the pine needles realistically discernible in the virtual environment.

[0122] Based on the projected data for the next 72 hours, the system established a time index structure. The index records data pointers for key frames such as: T+24h (initial spread), T+48h (outbreak phase), and T+72h (severe infection). The system packages and stores all "Massonia infection status data at T+48h: Tree_M_0150 and surrounding trees are severely infected, with reddish-brown crowns" and registers corresponding quick search keys in the index table. Users can drag the timeline on the system interface, position the pointer to the "next 48 hours" time node, and click "Start Roaming." The system captures the query command, instantly locks the T+48h data block through the index structure, and loads the disease distribution parameters at that time into memory.

[0123] The immersive engine updates the scene visuals; users see that the once green forest has undergone significant changes in the virtual landscape, with large patches of reddish-brown appearing along the northwest wind direction. The virtual camera displays the macroscopic spread of the disease to the southeast from a global overhead perspective; the scene automatically and smoothly transitions to a microscopic inspection node, under Tree_M_0150; users feel as if they are in the forest, clearly seeing the large areas of withered and drooping pine needles overhead, and can perceive the hot and humid atmosphere of the forest through changes in light and shadow, intuitively "experiencing" the severe damage of pine wilt disease in the next 48 hours.

[0124] Therefore, infection distribution data is extracted from the infection gradient evolution surface and multi-factor fusion is performed in combination with the distribution status of pests and diseases. In the fusion process, the boundaries of high-risk area, medium-risk area and low-risk area are divided, and a semi-transparent superimposed risk thermal early warning layer is generated. This constructs a forest pest and disease distribution map that can be dynamically demonstrated, further controlling the spatial diffusion distribution information and infection gradient evolution surface. It fully considers the rendering results and the preset three-dimensional space, and improves the accuracy of the forest pest and disease distribution map.

[0125] At this point, the system performs data analysis and feature extraction on the infection gradient evolution surface generated in step S133. This process transforms continuous three-dimensional field data into a structured dataset that can be used for spatial analysis, focusing on capturing key indicators that characterize the intensity of the disease.

[0126] On the three-dimensional grid of the infection gradient evolution surface, resampling is performed according to a predetermined "spatial resolution, such as a 2m×2m grid" to extract the infection severity value of each grid cell. This value is usually a normalized scalar, ranging from 0.0 to 1.0, where 1.0 represents that the trees are dead or severely infected, and 0.0 represents healthy.

[0127] The infection gradient vector of each grid point in three-dimensional space is calculated using the finite difference method. The magnitude of the gradient is extracted as the "infection change rate" to identify the front of disease spread. The extracted infection intensity values, gradient magnitudes, and their corresponding three-dimensional geographic coordinates (X, Y, Z) are integrated to generate an infection distribution dataset containing spatial attributes and attribute values.

[0128] The extracted infection distribution data is fused with the "distribution status of pests and diseases: i.e. the visual status of forest trees such as appearance, color and texture" determined in step S142. This process aims to correct the bias of relying solely on numerical extrapolation and improve the robustness of risk assessment.

[0129] Data on the distribution of pests and diseases is used to compare the inferred high-infection areas with the diseased characteristic areas "actually observed or simulated". If the two are highly overlapping in space, the confidence weight of the area is increased; if there is a deviation, an ecological adaptability factor is introduced for correction.

[0130] A fusion evaluation function R(x,y) = α·grad + β·Svisual + γ·Eenv is constructed; where Igrad is the infection gradient value, Svisual is the visual severity (e.g., reddening ratio), Eenv is the environmental suitability (e.g., temperature and humidity suitability), and α, β, and γ are preset weighting coefficients. The above fusion calculation is performed on each spatial unit within the target forest area, outputting a comprehensive risk score matrix that fully reflects the actual disease damage status at the current moment. Based on the comprehensive risk score matrix, spatial clustering and edge detection algorithms are used to discretize risk area boundaries with clear grade attributes in a continuous numerical field. A threshold system for risk judgment is established; for example, a comprehensive risk score > 0.75 is defined as a high-risk area, a score between 0.4 and 0.75 as a medium-risk area, and a score < 0.4 as a low-risk area.

[0131] Contour tracking or watershed algorithms from image processing are applied to identify the boundaries of different threshold regions in three-dimensional space. For three-dimensional volume data, the MarchingCubes algorithm is used to extract isosurfaces as boundary surfaces. The extracted boundary surfaces are projected onto a two-dimensional geographic plane, or the three-dimensional volume is preserved to generate closed vector polygons, which are labeled as "high-risk zone boundary", "medium-risk zone boundary", and "low-risk zone boundary", respectively.

[0132] For the defined risk zones, a semi-transparent heatmap layer with visual hierarchy is generated. This layer uses color coding and transparency control to intuitively display the risk distribution without obscuring the terrain information of the base map. Specific color gamuts are assigned to different risk levels: high-risk areas are mapped to red or dark red, medium-risk areas to orange or yellow, and low-risk areas to green or light blue. The transparency and alpha value of pixels are dynamically adjusted according to the risk intensity; high-risk areas are typically given higher opacity (e.g., 0.7) to indicate a warning, while low-risk areas are given lower opacity (e.g., 0.2) to avoid obscuring the background. Within the zone boundaries, a continuous color gradient effect is generated based on the numerical distribution of the comprehensive risk score matrix, forming a smoothly transitioning risk heatmap field.

[0133] The generated risk heat map layer is overlaid with the underlying "Geographic Information Base Map: Topography, Road Network, and Forest Stand Map," and a timeline control is introduced to construct a dynamic demonstration system. In the GIS rendering engine, the overlay is performed in the order of "Topographic Base Map > Vector Road Network > Forest Entity Model > Risk Heat Map Layer." The attribute data table of the heat map layer is associated with the timestamps of the four-dimensional rendering results in S141. Playback control logic is constructed so that, as the timeline progresses, the system updates the boundaries and heat map colors of risk areas in real time based on infection distribution data at different times. This allows users to see the dynamic evolution of disease from points to areas and the risk level from low to high, ultimately forming a complete distribution map of forest pests and diseases.

[0134] Specifically, the system analyzes the infection gradient evolution surface of the target forest area and extracts the three-dimensional infection values ​​of all Masson pine canopy locations. For example, at coordinates (XA, YA), the extracted infection gradient value is 0.85, indicating that the Masson pine at this location is in a dying state and the gradient of the disease spreading to the surrounding areas is extremely large.

[0135] The system combines the current "pest and disease distribution status: S142 shows that the trees in this area have turned red overall" and "environmental suitability: the current temperature is 28℃, which is extremely suitable for longhorn beetle activity." A fusion calculation is performed: Risk = 0.6 × 0.85 (numerical value) + 0.3 × 0.9 (visual reddening) + 0.1 × 0.95 (environmental suitability). The calculated comprehensive risk score for this area is as high as 0.88, confirming it as an extremely high-risk area.

[0136] The system runs a boundary tracking algorithm on the target forest area map. Centered on the first infected trees ("several Masson Pines: Tree_M_010 to Tree_M_015"), a closed polygon is drawn. All trees within this polygon have a risk score exceeding 0.75 and are marked as the "core epidemic area, high risk." The area surrounding the core epidemic area, extending southeastward, with scores between 0.4 and 0.75, is designated as the "buffer zone, medium risk." Further outwards, the forest area is designated as the "general monitoring area, low risk."

[0137] The system generates a semi-transparent layer; the core epidemic area is rendered as a deep red with high opacity, resembling a burning flame; the buffer zone is rendered as a semi-transparent orange; and the outer area is a light green. This layer is overlaid on the satellite map of the target forest area, allowing users to see both the undulating terrain and the red "infection cloud" moving "northwest: downwind".

[0138] The final interactive map of forest pest and disease distribution is generated. When the user clicks the play button, the map begins to evolve over time. The user can see that the red "high-risk core area" is initially small, but gradually expands over time as the simulated pine sawyer beetle migrates, while the orange "medium-risk area" continuously encroaches on the green area. This dynamic distribution map visually demonstrates the outbreak path of pine wilt disease over the next three days, helping managers accurately locate the "core area: high-risk area" requiring immediate felling and control, as well as the buffer area and medium-risk area requiring focused spraying for prevention.

[0139] Please see Figure 2 The system for constructing forest pest and disease distribution maps based on twin simulation is applied to the aforementioned method for constructing forest pest and disease distribution maps based on twin simulation; the system for constructing forest pest and disease distribution maps based on twin simulation includes: The simulation information module is used to acquire multimodal sensing data of the target forest area, perform cross-modal spatiotemporal alignment of the multimodal sensing data, and fuse it with the twin simulation space, outputting the corresponding twin simulation information during the fusion process; The twin simulation module is used to determine the state difference vector of forest trees based on the identification of twin simulation information, and trigger the dynamic evolution of the state difference vector. In this way, a corresponding dynamic infection feature field is constructed based on multiple pest and disease constraints. The dynamic infection feature field is then mapped to a preset twin simulation space to obtain the corresponding twin simulation object. The simulation module is used to load the corresponding environmental disturbance parameters into the twin simulation object with the current time as the simulation starting point, and to perform forward simulation of the dynamic infection feature field, thereby outputting the spatial diffusion distribution information of forest pests and diseases on each topological node and the infection gradient evolution surface within a preset time period in the future. The forest pest and disease distribution map module is used to mark the three-dimensional topological structure of the twin simulation object, and combine the spatial diffusion distribution information with the infection gradient evolution surface for corresponding rendering to output the corresponding rendering results. The rendering results are then mapped to a preset three-dimensional space to construct the corresponding forest pest and disease distribution map.

[0140] It should be noted that although multiple modules have been mentioned in the detailed description above, the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the protection scope of the present invention; therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

[0141] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for constructing a forest tree disease and pest distribution map based on twin simulation, characterized in that, include: Acquire multimodal sensing data of the target forest area, perform cross-modal spatiotemporal alignment of the multimodal sensing data, and fuse it with twin simulation space, outputting corresponding twin simulation information during the fusion process; Based on the identification of twin simulation information, the state difference vector of forest trees is determined, and the dynamic evolution of the state difference vector is triggered. Thus, a corresponding dynamic infection feature field is constructed on the basis of multiple pest and disease constraints. The dynamic infection feature field is mapped to the preset twin simulation space to obtain the corresponding twin simulation body. Starting from the current moment, the corresponding environmental disturbance parameters are loaded into the twin simulation, and the dynamic infection feature field is forward extrapolated to output the spatial diffusion distribution information of forest pests and diseases on each topological node and the infection gradient evolution surface within a preset time period in the future. The three-dimensional topological structure of the twin simulation object is marked, and the corresponding rendering is performed by combining the spatial diffusion distribution information and the infection gradient evolution surface to output the corresponding rendering result. The rendering result is then mapped to the preset three-dimensional space to construct the corresponding forest pest and disease distribution map.

2. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 1, characterized in that, The process involves acquiring multimodal sensing data of the target forest area, performing cross-modal spatiotemporal alignment on the multimodal sensing data, and fusing it with a twin simulation space. During the fusion process, corresponding twin simulation information is output, including: The target forest area is marked, the target forest area is detected, and multimodal perception data collected by satellite remote sensing, UAV hyperspectral imaging and ground IoT sensor array are acquired. Using a cross-modal alignment network based on spatiotemporal attention mechanism, the underlying semantic features of each submodal data in the multimodal perception data are extracted, and timestamp synchronization and three-dimensional spatial coordinate system registration are performed to output the registered data combination.

3. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 2, characterized in that, The process of acquiring multimodal sensing data of the target forest area, performing cross-modal spatiotemporal alignment of the multimodal sensing data, and fusing it with a twin simulation space, while outputting corresponding twin simulation information during the fusion process, also includes: The registered data combination is input into the twin simulation space, and multi-factor fusion is performed in the twin simulation space. During the fusion process, a forest ecological knowledge graph is introduced, thereby outputting twin simulation information in different dimensions, including forest canopy temperature anomaly index, forest abnormal morphology information, and canopy three-dimensional structural distortion parameters.

4. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 1, characterized in that, The process involves identifying the state difference vector of forest trees based on twin simulation information and triggering the dynamic evolution of the state difference vector. This constructs a corresponding dynamic infection feature field based on multiple pest and disease constraints. The dynamic infection feature field is then mapped to a preset twin simulation space to obtain the corresponding twin simulation entity, including: The twin simulation information is hierarchically divided. During the division process, the corresponding state recognition is triggered by the recognition mechanism, thereby marking the current physiological state of the trees and the standard state baseline. The corresponding state difference vector is determined based on the comparison between the current physiological state of the trees and the standard state baseline.

5. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 4, characterized in that, The process of determining the state difference vector of forest trees based on the identification of twin simulation information and triggering the dynamic evolution of the state difference vector to construct a corresponding dynamic infection feature field based on multiple pest and disease constraints, mapping the dynamic infection feature field to a preset twin simulation space to obtain the corresponding twin simulation entity, also includes: Multiple pest and disease constraints are obtained, including meteorological environmental parameters, tree species resistance index, and forest stand spatial pattern. The dynamic evolution of this state difference vector over a continuous time series is further combined with the multiple pest and disease constraints to construct a corresponding dynamic infection feature field, which characterizes the spatiotemporal transmission potential of pests and diseases. The dynamic infection feature field is then losslessly mapped to a preset twin simulation space, and a physically consistent forest twin simulation is constructed by combining multiple current images of the target forest area.

6. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 1, characterized in that, The process begins with the current moment as the simulation starting point, loads corresponding environmental disturbance parameters into the twin simulation, and performs forward extrapolation on the dynamic infection feature field. This outputs the spatial diffusion distribution information and infection gradient evolution surface of forest pests and diseases on each topological node within a preset future time period, including: Taking the current moment as the starting point of the simulation, historical multi-source meteorological time series data and ecological prior knowledge are extracted as condition vectors and input into a pre-trained spatiotemporal conditional generative adversarial network to generate environmental disturbance parameters within the future time window. These environmental disturbance parameters include extreme rainfall, sudden changes in wind direction and speed, and changes in the population density of natural enemies. They are then used as time-varying boundary conditions through the physics engine interface and dynamically loaded into the twin simulation.

7. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 6, characterized in that, The process of using the current moment as the simulation starting point, loading corresponding environmental disturbance parameters into the twin simulation entity, and performing forward extrapolation on the dynamic infection feature field to output the spatial diffusion distribution information and infection gradient evolution surface of forest pests and diseases on each topological node within a preset time period, also includes: In the twin simulation, driven by time-varying boundary conditions, a spatiotemporal graph neural network with a hybrid architecture including graph attention mechanism and temporal convolutional network is used to perform forward extrapolation of the dynamic infection feature field. The edge weights of the adjacency matrix of the graph structure are dynamically updated with environmental perturbation parameters to determine the pest and disease transmission information between each forest topology node based on the canopy overlap distance and wind direction propagation resistance. Then, the spatial diffusion distribution information within a future preset time period is output. This spatial diffusion distribution information characterizes the evolution of the infection status of each topology node over time. Spatial diffusion distribution information is used to determine corresponding spatial irregularities on a 3D mesh in a pre-defined twin simulation space. The spatial irregularities are iterated at multiple levels, and multiple target indicators are combined for multi-level progressive processing during the iteration process to generate the corresponding infection gradient evolution surface. The infection gradient evolution surface represents the infection severity as it continuously and nonlinearly changes in 3D space.

8. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 1, characterized in that, The three-dimensional topological structure of the labeled twin simulation object is combined with spatial diffusion distribution information and infection gradient evolution surface for corresponding rendering to output corresponding rendering results. The rendering results are then mapped to a preset three-dimensional space to construct a corresponding forest pest and disease distribution map, including: Based on the identification of the twin simulation object, the corresponding three-dimensional topology is determined. Semantic segmentation and node labeling are performed on the three-dimensional topology of the twin simulation object. Combined with spatial diffusion distribution information and infection gradient evolution surface, ray tracing and color mapping are performed using volume rendering mechanism to output the corresponding four-dimensional rendering result, which includes time dimension evolution features. The four-dimensional rendering result is then mapped to a preset three-dimensional space with fused geographic information system coordinates.

9. The method for constructing a forest tree disease and pest distribution map based on twin simulation according to claim 8, characterized in that, The three-dimensional topological structure of the labeled twin simulation object is combined with spatial diffusion distribution information and infection gradient evolution surface for corresponding rendering to output corresponding rendering results. The rendering results are then mapped to a preset three-dimensional space to construct a corresponding forest pest and disease distribution map. The method also includes: Virtual camera nodes are deployed in three-dimensional space, and an immersive view roaming engine is further introduced. At the same time, a corresponding index structure is built for the time-series data in the four-dimensional rendering results to respond to the time node query command input by the user, and the corresponding pest and disease distribution status is rendered in combination with the immersive view roaming engine. Infection distribution data are extracted from the infection gradient evolution surface and multi-factor fusion is performed in combination with the distribution status of pests and diseases. In the fusion process, the boundaries of high-risk area, medium-risk area and low-risk area are divided, and a semi-transparent superimposed risk thermal early warning layer is generated to construct a forest pest and disease distribution map that can be dynamically demonstrated.

10. A system for constructing forest tree disease and pest distribution maps based on twin simulation, characterized in that, The system for constructing forest pest and disease distribution maps based on twin simulation is applied to the method for constructing forest pest and disease distribution maps based on twin simulation as described in any one of claims 1-9; The system for constructing forest tree disease and pest distribution maps based on twin simulation includes: The simulation information module is used to acquire multimodal sensing data of the target forest area, perform cross-modal spatiotemporal alignment of the multimodal sensing data, and fuse it with the twin simulation space, outputting the corresponding twin simulation information during the fusion process; The twin simulation module is used to determine the state difference vector of forest trees based on the identification of twin simulation information, and trigger the dynamic evolution of the state difference vector. In this way, a corresponding dynamic infection feature field is constructed based on multiple pest and disease constraints. The dynamic infection feature field is then mapped to a preset twin simulation space to obtain the corresponding twin simulation object. The simulation module is used to load the corresponding environmental disturbance parameters into the twin simulation object with the current time as the simulation starting point, and to perform forward simulation of the dynamic infection feature field, thereby outputting the spatial diffusion distribution information of forest pests and diseases on each topological node and the infection gradient evolution surface within a preset time period in the future. The forest pest and disease distribution map module is used to mark the three-dimensional topological structure of the twin simulation object, and combine the spatial diffusion distribution information with the infection gradient evolution surface for corresponding rendering to output the corresponding rendering results. The rendering results are then mapped to a preset three-dimensional space to construct the corresponding forest pest and disease distribution map.

Images