A citrus huanglongbing diseased tree positioning and diffusion trend prediction management system

Abstract

This invention discloses a system for locating and predicting the spread of Huanglongbing (HLB) in citrus trees, belonging to the field of agricultural pest and disease control technology. The system achieves precise identification and centimeter-level location of diseased trees by fusing multi-source data from UAV hyperspectral remote sensing, ground-based IoT sensors, and manual inspections. It constructs a HLB spread trend prediction model based on spatiotemporal deep learning, comprehensively considering meteorological conditions, psyllid population dynamics, orchard management levels, and geographical environmental factors to predict the disease spread risk over the next 7-30 days. A hierarchical control decision support system is established to automatically generate targeted control plans and track their implementation effects. This invention solves the problems of late disease detection, inaccurate location, delayed prediction, and blind control in traditional HLB control methods. It can advance the detection time of diseased trees by 2-3 months, achieve a location accuracy of ±50cm, and a spread trend prediction accuracy rate exceeding 85%, significantly reducing the spread speed and severity of HLB, and providing strong technical support for the sustainable development of the citrus industry.

Description

Technical Field

[0001] This invention relates to the field of intelligent prevention and control technology for agricultural pests and diseases, and in particular to a system for predicting and controlling the location and spread trend of citrus Huanglongbing disease trees based on multi-source data fusion and spatiotemporal deep learning. Background Technology

[0002] Huanglongbing (HLB) is a devastating disease of citrus caused by bacteria of the genus *Hymenobacter*, often referred to as the "cancer" of the citrus industry. The disease is primarily transmitted by the citrus psyllid, but can also be spread through grafting. It is characterized by rapid spread, wide-ranging damage, and extreme difficulty in control. Currently, there is no effective treatment worldwide. The only feasible control strategy is "prevention first, integrated management," which involves timely detection and thorough removal of diseased trees, along with strict control of citrus psyllids.

[0003] Traditional methods of controlling Huanglongbing (HLB) mainly rely on manual inspections to identify diseased trees, which has the following significant drawbacks:

[0004] Delayed detection: Huanglongbing has a long incubation period (3-12 months), and manual inspection can only find diseased trees that show obvious symptoms. By this time, the disease has already spread in the orchard.

[0005] Inaccurate location: The location of diseased trees recorded manually has a large error, making it difficult to achieve accurate removal and easily leaving a source of infection;

[0006] Difficulty in prediction: There is a lack of scientific methods for predicting the spread trend, making it impossible to take preventive measures in advance;

[0007] Blind control: Prevention and control measures lack specificity and often adopt a "one-size-fits-all" approach, resulting in pesticide waste and environmental pollution;

[0008] Inefficient: Manual inspections are labor-intensive and costly, and it is difficult to cover large areas of orchards.

[0009] In recent years, with the development of remote sensing, Internet of Things (IoT), and artificial intelligence (AI) technologies, some scholars have begun to explore their application in monitoring Huanglongbing (HLB). For example, multispectral remote sensing using drones has been used to identify diseased trees, but the accuracy rate is low and it cannot distinguish HLB from other physiological diseases. IoT is used to monitor psyllid population dynamics, but correlation analysis with disease spread is lacking. Currently, there is no complete system integrating precise location of diseased trees, prediction of spread trends, and intelligent management.

[0010] Therefore, there is an urgent need to develop a comprehensive system that can enable early detection, precise location, scientific prediction, and intelligent management of Huanglongbing (HLB) in order to improve the efficiency and effectiveness of HLB prevention and control. Summary of the Invention

[0011] (a) Technical problems to be solved The purpose of this invention is to overcome the shortcomings of the prior art and provide a system for locating and predicting the spread of citrus Huanglongbing (HLB) diseased trees. This system can achieve early and accurate identification and location of diseased trees, scientifically predict the spread trend of the disease, and automatically generate intelligent control plans, thereby effectively controlling the spread and damage of HLB. (II) Technical Solution

[0012] To achieve the above objectives, the present invention adopts the following technical solution:

[0013] A system for locating and predicting the spread of citrus Huanglongbing (HLB) diseased trees, comprising a data acquisition module, a data processing and storage module, a diseased tree identification and location module, a spread trend prediction module, a control decision support module, and a visualization module;

[0014] The data acquisition module includes a UAV hyperspectral remote sensing submodule, a ground IoT monitoring submodule, and a manual inspection submodule, used to collect orchard hyperspectral image data, environmental meteorological data, psyllid population data, soil data, and diseased tree inspection data;

[0015] The data processing and storage module is used to preprocess, fuse and store the collected multi-source data to establish a spatiotemporal database of citrus Huanglongbing (HLB).

[0016] The diseased tree identification and positioning module is used to identify Huanglongbing diseased trees based on hyperspectral image data and deep learning algorithms, and to achieve centimeter-level positioning of diseased trees by combining UAV RTK positioning data and orchard GIS map;

[0017] The spread trend prediction module is used to construct a Huanglongbing spread trend prediction model based on spatiotemporal graph convolutional neural network, which comprehensively considers meteorological conditions, psyllid population dynamics, orchard management level and geographical environmental factors to predict the risk of disease spread in different time periods in the future.

[0018] The control and management decision support module is used to generate a hierarchical control plan based on the distribution of diseased trees and the prediction results of the spread risk, and to track the implementation effect of the control measures.

[0019] The visualization module is used to intuitively display the distribution of diseased trees, the risk level of spread, and control measures based on an electronic map, and supports users to perform interactive queries and operations.

[0020] Furthermore, the UAV hyperspectral remote sensing submodule is equipped with a hyperspectral imager and an RTK positioning module. It flies at an altitude of 50-100m, has a spatial resolution of 5-10cm, a spectral range of 400-1000nm, and a spectral resolution of 2-5nm. During the critical growth periods of citrus (spring shoots, summer shoots, and autumn shoots), it conducts remote sensing monitoring 1-2 times per month and intensifies monitoring of areas suspected of being diseased trees.

[0021] Furthermore, the ground-based IoT monitoring submodule includes a weather station, a psyllid trap, a soil sensor, and a video monitoring device; the weather station is used to monitor temperature, humidity, rainfall, wind speed, wind direction, and light intensity; the psyllid trap has a built-in counting sensor for real-time monitoring of the psyllid population; the soil sensor is used to monitor soil moisture, pH value, and nutrient content; and the video monitoring device is used to monitor orchard management activities and psyllid activity.

[0022] Furthermore, the specific working steps of the diseased tree identification and localization module are as follows:

[0023] Perform radiometric calibration, atmospheric correction, and geometric correction on hyperspectral images;

[0024] The improved U-Net semantic segmentation algorithm was used to extract the canopy region of a single citrus tree;

[0025] Spectral characteristics of the canopy layer of each citrus tree were extracted, including the original spectrum, first derivative spectrum, second derivative spectrum, and vegetation index;

[0026] Construct a convolutional neural network classification model based on the attention mechanism to distinguish between healthy trees, Huanglongbing disease trees, and other diseased trees;

[0027] By combining UAV RTK positioning data and orchard GIS map, the geographic coordinates of each diseased tree are calculated and marked on the electronic map;

[0028] The identified suspected diseased trees were manually reviewed, and the diseased tree database was updated.

[0029] Furthermore, the specific working steps of the diffusion trend prediction module are as follows:

[0030] The orchard was divided into several 10m×10m grid units, and a spatiotemporal graph structure was constructed, in which nodes represent grid units and edges represent the spatial proximity relationships between grid units.

[0031] Extract the feature vector of each grid cell, including the number of diseased trees, psyllid density, average temperature, average humidity, rainfall, orchard canopy closure, management level, and topographic features;

[0032] A spatiotemporal graph convolutional neural network model was constructed, which consists of spatial graph convolutional layers and temporal convolutional layers to capture the spatial and temporal dependencies of disease spread, respectively.

[0033] The model was trained and validated using historical Huanglongbing (HLB) occurrence data and environmental data.

[0034] Input current environmental data and diseased tree distribution data to predict the probability of disease occurrence in each grid cell over the next 7, 15, and 30 days;

[0035] Based on the forecast results, the orchard was divided into high-risk, medium-risk, and low-risk areas.

[0036] Furthermore, the specific working steps of the control and management decision support module are as follows:

[0037] Based on the severity of the disease and the risk of its spread, diseased trees are classified into three levels: Level 1, Level 2, and Level 3.

[0038] For diseased trees of different levels, corresponding removal plans are generated, including removal time, removal method and treatment requirements;

[0039] For areas with different risk levels, corresponding psyllid control plans are generated, including control time, pesticide type, pesticide dosage and application method;

[0040] Recommendations for orchard management, including pruning, fertilization, irrigation, and planting disease-free seedlings;

[0041] The management and control plan was communicated to orchard managers, and its implementation was tracked.

[0042] Based on feedback on the effectiveness of control measures, the prediction model and control plan will be dynamically adjusted. (III) Beneficial Effects

[0043] Compared with the prior art, the present invention has the following advantages:

[0044] Early detection: Hyperspectral remote sensing technology can identify Huanglongbing diseased trees that have not yet shown obvious symptoms, advancing the discovery time of diseased trees by 2-3 months and buying valuable time for timely removal of the source of infection;

[0045] Precise positioning: Combining drone RTK positioning technology and orchard GIS map, centimeter-level positioning of diseased trees is achieved with a positioning accuracy of ±50cm, ensuring that diseased trees can be completely removed;

[0046] Scientific prediction: A diffusion trend prediction model based on spatiotemporal graph convolutional neural network is constructed, which comprehensively considers multiple influencing factors and has a prediction accuracy of over 85%, enabling early warning of the risk of disease spread;

[0047] Intelligent management and control: Based on the distribution of diseased trees and the prediction of the risk of spread, a hierarchical management and control plan is automatically generated to achieve precise prevention and control and reduce pesticide use by more than 30%.

[0048] Highly efficient and convenient: It has achieved full automation and intelligence in the prevention and control of Huanglongbing, which has greatly reduced labor intensity and management costs and improved prevention and control efficiency. Attached Figure Description

[0049] To provide a more detailed description of the technical solutions of the embodiments of this invention, the accompanying drawings used in the following description of the embodiments are briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on such drawings without creative effort.

[0050] Figure 1 is an overall architecture diagram of the citrus Huanglongbing disease-causing tree location and spread trend prediction and control system of the present invention;

[0051] Figure 2 is a flowchart of the diseased tree identification and localization module;

[0052] Figure 3 is a flowchart of the diffusion trend prediction module;

[0053] Figure 4 is a flowchart of the control decision support module. Detailed Implementation

[0054] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0055] As shown in Figure 1, the citrus Huanglongbing disease tree location and spread trend prediction and control system disclosed in this invention includes a data acquisition module 1, a data processing and storage module 2, a disease tree identification and location module 3, a spread trend prediction module 4, a control decision support module 5, and a visualization display module 6.

[0056] Example 1

[0057] The data acquisition module 1 includes a UAV hyperspectral remote sensing submodule 11, a ground IoT monitoring submodule 12, and a manual inspection submodule 13.

[0058] The UAV hyperspectral remote sensing submodule 11 utilizes a DJI M300 RTK UAV equipped with a Headwall Nano-Hyperspec hyperspectral imager. The hyperspectral imager has a spectral range of 400-1000nm, a spectral resolution of 3nm, and a total of 270 bands. The UAV's flight altitude was set to 80m, its flight speed to 5m / s, with a forward overlap of 80%, a lateral overlap of 70%, and a spatial resolution of approximately 8cm. Remote sensing monitoring was conducted twice monthly during the spring shoot period (March-April), summer shoot period (June-July), and autumn shoot period (September-October) of citrus trees. Areas where suspected diseased trees were identified underwent intensive monitoring once a week.

[0059] The ground-based IoT monitoring submodule 12 deploys weather stations at a density of one station per 50 mu (approximately 3.3 hectares) within the orchard to monitor temperature, humidity, rainfall, wind speed, wind direction, and light intensity, with data collection frequency every 10 minutes. It also deploys intelligent psyllid traps at a density of one trap per 10 mu (approximately 0.67 hectares), each trap containing an infrared counting sensor to count psyllids in real time, with data uploaded hourly. Soil sensors are deployed in different areas of the orchard to monitor soil moisture, pH value, and nitrogen, phosphorus, and potassium content, with data collection frequency once daily. High-definition video surveillance equipment is deployed at the orchard entrance and key locations to monitor orchard management activities and psyllid activity in real time.

[0060] The manual inspection submodule 13 includes a mobile app. Orchard managers can use the app to record the location, symptoms, and severity of diseased trees during inspections, and upload photos. The app automatically obtains the phone's GPS location information and compares and verifies it with the results of drone remote sensing.

[0061] Example 2

[0062] As shown in Figure 2, the specific working steps of the diseased tree identification and localization module 3 are as follows:

[0063] Step S31: Data Preprocessing. Radiometric calibration is performed on the hyperspectral images acquired by the UAV, converting DN values ​​into radiance values; atmospheric correction is performed using the FLAASH model to eliminate the effects of atmospheric scattering and absorption; geometric correction is performed using UAV RTK positioning data and ground control points, with a correction accuracy of ±10cm.

[0064] Step S32: Individual Tree Extraction. The hyperspectral image is segmented using an improved U-Net semantic segmentation algorithm to extract the canopy region of individual citrus trees. The improved U-Net model incorporates residual connections in the encoder to enhance feature extraction capabilities and an attention mechanism in the decoder to strengthen the focus on the canopy region.

[0065] Step S33: Spectral Feature Extraction. For the canopy region of each citrus tree, calculate the average spectral curve and extract the following features:

[0066] Original spectral characteristics: reflectance values ​​in the range of 400-1000 nm;

[0067] Derivative spectral characteristics: first derivative spectrum and second derivative spectrum;

[0068] Vegetation indices include: Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), Red Edge Position (REP), Chlorophyll Index (CI), etc.

[0069] Step S34: Diseased Tree Classification. A convolutional neural network (CNN) classification model based on an attention mechanism is constructed. The input to this model is a spectral feature vector, which, after passing through multiple convolutional layers, pooling layers, and fully connected layers, outputs the classification results: healthy trees, Huanglongbing (HLB) diseased trees, canker diseased trees, and nutrient deficiency trees. During model training, transfer learning is employed, using a publicly available hyperspectral dataset for pre-training, followed by fine-tuning using labeled data from this orchard.

[0070] Step S35: Diseased Tree Location. Based on the UAV RTK positioning data and geometric correction parameters of the imagery, calculate the geographic coordinates (longitude and latitude) of each diseased tree. Combined with the orchard GIS map, mark the locations of the diseased trees on the electronic map and generate a diseased tree distribution map.

[0071] Step S36: Manual Verification. The system automatically generates a list of suspected diseased trees and pushes it to orchard managers for on-site verification. Managers confirm the diseased trees via a mobile app and supplement information such as the severity of the disease. The verification results are fed back to the system to update the diseased tree database and optimize the classification model.

[0072] Example 3

[0073] As shown in Figure 3, the specific working steps of the diffusion trend prediction module 4 are as follows:

[0074] Step S41: Spatiotemporal graph construction. Divide the orchard into several 10m×10m grid cells and construct a spatiotemporal graph G=(V, E), where V={v1, v2, ..., vn} is a set of nodes, each node representing a grid cell; E is a set of edges, and if two grid cells are adjacent, an edge is established between them.

[0075] Step S42: Feature Extraction. Extract the feature vector of each grid cell over the past T time steps, including:

[0076] Disease characteristics: number of diseased trees, density of diseased trees, severity of disease;

[0077] Vector characteristics: psyllid density, psyllid activity rate;

[0078] Meteorological characteristics: average temperature, average humidity, cumulative rainfall, and average wind speed;

[0079] Environmental characteristics: orchard canopy density, altitude, slope, and aspect;

[0080] Management characteristics: number of pesticide applications, pruning conditions, and fertilization conditions.

[0081] Step S43: Model Construction. Construct a Spatiotemporal Graph Convolutional Neural Network (ST-GCN) model. This model consists of multiple stacked spatiotemporal blocks, each containing a spatial graph convolutional layer and a temporal convolutional layer. The spatial graph convolutional layer captures the spatial dependency of disease spread, i.e., the mutual influence between adjacent grid cells; the temporal convolutional layer captures the temporal dependency of disease spread, i.e., the influence of historical states on future states. The model output is the probability of disease occurrence in each grid cell over the next K time steps.

[0082] Step S44: Model Training and Validation. Collect Huanglongbing (HLB) occurrence data, psyllid monitoring data, and meteorological data from the orchard over the past 3 years to construct training and validation datasets. Train the model using the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 100 training epochs. An early stopping strategy is employed to prevent overfitting; training is stopped when the validation set loss does not decrease for 10 consecutive epochs.

[0083] Step S45: Predict the risk of disease spread. Input the current environmental data and the distribution data of diseased trees, and use the trained model to predict the probability of disease occurrence in each grid cell in the next 7 days, 15 days, and 30 days.

[0084] Step S46: Risk Level Classification. Based on the predicted probability of disease occurrence, the orchard is divided into three risk levels:

[0085] High-risk areas: probability of occurrence ≥70%;

[0086] Medium-risk areas: 30% ≤ probability of occurrence < 70%;

[0087] Low-risk areas: probability of occurrence < 30%.

[0088] A disease spread risk level map of the orchard is generated and displayed in the visualization module.

[0089] Example 4

[0090] As shown in Figure 4, the specific working steps of the control decision support module 5 are as follows:

[0091] Step S51: Diseased Tree Classification. Based on the severity of disease and the risk of spread in the area, diseased trees are classified into three levels:

[0092] Level 1 diseased trees: Diseased trees with severe disease (leaf yellowing area ≥ 50%) or located in high-risk areas;

[0093] Level II diseased trees: Trees with moderate disease incidence (20% ≤ leaf yellowing area < 50%) or located in medium-risk areas;

[0094] Level 3 diseased trees: Diseased trees with mild disease (leaf yellowing area < 20%) and located in low-risk areas.

[0095] Step S52: Generation of Diseased Tree Removal Plans. For different levels of diseased trees, corresponding removal plans are generated:

[0096] Level 1 diseased trees: Remove them immediately, including uprooting them. Treat the stumps with herbicide and burn the diseased branches and leaves.

[0097] Second-level diseased trees: Remove within 3 days, and the removal requirement is the same as that of diseased trees of the same level;

[0098] Level 3 diseased trees: Remove them within 7 days, or isolate and observe them, monitor them once a week, and remove them immediately if the condition worsens.

[0099] Step S53: Psyllid Control Plan Generation. Generate corresponding psyllid control plans for areas with different risk levels:

[0100] High-risk areas: Immediately implement chemical control measures, using highly effective and low-toxicity pesticides such as thiamethoxam and imidacloprid, applying the pesticide once every 7 days for 2-3 consecutive applications; at the same time, hang yellow sticky traps to attract and kill adult insects.

[0101] Medium-risk areas: Chemical control should be carried out within 7 days, with one application every 10 days, for 1-2 consecutive applications;

[0102] Low-risk areas: Strengthen monitoring and implement chemical control when the psyllid density exceeds the threshold (5 psyllids per trap per week).

[0103] Step S54: Orchard Management Suggestions Generation. Based on the actual conditions of the orchard, the following management suggestions are generated:

[0104] Pruning: Remove diseased, weak, and overly dense branches in a timely manner to improve ventilation and light penetration in the orchard;

[0105] Fertilization: Increase the application of organic fertilizer and phosphorus and potassium fertilizer to improve the tree's disease resistance;

[0106] Irrigation: Irrigate properly and avoid waterlogging in the orchard;

[0107] Planting: Plant disease-free, healthy seedlings in the locations where diseased trees have been removed.

[0108] Step S55: Plan Push and Implementation Tracking. The generated control plan is pushed to orchard managers via a mobile app. During implementation, managers can upload photos and progress updates via the app. The system tracks the implementation of control measures in real time and provides reminders for tasks not executed on time.

[0109] Step S56: Effectiveness Evaluation and Model Update. After the control measures are implemented, the system evaluates the control effectiveness through drone remote sensing and manual inspections. Based on the feedback on the control effectiveness, the parameters of the diffusion trend prediction model and the thresholds of the control plan are dynamically adjusted to continuously improve the accuracy and effectiveness of the system.

[0110] Example 5

[0111] Visualization module 6 is developed based on WebGIS technology and uses electronic maps to intuitively display the following information:

[0112] Basic orchard information: orchard boundaries, plot divisions, roads, irrigation facilities, etc.;

[0113] Information on the distribution of diseased trees: location, number, and severity of disease in trees of different grades;

[0114] Disease spread risk information: Disease spread risk level map at different time periods in the future;

[0115] Psyllid monitoring information: Psyllid density change curves at each monitoring point;

[0116] Meteorological monitoring information: curves showing changes in meteorological elements such as temperature, humidity, and rainfall;

[0117] Control plan information: tasks to be executed, tasks already executed, and execution results.

[0118] Users can access the system via computer or mobile device to perform interactive queries and operations. For example, clicking on a diseased tree icon on the map will display detailed information about the diseased tree; clicking on a risk area will display the control plan for that area.

Claims

1. A system for locating and predicting the spread of citrus Huanglongbing (HLB) diseased trees, characterized in that, It includes a data acquisition module (1), a data processing and storage module (2), a disease tree identification and location module (3), a spread trend prediction module (4), a control decision support module (5), and a visualization display module (6); The data acquisition module (1) includes a UAV hyperspectral remote sensing submodule (11), a ground IoT monitoring submodule (12), and a manual inspection submodule (13), which are used to collect orchard hyperspectral image data, environmental meteorological data, psyllid population data, soil data, and diseased tree inspection data; The data processing and storage module (2) is used to preprocess, fuse and store the collected multi-source data to establish a spatiotemporal database of citrus Huanglongbing. The diseased tree identification and positioning module (3) is used to identify Huanglongbing diseased trees based on hyperspectral image data and deep learning algorithms, and to achieve centimeter-level positioning of diseased trees by combining UAV RTK positioning data and orchard GIS map; The spread trend prediction module (4) is used to construct a Huanglongbing spread trend prediction model based on spatiotemporal graph convolutional neural network, which comprehensively considers meteorological conditions, psyllid population dynamics, orchard management level and geographical environmental factors to predict the risk of disease spread in different time periods in the future. The control decision support module (5) is used to generate a hierarchical control plan based on the distribution of diseased trees and the prediction results of the spread risk, and to track the implementation effect of the control measures. The visualization module (6) is used to intuitively display the distribution of diseased trees, the risk level of spread, and the control plan based on an electronic map, and supports users to perform interactive queries and operations.

2. The system according to claim 1, characterized in that, The UAV hyperspectral remote sensing submodule (11) is equipped with a hyperspectral imager and an RTK positioning module. It flies at an altitude of 50-100m, has a spatial resolution of 5-10cm, a spectral range of 400-1000nm, and a spectral resolution of 2-5nm. It conducts remote sensing monitoring 1-2 times per month during the spring, summer, and autumn shoot growth periods of citrus trees, and intensifies monitoring in areas suspected of being diseased.

3. The system according to claim 1, characterized in that, The ground-based IoT monitoring submodule (12) includes a weather station, a smart psyllid trap, a soil sensor, and a video monitoring device; the weather station is used to monitor temperature, humidity, rainfall, wind speed, wind direction, and light intensity; the smart psyllid trap has a built-in infrared counting sensor for real-time monitoring of the psyllid population; the soil sensor is used to monitor soil moisture, pH value, and nutrient content; and the video monitoring device is used to monitor orchard management activities and psyllid activity.

4. The system according to claim 1, characterized in that, The specific working steps of the disease tree identification and localization module (3) are as follows: S31: Perform radiometric calibration, atmospheric correction and geometric correction on hyperspectral images; S32: Extract the canopy region of a single citrus tree using the improved U-Net semantic segmentation algorithm; S33: Extract the spectral characteristics of the canopy layer of each citrus tree, including the original spectrum, first derivative spectrum, second derivative spectrum, and vegetation index; S34: Construct a convolutional neural network classification model based on attention mechanism to distinguish healthy trees, Huanglongbing disease trees and other diseased trees; S35: Combine UAV RTK positioning data and orchard GIS map to calculate the geographic coordinates of each diseased tree and mark them on the electronic map; S36: Manually verify the identified suspected diseased trees and update the diseased tree database.

5. The system according to claim 1, characterized in that, The specific working steps of the diffusion trend prediction module (4) are as follows: S41: Divide the orchard into several 10m×10m grid units and construct a spatiotemporal graph structure, where nodes represent grid units and edges represent the spatial proximity relationships between grid units; S42: Extract the feature vector of each grid cell, including the number of diseased trees, psyllid density, average temperature, average humidity, rainfall, orchard canopy closure, management level, and topographic features; S43: Construct a spatiotemporal graph convolutional neural network model, which consists of spatial graph convolutional layers and temporal convolutional layers, to capture the spatial and temporal dependencies of disease spread, respectively. S44: Use historical Huanglongbing occurrence data and environmental data to train and validate the model; S45: Input the current environmental data and diseased tree distribution data, and predict the probability of disease occurrence in each grid cell in the next 7 days, 15 days and 30 days; S46: Based on the prediction results, the orchard is divided into high-risk, medium-risk, and low-risk areas.

6. The system according to claim 1, characterized in that, The specific working steps of the control decision support module (5) are as follows: S51: Based on the severity of the disease and the risk level of its spread in the area, diseased trees are classified into Level 1, Level 2, and Level 3 diseased trees. S52: Generate corresponding removal plans for diseased trees of different levels, including removal time, removal method and treatment requirements; S53: Generate corresponding psyllid control plans for areas with different risk levels, including control time, pesticide type, pesticide dosage and application method; S54: Recommendations for orchard management, including pruning, fertilization, irrigation, and planting disease-free seedlings; S55: Push the management and control plan to orchard managers and track its implementation; S56: Dynamically adjust the prediction model and control plan based on feedback on the control effect.

7. The system according to claim 6, characterized in that, The first-level diseased trees are those with severe disease (leaf yellowing area ≥ 50%) or located in high-risk areas, and must be removed immediately; the second-level diseased trees are those with moderate disease (20% ≤ leaf yellowing area < 50%) or located in medium-risk areas, and must be removed within 3 days; the third-level diseased trees are those with mild disease (leaf yellowing area < 20%) and located in low-risk areas, and must be removed or isolated for observation within 7 days.

8. The system according to claim 6, characterized in that, High-risk areas require immediate chemical control, with applications every 7 days for 2-3 consecutive applications; medium-risk areas require chemical control within 7 days, with applications every 10 days for 1-2 consecutive applications; low-risk areas require enhanced monitoring, with chemical control implemented when psyllid density exceeds the threshold.

9. The system according to claim 1, characterized in that, The visualization module (6) is developed based on WebGIS technology, supports access from computer and mobile devices, and can display basic orchard information, diseased tree distribution information, spread risk information, psyllid monitoring information, meteorological monitoring information, and management and control plan information.

10. The system according to claim 1, characterized in that, The positioning accuracy of the diseased tree identification and positioning module (3) reaches ±50cm, and the prediction accuracy of the diffusion trend prediction module (4) exceeds 85%.

Images