Greenhouse pest and disease diagnosis and decision support system fusing multi-source data

Abstract

The application provides a greenhouse pest and disease diagnosis and decision support system fusing multi-source data, and relates to the technical field of intelligent agriculture.The application fuses meteorological, video stream and spectral multi-source data, calculates migration risk index, group stress light flow divergence and disease course stage index respectively representing external threat, invasion behavior and internal pathology.Subsequently, a cellular automaton model is used to simulate disease space-time propagation by taking the aforementioned indexes as input, and to predict future disease outbreak risk score.The score is used to divide the greenhouse into high, medium and low risk areas, and to generate a dynamically updated three-dimensional risk map.Finally, a linkage intervention decision module determines the optimal and feasible final intervention strategy according to the risk map and real-time resource constraints through the closed-loop logic of "generation-evaluation-correction".The application solves the problems of lagging diagnosis and subjective decision-making, and realizes automatic management from active prediction to precise closed-loop intervention.

Description

Technical Field

[0001] This invention relates to the field of intelligent agriculture technology, specifically to a greenhouse pest and disease diagnosis and decision support system that integrates multi-source data. Background Technology

[0002] Facility agriculture, especially modern greenhouses, creates ideal production conditions for high-value-added crops through precise control of environmental factors. However, this stable, enclosed, and high-density planting environment also provides a breeding ground for the rapid outbreak of various pests and diseases, making it a key bottleneck restricting greenhouse yield and quality. To address this challenge, existing technologies mainly employ various methods. The most traditional method relies on manual inspections and experience-based judgment by agricultural technicians. However, this system suffers from significant lag, subjectivity, and instability. By the time obvious symptoms are observed with the naked eye, the disease has often progressed to the middle or late stages, and the judgment results heavily depend on personal experience, making quantification difficult and inefficient for large-scale greenhouses. Building on this, more automated methods utilize environmental sensors for risk warnings. By monitoring parameters such as temperature and humidity, alarms are triggered when environmental conditions are suitable for disease occurrence. However, its limitation lies in providing only non-specific macro-level risk warnings, failing to confirm the actual presence of pathogens, the specific location of infection, and the development stage, leading to high false alarm rates and unnecessary preventative interventions.

[0003] With the development of computer vision technology, image recognition-based disease diagnosis systems have emerged, using cameras and deep learning models to identify visible lesions or pests. This is essentially an "automated manual inspection," its core capability remaining the identification of existing symptoms. Therefore, it also suffers from problems such as delayed response, limited field of view (difficulty in detecting diseases on the underside of leaves or in the center of flowers), and inability to distinguish cause and effect (e.g., inability to differentiate between disease and nutrient imbalance causing yellowing leaves). The combined limitations of these existing technologies are particularly pronounced in a specific and highly challenging application scenario, such as the complex pest and disease management of high-yield cucumbers in modern glass greenhouses. In this scenario, the main threats come simultaneously from cucumber downy mildew induced by high humidity and thrips, tiny pests that are the primary vectors for virus transmission. A typical failure path is:

[0004] In late spring and early summer, large numbers of thrips invaded greenhouses through ventilation openings and quickly hid, completely missed by the image-based system due to their tiny size. Subsequently, the high humidity at night triggered a "high risk of downy mildew" alarm on the environmental sensors, leading to a potentially ineffective "blind" preventative spraying that was also ineffective against the lurking thrips. A few days later, when the inspection robot finally identified pale yellow downy mildew lesions on the upper surface of the leaves, the disease had already entered an exponential spread stage, and the system issued a delayed alarm. More seriously, the system could not effectively distinguish between the thrips active on these plants and the cucumber mosaic virus (CMV) they spread, which had similar symptoms, and downy mildew. As a result, while the managers were focusing on controlling downy mildew, the virus had already spread quietly due to the continued activity of the thrips, ultimately causing a catastrophic concurrence of downy mildew and viral diseases. This typical failure path clearly reveals the fundamental flaws of existing technologies: their core paradigms all remain at the level of passive response to isolated, static, and already occurred events, generally lacking a systematic forward-looking capability. They are unable to combine potential external threats (such as pest migration) with internal health conditions (plant immunity) to predict future "hotspot areas"; they cannot trace the causal chain between pests as "transmission vectors" and diseases as "final results"; and they cannot quantitatively assess the "development stages" of diseases to guide precise intervention. Summary of the Invention

[0005] The purpose of this invention is to provide a greenhouse pest and disease diagnosis and decision support system that integrates multi-source data to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A greenhouse pest and disease diagnosis and decision support system integrating multi-source data includes:

[0008] A multi-source heterogeneous data fusion module is used to obtain the basic dataset;

[0009] The status and risk assessment module is used to calculate the migration risk index, population stress divergence, and disease stage index based on the basic dataset and focusing on crop physiological status and external environmental stress, in order to construct a status and risk characteristic dataset.

[0010] The collaborative risk prediction module is used to predict the risk score of disease outbreaks in each crop planting area in the greenhouse in the future by using a preset disease transmission prediction model and taking the status and risk feature dataset as input.

[0011] The risk zone dynamic division module is used to divide the entire greenhouse into high-risk warning zones, medium-risk attention zones, and low-risk monitoring zones in three-dimensional space based on the disease outbreak risk score, and generate a dynamically updated three-dimensional risk map.

[0012] The linkage intervention decision-making module is used to receive intervention instructions generated based on the 3D risk map and specifically execute them as follows: generating a preliminary intervention strategy; simultaneously calculating the matching degree between the resources required to execute the preliminary intervention strategy and the currently available resources to construct a resource compliance score for quantifying the feasibility of the strategy; if the resource compliance score is lower than the preset compliance threshold, the preliminary intervention strategy is iteratively corrected through a local strategy adjustment mechanism until the resource compliance score corresponding to the corrected new strategy meets the requirements, thereby determining it as the final intervention strategy.

[0013] Furthermore, the basic dataset includes: regional weather forecasts and entomological model data obtained through external API interfaces; dynamic video stream data of crop canopies returned by high frame rate cameras; multi-band fluorescence image data of crop leaves returned by multispectral fluorescence imaging systems; temperature, humidity, and regional wind field vector data obtained by environmental sensor arrays deployed in the greenhouse; three-dimensional spatial division information of crop planting areas extracted from a pre-set greenhouse geographic information database, dividing the greenhouse into multiple continuous crop planting areas; and spatially and temporally associating the collected basic dataset with corresponding areas in multiple crop planting areas, thereby forming a structured basic dataset based on crop planting areas.

[0014] Furthermore, the status and risk assessment module includes a first risk index analysis unit and a second risk index analysis unit; the first risk index analysis unit is used to calculate the migration risk index for each crop planting area, and the specific steps are as follows:

[0015] S111. Extract the regional wind field vector, average temperature, and relative humidity for the next 24 hours from the structured basic dataset;

[0016] S112. Call the preset target pest biological model, take the wind field vector as the propagation power, and take the average temperature and relative humidity as the suitable conditions for survival and reproduction. Use the Lagrange particle diffusion model to perform simulation calculations to obtain the probability density of pests migrating to the current greenhouse area per unit time, which is defined as the migration risk index.

[0017] The second risk index analysis unit is used to calculate the population laser divergence for each crop planting area. The specific steps are: S211, extracting dynamic video stream data of the crop population from the structured basic dataset;

[0018] S212. An optical flow estimation algorithm is used to process dynamic video stream data to generate a pixel-level optical flow vector field characterizing the micro-motion of the crop population.

[0019] S213. Calculate the spatial divergence of the optical flow vector field, and perform time series analysis on the divergence values ​​within a preset time window to extract specific frequency band energy values ​​that match the frequency of insect settlement disturbances, which are defined as the population response optical flow divergence.

[0020] Furthermore, the status and risk assessment module also includes a third risk index analysis unit; the third risk index analysis unit is used to calculate the disease stage index for each crop planting area, and the specific steps are as follows: S311, extract multi-band fluorescence image data from the structured basic dataset; S312, use a non-negative matrix factorization algorithm to process the multi-band fluorescence image data to separate the relative concentration time-series curves of at least two preset endogenous fluorescent substances related to plant defense response from the mixed fluorescence spectrum; S313, fit the relative concentration time-series curves to a preset plant immune response biochemical kinetic model, and obtain a dimensionless value in the range [0,1] that characterizes the current disease development stage through analytical solution or numerical optimization, which is the disease stage index.

[0021] Furthermore, the collaborative risk prediction module employs a cellular automata model for disease transmission prediction. Specifically, the module executes the following steps: S411, constructing a two-dimensional or three-dimensional cellular grid corresponding to the spatial layout of the greenhouse crop planting area; S412, assigning the disease progression stage index calculated by the state and risk assessment module to each cell as a basic susceptibility attribute characterizing its initial health state; S413, within a sliding time window, calculating the cross-correlation function between the migration risk index and the population response laser divergence. When its peak value exceeds a preset invasion threshold, it is determined to be an invasion event, and... S414. Based on preset biological evolution rules, perform time-domain iteration on the cellular automata model. The evolution rules include: a cell marked as an infection source has a probability of spreading its infection status to neighboring cells in the next time step; after a healthy cell is spread, the probability of it becoming infected is an increasing function of its own basic susceptibility attribute. S415. After the preset number of iterations is completed, calculate the cumulative probability of each cell being marked as infected to form a set containing the risk score of future disease outbreaks in each crop planting area.

[0022] Furthermore, the collaborative risk prediction module also includes a risk score aggregation unit and a time-series risk tracking unit. The risk score aggregation unit is used to generate disease outbreak risk score values ​​as the current regional future risk score values ​​for each crop planting area. The time-series risk tracking unit is used to receive and cache multiple regional future risk score values ​​output by the risk score aggregation unit in a continuous prediction period, and construct a time-series risk score sequence with a preset length according to the timestamp order of the multiple regional future risk score values, so as to analyze the evolution trend of the risk.

[0023] Furthermore, the risk area dynamic division module includes a risk level mapping unit and a 3D map generation unit. The risk level mapping unit is used to preset a first risk threshold and a second risk threshold, and to determine the disease outbreak risk score for each crop planting area to obtain the risk level result. This includes: when the disease outbreak risk score is lower than the first risk threshold, the risk level is mapped to a low-risk monitoring level; when the disease outbreak risk score is between the first and second risk thresholds, the risk level is mapped to a medium-risk attention level; when the disease outbreak risk score is not lower than the second risk threshold, the risk level is mapped to a high-risk warning level. The 3D map... The generation unit, based on a pre-configured 3D geometric model of the greenhouse, traverses the risk level results of each crop planting area and performs the following spatial attribute assignment and visualization rendering operations: if the risk level is low-risk monitoring level, the corresponding spatial area is divided into a low-risk monitoring area and rendered as a green visual marker; if the risk level is medium-risk attention level, the corresponding spatial area is divided into a medium-risk attention area and rendered as a yellow visual marker; if the risk level is high-risk warning level, the corresponding spatial area is divided into a high-risk warning area and rendered as a red visual marker; all rendered areas are aggregated to generate a dynamically updated 3D risk map.

[0024] Furthermore, the coordinated intervention decision-making module includes a preliminary strategy generation subunit, which receives a 3D risk map as input and generates a preliminary intervention strategy for each crop planting area by consulting a preset intervention rule knowledge base. The specific steps are as follows: traverse each crop planting area in the 3D risk map; if the current area is a high-risk warning area, match and assign a first-type intervention action from the knowledge base, which includes precise spraying with therapeutic agents; if the current area is a medium-risk concern area, match and assign a second-type intervention action from the knowledge base, which includes enhancing ventilation and dehumidification in the area; if the current area is a low-risk monitoring area, match and assign a third-type intervention action from the knowledge base, which includes maintaining the regular monitoring frequency; combine all areas and their assigned intervention actions to form a preliminary intervention strategy.

[0025] Furthermore, the coordinated intervention decision-making module also includes a resource compatibility assessment unit. This unit is used to simultaneously calculate the degree of matching between the initial intervention strategy and currently available resources. Specific steps include:

[0026] S611. Analyze the initial intervention strategy and calculate the total amount of various resources required to execute all assigned first and second type intervention actions, including the total amount of drugs, equipment usage time, and man-hours.

[0027] S612. Query the real-time inventory of various resources available in the greenhouse at the current moment from the preset resource management database; S613. For each type of resource, calculate the ratio of its real-time inventory to the total amount required for consumption, and determine the minimum value among all ratios as the resource compliance score.

[0028] Furthermore, the coordinated intervention decision-making module also includes a strategy correction unit. This unit is used to: trigger and execute an iterative correction loop instruction when and only when the resource compliance score is lower than a preset compliance threshold. This includes: locating the bottleneck resource type that causes the lowest resource compliance score; adjusting the initial intervention strategy according to preset correction priority rules, which include: first attempting to downgrade some second-category intervention actions in medium-risk concern areas to third-category intervention actions; if resources are still insufficient, further replacing some first-category intervention actions in high-risk warning areas with second-category intervention actions; based on the adjusted strategy, re-executing all steps of the resource compliance assessment unit to calculate a new resource compliance score; if the new resource compliance score is not lower than the compliance threshold, terminating the current iterative correction loop and determining the current adjusted strategy as the final intervention strategy; otherwise, continuing the correction.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] This invention achieves the beneficial effect of transforming pest and disease management from a reactive, post-event response to an active, pre-event warning by integrating external migration risk prediction based on the Lagrange model with internal disease propagation extrapolation based on the cellular automata model.

[0031] This invention establishes a complete causal analysis chain from the introduction of external threats to the internal pathological evolution by constructing a logical relationship between the migration risk index, the population stress laser divergence, and the disease stage index. This overcomes the limitations of existing technologies in distinguishing between the root cause and the manifestation of disease.

[0032] This invention uses multispectral fluorescence unmixing technology to quantify disease stages and combines it with a spatiotemporal propagation model to generate probability scores for future outbreaks in each region. This results in transforming vague and subjective risk judgments into precise quantitative assessments of disease development and future risks, thus providing data support for tiered intervention.

[0033] This invention employs a closed-loop decision-making mechanism of "generation-evaluation-correction." After generating an intervention strategy, this mechanism verifies its compatibility with real-time available resources and iteratively optimizes it. This ensures that the final intervention plan has both spatial accuracy and resource feasibility, thereby avoiding ineffective or unenforceable extensive management and improving resource utilization efficiency. Attached Figure Description

[0034] Figure 1 This is a schematic diagram illustrating the actual application scenario and steps of the overall system of the present invention;

[0035] Figure 2 This is a schematic diagram of the overall system flow of the present invention. Detailed Implementation

[0036] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0037] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0038] Example 1:

[0039] Please see Figures 1 to 2 This invention provides a technical solution: a greenhouse pest and disease diagnosis and decision support system that integrates multi-source data, comprising:

[0040] A multi-source heterogeneous data fusion module is used to acquire and preprocess the data to form the basic dataset for greenhouse crop diagnosis.

[0041] The basic dataset includes: regional weather forecast and entomological model data obtained through external API interfaces; dynamic video stream data of crop populations returned by high frame rate cameras; multi-band fluorescence image data of crop leaves returned by multispectral fluorescence imaging systems; temperature, humidity and regional wind field vector data obtained by environmental sensor arrays deployed in greenhouses; three-dimensional spatial division information of crop planting areas extracted from a pre-set greenhouse geographic information database, dividing the greenhouse into multiple continuous crop planting areas; and spatially and temporally associating the collected basic dataset with corresponding areas in multiple crop planting areas, thereby forming a structured basic dataset based on crop planting areas.

[0042] The status and risk assessment module is used to extract migration risk index, population stress divergence, and disease stage index based on the basic dataset and focusing on crop physiological status and external environmental stress, in order to construct a status and risk characteristic dataset.

[0043] The status and risk assessment module includes a first risk index analysis unit and a second risk index analysis unit. The first risk index analysis unit is used to calculate the migration risk index, denoted as Mr, for each crop planting area. The specific steps are as follows:

[0044] S111. Extract the regional wind field vector, average temperature, and relative humidity for the next 24 hours from the structured basic dataset;

[0045] S112. Call the preset target pest biological model, take the wind field vector as the propagation power, and take the average temperature and relative humidity as the suitable conditions for survival and reproduction. Use the Lagrange particle diffusion model to simulate and calculate to obtain the probability density of pests migrating to the current greenhouse area per unit time, which is defined as the migration risk index Mr.

[0046] The second risk index analysis unit is used to calculate the population laser divergence for each crop planting area, denoted as Cf. The specific steps are: S211, extracting dynamic video stream data of the crop population from the structured basic dataset;

[0047] S212. An optical flow estimation algorithm is used to process dynamic video stream data to generate a pixel-level optical flow vector field characterizing the micro-motion of the crop population.

[0048] S213. Calculate the spatial divergence of the optical flow vector field, and perform time series analysis on the divergence value within a preset time window to extract specific frequency band energy values ​​that match the frequency of insect settlement disturbance, so as to generate the population response optical flow divergence Cf.

[0049] The status and risk assessment module also includes a third risk index analysis unit. The third risk index analysis unit is used to calculate the disease stage index, denoted as DPSI, for each crop planting area. The specific steps are as follows: S311, extract multi-band fluorescence image data from the structured basic dataset; S312, process the multi-band fluorescence image data using a non-negative matrix factorization algorithm to separate the relative concentration time-series curves of at least two preset endogenous fluorescent substances related to plant defense response from the mixed fluorescence spectrum; S313, fit the relative concentration time-series curves to a preset plant immune response biochemical kinetic model, and obtain a dimensionless value in the range [0,1] that characterizes the current disease development stage, which is the disease stage index DPSI, through analytical solution or numerical optimization.

[0050] In this embodiment, the application scenario is a modern intelligent glass greenhouse covering 5,000 square meters, used for the commercial production of a high-value-added crop—Dutch cluster cucumbers. The greenhouse is divided into 50 independent crop planting areas (100 square meters each), equipped with an automated environmental control system, an integrated water and fertilizer system, and a track-mounted intelligent inspection robot. Currently, the main threats to the greenhouse are thrips (Cucumber Mosaic Virus) and downy mildew, which is highly susceptible to outbreaks in high humidity environments. The goal of this system is to achieve early prediction, accurate diagnosis, and intelligent decision support for these two or more pests and diseases in the complex greenhouse environment, thereby replacing traditional manual inspections and extensive management.

[0051] The specific steps include:

[0052] S1. Multi-source heterogeneous data fusion module: This module enables the collection and structuring of information across all dimensions. This step forms the foundation for the entire system's perception capabilities, and its internal implementation is as follows:

[0053] S11. Acquire macro-environmental and biological prior data. Specifically, through an internet API interface, periodically (e.g., hourly), request accurate weather forecast data for the next 24 hours for the greenhouse's geographical location from a designated commercial weather service provider. This includes, but is not limited to, wind field vectors (wind speed and direction), ambient temperature, and relative humidity. Simultaneously, the system retrieves biological model parameters related to the target pest "thrips" from its internal agricultural knowledge base, including the suitable temperature range for its migration (set to 18-30°C), survival humidity threshold, and passive diffusion coefficient at different wind speeds.

[0054] S12. Crop population micro-dynamic behavior data: Specifically, at least one high frame rate industrial camera (120fps in this embodiment) is evenly distributed at the top of each crop planting area. The camera is vertically downward, continuously capturing dynamic video streams of the cucumber plant canopy below. The high frame rate setting aims to capture rapid micro-tremors of leaves that are imperceptible to the human eye, caused by minute disturbances (such as insects taking off, landing, or crawling).

[0055] S13. Collecting individual crop biochemical health data, specifically through a multispectral fluorescence imaging system mounted on a track-mounted intelligent inspection robot. This system scans each crop planting area at night (to avoid sunlight interference) along a preset path. During scanning, the system sequentially activates narrowband LED light sources of different wavelengths (set to 365nm ultraviolet light) to excite the crop leaves, and simultaneously acquires fluorescence images using narrowband filters (including 440nm and 525nm) corresponding to specific fluorescence emission peaks.

[0056] S14. Greenhouse Internal Environment Data Acquisition and Data Structuring: A sensor array deployed within the greenhouse collects real-time temperature, humidity, and regional wind field vector data for each area. All collected data (S11-S14) are precisely timestamped and spatially identified (crop planting area ID) and sent to a central database. The system aligns and correlates these data from different sources and in various formats, forming a structured basic dataset with "crop planting area" as the basic unit and time sequence as the data sequence, providing unified and standardized data input for subsequent analysis.

[0057] S2, Status and Risk Assessment Module: Extracts high-dimensional feature indices from structured basic datasets;

[0058] This step is the core analysis engine of the system. Based on the inherent physical and biological mechanisms of pest and disease occurrence and development, it transforms seemingly unrelated data into risk indices with clear implications. Its internal implementation is as follows:

[0059] Part 1, Migration Risk Index, denoted as Mr, is calculated to predict external threats;

[0060] S21. Threat Source Simulation and Calculation (Lagrange Particle Diffusion Model): This step aims to quantify the probability of external thrips populations migrating with airflow and invading greenhouses. The system loads the acquired regional wind field vector data into a two-dimensional or three-dimensional virtual weather field. Subsequently, at the upwind boundary of the weather field, a large number (set to 100,000) of "Lagrange particles" representing thrips are released. The Lagrange particle diffusion model simulates and calculates the trajectory of each particle time-step based on real-time wind field data, combined with gravity, air resistance, and the thrips' own biological parameters (such as the probability of inactivation under unfavorable temperature and humidity).

[0061] This step aims to quantify the probability of external thrips populations migrating with airflow and invading the greenhouse. Before the simulation begins, the system first performs detailed environmental modeling and initial settings: Greenhouse 3D geometric modeling and ventilation area division: The system first loads a high-precision 3D geometric model of the target greenhouse (e.g., imported via CAD or BIM files) from a pre-set greenhouse geographic information database. This model precisely defines the greenhouse's geographic coordinates, orientation, dimensions, and all structural details. Crucially, the model independently divides and marks all potential pest invasion channels—i.e., ventilation openings. For example, in a greenhouse with continuous roof skylights and staggered side wall windows, the ventilation openings are marked as multiple independent absorption surfaces, including Receptor-Roof-01, Receptor-Roof-02, ... (corresponding to various sections of the roof skylights);

[0062] Receptor-South-Wall-01, Receptor-South-Wall-02, ... (corresponding to the ventilation windows on the south wall);

[0063] ReceptorNorthWall01, ReceptorNorthWall02, ... (corresponding to the wind vents on the north wall); each absorption surface is a geometric polygon with a defined spatial location and area, serving as the target for subsequent particle collision detection.

[0064] The system constructs a three-dimensional grid computing domain covering a large surrounding area (e.g., 2 kilometers in length, width, and height) using the center of the greenhouse 3D model as the origin. Then, the system maps the regional wind field vector data, temperature, and humidity data for the next 24 hours, obtained via API, onto each node of this 3D grid using a spatial interpolation algorithm (such as Kriging interpolation), thus forming a virtual meteorological field that dynamically changes over time and reflects real atmospheric flow. At the upwind boundary of the virtual meteorological field, the system initializes and releases a large number (100,000 in this embodiment) of massless virtual particles representing "thrips," i.e., "Lagrange chiffres particles." Each particle is assigned an initial position upon release, and its distribution can be weighted according to prior knowledge (e.g., if a large area of ​​farmland is known to exist upwind, the particle release density in that direction will be higher).

[0065] S22. Invasion Flux Statistics and Index Generation: In the virtual space, the greenhouse is represented as an "absorber" with a specific geometry and vent locations. During the simulation, any particle whose trajectory collides with the greenhouse vent location is considered a successful "invasion event." By counting the total number of particles that successfully invade within a unit simulation time (set to 24 hours), the system can obtain a migration risk index Mr, which characterizes the pest invasion flux.

[0066] After the virtual environment is set up, the system initiates an iterative calculation process lasting 24 simulation hours, progressing in extremely small time steps (e.g., Δt = 0.1 seconds). Within each time step Δt, the system performs the following calculations and judgments for each still-active Lagrange particle:

[0067] Particle position update steps: For a particle at spatial position P(t) at time t, the instantaneous wind vector Vwind(P,t) at its location is calculated by interpolation. Then, combining the gravitational settling velocity Vgravity (a small, constant downward vector) and the random perturbation term Vrandom for simulating turbulence, the particle's displacement ΔP over time Δt is calculated. The particle's new position P(t+Δt) is determined by the following formula: P(t+Δt) = P(t) + (Vwind(P,t) + Vgravity + Vrandom) × Δt;

[0068] Collision detection and intrusion event determination: The system detects whether the trajectory line segment formed by connecting P(t) and P(t+Δt) intersects with any geometric polygon in the greenhouse model that is marked as an "absorbing surface".

[0069] If an intersection occurs, it is considered a successful "intrusion event." The system records which specific absorbing surface the particle hit (e.g., Receptor-South-Wall-02) and increments the intrusion counter associated with that absorbing surface. Subsequently, the particle is marked as "absorbed" and removed from subsequent calculations to save computational resources.

[0070] Biological inactivation determination: The system acquires the ambient temperature T and relative humidity H at the particle's new position P(t+Δt). If T or H falls within a preset "unsuitable survival zone" (e.g., temperature below 5°C or above 40°C), the system activates the particle's "inactivation timer." If the particle remains within the unsuitable zone for more than a preset duration (e.g., 30 minutes), it is determined to be biologically inactivated. The particle is marked as "inactivated" and removed from subsequent calculations.

[0071] Boundary condition judgment: If the new position P(t+Δt) of the particle exceeds the boundary of the entire three-dimensional mesh computational domain, it is determined that it has "escaped" and is also removed from subsequent calculations.

[0072] This iterative process will continue until the simulation time reaches the set 24 hours. After the simulation ends, the system counts the total number of particles marked as "absorbed," Nabsor. Finally, the migration risk index Mr, which characterizes the pest invasion flux, is calculated as follows: Mr = Nabsor / Ntotal; where Ntotal is the initial total number of released particles (100,000).

[0073] For example, if in a 24-hour simulation, a total of 3,500 particles enter the greenhouse through various vents, then Mr = 3500 / 100000 = 0.035.

[0074] An example of the three-level risk threshold system for the migration risk index Mr is shown in Table 1:

[0075] Table 1: Three-level risk thresholds for the migration risk index Mr

[0076]

[0077] According to the threshold system mentioned above, the calculated result Mr=0.035 (i.e., 3.5%) in the example falls within the "Third Risk Level: High-Level Invasion Risk" range and significantly exceeds the 2.0% threshold. This means that the system predicts that the current meteorological conditions have created an extremely efficient invasion channel. The potential thrips population in the upwind area has a 3.5% probability of directly invading the greenhouse with the airflow. This is no longer a matter of "possibility," but a high-probability, deterministic event that is about to occur or is already occurring. At this point, the system's decision support module will immediately trigger the highest level of alert and propose a series of decisive proactive defense measures before the pests establish themselves and reproduce, such as:

[0078] Alert: "Level 3 migration risk detected, Mr=0.035. Large-scale thrips invasion is highly probable. Immediate action recommended." Instruction 1 (Automation): "Close south-side wind windows 1-5 and north-side skylights 2-4." Instruction 2 (Task Assignment): "Assign task to crop planting areas C7, C8, C15 (south-side window areas): Release 2000 predatory mites per area." Data Linkage: "The system has increased the Cf index monitoring frequency to the highest level, tracking invasion confirmation signals in real time." In this way, the system successfully transforms an abstract probability value (Mr) into a clear, hierarchical action guide directly linked to specific countermeasures, thus achieving true intelligent decision support.

[0079] Part Two: Population Response Dispersion (Cf) Calculation – Sensing Microscopic Intrusions; The goal of this step is to accurately identify the unique “stress signals” triggered by the collective activity of microscopic insect populations such as thrips from the microscopic dynamics of the crop canopy. The core logic is that the activity of a single insect is noise, while the coordinated activity of a micro-population generates a detectable “stress field” with specific spatiotemporal characteristics on the crop leaves. The following is a detailed calculation process and a specific data example demonstrating how the system processes data from crop planting area C7 within one second at 2:15:30 PM:

[0080] S23. Population Micro-motion Field Calculation (Optical Flow Method): This step aims to detect the population micro-stress response triggered by thrips alighting or crawling on crop leaves. The system employs a robust optical flow estimation algorithm (such as the Farnebäck algorithm) to process the high frame rate video stream for each crop planting area. This algorithm calculates the motion vector of each pixel between two adjacent frames in the video, thereby generating a dense optical flow vector field that can finely describe how the entire crop canopy "flows" or "trembles".

[0081] S24. Stress Signal Purification and Energization: The core principle is that the leaf movement triggered by the activity of a single thrips is random and weak. However, when a small population moves simultaneously, it triggers synchronized, coordinated movements of multiple leaves within a region, exhibiting a specific frequency and spatially manifested as "collective extension" or "avoidance." The system first calculates the spatial divergence of the optical flow vector field. A positive divergence pulse indicates that the pixels in that region are collectively "diverging" outward, which closely matches the extension movement of leaves when startled. Subsequently, the system performs a Fast Fourier Transform (FT) or Wavelet Transform on the time series of divergence values ​​in that region, transforming it from the time domain to the frequency domain. According to insect behavioral studies, the disturbances caused by thrips crawling or feeding have a characteristic frequency range (set to 5-15Hz). The characteristic frequency range of 5-15Hz is an empirical parameter derived from the prior knowledge base of insect behavior. This value is not arbitrary but based on fundamental research data obtained through high-precision experiments (such as laser vibrometers and high-speed photography) in the field of agricultural entomology. Studies have shown that when tiny insects like thrips crawl, feed, or land on leaves, the interaction between their legs and the leaf surface triggers weak vibrations within a specific frequency range. The 5-15Hz range is identified as a biophysical fingerprint that distinguishes thrips activity from other sources of interference such as wind (typically <2Hz) and equipment vibration. By designing a bandpass filter in the frequency domain, the system can accurately extract the characteristic frequency band energy (Efreqband) of this range, effectively filtering out low-frequency or high-frequency noise caused by wind or equipment vibration.

[0082] The system retrieves video clips recorded from time t to t+1 seconds from a camera deployed on top of crop planting area C7. Since the frame rate is 120fps, this 1-second clip contains 120 consecutive frames: Frame0, Frame1, ..., Frame119.

[0083] Data example (a 3x3 pixel neighborhood of FlowField0):

[0084] (0.1,0.2)(0.3,0.1)(0.2,-0.1);

[0085] (-0.1,0.4)(0.2,0.3)(0.4,0.2) / / Center pixel (i,j);

[0086] (-0.2,0.2)(-0.1,0.1)(0.1,0.0);

[0087] This process is continuous, generating 119 optical flow fields: FlowField0, FlowField1, ..., FlowField118. The system processes the video frame by frame. Taking the first pair of adjacent frames, Frame0 and Frame1, as an example, the Farnebäck algorithm is used to calculate a dense optical flow field, FlowField0. This is a two-dimensional matrix with the same resolution as the image, where each element is a two-dimensional motion vector (vx, vy), representing the direction and velocity of the pixel's movement between the two frames.

[0088] For each optical flow field, the system calculates its spatial divergence, generating a divergence field. Divergence measures the intensity of the "source" or "sink" at each point in a vector field. In the current scene, a positive divergence value means that a pixel is "diverging" outward from that point, which closely matches the stretching or vibrating motion of a leaf when startled. The discrete calculation formula for divergence is: Div(i,j)=(vx(i+1,j)-vx(i-1,j)) / 2+(vy(i,j+1)-vy(i,j-1)) / 2; where (i,j) represents the coordinates in the image pixel grid, i is the horizontal coordinate (column), and j is the vertical coordinate (row). We are currently calculating the divergence value of the center pixel (i,j). vx(i,j) refers to the horizontal component (x-axis) of the motion vector at pixel (i,j). A positive value indicates movement to the right, and a negative value indicates movement to the left. vy(i,j) refers to the vertical component (y-axis) of the motion vector at pixel (i,j). A positive value indicates downward movement, and a negative value indicates upward movement.

[0089] The first part of the formula is the horizontal divergence contribution, (vx(i+1,j)-vx(i-1,j)) / 2; where vx(i+1,j) is the horizontal motion component of the right neighbor (coordinate i+1,j) of the center pixel, and vx(i-1,j) is the horizontal motion component of the left neighbor (coordinate i-1,j) of the center pixel. If the right neighbor moves to the right at a greater speed than the left neighbor moves to the right at a greater speed than the left neighbor, then the result of vx(i+1,j)-vx(i-1,j) is a positive number. This indicates that the pixel flow is being "stretched" in the horizontal direction, exhibiting a horizontal divergence. Conversely, if the result is negative, it indicates that the pixel flow is being "compressed" in the horizontal direction.

[0090] The second part of the formula is the vertical divergence contribution: (vy(i,j+1)-vy(i,j-1)) / 2; vy(i,j+1) is the vertical motion component of the neighbor below the center pixel (coordinate i,j+1), and vy(i,j-1) is the vertical motion component of the neighbor above the center pixel (coordinate i,j-1). If the downward movement speed of the lower neighbor vy(i,j+1) is greater than the downward movement speed of the upper neighbor vy(i,j-1) (or in other words, the upward movement speed of the upper neighbor is faster), then the result of vy(i,j+1)-vy(i,j-1) is a positive number. This indicates that the pixel flow is also being "stretched" in the vertical direction, exhibiting a vertical divergence. Conversely, if the result is negative, it indicates that the pixel flow is being "compressed" in the vertical direction. Div(i,j) adds the horizontal divergence / contraction degree to the vertical divergence / contraction degree to obtain the overall divergence intensity of the center pixel (i,j).

[0091] Div(i,j) > 0 (positive divergence) means that the amount flowing out of point (i,j) is greater than the amount flowing in. On the graph, this represents an expanding or diverging pattern, like the "source" of a water flow. Div(i,j) < 0 (negative divergence) means that the amount flowing into point (i,j) is greater than the amount flowing out. This represents a contracting or converging pattern, like the "sinking point" of a water flow.

[0092] Div(i,j)=0 (dispersion) means that the inflow and outflow are equal, and the flow at that point is "incompressible", like uniform translation of the whole.

[0093] Example data (calculate the divergence value of the center pixel (i,j) of the above FlowField0): Div(i,j)≈(0.4-(-0.1)) / 2+(0.1-0.4) / 2=0.25-0.15=0.1;

[0094] The positive value of 0.1 indicates that at this moment, the central pixel region has a slight expansion trend. The system performs this calculation for all pixels of each optical flow field, generating 119 divergence fields DivField0, DivField1,..., DivField118.

[0095] Instead of analyzing individual pixels, the system divides the image into multiple small regions (e.g., a 10x10 pixel grid). For each small region, the average divergence value of all pixels within it is calculated. This results in a "mean divergence time series" of 119 data points for each small region. Example data (time series DivTimeSeriesA for a small region ROIA): DivTimeSeriesA=[0.10,0.15,0.08,-0.05,-0.12,0.09,0.18,...] (119 points in total). This sequence reflects the "breathing" or "pulsating" pattern of the ROIA region within 1 second. Applying a Fast Fourier Transform (FFT) to DivTimeSeriesA converts it from a time-domain signal to a frequency-domain spectrum. The frequency-domain spectrum shows the distribution of signal energy at different frequencies. A peak appears in the 0-2Hz band. This corresponds to the slow, regular airflow (wind disturbance noise) caused by the greenhouse ventilation system. In the 5-15Hz frequency band, a clearly discernible "energy spike," though not as high as the low-frequency peak, appears. Based on prior knowledge of insect behavior, this is precisely the characteristic frequency band (target signal) of the high-frequency micro-vibrations in leaves caused by thrips and other tiny insects crawling or feeding. In other frequency bands, the energy distribution is relatively flat, belonging to random noise. A digital bandpass filter is applied to retain only the signal within the 5-15Hz range of the frequency spectrum, filtering out signals from all other frequencies. Then, the total energy within this characteristic frequency band is calculated.

[0096] Data Example: The total signal energy within the 5-15Hz characteristic frequency band is calculated and defined as the characteristic frequency band energy, Efreqband = 85.3; simultaneously, the total energy of the original signal before filtering, Etotal, is calculated to be 110.2. Finally, by normalizing the characteristic frequency band energy and the total energy, a dimensionless exponent Cf in the range [0,1] is obtained. Data Example: Cf = Efreqband / Etotal = 85.3 / 110.2 ≈ 0.774; the calculated Cf = 0.774. This value exceeds the system's preset action threshold of 0.7. Therefore, the system made the following judgment: "At 2:15:30 PM, a significant population thrips divergence signal (Cf=0.774>0.7) was detected in crop planting area C7. It is highly suspected that a small population of thrips is conducting group activities on the crop canopy in this area. It is recommended to immediately dispatch an orbital robot to area C7 and use a multispectral fluorescence imaging system for close-range scanning to confirm whether there are any early pathological reactions." Through this series of precise calculations, the system successfully identified insect activity events in the microscopic world from a macroscopic, noisy video stream, achieving real-time perception of invasion events.

[0097] Part Three: Calculation of the Disease Progression Index (DPSI) – A Perspective on Plant Intrinsic Health;

[0098] S25. Characteristic Fluorescent Substance Demixing (Non-negative Matrix Factorization): This step aims to quantify the health status and disease development stage of plants through "biochemical fingerprinting." The biological basis is that healthy plants are rich in a class of phenolic substances with defensive functions (denoted as healthy component P), which emit strong blue fluorescence under ultraviolet excitation (with a peak at 440nm). When infected by pathogens such as downy mildew, the plant initiates a defensive response, producing another type of substance (such as certain phytotoxicants, denoted as pathological component D), which emits green fluorescence (with a peak at 525nm). Simultaneously, the metabolic activity of the pathogen decomposes healthy component P. Therefore, as the disease progresses from healthy to infected and then to severely infected, the blue fluorescence gradually weakens, while the green fluorescence gradually strengthens. The multispectral camera acquires a mixed spectrum of these two (or more) fluorescences. The system uses a non-negative matrix factorization (NMF) algorithm to decompose the acquired mixed spectral matrix into the product of two matrices: one representing the characteristic spectrum of the "pure" healthy component P and pathological component D, and the other representing their relative concentration at each pixel.

[0099] S26. Biochemical Kinetic Model Fitting and Exponential Output: Using the NMF algorithm, the system obtained the average concentration C(P) of the healthy component P in each crop planting area. avg The average concentration C(D) of pathological component D. avg The Disease Progression Index (DPSI) is defined as a dimensionless value that intuitively reflects the proportion of pathological components. Its calculation formula is as follows: DPSI = C(D) avg / (C(P) avg +C(D) avg +ε) where ε is a very small positive number used to prevent the denominator from being zero. This formula ensures that the DPSI value is always within the range of [0,1]. A DPSI value close to 0 represents a very healthy plant; while a value close to 1 represents a plant that is critically ill, with almost all healthy components depleted. Through extensive experimental calibration, a correspondence between DPSI and actual disease progression can be established;

[0100] Example data: A track-guided inspection robot scans a target leaf in zone C7. The system uses a 365nm ultraviolet LED to excite the leaf and simultaneously acquires fluorescence intensity images in two key bands: Image 440nm: Blue fluorescence image, mainly reflecting the content of the healthy component P. Image 525nm: Green fluorescence image, mainly reflecting the content of the pathological component D. Example data (fluorescence intensity values ​​of a 2x2 pixel area, in relative grayscale values): Image 440nm (blue): specifically [180, 175]; [150, 90]; Image 525nm (green): [25, 30]; [55, 110];

[0101] After data interpretation, it was observed that the upper left corner region exhibited strong blue fluorescence (180) and weak green fluorescence (25), indicating a healthy condition. Conversely, the lower right corner region showed significantly weakened blue fluorescence (90) and a sharp increase in green fluorescence (110), exhibiting clear disease characteristics. The system constructed an input matrix from the multispectral data of these four pixels (each pixel having a 440nm value and a 525nm value) and then applied the NMF algorithm. The algorithm aims to decompose the mixed fluorescence signal into the contribution of the "pure" healthy component P and the pathological component D. After processing with the NMF algorithm, the relative concentration coefficients of the two components at each pixel were output. The output results are as follows:

[0102] Pixel (0,0): C(P)=0.95, C(D)=0.05 (Very healthy);

[0103] Pixel (0,1): C(P)=0.92, C(D)=0.08 (slight stress);

[0104] Pixel (1,0): C(P)=0.70, C(D)=0.30 (early infection);

[0105] Pixel (1,1): C(P)=0.40, C(D)=0.60 (increased infection); The system calculates the average relative concentration of the two components in the tested leaf area.

[0106] Data Example: Average Concentration C(P) of Healthy Components avg =(0.95+0.92+0.70+0.40) / 4=0.7425;

[0107] Average concentration of pathological components C(D) avg =(0.05+0.08+0.30+0.60) / 4=0.2575; Substitute the calculated average concentration into the DPSI formula DPSI=C(D) avg / (C(P) avg +C(D) avg+ε)=0.2575 / 1.0≈0.258;

[0108] Through the two major steps S1 and S2 mentioned above, the system successfully abstracts the operating status of the greenhouse into three key, high-dimensional characteristic indices: Mr (how great is the external threat), Cf (whether the threat has appeared locally), and DPSI (how strong is the plant's own resistance).

[0109] Compared with existing technologies, the beneficial effects of this embodiment lie in its ability to solve the core problems of traditional pest and disease management methods, such as lack of predictive ability, delayed diagnosis, and inability to distinguish cause and effect, through a comprehensive perception and analysis process that is "from macro to micro" and "from external to internal." Specifically, this method can deeply integrate data from three completely different modalities and scales: external macro-level meteorological data, micro-level stress behavior of crop populations, and internal biochemical states of individual crops, to construct a three-dimensional cognitive framework capable of dynamically assessing pest and disease risks. The underlying technology lies in this embodiment, which provides a profound understanding of the entire chain of pest and disease outbreaks: First, by using the migration risk index Mr based on the Lagrange model, predictive perception of the "pathogenic factor"—external pest invasion—is achieved; second, by using the swarm response laser divergence Cf based on photofluid field analysis, highly sensitive real-time capture of the "invasive behavior"—the collective activity of tiny pests on the crop surface—is achieved; and finally, by using the disease progression index DPSI based on multispectral fluorescence unmixing, precise quantitative diagnosis of the "pathological result"—the internal biochemical state of the plant after infection—is achieved. These three indices are independent yet complementary, and their combination provides solid, quantifiable data input for the subsequent collaborative risk prediction module (such as in Embodiment 2) to determine "to what extent an external threat will lead to a disaster at an internal vulnerability," thereby elevating pest and disease management from a passive "locking the stable door after the horse has bolted" to a proactive "prevention before the event" approach.

[0110] Example 2

[0111] The collaborative risk prediction module is used to predict the risk score of disease outbreaks in each crop planting area in the greenhouse in the future by using a preset disease transmission prediction model and taking the status and risk feature dataset as input.

[0112] The collaborative risk prediction module uses a cellular automata model to predict disease transmission. Specifically, this module executes the following steps: S411, constructing a two-dimensional or three-dimensional cellular grid corresponding to the spatial layout of the greenhouse crop planting area; S412, assigning the disease progression index DPSI calculated by the status and risk assessment module to each cell as a basic susceptibility attribute characterizing its initial health state; S413, within a sliding time window, calculating the cross-correlation function between the migration risk index Mr and the population response laser divergence Cf. When its peak value exceeds a preset invasion threshold, it is determined to be an invasion event, and... Cells in the crop planting area corresponding to the event are marked as the initial source of infection; S414, based on the preset biological evolution rules, the cellular automata model is iterated in the time domain; the evolution rules include: a cell marked as a source of infection has a probability of spreading its infection status to neighboring cells in the next time step; after a healthy cell is spread, the probability of it becoming infected is an increasing function of its own basic susceptibility attribute; S415, after the preset number of iterations is completed, the cumulative probability of each cell being marked as infected is calculated to form a set containing the risk score of future disease outbreaks in each crop planting area.

[0113] The collaborative risk prediction module also includes a risk score aggregation unit and a time-series risk tracking unit. The risk score aggregation unit is used to take the disease outbreak risk score value generated in step S415 as the regional future risk score value at the current moment for each crop planting area. The time-series risk tracking unit is used to receive and cache multiple regional future risk score values ​​output by the risk score aggregation unit in a continuous prediction period, and construct a time-series risk score sequence with a preset length according to the timestamp order of the multiple regional future risk score values, so as to analyze the evolution trend of the risk.

[0114] Construct and initialize the cellular automaton mesh (S411 and S412).

[0115] The system first constructs a cellular grid that completely corresponds to the spatial layout of the crop planting area in greenhouse A. To simplify the example, we assume that area A is a 3x3 layout with a total of 9 planting areas (C1 to C9).

[0116] Then, the system converts the DPSI values ​​of each region calculated in real time by the "State and Risk Assessment Module" into the "Basic Susceptibility" attribute of each cell. In this example, S is set to DPSI. Data example: Assuming that at the current moment, the DPSI values ​​of each region evaluated by the system are as follows, this constitutes the initial susceptibility map of the cellular automata, as shown in Table 2.

[0117] Table 2: Initial Susceptibility Map of Cellular Automata

[0118]

[0119] S413. Determine the initial source of infection and continuously analyze the time series of Mr and Cf indices. When a high Mr value (external threat) is detected, and a significant peak in the Cf value of a certain area occurs within a short period of time, the system determines a successful "intrusion event" and marks the corresponding cell as the initial source of infection.

[0120] Data example: At time T=0: The system detected a high migration risk Mr=0.035 (third risk level).

[0121] After 1 hour (T=1 hour): The system detected that the population laser divergence Cf in region C7 surged to 0.774, exceeding the intrusion confirmation threshold. The system confirmed an intrusion event in region C7 and marked its state in the cellular automaton as "Infected (I)". The initial state of other cells was "Healthy (H)".

[0122] The initial state map StatusGrid(t=0) is shown in Table 3.

[0123] Table 3: Initial State Map StatusGrid (t=0)

[0124]

[0125] S414, Temporal Iteration and Disease Spread Simulation Steps: Specifically, an iterative simulation process is initiated to predict disease spread over the next 12 hours (assuming one iteration per hour, for a total of 12 steps). The evolutionary rules are as follows: At each time step, an "infected" cell has a base probability of Pspread = 15% to attempt to infect its neighboring "healthy" cells (in this example, the four neighbors: top, bottom, left, and right). When a "healthy" cell is attempted to be infected, its eventual infection probability, Pinfect, is positively correlated with its susceptibility, S. The formula is: Pinfect = Pspread × (1 + S). This means that cells with higher DPSI are more susceptible to infection.

[0126] Data Example (simulating the first two time steps): At iteration step t=1, infection source C7 (I) attempts to infect its neighbors C4 and C8. For C4 (S=0.04): Pinfect=0.15×(1+0.04)=15.6%. The system makes a random decision using the Monte Carlo method. Assuming this decision fails, C4 remains healthy. For C8 (S=0.31): Pinfect=0.15×(1+0.31)=19.65%. Due to its higher susceptibility, the probability of infection is also higher. Assuming this decision succeeds, C8's state changes to I. As shown in Table 4.

[0127] Table 4: Initial state map StatusGrid (t=1) at iteration step t=1

[0128]

[0129] Perform iteration step t=2:

[0130] Now both C7 and C8 are sources of infection. C7 attempts to infect C4 again; C8 attempts to infect its neighbors C5 and C9. For C9 (S=0.28): Pinfect=0.15×(1+0.28)=18.9%. Assuming the determination is successful, C9's state changes to I. Other determinations assume failure. See Table 5.

[0131] Table 5: Initial state map StatusGrid (t=2) for iteration step t=1

[0132]

[0133] The system will independently repeat the above simulation process a large number of times (e.g., 1000 times) to obtain statistically reliable probability results. Step 4: Generate disease outbreak risk score (S415 & Risk Score Aggregation Unit)

[0134] After all 1000 simulations were completed, the system counted how many times each cell was marked as "infected" at the end of the 12 time steps. This number divided by the total number of simulations yielded the risk score for future disease outbreaks in that area, as shown in Table 6.

[0135] Table 6: Final Disease Outbreak Risk Score

[0136]

[0137] C7, as the initial source of infection, has a risk score of 1.0 (100%). Its neighboring and highly susceptible areas, C8 and C9, also have extremely high risk scores. It is noteworthy that even initially healthy areas such as C4, C5, and C6, due to their location along the transmission path, exhibit significant future risks.

[0138] S413, Time Series Risk Tracking and Trend Analysis (Time Series Risk Tracking Unit); This unit is responsible for collecting the RiskScoreGrid generated for each forecast period (e.g., a complete forecast every 6 hours) and performing trend analysis on the risk evolution of a specific region, as shown in Table 7.

[0139] Table 7: Time series of risk scores for region C6:

[0140]

[0141] T = -12 hours (previous forecast): RiskScore (C6) = 0.12; T = -6 hours (previous forecast): RiskScore (C6) = 0.25; T = 0 (current forecast): RiskScore (C6) = 0.45; Time-series risk score sequence: [0.12, 0.25, 0.45]; After analyzing this time-series sequence, the system identified that the risk score of region C6 is accelerating. This indicates that the "spreading vanguard" of the disease is rapidly advancing into this region. The system will immediately generate a high-level warning: Warning: The disease outbreak risk score (0.45) in region C6 is showing an exponential growth trend in the next 12 hours. It is predicted that the disease will spread from region C9 to C6 within 6-8 hours. It is recommended to immediately strengthen the intervention in the isolation zone between C6 and C9, and consider the preventive application of biological agents to crops in region C6.

[0142] Through this entire process, the system not only "sees" the current state of the disease, but also "foresees" its future trajectory, elevating defensive measures from passive "locking the stable door after the horse has bolted" to proactive "planning ahead".

[0143] The core technology of this embodiment lies in the use of a cellular automata model to construct a dynamic, multi-dimensional feature-based disease spatiotemporal propagation prediction system. This system maps the physical space of the greenhouse into a cellular grid and uses the Disease Development Stage Index (DPSI) to assign a quantified initial susceptibility to each cell. Its operating mechanism does not involve random simulation initiation; instead, it accurately determines the "initial source of infection" by calculating the cross-correlation function between migration risk (Mr) and population stress (Cf), thus linking the prediction initiation point to the actual invasion event. In the time-domain iteration, the model extrapolates according to biological rules, where the probability of infection transmission is an increasing function of the cell's own susceptibility, making the prediction process closer to biological reality. The resulting benefit is that the system elevates diagnosis from assessing the static status quo to dynamically predicting future risks. It can generate clear future disease outbreak risk scores for each planting area, replacing vague macro-level early warnings and providing data support for prioritizing intervention measures. Furthermore, through time-series risk tracking, the system can also identify the evolution trend of risks, such as "accelerated rise," thereby enabling proactive deployment in areas that are about to become high-risk "spread fronts." This achieves a shift from passive response to proactive and precise defense, providing a reliable basis for optimizing resource allocation and improving prevention and control efficiency.

[0144] Example 3

[0145] The dynamic risk zone division module is used to divide the entire greenhouse into high-risk warning zones, medium-risk attention zones, and low-risk monitoring zones in three-dimensional space based on disease outbreak risk scores, thereby generating a dynamically updated three-dimensional risk map. The module includes a risk level mapping unit and a three-dimensional map generation unit. The risk level mapping unit is used to preset a first risk threshold and a second risk threshold, and to determine the disease outbreak risk score for each crop planting area to obtain a risk level result, including: when the disease outbreak risk score is lower than the first risk threshold, the risk level is mapped to low-risk monitoring; when the disease outbreak risk score is between the first and second risk thresholds, the risk level is mapped to medium-risk attention; when a disease outbreak occurs... When the risk score is not lower than the second risk threshold, the risk level is mapped to a high-risk warning level. The 3D map generation unit is used to traverse the risk level results of each crop planting area based on the pre-configured greenhouse 3D geometric model, and perform the following spatial attribute assignment and visualization rendering operations: if the risk level is low-risk monitoring level, the corresponding spatial area is divided into a low-risk monitoring area and rendered as a green visual marker; if the risk level is medium-risk attention level, the corresponding spatial area is divided into a medium-risk attention area and rendered as a yellow visual marker; if the risk level is high-risk warning level, the corresponding spatial area is divided into a high-risk warning area and rendered as a red visual marker; all rendered areas are aggregated to generate a dynamically updated 3D risk map.

[0146] S41. The risk level mapping unit sets risk thresholds, and then maps the RiskScores calculated in the previous embodiment (see Table 6) to the corresponding risk levels one by one. Preset thresholds: First risk threshold (low-medium risk boundary), preferably set to 0.30; Second risk threshold (medium-high risk boundary), preferably set to 0.70; Map Table 6 to the corresponding risk levels one by one to obtain the risk level mapping table, as shown in Table 8.

[0147] Table 8: Risk Level Mapping Table

[0148]

[0149] S42. The 3D map generation unit renders the aforementioned risk levels and colors to the corresponding spatial areas based on the geometric model of the greenhouse, generating a dynamically updated 3D risk map.

[0150] Data Example: 3D Risk Map (Conceptual Description). The system presents a 3D model of the greenhouse on the user interface. Specifically, crop racks or plots in areas C1, C2, and C3 are rendered in green and marked as "Low-Risk Monitoring Zones".

[0151] The spaces representing zones C4, C5, and C6 are rendered in yellow and marked as "medium-risk concern zones".

[0152] The spaces representing zones C7, C8, and C9 were rendered in a striking red and marked as "high-risk warning zones".

[0153] This map is dynamically refreshed at the end of each forecast period, visually showing the contraction or expansion of risk areas.

[0154] In this embodiment, the core technology lies in establishing a bridge that transforms quantitative risk data into intuitive spatial situational awareness. The principle is based on a two-stage processing flow: First, by using preset dual thresholds, the continuous disease outbreak risk score output by the collaborative risk prediction module is discretized into three distinct risk levels: low, medium, and high, effectively reducing the dimensionality and qualitatively defining complex data. Second, based on the pre-configured three-dimensional geometric model of the greenhouse, these risk level results are bound to the specific spatial location of the crop planting area and rendered with differentiated visual identifiers such as red, yellow, and green. The beneficial effect of this system is that it transforms an abstract numerical matrix into a clear three-dimensional risk map, allowing managers to quickly grasp the risk distribution, key areas, and spatial relationships of the entire greenhouse without interpreting complex data. This intuitive visualization reduces the cognitive load of decision-making and improves the efficiency of situational awareness. Simultaneously, the map's dynamic updating capability reflects the evolution of risk areas in real time, providing clear spatial guidance and operational scope definition for subsequent coordinated intervention decisions, serving as a crucial link between predictive analysis and precise execution.

[0155] Example 4

[0156] The linkage intervention decision-making module receives intervention instructions generated based on a 3D risk map and executes them as follows: generating a preliminary intervention strategy; simultaneously calculating the matching degree between the resources required for the execution of the preliminary intervention strategy and the currently available resources to construct a resource compliance score for quantifying the feasibility of the strategy; evaluating the resource compliance score, and if the resource compliance score is lower than a preset compliance threshold, iteratively revising the preliminary intervention strategy through a local strategy adjustment mechanism until the new resource compliance score corresponding to the revised strategy meets the requirements, thereby determining it as the final intervention strategy.

[0157] The coordinated intervention decision-making module includes a preliminary strategy generation subunit. This subunit receives a 3D risk map as input and generates preliminary intervention strategies for each crop planting area by consulting a pre-defined intervention rule knowledge base. The specific steps are as follows: traverse each crop planting area in the 3D risk map; if the current area is a high-risk warning area, match and assign a first-type intervention action from the knowledge base, which includes precise spraying with therapeutic agents; if the current area is a medium-risk concern area, match and assign a second-type intervention action from the knowledge base, which includes enhancing ventilation and dehumidification in the area; if the current area is a low-risk monitoring area, match and assign a third-type intervention action from the knowledge base, which includes maintaining regular monitoring frequency; combine all areas and their assigned intervention actions to form a preliminary intervention strategy.

[0158] The coordinated intervention decision-making module also includes a resource compliance assessment unit. This unit is used to simultaneously calculate the degree of matching between the preliminary intervention strategy and the currently available resources. The specific steps include: S611, analyzing the preliminary intervention strategy and calculating the total amount of various resources required to execute all assigned first and second type intervention actions, including the total amount of medicine, equipment usage time, and man-hours; S612, querying the real-time inventory of various resources available in the greenhouse at the current moment from the preset resource management database; S613, for each type of resource, calculating the ratio of its real-time inventory to the total amount required for consumption, and determining the minimum value among all ratios as the resource compliance score.

[0159] The coordinated intervention decision-making module also includes a strategy correction unit. This unit is used to: trigger and execute an iterative correction loop instruction when and only when the resource compliance score is lower than a preset compliance threshold. This includes: locating the bottleneck resource type that causes the lowest resource compliance score; adjusting the initial intervention strategy according to preset correction priority rules, which include: first attempting to downgrade some second-type intervention actions in medium-risk concern areas to third-type intervention actions; if resources are still insufficient, further replacing some first-type intervention actions in high-risk warning areas with second-type intervention actions that consume less resources; based on the adjusted strategy, re-executing all steps of the resource compliance assessment unit to calculate a new resource compliance score; if the new resource compliance score is not lower than the compliance threshold, terminating the current iterative correction loop and determining the current adjusted strategy as the final intervention strategy; otherwise, continuing the correction.

[0160] The linkage intervention decision-making module is the "executor" of the entire system. It receives the three-dimensional risk map, automatically generates a set of intervention strategies with optimal resources, and can directly issue them to automated equipment or management personnel.

[0161] S51. Generate preliminary intervention strategies. The preliminary strategy generation sub-unit matches intervention actions for each region based on the newly generated 3D risk map and the preset intervention rule knowledge base.

[0162] Pre-set intervention rule knowledge base (partial):

[0163] Category 1 Action (High-Risk Area): "Precision Spraying of Therapeutic Agent A", Resource Consumption: 100ml of agent / area, equipment 0.5h / area, manual labor 0.5h / area.

[0164] Second type of action (medium risk area): "Turn on powerful ventilation and dehumidification", resource consumption: equipment 1.0h / area.

[0165] The third type of action (low-risk area): "Maintain routine monitoring", resource consumption: 0.1 h / area. As shown in Table 9.

[0166] Table 9: List of Preliminary Intervention Strategies

[0167]

[0168] S52. The resource compliance assessment unit calculates the total resources required to implement the above preliminary strategy and compares them with the currently available resources.

[0169] S511: Calculate the total amount of resources required. Specific data examples are as follows: Total amount of medicine: 3 high-risk areas × 100ml / area = 300ml; Equipment usage time: (3 high-risk areas × 0.5h / area) + (3 medium-risk areas × 1.0h / area) = 1.5h + 3.0h = 4.5h; Labor hours: (3 high-risk areas × 0.5h / area) + (3 low-risk areas × 0.1h / area) = 1.5h + 0.3h = 1.8h;

[0170] S512: Query the real-time inventory of available resources. Available drug A: 500ml; Total available equipment time (within the next 3 hours): 4.0h (a bottleneck resource is intentionally set here); Total available manpower hours (within the next 3 hours): 8.0h;

[0171] S513: Calculate the resource compliance score, specifically including the compliance of the medicine: 500ml / 300ml=1.67; the compliance of the equipment: 4.0h / 4.5h=0.89; the compliance of the labor: 8.0h / 1.8h=4.44; the final resource compliance score is: min(1.67,0.89,4.44)=0.89;

[0172] The strategy correction unit evaluates the resource compliance score and iteratively corrects it according to the rules.

[0173] Preset compliance threshold: 0.95; current score 0.89 < compliance threshold 0.95. The conclusion is that the initial strategy is infeasible due to insufficient resources. A strategy correction loop is triggered. During the first iteration, the bottleneck resource is identified: "Equipment usage time" (lowest compliance). The correction rule (priority 1) is executed: some second-category actions in medium-risk areas are downgraded to third-category actions. The system selects to downgrade the intervention action for a medium-risk area (e.g., C4) from "activating powerful ventilation and dehumidification" (consuming 1.0h of equipment) to "maintaining routine monitoring" (consuming 0.1h of manual labor). New strategies are generated: C1-C3, C4 (4 areas) → routine monitoring; C5, C6 (2 areas) → ventilation and dehumidification; C7-C9 (3 areas) → spraying chemicals.

[0174] Recalculate the required resources: Medicine: 300ml (unchanged); Equipment: (2×1.0h)+(3×0.5h)=3.5h; Labor: (4×0.1h)+(3×0.5h)=1.9h; Recalculate the resource compliance score: Equipment compliance: 4.0h / 3.5h=1.14; Labor compliance: 8.0h / 1.9h=4.21; New resource compliance score: min(1.67,1.14,4.21)=1.14;

[0175] The loop terminates and the final strategy is determined. The new score of 1.14 is greater than or equal to the compliance threshold of 0.95. Therefore, the revised strategy has sufficient resources and is feasible. The revision loop terminates.

[0176] The final intervention strategy was determined, as shown in Table 10.

[0177] Table 10: Final Intervention Strategy Table

[0178]

[0179] The system ultimately outputs this resource-optimized and feasible strategy list and distributes it to the corresponding execution units (such as spraying robots, environmental control systems, and personnel task terminals) to achieve closed-loop automated management from perception and prediction to decision-making and execution.

[0180] Figure 1 The isometric view on the left illustrates the application scenario of this invention: a modern intelligent greenhouse equipped with various sensors and data acquisition devices. The crops, environment, and technical equipment inside the greenhouse together constitute the physical basis for the system's data acquisition and analysis. This corresponds to the function of the multi-source heterogeneous data fusion module, which acquires and preprocesses data within this environment to form a basic dataset. (Appendix) Figure 1 The technology roadmap on the right illustrates the system's data processing and decision-making process in detail through four logical block diagrams and their connections.

[0181] The first box, “Multi-source data fusion and index calculation,” represents the initial steps of the system execution. It corresponds to the in-depth processing of the basic dataset by the multi-source heterogeneous data fusion module, and the process by which the status and risk assessment module extracts migration risk index, population stress laser divergence, and disease stage index based on the dataset to construct a status and risk characteristic dataset.

[0182] The second box, "Simulation of Spatiotemporal Propagation of Diseases," represents the core link in the system's risk prediction process. It corresponds to the collaborative risk prediction module, which uses a pre-set disease propagation prediction model and takes the feature dataset generated in the previous step as input to predict the risk of disease outbreaks in future periods.

[0183] The third box, "3D Risk Map Generation," visually displays the results of risk visualization. It corresponds to the dynamic risk area division module, which dynamically divides the greenhouse space into areas with different risk levels based on the predicted risk score and generates a 3D risk map.

[0184] The fourth section, “Closed-Loop Intervention Strategy Decision-Making,” clarifies the system’s final output and decision-making mechanism. It corresponds to the linkage intervention decision-making module, which, based on the risk map, uses a closed-loop logic of “generation-evaluation-correction” and combines resource compliance verification to ultimately determine an optimal intervention strategy that is both accurate and feasible.

[0185] In this embodiment, the core technology lies in constructing a closed-loop decision engine that transforms risk perception into the optimal feasible strategy under resource constraints. Its principle employs an iterative logic of "generation-evaluation-correction": First, the system generates an idealized preliminary intervention strategy from a knowledge base based on a 3D risk map. Then, by calculating the ratio of required resources to real-time available resources, a resource compliance score determined by "bottleneck resources" is quantified to evaluate the strategy's feasibility. When the score falls below a threshold, the system does not simply abandon the strategy but initiates a correction loop running according to preset priority rules. This dynamically adjusts the strategy until its resource consumption matches the available resources by locally and systematically downgrading secondary intervention actions. The beneficial effect of this system is that it ensures the final intervention command output by the system is actually executable, avoiding decision-making stagnation due to insufficient resources. Through automated iterative correction, the system can form an optimized action plan by weighing the pros and cons under limited resource conditions, prioritizing the allocation of limited resources to the highest-risk areas. This not only improves resource utilization efficiency but also automates the entire process from risk prediction to specific task assignment, forming the decision-making and execution center of the entire intelligent management system.

[0186] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization.

[0187] The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used pandas library in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this in-memory data structure.

[0188] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A greenhouse pest and disease diagnosis and decision support system that fuses multi-source data, characterized in that, include: A multi-source heterogeneous data fusion module is used to obtain the basic dataset; The basic dataset includes: regional weather forecast and entomological model data obtained through external API interfaces; dynamic video stream data of crop populations returned by high frame rate cameras; multi-band fluorescence image data of crop leaves returned by multispectral fluorescence imaging systems; temperature, humidity, and regional wind field vector data obtained by environmental sensor arrays deployed in greenhouses; three-dimensional spatial division information of crop planting areas is extracted from a pre-set greenhouse geographic information database to divide the greenhouse into multiple continuous crop planting areas; and the collected basic dataset is spatially and temporally correlated with corresponding areas in multiple crop planting areas to form a structured basic dataset based on crop planting areas. The status and risk assessment module is used to calculate the migration risk index, population stress divergence, and disease stage index based on the basic dataset and focusing on crop physiological status and external environmental stress, in order to construct a status and risk characteristic dataset. The migration risk index is obtained by extracting the regional wind field vector, average temperature and relative humidity for the next 24 hours, and simulating the calculation using the Lagrange particle diffusion model to obtain the probability density of pests migrating to the current greenhouse area per unit time, which is defined as the migration risk index. The method for obtaining the swarm response laser divergence is as follows: extract dynamic video stream data of crop swarms, use optical flow estimation algorithm to calculate the spatial divergence of optical flow vector field, and perform time series analysis on the divergence value within a preset time window to extract specific frequency band energy values ​​that match the frequency of insect settlement disturbance, which are defined as swarm response laser divergence. The disease progression index is obtained by extracting multi-band fluorescence image data, separating and obtaining the concentration time-series curve, fitting the concentration time-series curve into a preset plant immune response biochemical kinetic model, and analyzing and obtaining the disease progression index. The collaborative risk prediction module is used to predict the risk score of disease outbreaks in each crop planting area in the greenhouse in the future by using a preset disease transmission prediction model and taking the status and risk feature dataset as input. The risk zone dynamic division module is used to divide the entire greenhouse into high-risk warning zones, medium-risk attention zones, and low-risk monitoring zones in three-dimensional space based on the disease outbreak risk score, and generate a dynamically updated three-dimensional risk map. The linkage intervention decision-making module is used to receive intervention instructions generated based on the 3D risk map and specifically execute them as follows: generating a preliminary intervention strategy; simultaneously calculating the matching degree between the resources required to execute the preliminary intervention strategy and the currently available resources to construct a resource compliance score for quantifying the feasibility of the strategy; if the resource compliance score is lower than the preset compliance threshold, the preliminary intervention strategy is iteratively corrected through a local strategy adjustment mechanism until the resource compliance score corresponding to the corrected new strategy meets the requirements, thereby determining it as the final intervention strategy.

2. The greenhouse pest and disease diagnosis and decision support system that fuses multi-source data according to claim 1, characterized in that: The status and risk assessment module includes a first risk index analysis unit and a second risk index analysis unit. The first risk index analysis unit is used to calculate the migration risk index for each crop planting area. The specific steps are as follows: S111. Extract the regional wind field vector, average temperature, and relative humidity for the next 24 hours from the structured basic dataset; S112. Call the preset target pest biological model, take the wind field vector as the propagation power, and take the average temperature and relative humidity as the suitable conditions for survival and reproduction. Use the Lagrange particle diffusion model to perform simulation calculations to obtain the probability density of pests migrating to the current greenhouse area per unit time, which is defined as the migration risk index. The second risk index analysis unit is used to calculate the population laser divergence for each crop planting area. The specific steps are: S211, extracting dynamic video stream data of the crop population from the structured basic dataset; S212. An optical flow estimation algorithm is used to process dynamic video stream data to generate a pixel-level optical flow vector field characterizing the micro-motion of the crop population. S213. Calculate the spatial divergence of the optical flow vector field, and perform time series analysis on the divergence values ​​within a preset time window to extract specific frequency band energy values ​​that match the frequency of insect settlement disturbances, which are defined as the population response optical flow divergence.

3. The greenhouse pest and disease diagnosis and decision support system that fuses multi-source data according to claim 2, characterized in that: The status and risk assessment module also includes a third risk index analysis unit. The third risk index analysis unit is used to calculate the disease stage index for each crop planting area. The specific steps are as follows: S311, extract multi-band fluorescence image data from the structured basic dataset; S312, process the multi-band fluorescence image data using a non-negative matrix factorization algorithm to separate the relative concentration time-series curves of at least two preset endogenous fluorescent substances related to plant defense response from the mixed fluorescence spectrum; S313, fit the relative concentration time-series curves to a preset plant immune response biochemical kinetic model, and obtain a dimensionless value in the range [0,1] that characterizes the current disease development stage through analytical solution or numerical optimization. This value is the disease stage index.

4. The greenhouse pest and disease diagnosis and decision support system integrating multi-source data according to claim 3, characterized in that: The collaborative risk prediction module uses a cellular automata model to predict disease transmission. Specifically, the module executes the following steps: S411, constructing a two-dimensional or three-dimensional cellular grid corresponding to the spatial layout of the greenhouse crop planting area; S412, assigning the disease progression index calculated by the status and risk assessment module to each cell as a basic susceptibility attribute characterizing its initial health status; S413, within a sliding time window, calculating the cross-correlation function between the migration risk index and the population response laser divergence. When its peak value exceeds a preset invasion threshold, it is determined as an invasion event, and the corresponding crop planting area cell is marked as the initial source of infection. S414. Based on preset biological evolution rules, perform time-domain iteration on the cellular automata model; The evolutionary rules include: a cell marked as an infection source has a probability of spreading its infection status to neighboring cells in the next time step; After a healthy cell is spread, the probability of it becoming infected is an increasing function of its own basic susceptibility attribute; S415, after the preset number of iterations is completed, the cumulative probability of each cell being marked as infected is calculated to form a set containing the risk score of future disease outbreaks in each crop planting area.

5. The greenhouse pest and disease diagnosis and decision support system that fuses multi-source data according to claim 4, characterized in that: The collaborative risk prediction module also includes a risk scoring aggregation unit and a time-series risk tracking unit; The risk score aggregation unit is used to generate a disease outbreak risk score for each crop planting area and use the generated risk score as the area's future risk score at the current moment. The time-series risk tracking unit is used to receive and cache the future risk score values ​​of multiple regions output by the risk score aggregation unit in a continuous prediction period, and construct the future risk score values ​​of multiple regions into a time-series risk score sequence of preset length according to the timestamp order, so as to analyze the evolution trend of risk.

6. The greenhouse pest and disease diagnosis and decision support system that fuses multi-source data according to claim 5, characterized in that: The risk zone dynamic division module includes a risk level mapping unit and a 3D map generation unit; The risk level mapping unit is used to preset a first risk threshold and a second risk threshold, and to determine the disease outbreak risk score for each crop planting area to obtain the risk level result, including: when the disease outbreak risk score is lower than the first risk threshold, the risk level is mapped to a low-risk monitoring level; when the disease outbreak risk score is between the first and second risk thresholds, the risk level is mapped to a medium-risk attention level; when the disease outbreak risk score is not lower than the second risk threshold, the risk level is mapped to a high-risk warning level; the 3D map generation unit is used to generate a 3D map of the greenhouse based on a pre-configured 3D map. The model iterates through the risk level results for each crop planting area and performs the following spatial attribute assignment and visualization rendering operations: if the risk level is low-risk monitoring level, the corresponding spatial area is divided into low-risk monitoring area and rendered as a green visual marker; if the risk level is medium-risk attention level, the corresponding spatial area is divided into medium-risk attention area and rendered as a yellow visual marker; if the risk level is high-risk warning level, the corresponding spatial area is divided into high-risk warning area and rendered as a red visual marker; all rendered areas are aggregated to generate a dynamically updated 3D risk map.

7. The greenhouse pest and disease diagnosis and decision support system that fuses multi-source data according to claim 6, characterized in that: The coordinated intervention decision-making module includes a preliminary strategy generation subunit, which receives a 3D risk map as input and generates a preliminary intervention strategy for each crop planting area by consulting a pre-set intervention rule knowledge base. The specific steps are as follows: traverse each crop planting area in the 3D risk map; if the current area is a high-risk warning area, match and assign a first-type intervention action from the knowledge base, which includes precise spraying with therapeutic agents; if the current area is a medium-risk concern area, match and assign a second-type intervention action from the knowledge base, which includes enhancing ventilation and dehumidification in the area; if the current area is a low-risk monitoring area, match and assign a third-type intervention action from the knowledge base, which includes maintaining the regular monitoring frequency; combine all areas and their assigned intervention actions to form a preliminary intervention strategy.

8. The greenhouse pest and disease diagnosis and decision support system that fuses multi-source data according to claim 7, characterized in that: The coordinated intervention decision-making module also includes a resource suitability assessment unit, used to simultaneously calculate the degree of matching between the initial intervention strategy and currently available resources. Specific steps include: S611. Analyze the initial intervention strategy and calculate the total amount of various resources required to execute all assigned first and second type intervention actions, including the total amount of drugs, equipment usage time, and man-hours. S612. Query the real-time inventory of various resources available in the greenhouse at the current moment from the preset resource management database; S613. For each type of resource, calculate the ratio of its real-time inventory to the total amount required for consumption, and determine the minimum value among all ratios as the resource compliance score.

9. The greenhouse pest and disease diagnosis and decision support system that fuses multi-source data according to claim 7, characterized in that: The coordinated intervention decision-making module also includes a strategy correction unit, used to: trigger and execute an iterative correction loop instruction when and only when the resource compliance score is lower than a preset compliance threshold, including: locating the bottleneck resource type that causes the lowest resource compliance score; adjusting the initial intervention strategy according to preset correction priority rules, which include: firstly attempting to downgrade some second-type intervention actions in medium-risk concern areas to third-type intervention actions; if resources are still insufficient, further replacing some first-type intervention actions in high-risk warning areas with second-type intervention actions; based on the adjusted strategy, re-executing all steps of the resource compliance assessment unit to calculate a new resource compliance score; if the new resource compliance score is not lower than the compliance threshold, terminating the current iterative correction loop and determining the current adjusted strategy as the final intervention strategy; otherwise, continuing to execute the correction.

Images