An AI-based method for early warning of crop diseases and pests

Abstract

This invention discloses an artificial intelligence-based method for early warning of crop diseases and pests, relating to the field of artificial intelligence technology. It includes an intelligent analysis terminal performing multimodal feature fusion analysis on a standardized training dataset to obtain a meteorological sensitivity coefficient, a crop resistance coefficient, and a disease and pest outbreak probability coefficient. These three coefficients are then input into a disease and pest risk quantification model, outputting a disease and pest risk index. An early warning decision terminal triggers multi-level early warning rules based on the disease and pest risk index, generating a customized early warning plan including emergency response suggestions. A feedback verification terminal receives feedback data from users after implementing control measures and dynamically updates the model parameters. This invention, through the early warning decision terminal triggering multi-level early warning rules based on the disease and pest risk index to generate a customized early warning plan including emergency response suggestions, achieves tiered response and precise policy implementation for disease and pest early warning, improving the targeting and effectiveness of control measures.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based method for early warning of crop diseases and pests. Background Technology

[0002] With global population growth and accelerated agricultural modernization, the importance of crop pest and disease control for ensuring food security and sustainable agricultural development is becoming increasingly prominent. Factors such as intensified climate change and more complex planting structures have led to a significant increase in the frequency and severity of pest and disease outbreaks. Traditional control models that rely on manual inspections and experience-based judgment are no longer sufficient to meet the needs of precise and intelligent management.

[0003] Currently, crop pest and disease monitoring and early warning technologies are gradually evolving from single-parameter monitoring to multi-source data integration. Some studies are attempting to use IoT sensors to collect environmental data such as temperature, humidity, and light intensity, and combine this data with statistical models for risk prediction. However, existing technologies generally suffer from limited data dimensions and insufficient depth of multimodal feature fusion, particularly lacking real-time dynamic monitoring of crop physiological states and pest and disease characteristics, resulting in poor model adaptability to complex field environments. Furthermore, traditional early warning methods are mostly based on fixed threshold rules, making it difficult to dynamically reflect the coupling effect of meteorological conditions, crop resistance, and pest and disease population dynamics, thus limiting the accuracy and timeliness of early warning results.

[0004] The main shortcomings of existing technologies are as follows: First, at the data acquisition level, there is a lack of systematic monitoring of key physiological and pathological indicators such as crop chlorophyll content, stem water potential, and pathogen spore concentration, making it impossible to construct a complete map of pest and disease causative factors. Second, at the model analysis level, a quantitative correlation mechanism between meteorological sensitivity, crop resistance, and the probability of pest and disease outbreaks has not been established, making it difficult to accurately quantify the risks under the synergistic effect of multiple factors. Third, at the system closed-loop level, there is a lack of a model iterative optimization mechanism driven by the feedback of control effects, resulting in the inability to dynamically adjust the early warning plan according to the actual control effect, which reduces the practicality and sustainability of the technology.

[0005] Therefore, it is essential to invent an artificial intelligence-based early warning method for crop diseases and pests to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide an artificial intelligence-based method for early warning of crop diseases and pests, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based method for early warning of crop diseases and pests, comprising an intelligent monitoring terminal, a data processing terminal, an intelligent analysis terminal, an early warning decision-making terminal, and a feedback verification terminal, specifically including the following steps:

[0008] S1. The intelligent monitoring terminal collects multi-dimensional environmental data at regular intervals during the crop growth cycle through an IoT sensor array deployed in farmland, forming a multi-source fusion dataset.

[0009] S2. The data processing terminal performs outlier removal and normalization on the multi-source fusion dataset to generate a standardized training dataset.

[0010] S3, the intelligent analysis terminal performs multimodal feature fusion analysis on the standardized training dataset to obtain the meteorological sensitivity coefficient, crop resistance coefficient and pest outbreak probability coefficient, and inputs the three coefficients into the pest risk quantification model to output the pest risk index;

[0011] S4. The early warning decision terminal triggers multi-level early warning rules based on the pest and disease risk index and generates a customized early warning plan that includes emergency response suggestions.

[0012] S5. The feedback verification terminal receives feedback data from users after they have implemented prevention and control measures, and dynamically updates the model parameters.

[0013] Preferably, the multi-source fusion dataset includes a meteorological environment dataset, a crop physiology dataset, and a pest and disease feature dataset; the meteorological environment dataset includes farmland three-dimensional coordinates, air temperature, relative humidity, light intensity, and precipitation; the crop physiology dataset includes relative chlorophyll content, initial fluorescence, maximum fluorescence, and stem water potential; and the pest and disease feature dataset includes pest population density, pathogen spore concentration, and the percentage of typical lesion area.

[0014] Preferably, the meteorological sensitivity coefficient is specifically:

[0015] ,

[0016] Where t0 is the data acquisition start time, t is the current time, k is the temperature response decay coefficient, and T air (t) represents the air temperature at time t, T opt M is the optimal temperature for crop growth, M(t) is the air humidity at time t, and M base For optimal humidity for crop growth, I light (t) represents the light intensity at time t, I thresh P is the light intensity threshold. rain，max Let be the maximum single precipitation, e be the natural constant, and ln be the logarithm to the base e.

[0017] Preferably, the crop resistance coefficient is specifically:

[0018] ,

[0019] Among them, α, β, γ, and δ are weighting factors, where α + β + γ + δ = 1 and α, β, γ, and δ ∈ [0, 1], Chl is the relative chlorophyll content, Ψ stem is the stem water potential, Ψ wilt is the critical value of wilting water potential, F min is the minimum fluorescence under dark adaptation, A lesion is the proportion of the area of typical lesions, F v is the variable fluorescence, and the calculation formula is F v =F m -F o , where F m is the maximum fluorescence, and F o is the initial fluorescence.

[0020] Preferably, the probability coefficient of pest and disease outbreak is specifically:

[0021] ,

[0022] Among them, λ, η, and ρ are adjustment factors, D pest is the pest population density, S spores is the concentration of pathogen spores, A lesion is the proportion of the area of typical lesions, Chl is the relative chlorophyll content, Chl crit is the chlorophyll critical value, T air is the air temperature, T pest is the optimum temperature for pest activity, and e is the natural constant.

[0023] Preferably, the quantitative model of pest and disease risk is specifically:

[0024] ,

[0025] Among them, , ω1, ω2, and ω3 are dynamic weighting coefficients, C is the meteorological sensitivity coefficient, R is the crop resistance coefficient, P is the probability coefficient of pest and disease outbreak, τ is the historical data correction factor, Q hist is the pest and disease risk index in the same historical period, and e is the natural constant.

[0026] Preferably, the multi-level early warning rules include:

[0027] If the pest and disease risk index Q ≥ Q1, trigger a first-level early warning and take chemical emergency measures;

[0028] If Q2 ≤ Q < Q1, trigger a second-level early warning and implement biological control;

[0029] If Q3 ≤ Q < Q2, trigger a third-level early warning and start enhanced inspections;

[0030] If the pest and disease risk index satisfies \(0\leq Q<Q_3\), a level - 4 warning is triggered, and routine monitoring is recommended.

[0031] Preferably, the customized warning scheme includes:

[0032] Matching a list of recommended pesticides for different types of pests and diseases;

[0033] Generating a drone pesticide application path plan based on the three - dimensional coordinates of the farmland;

[0034] Providing the best prevention and control window period in combination with weather forecast data.

[0035] Preferably, the specific implementation method of the feedback verification terminal is as follows:

[0036] A1. Obtaining the lesion regression rate and pest population survival rate after pesticide application through the intelligent monitoring terminal and the data processing terminal;

[0037] A2. The feedback verification terminal reversely modifies the model parameters according to the lesion regression rate and pest population survival rate, and iteratively optimizes the meteorological sensitivity coefficient, crop resistance coefficient, and pest and disease outbreak probability coefficient;

[0038] A3. According to the correction result, re - allocate the weights of the meteorological sensitivity coefficient, crop resistance coefficient, and pest and disease outbreak probability coefficient.

[0039] Preferably, the rules for iterative optimization include:

[0040] Correction of the meteorological sensitivity coefficient:

[0041] C new =C old ×(1 - η C ×S surv )

[0042] where η C is the adjustment factor, S surv is the pest survival rate, C old is the meteorological sensitivity coefficient before iterative optimization, and C new is the meteorological sensitivity coefficient after iterative optimization;

[0043] Correction of the crop resistance coefficient:

[0044] R new =R old ×(1 + η R ×(1 - S surv ))

[0045] where η R is the adjustment factor, S surv is the pest survival rate, R old is the crop resistance coefficient before iterative optimization, and Rnew The crop resistance coefficient after iterative optimization;

[0046] Correction for the probability coefficient of pest and disease outbreaks:

[0047] P new =P old ×(1+η P ×(1-R rec )),

[0048] Where, η P As a regulating factor, R rec P represents the rate of lesion regression. old To iteratively optimize the probability coefficient of pest and disease outbreaks, P new This refers to the probability coefficient of pest and disease outbreaks after iterative optimization.

[0049] The implementation method of step A3 is as follows:

[0050] Compare the corrected model values ​​with the model predictions, and calculate the error:

[0051] ,

[0052] Among them, Q predicted Q is the model's predicted value. actual These are the corrected model values;

[0053] The weighting coefficients are updated by combining the error and the corrected coefficient values:

[0054] , , ,

[0055] Among them, E total =(C new +R new +P new )×(1+E).

[0056] The technical effects and advantages of this invention are as follows:

[0057] This invention uses an intelligent analysis terminal to perform multimodal feature fusion analysis on a standardized training dataset to obtain meteorological sensitivity coefficient, crop resistance coefficient, and pest and disease outbreak probability coefficient. These coefficients are then input into a pest and disease risk quantification model to output a risk index, thereby achieving multidimensional and quantitative analysis of pest and disease risk and accurately assessing the risk of pest and disease occurrence.

[0058] This invention triggers multi-level early warning rules based on the pest and disease risk index through an early warning decision terminal, generating a customized early warning plan that includes emergency response suggestions. This enables graded response and precise policy implementation for pest and disease early warning, improving the pertinence and effectiveness of prevention and control measures.

[0059] This invention receives feedback data from users after implementing prevention and control measures through a feedback verification terminal, dynamically updates model parameters, and forms a self-optimizing closed loop of the system, continuously improving the model's prediction accuracy and early warning capabilities, and adapting to changes in different farmland environments and prevention and control needs. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the device connection according to the present invention.

[0061] Figure 2 This is a flowchart of the method steps of the present invention.

[0062] Figure 3 This is a flowchart illustrating the specific implementation of the feedback verification terminal of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] This invention provides, for example Figure 1 The method for early warning of crop diseases and pests based on artificial intelligence includes an intelligent monitoring terminal, a data processing terminal, an intelligent analysis terminal, an early warning decision terminal, and a feedback verification terminal.

[0065] It is important to understand that the core function of the intelligent monitoring terminal is to continuously collect multi-dimensional environmental, physiological, and pest and disease characteristic data throughout the entire crop growth cycle through a diverse array of IoT sensors deployed in farmland, including meteorological sensors, crop physiological sensors such as chlorophyll meters, fluorescence meters, and stem water potential probes, as well as pest and disease monitoring equipment such as insect monitoring lamps, spore traps, and image recognition cameras. This data is ultimately fused to form a multi-source fusion dataset that includes three major categories: meteorological environment, crop physiology, and pest and disease characteristics. In terms of physical implementation, the terminal mainly relies on various dedicated sensor nodes deployed in the farmland and performs initial data aggregation through edge computing gateways or direct connection to the cloud platform.

[0066] The data processing terminal is responsible for performing professional data preprocessing on the original multi-source fusion dataset from the intelligent monitoring terminal. Its core tasks include removing outliers and normalizing the dataset to generate a standardized dataset suitable for model training. Its physical carrier can be an edge computing device for preliminary processing, but it mainly relies on a cloud-based big data processing platform, such as a Spark or Hadoop framework or a real-time data stream processing engine, such as Apache Flink, to complete the data cleaning, transformation and standardization work using these powerful computing resources.

[0067] The intelligent analysis terminal relies heavily on powerful computing capabilities and is usually deployed on GPU server clusters in the cloud for complex model training and inference, or uses AI inference chips at the edge, such as NVIDIA Jetson, to run lightweight models for real-time analysis.

[0068] The early warning decision terminal automatically triggers the corresponding level of early warning based on the pest and disease risk index output by the intelligent analysis terminal and combined with preset multi-level early warning rules. It also generates a highly customized early warning plan that includes specific action suggestions. Its implementation relies on a business logic engine deployed in the cloud, such as the Drools rule engine, geographic information system platform, and agricultural knowledge graph database. It can accurately match the recommended pesticide list, generate drone spraying path planning based on the three-dimensional coordinates of farmland, and provide the best prevention and control window period in combination with weather forecasts. Finally, it pushes early warning information and plans to users through the Web console.

[0069] The feedback verification terminal forms a self-optimizing closed loop in the system. It acquires data on the effectiveness of user-implemented control measures through intelligent monitoring terminals, such as the rate of lesion regression and pest survival rate after pesticide application. This feedback data drives the dynamic update of model parameters. The specific process includes revising the calculation formulas of meteorological sensitivity coefficient, crop resistance coefficient, and pest outbreak probability coefficient based on feedback indicators, and redistributing the dynamic weight coefficients of these three coefficients in the risk quantification model in combination with prediction errors. Its physical implementation mainly relies on a cloud-deployed machine learning operation and maintenance platform, such as MLflow, to manage the iterative training and version updates of the model. It uses online learning or incremental learning techniques to continuously optimize model performance and improve the accuracy of subsequent early warnings.

[0070] This invention provides, for example Figure 2 The method for early warning of crop diseases and pests based on artificial intelligence, as shown, specifically includes the following steps:

[0071] S1. The intelligent monitoring terminal collects multi-dimensional environmental data at regular intervals during the crop growth cycle through an IoT sensor array deployed in farmland, forming a multi-source fusion dataset.

[0072] Furthermore, in the above technical solution, the multi-source fusion dataset includes a meteorological environment dataset, a crop physiology dataset, and a pest and disease characteristic dataset; the meteorological environment dataset includes farmland three-dimensional coordinates, air temperature, relative humidity, light intensity, and precipitation; the crop physiology dataset includes relative chlorophyll content, initial fluorescence, maximum fluorescence, and stem water potential; the pest and disease characteristic dataset includes pest population density, pathogen spore concentration, and the proportion of typical lesion area.

[0073] It should be noted that the three-dimensional coordinates of the farmland are obtained by using global navigation satellite systems, such as GPS and BeiDou, combined with differential positioning technology to acquire the longitude, latitude, and altitude of the farmland plots, achieving three-dimensional coordinate acquisition with centimeter-level to meter-level accuracy;

[0074] The air temperature is collected periodically by a protective temperature sensor, such as a platinum resistance thermometer (PT100), which is deployed about 1.5 meters above the ground in the farmland and equipped with a radiation shield. Data is recorded every 5-10 minutes to accurately reflect the temperature environment near the crop canopy.

[0075] The relative humidity is measured using a capacitive humidity sensor, which works based on the principle of the change in dielectric constant after the polymer film absorbs moisture. To ensure data accuracy, it needs to be calibrated periodically using a saturated salt solution, such as 75% NaCl.

[0076] The light intensity is collected using a quantum light sensor, such as a PAR sensor. This device is specifically designed to measure photosynthetically active radiation in the 400-700nm wavelength band. When installing, it should be placed horizontally above the crop canopy to avoid shadows interfering with the readings.

[0077] The rainfall is monitored using a tipping bucket rain gauge. Its working principle is that the tipping bucket will tip once for every 0.2mm of accumulated rainfall. The total rainfall is calculated by recording the number of tipping times through a reed switch. The equipment is usually equipped with bird-proof pins and filters to prevent debris from clogging and affecting accuracy.

[0078] The relative chlorophyll content was measured using a handheld chlorophyll meter, such as SPAD-502. This instrument calculates the relative chlorophyll index by measuring the transmittance of the leaf to 650nm red light and 940nm infrared light. When sampling, healthy leaves that are fully unfolded at the top of the plant should be selected, avoiding the main veins, and multiple measurements should be taken and the average value should be taken.

[0079] The initial and maximum fluorescence were acquired using a modulated chlorophyll fluorometer, such as the PAM-2500. Before measurement, the leaves underwent 20 minutes of dark adaptation, and then the initial fluorescence was excited under weak measurement light, typically below 0.1 μmol photons / m². -2 s -1Then, a saturated pulse of light is applied, the intensity of which typically reaches approximately 3000 μmol photonsm. -2 s -1 This is used to stimulate the leaves to produce maximum fluorescence;

[0080] The stem water potential is measured using a pressure chamber water potential meter, such as the PMS Model 1505D. The operation requires cutting leafy branches before sunrise and quickly sealing them in a pressure chamber. The pressure is gradually increased until sap appears at the cut of the xylem vessels. The corresponding pressure value at this time is the stem water potential. Sampling in the early morning can minimize the interference of transpiration on the results.

[0081] The pest population density is mainly determined by automatic insect monitoring lamps, such as insect-attracting lamps with specific wavelengths or sex pheromone traps, which capture insects. Combined with a built-in image recognition system, the captured insects are automatically counted and classified, and the number of captured insects is finally converted into the pest population density per unit area, such as per hectare.

[0082] The concentration of pathogenic spores was collected using a spore trapping device, such as the TPBZ3 model. This device draws in spores from the air through airflow and causes them to collide and adhere to a glass slide. Subsequently, an image is acquired using a microscopic imaging system, and then an AI recognition model based on deep learning, such as YOLOv5, is applied to automatically identify and count the spores in the image. The final result is expressed as the number of spores per cubic meter of air.

[0083] The percentage of typical lesion area is calculated by using high-definition cameras deployed in the field, especially multispectral cameras such as RedEdge-MX, to capture crop leaf images. Image processing techniques, such as segmentation based on HSV color space, are used to identify and segment the lesion areas on the leaves, and finally, the percentage of lesion pixel area to the total pixel area of ​​the leaf is calculated.

[0084] S2. The data processing terminal performs outlier removal and normalization on the multi-source fusion dataset to generate a standardized training dataset.

[0085] It is important to understand that the specific implementation process of the data processing terminal is as follows: First, outlier removal is performed on the original multi-source fusion dataset from the intelligent monitoring terminal. Strict data verification rules are applied, and for the meteorological environment dataset, a physically reasonable range is used for filtering, such as excluding abnormal temperature values ​​exceeding -10℃ to 50℃. For the crop physiological dataset, a dynamic threshold algorithm is applied to identify and remove data points that significantly deviate from normal fluctuations, such as chlorophyll readings exceeding 120% of the historical highest value. For the pest and disease feature dataset, the isolated forest algorithm is used to detect statistically significant outliers. For data gaps resulting from outlier removal, linear interpolation is used to fill in missing values ​​in the continuous time series data. After anomaly processing, data normalization is then performed to eliminate differences in the dimensions of different sensors. For numerical data, such as the meteorological environment dataset, crop physiological dataset, and pathogen spore concentration, Min-Max normalization is used to linearly map them to the [0, 1] interval. For pest and disease data with high skewness, such as pest population density and the proportion of typical lesion area, logarithmic transformation is applied to reduce the data distribution skewness. Finally, a standardized training dataset is obtained.

[0086] S3, the intelligent analysis terminal performs multimodal feature fusion analysis on the standardized training dataset to obtain the meteorological sensitivity coefficient, crop resistance coefficient and pest outbreak probability coefficient, and inputs the three coefficients into the pest risk quantification model to output the pest risk index;

[0087] Furthermore, in the above technical solution, the meteorological sensitivity coefficient is specifically:

[0088] ,

[0089] Where t0 is the data acquisition start time, t is the current time, k is the temperature response decay coefficient, and T air (t) represents the air temperature at time t, T opt M is the optimal temperature for crop growth, M(t) is the air humidity at time t, and M base For optimal humidity for crop growth, I light (t) represents the light intensity at time t, I thresh P is the light intensity threshold. rain，max Let be the maximum single precipitation, e be the natural constant, and ln be the logarithm to the base e.

[0090] It is important to understand that the formula for the meteorological sensitivity coefficient is designed based on the dynamic coupling mechanism and time-cumulative effect of environmental factors on pests and diseases. Its logical structure is progressive: First, the numerator of the formula captures the continuous effect of environmental factors during the crop growth cycle through integral operations, emphasizing the difference between short-term fluctuations and long-term trends; second, a triple nonlinear coupling is designed in the integral term, and the temperature response adopts an exponential decay function. This expresses the sensitivity of pests and diseases to temperature deviations; for example, activity is highest in the optimal temperature range and rapidly declines when deviating from the optimal temperature range. Humidity also affects [the situation]. To achieve logarithmic scaling, such as high humidity promoting pathogen spore germination and low humidity inhibiting pest activity, while the light intensity threshold I... thresh Penalty items This relates to the weakening of crop resistance in low-light environments; finally, the denominator is the maximum single rainfall P. rain，max +1 represents the precipitation inhibition effect, forming a dynamic balance between numerator promotion and denominator inhibition.

[0091] Furthermore, in the above technical solution, the crop resistance coefficient is specifically:

[0092] ,

[0093] Where α, β, γ, and δ are weighting factors, α + β + γ + δ = 1 and α, β, γ, and δ ∈ [0, 1], Chl is the relative chlorophyll content, and Ψ stem For stem water potential, Ψ wilt F is the critical value of wilting water potential. min For dark-adapted minimum fluorescence, A lesion F represents the percentage of typical lesion area. v For variable fluorescence, the calculation formula is F. v =F m -F o , of which F m For maximum fluorescence, F o This is the initial fluorescence.

[0094] It is important to understand that in the formula for the crop resistance coefficient, the numerator and denominator represent the dynamic balance between positive driving factors and negative inhibiting factors of resistance, respectively. The numerator is composed of the relative chlorophyll content and the difference between stem water potential and wilting critical water potential: chlorophyll content reflects the integrity of photosynthetic organ function through weight α, while the water potential difference, after weighting by β, quantifies the crop's buffering capacity against water stresses such as drought. Together, they reflect the crop's potential to maintain growth through photosynthesis and water homeostasis; the denominator contains a variation term in photosynthetic efficiency. and disease damage item δ×A lesion : where F v =F m -F o The defined ratio of variable fluorescence to maximum fluorescence, after γ-modulation, characterizes the sensitivity of photosystem II to biotic or abiotic stresses; the percentage of typical lesion area A. lesionThe degree of pathological damage is directly correlated with the weight δ, and the two work together to reflect the weakening effect of functional impairment and energy dissipation resistance under adversity. The four weighting factors α, β, γ, and δ are set as follows: α represents the contribution of relative chlorophyll content (Chl) to the target indicator R. If chlorophyll is the main influencing factor, such as in the assessment of photosynthesis in healthy plants, α can be set between 0.3 and 0.5. If the plant is under stress, such as drought or disease, the chlorophyll content may decrease, and α needs to be reduced. β reflects the influence of plant water status, i.e., the difference between stem water potential and wilting threshold. In drought or water stress studies, β can be set between 0.2 and 0.4. If the research focus is on water relationships, β needs to be increased; if water conditions are stable, β can be reduced. γ measures photosynthetic efficiency. In photosynthesis-related studies, γ can be set between 0.2 and 0.3. The parameters in the model are sensitive to the target indicators, such as early detection of stress, which can increase γ; δ characterizes the impact of diseases on the target indicators. In studies of healthy plants, δ can be set to a lower value, between 0 and 0.1. In studies of diseases, δ needs to be increased, between 0.3 and 0.5.

[0095] Furthermore, in the above technical solution, the probability coefficient of pest and disease outbreak is specifically as follows:

[0096] ,

[0097] Where λ, η, and ρ are adjustment factors, and D pest For pest population density, S spores A represents the concentration of pathogenic spores. lesion Chl represents the percentage of typical lesion area, and Chl represents the relative chlorophyll content. crit T is the critical value for chlorophyll. air For air temperature, T pest The temperature is the optimal temperature for pest activity, and e is a natural constant.

[0098] It's important to know that in the formula for the probability coefficient of pest and disease outbreaks, the core driving factor is the pest population density D. pest With the concentration of pathogenic spores S spores A. Percentage of typical lesion area lesion The system consists of two parts. The basic driving term is formed by directly multiplying the pest population density and pathogen spore concentration, reflecting their synergistic effect on pest outbreaks. As pests act as vectors for pathogens, the higher the product of their population density and pathogen spore concentration, the easier it is to create the basic conditions for pest epidemics. Meanwhile, the percentage of typical lesion area is expressed as an exponential function. Nonlinear amplification was used to reflect the spatial expansion effect of the disease: a higher proportion of lesion area not only indicates that the host tissue has been severely infected, but also may accelerate the outbreak process through secondary transmission or the cumulative effect of tissue damage. This exponential growth characteristic highlights the significant strengthening effect of the disease diffusion stage on the outbreak probability; the plant resistance regulation term is composed of the relative chlorophyll content Chl and the critical value Chl. crit Composition, denominator Using the Sigmoid function form implies the following logic: when Chl > Chl crit When the plant is healthy, the denominator approaches 1, the overall term approaches 1, and the probability of an outbreak is suppressed; when Chl <Chl crit When plants are weak, the denominator approaches 0, and the overall term approaches 0, significantly reducing the probability of an outbreak; the temperature adaptation correction term is determined by the air temperature T. air The optimal temperature for pests T pest Composition, Modifications Similarly, the Sigmoid function is used, with the following effect: when T... air ≈T pest When T approaches 1, the correction term indicates a suitable temperature for pest activity; when T... air < <T pest or T air >>T pest When the correction term approaches 0, the temperature is unsuitable and suppresses pests; the roles of regulating factors λ, η, and ρ: λ controls the sensitivity of the influence of lesion area; the larger the value, the stronger the amplification effect of lesion area changes on probability; η regulates the steepness of the transition near the plant resistance threshold; the larger the value, the more drastic the probability change when the relative chlorophyll content approaches the critical value; ρ controls the response width of temperature adaptability; the larger the value, the faster the probability decreases when the temperature deviates from the optimum value; the setting of regulating factors λ, η, and ρ: if the probability increases by n times for every doubling of the lesion area of ​​a certain disease, then λ = ln(n), where λ ∈ [0.1, 5]; if the relative chlorophyll content is lower than the critical value by m times, then η = 10 × m, where η ∈ [0.5, 10]; if pests are active within the range of suitable temperature ± T℃, then ρ = , where ρ∈[0.1,1].

[0099] Furthermore, in the above technical solution, the pest and disease risk quantification model is specifically as follows:

[0100] ,

[0101] in, ω1, ω2, and ω3 are dynamic weighting coefficients, C is the meteorological sensitivity coefficient, R is the crop resistance coefficient, P is the probability coefficient of pest and disease outbreaks, τ is the historical data correction factor, and Q... hist , where is the historical pest and disease risk index for the same period, and e is a natural constant.

[0102] It should be noted that if the historical data is stable for a long time, such as the pest and disease patterns in the same period in a certain area are consistent for many years, and the meteorological and crop varieties change little, τ can be set between 2 and 5, aiming to make the historical rules have a more significant impact on the current prediction; if the historical data fluctuates greatly, such as abnormal climate and frequent variety updates in recent years, and the historical rules fail, τ can be set between 0.1 and 1, aiming to reduce the interference of historical data;

[0103] It should be noted that the setting of the initial values of the dynamic weight coefficients ω1, ω2 and ω3: if it is a meteorological-driven pest and disease scenario, the initial values can be set as ω1 = 0.4, ω2 = 0.3, ω3 = 0.3; if it is a pest-source-dominated scenario, the initial values can be set as ω1 = 0.3, ω2 = 0.3, ω3 = 0.4; if it is a scenario in the promotion area of disease-resistant varieties, the initial values can be set as ω1 = 0.3, ω2 = 0.4, ω3 = 0.3; if it is a scenario in the area with frequent extreme climates, the initial values can be set as ω1 = 0.5, ω2 = 0.25, ω3 = 0.25.

[0104] S4. The early warning decision-making terminal triggers multi-level early warning rules according to the pest and disease risk index, and generates a customized early warning plan including suggestions for emergency measures;

[0105] Furthermore, in the above technical solution, the multi-level early warning rules include:

[0106] If the pest and disease risk index Q≥Q1, trigger a first-level early warning and take chemical emergency measures;

[0107] If Q2≤Q<Q1 for the pest and disease risk index, trigger a second-level early warning and implement biological control;

[0108] If Q3≤Q<Q2 for the pest and disease risk index, trigger a third-level early warning and start enhanced inspections;

[0109] If 0≤Q<Q3 for the pest and disease risk index, trigger a fourth-level early warning and recommend routine monitoring.

[0110] It should be noted that Q1, Q2 and Q3 are risk index critical values, Q1 = 0.8, Q2 = 0.6, Q3 = 0.3;

[0111] It should be noted that the chemical emergency measures are based on the pest and disease types to match a list of recommended highly effective and low-toxic chemical agents. For example, chlorantraniliprole for Lepidoptera pests and pyraclostrobin for fungal diseases. The three-dimensional coordinates of the farmland are used to generate a drone spraying path plan to ensure uniform coverage of the agent, and the weather forecast data is combined to provide the best prevention and control window period such as no rain in the next 24 hours to maximize the drug effect;

[0112] The biological control measures specifically include releasing natural enemies such as Trichogramma wasps and aphid wasps, applying microbial agents such as Bacillus subtilis and Trichoderma, using plant-derived extracts such as azadirachtin and matrine, and combining farmland vegetation structure adjustments, such as intercropping with insect-repelling plants, to enhance natural pest control capabilities.

[0113] The enhanced patrols require increasing the frequency of manual field patrols from once a week to once every 2 to 3 days, focusing on observing and recording data on hidden parts of the crop, such as the underside of leaves and the base of stems. At the same time, intelligent devices such as insect monitoring lamps and multispectral cameras are used to collect data in real time, automatically identify indicators such as the number of adult pests and the trend of lesion expansion, issue warnings immediately upon discovering abnormalities, and compare the patrol data with historical models to assess the development trend of pests and diseases.

[0114] Furthermore, in the above technical solution, the customized early warning solution includes:

[0115] A recommended list of pesticides to match different types of pests and diseases;

[0116] UAV pesticide application path planning is generated based on the three-dimensional coordinates of farmland.

[0117] The best window of opportunity for prevention and control can be determined by combining weather forecast data.

[0118] It's important to understand that the recommended pesticide list is matched to different types of pests and diseases. Specifically, based on the pest and disease types output by the intelligent analysis terminal, such as pest species and pathogen categories, the system automatically retrieves and matches corresponding highly effective and low-toxicity chemical or biological agents from the agricultural knowledge graph database built into the early warning and decision-making terminal. For example, for lepidopteran pests such as cotton bollworm and corn borer, selective insecticides such as chlorantraniliprole and emamectin benzoate are recommended; for fungal diseases such as powdery mildew and downy mildew, fungicides such as pyraclostrobin and tebuconazole are recommended; and for bacterial diseases, agents such as thiamethoxam and streptomycin are recommended. This process achieves precise matching by associating pest and disease characteristic data, such as pest population density and typical lesion morphology, with tags in the pesticide database. Furthermore, the system generates drone application path planning based on the three-dimensional coordinates of the farmland. Specifically, it utilizes the three-dimensional geographic coordinate data of the farmland plots, combined with crop planting row spacing, crown spacing, and other factors. Based on parameters such as layer height, the farmland is divided into several grid units through the geographic information system platform of the early warning decision terminal. Differentiated pesticide application rates are set according to the pest and disease risk index of each unit. Then, the Dijkstra algorithm or genetic algorithm is used to plan the shortest flight route to ensure that the drone evenly covers all high-risk areas. The path is dynamically corrected in combination with the real-time image data transmitted by the drone to improve the control efficiency and reduce pesticide waste. The best control window period is provided by combining weather forecast data. Specifically, the early warning decision terminal connects to the real-time data interface of the meteorological department to obtain weather forecast information such as air temperature, relative humidity, precipitation probability, and wind speed for the next 3-7 days. Combined with the analysis of rules such as avoiding the period of rainfall in the next 24 hours, choosing the appropriate application temperature of 15-28℃ for most chemical agents, and wind speed not exceeding level 4, the control time window suggestion containing specific dates and times is generated and pushed to the user.

[0119] S5. The feedback verification terminal receives feedback data from users after they have implemented prevention and control measures, and dynamically updates the model parameters.

[0120] Furthermore, in the above technical solution, the specific implementation method of the feedback verification terminal is as follows: Figure 3 As shown:

[0121] A1. Obtain the rate of lesion regression and pest population survival rate after pesticide application through intelligent monitoring terminals and data processing terminals;

[0122] A2. The feedback verification terminal corrects the model parameters in reverse based on the lesion regression rate and the survival rate of pest populations, and iteratively optimizes the meteorological sensitivity coefficient, crop resistance coefficient and pest outbreak probability coefficient.

[0123] A3. Based on the revised results, the weights of the meteorological sensitivity coefficient, crop resistance coefficient, and pest and disease outbreak probability coefficient are reallocated.

[0124] It is important to know the rate of lesion regression. , where A lesion，before A represents the percentage of lesion area before application of medication. lesion，after The percentage of lesion area after pesticide application; the survival rate of the pest population. , where N surv To determine the initial population size of pests before applying pesticides, N init To count the number of survivors after application of the medication;

[0125] Furthermore, in the above technical solution, the rules for iterative optimization include:

[0126] Correction for meteorological sensitivity coefficient:

[0127] C new =C old ×(1-η C ×S surv ),

[0128] Where, η C S is a regulating factor. surv For pest survival rate, C old To optimize the meteorological sensitivity coefficient before iterative optimization, C new The meteorological sensitivity coefficient after iterative optimization;

[0129] Correction for crop resistance coefficient:

[0130] R new =R old ×(1+η R ×(1-S surv )),

[0131] Where, η R S is a regulating factor. surv For pest survival rate, R old To iteratively optimize the crop resistance coefficient before the change, R new The crop resistance coefficient after iterative optimization;

[0132] Correction for the probability coefficient of pest and disease outbreaks:

[0133] P new =P old ×(1+η P ×(1-R rec )),

[0134] Where, η P As a regulating factor, R rec P represents the rate of lesion regression. old To iteratively optimize the probability coefficient of pest and disease outbreaks, P new This refers to the probability coefficient of pest and disease outbreaks after iterative optimization.

[0135] The implementation method of step A3 is as follows:

[0136] Compare the corrected model values ​​with the model predictions, and calculate the error:

[0137] ,

[0138] Among them, Q predicted Q is the model's predicted value. actual These are the corrected model values;

[0139] The weighting coefficients are updated by combining the error and the corrected coefficient values:

[0140] , , ,

[0141] Among them, E total =(C new +R new +P new )×(1+E).

[0142] It should be noted that the adjustment factor η C Setting: If meteorological conditions significantly affect the control effect, such as in rainy or high-temperature areas, then increase η. C It can be set between 0.3 and 0.5; if the weather is stable or the impact is minor, then reduce η. C It can be set between 0.05 and 0.15;

[0143] The η R Setting: If crop resistance is critical to the control effect, such as in areas where disease-resistant varieties are promoted, then increase η. R It can be set between 0.3 and 0.5; if the crop growth is generally weak, then reduce η. R It can be set to between 0.1 and 0.2;

[0144] The η P Setting: If the disease spreads rapidly, such as when the concentration of pathogen spores is high, then increase η. P It can be set between 0.4 and 0.6; if pests are the main problem or diseases are stable, then reduce η. P It can be set to between 0.1 and 0.3.

[0145] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early warning of crop diseases and pests based on artificial intelligence, characterized in that, It includes an intelligent monitoring terminal, a data processing terminal, an intelligent analysis terminal, an early warning decision-making terminal, and a feedback verification terminal. Specifically, it includes the following steps: S1. The intelligent monitoring terminal regularly collects multi-dimensional environmental data through the Internet of Things sensor array deployed in the farmland during the crop growth cycle to form a multi-source fusion data set; The multi-source fusion data set includes a meteorological environment data set, a crop physiology data set, and a pest and disease characteristic data set; the meteorological environment data set includes the three-dimensional coordinates of the farmland, air temperature, relative humidity, light intensity, and precipitation; the crop physiology data set includes the relative chlorophyll content, initial fluorescence, maximum fluorescence, and stem water potential; the pest and disease characteristic data set includes the pest population density, the concentration of pathogen spores, and the proportion of the area of typical disease spots; S2. The data processing terminal performs outlier removal and normalization processing on the multi-source fusion data set to generate a standardized training data set; S3. The intelligent analysis terminal performs multi-modal feature fusion analysis on the standardized training data set to obtain the meteorological sensitivity coefficient, the crop resistance coefficient, and the pest and disease outbreak probability coefficient, and inputs the three coefficients into the pest and disease risk quantification model to output the pest and disease risk index; The meteorological sensitivity coefficient is specifically: ， Where t0 is the data acquisition start time, t is the current time, k is the temperature response decay coefficient, and T air (t) represents the air temperature at time t, T opt M is the optimal temperature for crop growth, M(t) is the air humidity at time t, and M base For optimal humidity for crop growth, I light (t) represents the light intensity at time t, I thresh P is the light intensity threshold. rain，max Let be the maximum single precipitation, e be the natural constant, and ln be the logarithm to the base e; The crop resistance coefficient is specifically: ， Where α, β, γ, and δ are weighting factors, α + β + γ + δ = 1 and α, β, γ, and δ ∈ [0, 1], Chl is the relative chlorophyll content, and Ψ stem For stem water potential, Ψ wilt F is the critical value of wilting water potential. min For dark-adapted minimum fluorescence, A lesion F represents the percentage of typical lesion area. v For variable fluorescence, the calculation formula is F. v =F m -F o , of which F m For maximum fluorescence, F o Initial fluorescence; The pest and disease outbreak probability coefficient is specifically: ， Where λ, η, and ρ are adjustment factors, and D pest For pest population density, S spores A represents the concentration of pathogenic spores. lesion Chl represents the percentage of typical lesion area, and Chl represents the relative chlorophyll content. crit T is the critical value for chlorophyll. air For air temperature, T pest The temperature is the optimal temperature for pest activity, and e is a natural constant. The pest and disease risk quantification model is specifically: ， Where Q represents the pest and disease risk index. ω1, ω2, and ω3 are dynamic weighting coefficients, C is the meteorological sensitivity coefficient, R is the crop resistance coefficient, P is the probability coefficient of pest and disease outbreaks, τ is the historical data correction factor, and Q... hist The risk index of pests and diseases in the same period of history, where e is a natural constant; S4. The early warning decision-making terminal triggers multi-level early warning rules according to the pest and disease risk index and generates a customized early warning plan including suggestions for emergency measures; S5. The feedback verification terminal receives the feedback data after the user implements the prevention and control measures and dynamically updates the model parameters.

2. The method for early warning of crop diseases and pests based on artificial intelligence according to claim 1, characterized in that, The multi-level early warning rules include: If the pest and disease risk index Q≥Q1, trigger a first-level early warning and take chemical emergency measures; ​ ​ ​ ​ 3. The method for early warning of crop diseases and pests based on artificial intelligence according to claim 1, characterized in that, ​ ​ ​ ​ 4. The method for early warning of crop diseases and pests based on artificial intelligence according to claim 1, characterized in that, ​ ​ ​ ​ 5. The method for early warning of crop diseases and pests based on artificial intelligence according to claim 4, characterized in that, ​ ​ C new =C old ×(1-n) C ×S surv ), Where, η C S is a regulating factor. surv For pest survival rate, C old To optimize the meteorological sensitivity coefficient before iterative optimization, C new The meteorological sensitivity coefficient after iterative optimization; ​ R new =R old ×(1+n) R ×(1-S surv )), Where, η R S is a regulating factor. surv For pest survival rate, R old To iteratively optimize the crop resistance coefficient before the change, R new The crop resistance coefficient after iterative optimization; ​ P.S new =P old ×(1+η P ×(1-R rec )), Where, η P As a regulating factor, R rec P represents the rate of lesion regression. old To iteratively optimize the probability coefficient of pest and disease outbreaks, P new This refers to the probability coefficient of pest and disease outbreaks after iterative optimization. ​ ​ ， Among them, Q predicted Q is the model's predicted value. actual These are the corrected model values; ​ ， ， ， Among them, E total =(C new +R new +P new )×(1+E).

Images