Rice disease model prediction method and system based on remote sensing monitoring

Abstract

This invention relates to the field of agricultural technology. It discloses a method and system for predicting rice diseases based on remote sensing monitoring. The method includes: S1: receiving remote sensing parameters of rice diseases in a target area generated by a remote sensing device, and pre-setting estimated mitigation parameters; S2: calculating the first expected disease spread rate based on the first two parameters using a model; S3: calculating the expected natural disease development rate without intervention, and the difference between this and the estimated mitigation parameters is the remaining disease parameter; S4: calculating the second expected rate by combining soil condition parameters and the remaining disease parameters; and S5: adjusting the estimated parameters to make the two rates equal, using this as the target parameter to predict future disease severity. This invention ensures the reliability of initial data by acquiring multispectral and hyperspectral data through remote sensing equipment, pre-processing to eliminate interference and avoid human error, ensuring the data reflects the actual disease; and incorporating soil and environmental factors, adjusting parameters to make the two rates equal, thus improving the accuracy of the prediction.

Description

Technical Field

[0001] This invention relates to the field of agricultural technology, and in particular to a method and system for predicting rice diseases based on remote sensing monitoring models. Background Technology

[0002] Traditional rice disease monitoring and forecasting methods have significant limitations in data acquisition. Most solutions rely on manual field surveys, requiring staff to visually assess lesion size and count diseased panicles to determine disease severity. This approach is not only inefficient, especially in large rice-growing areas, often requiring substantial manpower and time, making real-time monitoring difficult; furthermore, manual judgment is susceptible to subjective errors due to experience differences and visual fatigue. Even when some solutions incorporate simple remote sensing data, they often lack systematic preprocessing of multispectral or hyperspectral data, failing to effectively eliminate interference from sensor noise, light intensity variations, and drone flight attitude deviations. This results in discrepancies between extracted disease characteristics and actual field conditions, leading to insufficient reliability of the foundational data for subsequent disease prediction.

[0003] Existing methods for predicting rice diseases suffer from significant biases in their consideration of influencing factors. Most schemes only perform simple extrapolations based on the current state of the disease, neglecting the role of key environmental factors in disease development. Specifically, regarding soil conditions, soil pH and fertility directly affect rice's disease resistance; alkaline or high-nitrogen soils exacerbate pathogen infection, yet existing methods often fail to incorporate these factors into their prediction models. Regarding environmental parameters, temperature, humidity, and precipitation are crucial for pathogen spore germination and spread; high temperature and humidity significantly accelerate disease diffusion, but current technologies often fail to deeply integrate these parameters with disease prediction. Furthermore, these methods fail to clearly distinguish between the natural progression of the disease and the effects of intervention measures, often directly deriving predictions from idealized intervention outcomes. This leads to a disconnect between predicted values ​​and actual field disease development, making it difficult to support precise disease control decisions.

[0004] In practical application, existing rice disease prediction technologies lack practicality and systematicity. On the one hand, prediction results are mostly presented as single data on disease spread rates, failing to transform them into intuitive quantitative indicators of intervention effects. Growers find it difficult to quickly assess the actual effectiveness of their proposed control measures, nor can they clearly understand the disease risks that still need to be addressed after intervention. On the other hand, there is a lack of supporting automated systems. Data collection, model calculation, and result output often require manual step-by-step operations, which are not only cumbersome and inefficient but also prone to data loss or calculation errors due to human error. Furthermore, most technical solutions can only predict short-term disease status and do not combine environmental parameter prediction data for future cycles to generate long-term disease development trend reports. Growers cannot predict disease trends in advance and often find themselves in a predicament of reactive response only after disease outbreaks, ultimately leading to reduced rice yields and damaged quality. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application provides a method and system for predicting rice diseases based on remote sensing monitoring, which offers reliable data, accurate predictions, and high application efficiency, and can deeply integrate remote sensing technology, environmental and soil parameters, and deep learning models.

[0006] This invention discloses a method for predicting rice diseases based on remote sensing monitoring, which includes the following steps:

[0007] This invention discloses a method for predicting rice diseases based on remote sensing monitoring, characterized by the following steps:

[0008] S1: Receive remote sensing disease parameters of the target rice planting area. The remote sensing disease parameters are generated by collecting disease-related image data of rice in the target area using remote sensing equipment.

[0009] Preset mitigation parameters for rice diseases in the target area. The predicted mitigation parameters represent the expected reduction in rice diseases in the target area in each unit time period within a single cycle after intervention measures are taken.

[0010] S2: Using a trained disease prediction model, the first expected disease spread rate is calculated based on remote sensing disease parameters and estimated mitigation parameters.

[0011] S3: Using a trained disease prediction model, calculate the expected natural disease development rate in the target rice planting area. The expected natural disease development rate represents the actual development of rice diseases in the target area in each unit time period within a single cycle when no intervention measures are taken.

[0012] The difference between the expected natural disease development rate and the estimated mitigation parameter is used as the remaining disease parameter;

[0013] S4: Using a trained disease prediction model, the second expected disease spread rate is calculated based on soil condition parameters and remaining disease parameters in the target rice planting area.

[0014] S5: Adjust the estimated mitigation parameters so that the first expected disease spread rate is equal to the second expected disease spread rate. Use the adjusted estimated mitigation parameters as the target disease mitigation parameters and use the target disease mitigation parameters to predict the future disease status of rice.

[0015] Furthermore, in step S1, the process of acquiring remote sensing disease parameters includes:

[0016] Acquire multispectral and hyperspectral remote sensing data of the target rice planting area;

[0017] Denoising, radiometric correction, and geometric correction preprocessing are performed on multispectral and hyperspectral remote sensing data.

[0018] The spectral, texture, and morphological features of rice lesions are extracted from the preprocessed remote sensing data, and remote sensing disease parameters are generated based on the extracted features.

[0019] Furthermore, step S4 also includes:

[0020] Using a trained disease prediction model, the second expected disease spread rate is calculated and updated based on soil condition parameters, remaining disease parameters, and environmental parameters of the target rice planting area.

[0021] Soil condition parameters include soil pH, soil fertility, and soil moisture content data.

[0022] Furthermore, the disease prediction model is a deep learning model, including an input layer, a hidden layer, and an output layer;

[0023] The input layer is used to receive input remote sensing disease parameters, estimated mitigation parameters, soil condition parameters and / or environmental parameters, perform standardized preprocessing, and then pass them to the hidden layer;

[0024] Each neuron in the hidden layer receives the output from the neurons in the previous layer, processes it through the ReLU activation function, and then outputs it to the neurons in the next layer. The neurons calculate the result based on the learned weights and pass it to the output layer.

[0025] The output layer is used to output the predicted disease spread rate results.

[0026] Further, step S3 includes:

[0027] S31: Divide the target rice planting area into multiple monitoring units according to the preset area, and treat each monitoring unit as an independent node;

[0028] S32: Calculate the degree of rice disease corresponding to each node. Based on the node disease degree value of semantic segmentation, calculate the expected natural development rate of disease through a trained disease prediction model.

[0029] S33: The difference between the expected natural development rate of the disease and the estimated mitigation parameter is used as the remaining disease parameter.

[0030] Furthermore, in step S32, the process of calculating the rice disease severity value corresponding to each node includes:

[0031] Based on the preprocessed remote sensing data, a semantic segmentation model was used to identify the diseased areas of rice plants within the nodes;

[0032] Calculate the ratio of the area of ​​the lesion region to the total coverage area of ​​rice plants within the node, and use this ratio as the degree of rice disease corresponding to the node.

[0033] Furthermore, S6: Calculate the disease relief rate based on the target disease relief parameters and the expected natural disease development rate. The disease relief rate is the ratio of the target disease relief parameters to the expected natural disease development rate.

[0034] Furthermore, S7: Combining disease mitigation rate and environmental parameter prediction data for future cycles, generate a report on the disease development trend of the target rice planting area within the next 1-2 growth cycles.

[0035] This invention discloses a rice disease prediction system based on remote sensing monitoring, using any of the rice disease prediction methods based on remote sensing monitoring described above, comprising:

[0036] The processor module, input device module, output device module, and memory module are interconnected.

[0037] The memory module is used to store computer programs, which include program instructions;

[0038] The input device module is used to receive remote sensing disease parameters, soil condition parameters, and environmental parameters of the target rice planting area;

[0039] The processor module is configured to call program instructions to execute a rice disease prediction method based on remote sensing monitoring.

[0040] The output device module is used to output the target disease relief parameters and disease development trend reports.

[0041] Furthermore, the processor module is further configured as follows:

[0042] The estimated mitigation parameters are adjusted iteratively, and the first and second expected disease spread rates are recalculated until the absolute value of the difference between the two rates is less than or equal to the preset error threshold.

[0043] Beneficial effects:

[0044] This invention achieves dynamic balance optimization of rice disease intervention effects by constructing a dual-track prediction mechanism with a first and a second expected disease spread rate, and using their equality as a parameter calibration target. This design overcomes the bias problem caused by traditional prediction models that rely solely on a single path. The first expected rate is an idealized prediction based on remote sensing data and a preset intervention intensity, reflecting the control potential of control measures under ideal conditions; the second expected rate integrates soil conditions, environmental factors, and residual disease pressure, reflecting the realistic constraints of disease development in the actual field environment. Only when the two rates converge can the corresponding estimated mitigation parameters simultaneously satisfy the objectivity of remote sensing monitoring and the authenticity of environmental impact, thus ensuring that the prediction results are neither overly optimistic nor blindly pessimistic. This cross-validation calibration mechanism effectively avoids decision-making errors caused by ignoring natural development trends or environmental adaptability, significantly improving the reliability and operability of disease control recommendations, upgrading the prediction model from static fitting to dynamic adaptation, and truly serving scientific treatment and risk pre-management in precision agriculture.

[0045] This invention utilizes remote sensing equipment such as drones to collect multispectral and hyperspectral remote sensing data of target rice-growing areas, replacing the traditional method of visual estimation through manual field surveys. This avoids subjective errors caused by differences in human experience and visual fatigue from the data acquisition source. Simultaneously, the collected remote sensing data undergoes denoising, radiometric correction, and geometric correction preprocessing to effectively eliminate interference factors such as sensor noise, changes in light intensity, and drone flight attitude deviations. This allows for the accurate extraction of spectral, textural, and morphological features of rice lesions to generate remote sensing disease parameters. This process ensures that the initial data accurately reflects the actual state of rice field diseases, completely resolving the problem of insufficient reliability of basic data in traditional technologies and providing accurate data support for all subsequent prediction and calculation stages.

[0046] This invention overcomes the limitations of traditional prediction methods that neglect key environmental factors by improving prediction accuracy through multi-dimensional technical design. On one hand, a trained disease prediction model calculates the expected natural disease progression rate without intervention, allowing for early understanding of the disease's natural deterioration trend and avoiding misjudgments that focus solely on intervention effects while ignoring the risks of non-intervention. On the other hand, soil condition parameters and environmental parameters are deeply integrated into the prediction process. Soil condition parameters include soil pH, fertility, and moisture content, which are combined with remaining disease parameters to calculate a second expected disease spread rate. By iteratively adjusting the estimated mitigation parameters, the first and second expected disease spread rates are made equal. The resulting target disease mitigation parameters balance the objectivity of remote sensing data with adaptability to the actual environment, ensuring that the prediction results do not deviate from actual field growth conditions and significantly improving the accuracy of disease prediction. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating a method for predicting rice diseases based on remote sensing monitoring, as described in this application. Detailed Implementation

[0048] To enable those skilled in the art to better understand the present invention, the technical solutions in the specific embodiments of the present invention will be clearly and completely described below.

[0049] This invention discloses a method for predicting rice diseases based on remote sensing monitoring, which includes the following steps:

[0050] S1: Receive remote sensing disease parameters of the target rice planting area. The remote sensing disease parameters are generated by collecting disease-related image data of rice in the target area using remote sensing equipment.

[0051] Preset mitigation parameters for rice diseases in the target area. The predicted mitigation parameters represent the expected reduction in rice diseases in the target area in each unit time period within a single cycle after intervention measures are taken.

[0052] S2: Using a trained disease prediction model, the first expected disease spread rate is calculated based on remote sensing disease parameters and estimated mitigation parameters.

[0053] S3: Using a trained disease prediction model, calculate the expected natural disease development rate in the target rice planting area. The expected natural disease development rate represents the actual development of rice diseases in the target area in each unit time period within a single cycle when no intervention measures are taken.

[0054] The difference between the expected natural disease development rate and the estimated mitigation parameter is used as the remaining disease parameter;

[0055] S4: Using a trained disease prediction model, the second expected disease spread rate is calculated based on soil condition parameters and remaining disease parameters in the target rice planting area.

[0056] S5: Adjust the estimated mitigation parameters so that the first expected disease spread rate is equal to the second expected disease spread rate. Use the adjusted estimated mitigation parameters as the target disease mitigation parameters and use the target disease mitigation parameters to predict the future disease status of rice.

[0057] In step S1, disease-related image data of the target rice planting area is first collected using remote sensing equipment such as drones. After analyzing the image data, remote sensing disease parameters are generated. These parameters directly reflect the actual state of rice diseases in the current planting area, such as the current spread rate and coverage of lesions. Simultaneously, based on past experience or preliminary experimental results in rice disease control, estimated mitigation parameters are preset. These parameters are initialized using average values ​​or regression models based on historical intervention data of the same disease type and planting conditions in the target area over the past 3-5 years. This clarifies the expected reduction in rice diseases in the target area within each unit time period after intervention measures are implemented, providing an initial standard for subsequent evaluation of the intervention effect. Intervention measures include pesticide application and fertilization. The initial values ​​of the estimated mitigation parameters are determined based on historical rice disease intervention data of the target area over the past 3-5 years, specifically referring to the average reduction in disease severity within a single unit time period after implementing conventional intervention measures under similar disease types and planting conditions.

[0058] In step S2, the remote sensing disease parameters and the estimated mitigation parameters obtained in S1 are input into the trained disease prediction model. The model performs calculations based on the preset disease spread logic and finally obtains the first expected disease spread rate. The first expected disease spread rate is a prediction based on the ideal state of the current actual disease and the preset intervention effect, reflecting the control effect of intervention measures on disease spread without considering other environmental influences.

[0059] Step S3 is implemented in two parts: First, using a trained disease prediction model, combined with historical disease data from the paddy field and current basic growth conditions, the expected natural disease development rate without any intervention is calculated. This expected natural disease development rate represents the actual development of rice diseases in each unit time period within a single cycle without intervention, such as how much more the lesions would spread each day without pesticide application. Then, the estimated mitigation parameter preset in S1 is subtracted from the expected natural disease development rate to obtain the remaining disease parameter. This remaining parameter represents the net disease development rate after intervention. Both the expected natural disease development rate and the estimated mitigation parameter are defined using a unified dimension. The expected natural disease development rate is quantified as the absolute increase in disease severity per unit time without intervention, such as the daily increase in the percentage of lesion area. The estimated mitigation parameter is correspondingly quantified as the absolute amount of disease severity expected to decrease per unit time after intervention, such as the daily decrease in the percentage of lesion area. Both have the same dimensions, and their difference represents the residual disease parameter, which quantifies the baseline development of the disease that remains uncontrolled even after intervention measures are implemented. In step S3, the calculation of the expected natural disease development rate is not based on the state after intervention, but rather on simulations of the target area using a trained disease prediction model. During training, this model uses historical remote sensing data from periods without intervention, synchronized environmental parameters, and corresponding field-measured disease data as learning samples. Therefore, when the model is applied, it can simulate the most likely development of the disease without any human intervention, based on currently collected remote sensing disease parameters and environmental conditions, thus eliminating the influence of intervention measures and ensuring that the expected natural disease development rate truly reflects the natural development trend of the disease.

[0060] The difference between the expected natural disease development rate and the remote sensing disease parameters in step S1 lies in the fact that the remote sensing disease parameters reflect the static state of rice diseases in the target area at the current time point, such as the proportion of lesion area and the proportion of diseased panicles, while the expected natural disease development rate reflects the dynamic development degree of the disease within a single unit of time when no intervention measures are taken, such as the daily average growth rate of the proportion of lesion area. The calculation of the expected natural disease development rate is achieved through a disease prediction model. For example, the model input data includes remote sensing disease parameters, data collected in the same period of the past three years in the target area during the early stage of the disease and before the implementation of intervention measures as an approximate sample of the uninterventional state, and information on the current growth stage of rice. The model is based on the correlation between the static disease state and the dynamic development degree in historical data, and outputs the disease development value within a single unit of time, i.e., the expected natural disease development rate. The remote sensing disease parameters represent the static state of the disease at the current moment, while the expected natural disease development rate represents the dynamic development rate predicted by the model in the future without intervention.

[0061] In step S4, soil condition parameters of the target rice planting area, such as soil fertility and acid-base characteristics, are input together with the remaining disease parameters obtained in S3 into the disease prediction model. The model calculates the second expected disease spread rate by combining the influence of soil conditions on rice disease resistance and disease spread rate. Compared with the first expected disease spread rate, the second expected disease spread rate incorporates the influence of the actual soil environment and is closer to the actual disease development in the paddy field.

[0062] Step S5 is the core calibration step. By gradually adjusting the preset mitigation parameters in S1, the first expected disease spread rate in S2 and the second expected disease spread rate in S4 are recalculated after each adjustment. The changes in the two rates are observed until the two rates are completely equal. In practice, errors are allowed. When the difference between the two rates is ≤0.01% / day, the two rates are considered to be completely equal. At this time, the adjusted predicted mitigation parameters are the target disease mitigation parameters. When adjusting the predicted mitigation parameters, users can associate them with a pre-established intervention effect lookup table. This lookup table stores the typical disease mitigation parameter ranges corresponding to different agents and dosages under different environmental conditions. Growers can choose the most suitable intervention plan based on this. Finally, based on the target disease mitigation parameters, the future development of the rice disease is predicted, such as how much the lesions will spread in the next week and whether they can be controlled.

[0063] The calibration process in step S5, which equates the first expected rate with the second expected rate, aims to find a balance between the predicted intervention effect and the actual environmental impact. The first expected rate is a prediction based on the ideal intervention effect, while the second expected rate is a prediction that comprehensively considers the constraints of the actual soil and environment. When the two are equal, it means that the preset intervention effect can just adapt to the disease development pressure under the current environment. At this point, the target disease mitigation parameter meets both the control expectation and the realistic conditions, thus becoming a reliable and achievable prediction benchmark. The calibration process is implemented through an iterative optimization algorithm, ensuring the availability of this balance point and the rationality of the results.

[0064] The advantages of this invention revolve around data accuracy, precise prediction, and close alignment with reality, effectively solving the problems of large subjective biases and neglect of key factors in traditional predictions. Initial data is more reliable. Step S1 acquires remote-sensing disease parameters through remote sensing equipment, rather than manual estimation, avoiding subjective errors from human observation and ensuring that the data reflecting the current disease state is accurate and objective. This lays the foundation for all subsequent calculations and predictions, reducing the chain reaction of biases caused by inaccurate initial data. Step S3 calculates the expected natural disease development rate without intervention, enabling early understanding of the natural deterioration trend of the disease and avoiding misjudgments of disease development due to focusing only on the intervention effect and ignoring the risks of non-intervention. Step S4 introduces soil condition parameters, considering environmental factors of rice growth and the impact of soil on disease spread, ensuring that the prediction is no longer based on ideal conditions but closely reflects the actual planting scenario. Step S5 calibrates the two rates to equalize, resulting in the target disease mitigation parameter. This parameter balances the objectivity of the remote sensing data in S1 with the actual impact of soil conditions in S4. It is equivalent to a verification process involving ideal prediction, actual correction, and parameter optimization, ensuring an unbiased assessment of the intervention effect. Using this parameter to predict future disease progression allows for a more accurate assessment of disease development, helping growers formulate more reasonable control plans and reducing errors in control decisions due to misjudgments of intervention effects or neglect of soil factors. When adjusting the estimated mitigation parameter, the first expected disease spread rate is made equal to the second expected disease spread rate. The first expected rate reflects the ideal intervention effect of the remote sensing disease parameters and the estimated mitigation parameter, while the second expected rate reflects the actual intervention effect of soil conditions, remaining disease parameters, and environmental parameters. Equal rates mean that the intervention effect fully adapts to the actual environmental impact on disease spread, avoiding a disconnect between ideal prediction and reality. The adjusted estimated mitigation parameter is then used as the target disease mitigation parameter to predict the future disease condition of rice.

[0065] First, based on historical intervention experience, a reasonable initial value and range are set for the estimated mitigation parameters. Then, the estimated mitigation parameters are iteratively adjusted using a preset fixed or adaptive step size. In each iteration, steps S2 and S4 need to be re-executed to calculate the corresponding first and second expected disease spread rates. Iteration continues, and specific iterative adjustments can be made using optimization algorithms such as the bisection method or gradient descent method to adjust the estimated mitigation parameters until the absolute value of the difference between the two rates is less than or equal to the error threshold. The error threshold is determined comprehensively based on the accuracy requirements of agricultural disease prediction and the stability of the model calculation.

[0066] In step S5, during the adjustment of the estimated mitigation parameters to achieve equality between the first and second expected disease spread rates, an iterative optimization method is used to calibrate the parameters. First, based on historical intervention data of rice diseases in the target area and initial remote sensing disease parameters, an initial range for the estimated mitigation parameters is set. This range must cover a reasonable interval of disease mitigation in historical intervention measures to ensure the initial values ​​have practical reference value. Then, starting from the initial estimated mitigation parameters, the parameter values ​​are gradually adjusted according to a preset step size. The step size is determined based on the fluctuation range of the disease development rate to ensure a balance between adjustment accuracy and efficiency. Each adjustment is simultaneously input into the trained disease prediction model to calculate the corresponding first and second expected disease spread rates. During the iteration process, the difference between the two rates is compared in real time. If the first expected rate is greater than the second expected rate, it indicates that the current estimated mitigation parameter setting is too high and needs to be lowered by the step size; if the first expected rate is less than the second expected rate, it indicates that the current estimated mitigation parameter setting is too low and needs to be raised by the step size. The calibration process involves continuous iterative adjustments until the difference between the two rates falls within a preset error threshold. This threshold is determined based on the accuracy requirements for disease prediction in the target area, ensuring that the calibration results meet practical application needs. The corresponding estimated mitigation parameters at this point are the target disease mitigation parameters. Through this iterative optimization process, parameter values ​​that are suitable for the current remote sensing data, soil conditions, and environmental factors can be found, ensuring the operability and reliability of the dual-rate calibration.

[0067] The training and validation process of the disease prediction model used in this invention meets the following requirements: the training data source must cover multiple scenarios and multi-dimensional information, specifically including measured data from 12 typical rice-growing areas in southern and northern rice-growing areas, data dimensions covering 5-10 years of multispectral remote sensing data, a sample size of no less than 5000 sets, resolution of 10m, hyperspectral remote sensing data, wavelength range of 400-1000nm, soil condition parameters, soil pH value of 4.5-8.5, soil fertility available nitrogen content of 20-150mg / kg, soil moisture content of 15%-40%, environmental parameters, daily average temperature of 15-35℃, daily average relative humidity of 60%-95%, daily precipitation of 0-50mm, and field disease survey data of the corresponding areas, lesion type, lesion area ratio, disease spread rate, and the sample size of each data category is no less than 5000 sets. The training data annotation process involved three technicians with over five years of experience in rice disease research manually annotating lesion areas based on high-resolution remote sensing imagery. A pre-trained semantic segmentation model was then used to optimize the edges of the annotations. Finally, agricultural plant protection experts verified the annotation accuracy, requiring an accuracy of at least 95% to be used as training labels. The model training process employed a five-fold cross-validation method, using prediction accuracy, mean absolute error, and root mean square error as validation metrics.

[0068] The remaining disease parameter is the difference between the expected natural disease development rate and the estimated mitigation parameter. The expected natural disease development rate reflects the actual development of rice diseases in the target area within each unit time period of a single cycle when no intervention measures are taken. The estimated mitigation parameter reflects the expected reduction of diseases within each unit time period of a single cycle after intervention measures are taken. Therefore, the remaining disease parameter accurately represents the basic degree of disease development that will continue to develop in the target area after intervention measures are taken. This parameter is an intermediate variable that connects the intervention effect with the actual environmental impact. In step S4, the second expected disease spread rate is calculated based on the remaining disease parameters. This is not a recalculation of the disease development after intervention, but rather a further integration of the influence of soil condition parameters and environmental parameters on disease spread. The remaining disease parameters provide an initial benchmark for disease development after intervention, soil condition parameters directly affect the rice's own disease resistance and pathogen reproduction rate, and environmental parameters determine the suitability for pathogen spore germination and spread. The second expected disease spread rate calculated by combining these three factors can truly reflect the actual spread trend of the disease in the actual growth environment of the target area after the implementation of intervention measures. This makes up for the shortcomings of the first expected rate calculated based solely on remote sensing data and intervention effects, which did not consider environmental adaptability. It provides a reference basis that fits the actual planting scenario for subsequent dual-rate calibration, ensuring the accuracy of the final prediction results.

[0069] As one implementation method, in step S1, the process of acquiring remote sensing disease parameters includes: acquiring multispectral and hyperspectral remote sensing data of the target rice planting area; performing noise reduction, radiometric correction, and geometric correction preprocessing on the multispectral and hyperspectral remote sensing data; extracting spectral features, texture features, and morphological features of rice lesions from the preprocessed remote sensing data, and generating remote sensing disease parameters based on the extracted features.

[0070] Multispectral remote sensing data was collected using a drone equipped with a 6-band multispectral sensor. Key bands selected included the 560nm green band (adapted for identifying the dark green spectral distribution of rice false smut sclerotia) and the 668nm red band (distinguishing between diseased and normal panicles by chlorophyll differences). The drone flew at an altitude of 80 meters, following a serpentine flight path to ensure image coverage of every square meter of paddy field, balancing data acquisition efficiency and spatial resolution. Hyperspectral remote sensing data was supplemented by a hyperspectral imager, acquiring continuous spectral curves in the 400-1000nm range to accurately capture the unique reflection peaks of the rice false smut sclerotia at specific wavelengths. This combined acquisition of both types of data leverages the rapid coverage advantage of multispectral data while utilizing the fine resolution of hyperspectral data, avoiding the inaccurate identification of early-stage rice false smut panicles by a single data type.

[0071] The denoising process employs a combination of Gaussian filtering and median filtering. Gaussian filtering smooths the overall noise in the image, while median filtering removes localized abnormal noise, ensuring that image details in the rice blast diseased panicle area are not obscured by noise. Radiometric correction uses absolute radiometric correction based on sensor calibration coefficients to convert image grayscale values ​​into true surface reflectance, eliminating the impact of light intensity differences at different acquisition times on the spectral characteristics of diseased panicles. Geometric correction uses 12 pre-set ground control points in the field as a benchmark, employing a polynomial fitting algorithm to correct image geometric distortions caused by UAV flight attitude deviations, ensuring that every pixel in the image accurately corresponds to the actual location in the paddy field, avoiding errors in subsequent diseased panicle location and area statistics due to coordinate offset.

[0072] Spectral feature extraction was performed by comparing the spectral curves of rice blast fungal sclerotia and normal rice panicles, identifying three key spectral indicators: 560nm reflectance, 668nm reflectance, and the difference in reflectance between 580nm and 620nm, quantifying the spectral differences between diseased and normal panicles. Texture feature extraction employed a gray-level co-occurrence matrix algorithm to calculate three indicators—texture density, contrast, and correlation—in the diseased panicle region, capturing the rough and irregular texture characteristics of the rice blast fungal sclerotia surface. Morphological feature extraction utilized an improved U-Net semantic segmentation model to automatically segment the diseased panicle region, statistically analyzing indicators such as the equivalent diameter, area, and roundness of the diseased panicles to distinguish between single diseased panicles and clusters of diseased panicles. Based on these three types of features, the initial diseased panicle rate was calculated through weighted summation. Combining this with image data from two 24-hour intervals, the daily growth rate of the diseased panicle rate was calculated, ultimately generating remote sensing disease parameters including the initial diseased panicle rate and the daily growth rate of the diseased panicle rate.

[0073] As one implementation method, step S4 further includes: using a trained disease prediction model, calculating and updating the second expected disease spread rate based on soil condition parameters, remaining disease parameters, and environmental parameters of the target rice planting area; the soil condition parameters include soil pH, soil fertility, and soil moisture content data.

[0074] In the planting area, eight soil sampling points were set up at a density of one sampling point per 10 mu (approximately 6.7 hectares). Soil samples from the 0-20 cm topsoil layer were collected at each sampling point using the soil profile sampling method. Soil pH was measured using a portable soil pH meter, soil fertility was measured using a soil fertility rapid tester, and soil moisture content was measured using a soil moisture sensor. After collection, abnormal samples were removed, and the average value of the data from the eight sampling points was taken as the overall soil condition parameters for the target area.

[0075] When the three types of parameters are input into the trained improved U-Net semantic segmentation and GRU temporal prediction model, the model calculates the second expected disease spread rate through a multi-dimensional feature fusion mechanism. First, all parameters are standardized and preprocessed, mapping soil pH, available nitrogen, soil moisture content, remaining disease parameters, and environmental parameters to the [0, 1] interval to eliminate the interference of different parameter dimensions on the calculation results. Then, the model assigns parameter weights according to the disease incidence pattern of rice false smut. Since alkaline soil reduces rice disease resistance and high nitrogen promotes pathogen growth, the soil condition parameter weight is set to 0.3, the remaining disease parameter weight is set to 0.4 as a basic spread index, and the environmental parameter weight is set to 0.3 to ensure that key influencing factors are fully reflected. Finally, the interaction between parameters is captured by the GRU temporal unit, and combined with the spatial distribution characteristics of diseased panicles extracted by the U-Net model, the second expected disease spread rate is calculated.

[0076] As one implementation method, the disease prediction model is a deep learning model, including an input layer, a hidden layer, and an output layer. The input layer receives remotely sensed disease parameters, estimated mitigation parameters, soil condition parameters, and / or environmental parameters, performs standardized preprocessing, and then passes them to the hidden layer. Each neuron in the hidden layer receives the output from the neurons in the previous layer, processes it through the ReLU activation function, and then outputs it to the neurons in the next layer. The neurons calculate the result based on the learned weights and pass it to the output layer. The output layer outputs the predicted disease spread rate.

[0077] In the rice false smut prediction scenario, the input layer receives parameters including remote sensing disease parameters generated in step S1, preset estimated mitigation parameters, soil condition parameters involved in step S4, and environmental parameters in step S4. The input layer performs standardized preprocessing on these parameters: using the max-min normalization method, parameters of different dimensions are uniformly mapped to the range of 0 to 1. For example, a daily average temperature of 26℃ is mapped to 0.5, and a daily increase in diseased panicle rate of 3% is mapped to 0.3, eliminating the interference of parameter numerical range differences on model calculations. The preprocessed parameters are then passed to the hidden layer via a fully connected layer, laying the foundation for subsequent feature extraction.

[0078] Hidden layers are the core of the model's ability to learn complex features. They capture the nonlinear relationships between parameters through the synergistic effect of multiple neurons and the ReLU activation function. This deep learning model uses three hidden layers, each containing 64 neurons. The first layer of neurons receives standardized parameters from the input layer and performs a weighted summation based on preset initial weights. For example, a weight of 0.2 is assigned to soil pH of 8.0, and a weight of 0.3 is assigned to average daily humidity of 85%. This calculates the input value for the neuron. The ReLU activation function then processes the input, retaining only signals with positive input values, and the result is passed to the next layer of neurons. The second and third layers repeat this process, with weights dynamically adjusted during model training. This process gradually extracts multi-level information from basic parameters to complex correlated features, ultimately passing the integrated features to the output layer.

[0079] The output layer transforms the features extracted by the hidden layer into specific disease spread rate predictions, directly serving the rate calculation needs of steps S2 and S4. The output layer contains one neuron that receives the final features from the third hidden layer and maps the feature values ​​back to the actual rate range through a linear transformation, outputting either the first or second expected disease spread rate. This deep learning model addresses the challenge of handling complex interactions among multiple parameters in traditional models: the standardized preprocessing of the input layer ensures that different types of parameters can participate in the calculation collaboratively, avoiding a single parameter dominating the result; the multi-layered structure of the hidden layer and the ReLU activation function accurately capture the nonlinear relationships between soil, environment, and disease status in rice blast disease spread, improving prediction accuracy compared to traditional linear regression models; and the direct mapping of the output layer ensures that the prediction results can be directly used for parameter calibration in step S5.

[0080] As one implementation method, step S3 includes: S31: Dividing the target rice planting area into multiple monitoring units according to a preset area, and treating each monitoring unit as an independent node; S32: Calculating the rice disease severity value corresponding to each node, and calculating the expected natural disease development rate based on the node disease severity value of semantic segmentation and a trained disease prediction model; S33: Using the difference between the expected natural disease development rate and the estimated mitigation parameter as the remaining disease parameter.

[0081] S31: Regional Division and Node Layout. The target rice planting area is divided into multiple monitoring units according to a preset area, and each monitoring unit is used as an independent analysis node. Decomposing the overall prediction of a large area into multiple local predictions of smaller units and then integrating them helps to capture the spatial heterogeneity of the disease and avoids the impact of severely affected local areas being masked by the overall average.

[0082] S32: Calculation of Node Disease Severity Based on Semantic Segmentation. For each node in S31, based on preprocessed high-resolution remote sensing data, a pre-trained semantic segmentation model automatically identifies and segments the lesion areas of rice plants within the node. Subsequently, the ratio of the area of ​​the lesion area to the total coverage area of ​​rice plants within the node is calculated, and this ratio is used as the real-time disease status value of the node. This step utilizes computer vision technology to replace manual interpretation, achieving objective and quantitative measurement of disease severity, and providing reliable, fine-grained spatial input data for subsequent model prediction. Integrating Node Data and Executing Model Prediction. The real-time disease status values ​​of all nodes obtained in S32, combined with the corresponding environmental parameters and soil condition parameters of each node, are input into the trained disease prediction model. During training, the model has learned the complex mapping relationship between this type of fine-grained spatial data and the future disease development rate under uninterrupted conditions. By processing this integrated node data, the model ultimately outputs a more accurate representation of the expected natural disease development rate of the entire target area within a future unit of time.

[0083] S33: Calculate the remaining disease parameters. The difference between the expected natural disease development rate and the preset estimated mitigation parameter is used as the remaining disease parameter.

[0084] It should be noted that S31 to S32 above are specific data preprocessing methods to provide optimized input for model prediction. Those skilled in the art should understand that calculating the expected natural development rate of the disease is essentially the model prediction behavior, and node partitioning and semantic segmentation are optional preliminary steps to improve the accuracy of this prediction, not the only way.

[0085] In one implementation method, step S32, the process of calculating the degree of rice disease corresponding to each node includes using a semantic segmentation model to identify the lesion area of ​​rice plants within the node based on preprocessed remote sensing data; calculating the ratio of the lesion area area to the total coverage area of ​​rice plants within the node, and using this ratio as the degree of rice disease corresponding to the node.

[0086] Semantic segmentation models can distinguish between lesions and healthy plants based on the spectral and morphological features of remote sensing data, especially capturing early, minute lesions. This avoids subjective errors or missed detections inherent in manual identification, making the identification of diseased areas at each node more reliable. It ensures the comparability and quantification of disease severity. By using the ratio of lesion area to total plant coverage area, the influence of planting density differences at different nodes is eliminated, allowing for direct comparison of disease severity values ​​across nodes. Simultaneously, visual features are transformed into specific numerical values, providing a unified and stable local data foundation for subsequent averaging to calculate the expected natural disease development rate.

[0087] As one implementation method, the method further includes the following steps: S6: Calculate the disease relief rate based on the target disease relief parameters and the expected natural disease development rate, wherein the disease relief rate is the ratio of the target disease relief parameters to the expected natural disease development rate; S7: Combine the disease relief rate and environmental parameter prediction data for future cycles to generate a report on the disease development trend of the target rice planting area within the next 1-2 growth cycles.

[0088] The disease mitigation rate is the ratio of the target disease mitigation parameter to the expected natural disease progression rate. When the mitigation rate > 100%, it indicates that the intervention measures have reduced the disease severity beyond the natural disease progression rate without intervention, and the disease will show a decreasing trend. The disease mitigation rate provides a direct and quantifiable measure of the intervention effect. By comparing the target disease mitigation parameter with the expected natural disease progression rate, abstract parameters are transformed into concrete efficiency indicators, allowing growers to clearly understand the percentage of disease that the intervention measures can control and quickly determine whether the plan has achieved its intended goals. Combining the disease mitigation rate with future environmental prediction data to generate trend reports allows for the prediction of disease trends 1-2 growth cycles in advance. Each growth cycle is set at 30-45 days, helping growers adjust measures in advance, such as whether to strengthen interventions to avoid losses caused by sudden disease outbreaks.

[0089] The calculation of disease mitigation rate is based on the premise of unified parameter dimensions. The expected natural disease development rate represents the amount of disease that will develop naturally per unit time without intervention, while the target disease mitigation parameter represents the effective reduction that can be achieved per unit time after calibration. The disease mitigation rate is the ratio of the two. This ratio directly reflects the control capability of intervention measures relative to the natural development of the disease. For example, if the ratio is 120%, it indicates that the amount of disease that the intervention measures can suppress exceeds the natural growth of the disease at different expected times by 20%, suggesting that the disease will be effectively controlled and show a decreasing trend.

[0090] This invention discloses a rice disease prediction system based on remote sensing monitoring, using the rice disease prediction method based on remote sensing monitoring as described above. The system includes: a processor module, an input device module, an output device module, and a memory module, all interconnected. The memory module stores a computer program, which includes program instructions. The input device module receives remote sensing disease parameters, soil condition parameters, and environmental parameters of the target rice planting area. The processor module is configured to invoke the program instructions and execute the rice disease prediction method based on remote sensing monitoring. The output device module outputs target disease mitigation parameters and a report on disease progression trends.

[0091] The input module centrally receives various data, including remote sensing disease parameters and soil condition parameters, avoiding errors and inefficiencies from scattered manual collection. The processor module executes prediction methods precisely by calling program instructions, eliminating the need for manual calculations and reducing biases caused by human intervention. The memory module stores programs and data, ensuring the stability of method execution and preventing data loss. The collaborative work of all modules makes the prediction process more efficient and the results more reliable. Simultaneously, the system output directly addresses practical application needs, providing growers with clear decision-making support. The output module specifically outputs reports on target disease mitigation parameters and disease development trends, eliminating the need for growers to extract information from complex data. This allows for rapid understanding of core indicators and future disease trends in intervention plans, directly enabling the formulation or adjustment of control measures. This transforms prediction results into actual control capabilities, effectively reducing yield and quality losses caused by rice diseases.

[0092] The input device module specifically includes interfaces for receiving multispectral and hyperspectral remote sensing data, data transmission links for a network of small automatic weather stations deployed in the field, and input channels for soil parameter acquisition equipment, used for centralized reception of multi-source heterogeneous data. The processor module, after calling program instructions from the memory module, specifically executes data standardization preprocessing, loading and computation of the disease prediction model, and control flow for iterative calibration algorithms. The output device module encompasses a graphical user interface and a data report generation unit, used to intuitively display target disease mitigation parameters and disease progression trends to the user. All modules work collaboratively via a system bus to form a complete automated prediction system.

[0093] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for predicting rice diseases based on remote sensing monitoring, characterized in that, Includes the following steps: S1: Receive remote sensing disease parameters of the target rice planting area. The remote sensing disease parameters are generated by collecting disease-related image data of rice in the target area using remote sensing equipment. Preset mitigation parameters for rice diseases in the target area. The predicted mitigation parameters represent the expected reduction in rice diseases in the target area in each unit time period within a single cycle after intervention measures are taken. S2: Using a trained disease prediction model, the first expected disease spread rate is calculated based on remote sensing disease parameters and estimated mitigation parameters. S3: Using a trained disease prediction model, calculate the expected natural disease development rate in the target rice planting area. The expected natural disease development rate represents the actual development of rice diseases in the target area in each unit time period within a single cycle when no intervention measures are taken. The difference between the expected natural disease development rate and the estimated mitigation parameter is used as the remaining disease parameter; S4: Using a trained disease prediction model, the second expected disease spread rate is calculated based on soil condition parameters and remaining disease parameters in the target rice planting area. S5: Adjust the estimated mitigation parameters so that the first expected disease spread rate is equal to the second expected disease spread rate. Use the adjusted estimated mitigation parameters as the target disease mitigation parameters and use the target disease mitigation parameters to predict the future disease status of rice.

2. The method for predicting rice diseases based on remote sensing monitoring according to claim 1, characterized in that, In step S1, the process of acquiring remote sensing disease parameters includes: Acquire multispectral and hyperspectral remote sensing data of the target rice planting area; Denoising, radiometric correction, and geometric correction preprocessing are performed on multispectral and hyperspectral remote sensing data. The spectral, texture, and morphological features of rice lesions are extracted from the preprocessed remote sensing data, and remote sensing disease parameters are generated based on the extracted features.

3. The method for predicting rice diseases based on remote sensing monitoring according to claim 1, characterized in that, Step S4 also includes: Using a trained disease prediction model, the second expected disease spread rate is calculated and updated based on soil condition parameters, remaining disease parameters, and environmental parameters of the target rice planting area. Soil condition parameters include soil pH, soil fertility, and soil moisture content data.

4. The method for predicting rice diseases based on remote sensing monitoring according to claim 3, characterized in that, The disease prediction model is a deep learning model, consisting of an input layer, a hidden layer, and an output layer; The input layer is used to receive input remote sensing disease parameters, estimated mitigation parameters, soil condition parameters and / or environmental parameters, perform standardized preprocessing, and then pass them to the hidden layer; Each neuron in the hidden layer receives the output from the neurons in the previous layer, processes it through the ReLU activation function, and then outputs it to the neurons in the next layer. The neurons calculate the result based on the learned weights and pass it to the output layer. The output layer is used to output the predicted disease spread rate results.

5. The method for predicting rice diseases based on remote sensing monitoring according to claim 1, characterized in that, Step S3 includes: S31: Divide the target rice planting area into multiple monitoring units according to the preset area, and treat each monitoring unit as an independent node; S32: Calculate the degree of rice disease corresponding to each node. Based on the node disease degree value of semantic segmentation, calculate the expected natural development rate of disease through a trained disease prediction model. S33: The difference between the expected natural development rate of the disease and the estimated mitigation parameter is used as the remaining disease parameter.

6. The method for predicting rice diseases based on remote sensing monitoring according to claim 5, characterized in that, Step S32, the process of calculating the rice disease severity value corresponding to each node includes: Based on the preprocessed remote sensing data, a semantic segmentation model was used to identify the diseased areas of rice plants within the nodes; Calculate the ratio of the area of ​​the lesion region to the total coverage area of ​​rice plants within the node, and use this ratio as the degree of rice disease corresponding to the node.

7. The method for predicting rice diseases based on remote sensing monitoring according to claim 1, characterized in that, The method also includes the following steps: S6: Calculate the disease relief rate based on the target disease relief parameters and the expected natural disease development rate. The disease relief rate is the ratio of the target disease relief parameters to the expected natural disease development rate.

8. The method for predicting rice diseases based on remote sensing monitoring according to claim 7, characterized in that, The method also includes the following steps: S7: Combine disease mitigation rate and environmental parameter prediction data for future cycles to generate a report on the disease development trend of the target rice planting area within the next 1-2 growth cycles.

9. A rice disease prediction system based on remote sensing monitoring, using the rice disease prediction method based on remote sensing monitoring as described in any one of claims 1-8, characterized in that, include: The processor module, input device module, output device module, and memory module are interconnected. The memory module is used to store computer programs, which include program instructions; The input device module is used to receive remote sensing disease parameters, soil condition parameters, and environmental parameters of the target rice planting area; The processor module is configured to call program instructions to execute a rice disease prediction method based on remote sensing monitoring. The output device module is used to output the target disease relief parameters and disease development trend reports.

10. A rice disease prediction system based on remote sensing monitoring according to claim 9, characterized in that: The processor module is further configured as follows: The estimated mitigation parameters are adjusted iteratively, and the first and second expected disease spread rates are recalculated until the absolute value of the difference between the two rates is less than or equal to the preset error threshold.

Images