A crop water-nitrogen stress diagnosis and self-adaptive correction method based on physical model driving and multi-source data fusion

Abstract

This invention discloses a method for diagnosing and adaptively correcting crop water-nitrogen stress based on a physical model-driven approach and multi-source data fusion. First, simulated spectral-canopy parameter samples are generated and trained to obtain an initial prediction model after sensor band response uniformity and observational perturbation simulation. Then, multi-source observation data, including remote sensing images, meteorological data, soil data, and crop growth stage data, are collected, cleaned, quality controlled, spatiotemporally aligned, and standardized to construct a unified data foundation. Based on the canopy energy balance mechanism, different water constraint scenarios are set to calculate the dynamic upper and lower limits of canopy temperature, and the temperature-normalized water stress index (CWSI) is calculated by combining canopy observation temperatures obtained from thermal infrared remote sensing. Simultaneously, canopy spectral features are extracted based on hyperspectral remote sensing and input into the initial prediction model to obtain estimated values ​​of canopy nitrogen concentration and aboveground dry matter. The nitrogen nutrient index (NNI) is calculated by combining the critical nitrogen dilution curve, thereby achieving joint diagnosis of crop water-nitrogen stress. By continuously adding new measured data, the prediction model and key diagnostic parameters are iteratively corrected and updated, achieving adaptive specialization and improved diagnostic accuracy for different crops, regions, and growth stages.

Description

Technical Field

[0001] This invention belongs to the field of crop water and nitrogen stress diagnosis technology, specifically relating to a water-nitrogen stress diagnosis method for different crops throughout their entire growth cycle. Background Technology

[0002] Water and nitrogen are key environmental factors affecting crop growth, development, and yield. Accurate and timely diagnosis of crop water-nitrogen status is a crucial foundation for optimizing irrigation and nitrogen management strategies. Quantitative identification of crop water and nitrogen stress levels allows for precision irrigation and nitrogen application while ensuring normal crop growth and yield. This improves water and fertilizer resource utilization efficiency, reduces agricultural inputs, and minimizes nitrogen loss and environmental pollution caused by excessive nitrogen application, as well as water waste from excessive groundwater extraction. This is of great significance for sustainable agricultural development.

[0003] However, existing methods for diagnosing crop water-nitrogen stress still have limitations. On the one hand, existing methods often use a single indicator or fixed threshold for judgment, while crop water-nitrogen status is affected by multiple factors such as climate conditions, growth stage, soil characteristics, and field management. Single indicator and fixed threshold methods are difficult to adapt to the differences in different crop varieties, regions, and production stages, and usually require repeated model construction for specific crops or regions, resulting in insufficient universality. On the other hand, the construction of existing thresholds or models often relies on the accumulation of a large amount of preliminary field trials and measured data, which is time-consuming and costly. When measured data is insufficient or environmental conditions change, it is difficult to establish or dynamically update effective diagnostic models in a timely manner. Especially in the early stages of application for new crop varieties or new production areas, there is a lack of available diagnostic models, which limits the promotion and application of related technologies.

[0004] Based on the above problems, it is necessary to propose a crop water-nitrogen stress diagnosis method that has a low dependence on measured data, can quickly establish initial diagnostic capabilities under data-scarce conditions, and can be continuously updated and optimized as measured data accumulates.

[0005] A search revealed no patent applications that are identical or similar to the technical solution of this invention. Summary of the Invention

[0006] The purpose of this invention is to provide a crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion. This method is used to construct a universally applicable initial diagnostic model under conditions of lack of or limited measured data, and to continuously correct and adapt it to regionalization through an incremental learning mechanism during agricultural production and experimentation. This enables the diagnosis and visualization of water-nitrogen stress throughout the entire growth cycle under different crops and regional conditions.

[0007] The present invention achieves the above objectives through the following technical means:

[0008] A method for diagnosing and adaptively correcting crop water-nitrogen stress based on a physical model and multi-source data fusion includes the following steps:

[0009] S1: Simulate and generate a simulated spectral-canopy parameter sample set covering different crop types, different growth stages and different canopy conditions. The simulated spectral-canopy parameter sample set includes simulated canopy reflectance spectra and corresponding canopy parameter samples.

[0010] S2: Using the simulated spectrum-canopy parameter sample set, an initial prediction model for the spectrum-canopy parameters is trained. The simulated spectrum features are used as the model input, and the canopy parameters that correspond one-to-one with the simulated spectrum are used as the model output to predict the crop canopy nitrogen concentration (CNC) and the crop aboveground dry matter (DM).

[0011] S3: Acquire multi-source observation data related to the target crop plot and construct a unified data base. The multi-source observation data includes at least remote sensing image data, meteorological data, soil data, and crop growth period information. The multi-source observation data needs to be processed, including but not limited to additional metadata, cleaning, spatiotemporal alignment, standardization, and indexing. The processed multi-source data is organized and stored according to plot identification, time dimension, and crop type to form a unified data base.

[0012] S4: Based on the physical model of canopy energy balance mechanism, and with the support of a unified data base, calculate the dynamic upper and lower limits of canopy temperature of the target crop plot on the corresponding time axis.

[0013] S5: Obtaining crop canopy observation temperature T based on thermal infrared remote sensing images c ,obs, and combined with the dynamic upper and lower limits of canopy temperature, calculate the temperature-normalized crop water stress index (CWSI);

[0014] S6: Extract crop canopy spectral features based on hyperspectral remote sensing images, input them into the initial prediction model of spectral-canopy parameters, obtain the estimated values ​​of crop CNC and DM, and calculate the crop nitrogen nutrient index NNI by combining the critical nitrogen dilution curve, which is used to identify the degree of crop nitrogen stress and nitrogen deficiency status.

[0015] S7: Continuously acquire new measured data and introduce them as new samples into the incremental learning mechanism to continuously correct and optimize the spectral-canopy parameter prediction model of S2 and the key parameters related to diagnosis in S4 to S6, so as to achieve adaptive specialization of the model for different crops, different regions and different growth stages.

[0016] Furthermore, S1 uses a physical model of the leaf-canopy radiative transfer mechanism for simulation, generating canopy reflectance spectra covering different crop types, different growth stages, and different canopy conditions within a preset parameter space.

[0017] Furthermore, the simulated spectrum is processed to ensure that its band settings are consistent with the data collected by the UAV hyperspectral remote sensing equipment. This includes: firstly, cropping the simulated spectrum according to the band range of the target equipment; then, based on the center wavelength and bandwidth information of each band of the target equipment, resampling and band-aligning the cropped simulated spectrum is performed using a band response consistency method.

[0018] Furthermore, noise perturbation is superimposed on the resampled simulated spectrum.

[0019] Furthermore, the initial prediction model employs a machine learning model or a deep learning model.

[0020] Furthermore, S4 calculates the dynamic upper and lower limits of canopy temperature for the target crop plot on the corresponding time axis:

[0021] Based on the unified data base, input data related to energy balance solution is called according to plot identification and time axis. The input data includes at least meteorological driving quantities, soil moisture state quantities and crop growth period information, and the canopy structure parameters used for energy balance calculation are determined.

[0022] Under the premise of keeping the meteorological driving force and canopy structure conditions unchanged, at least two types of water constraint scenarios are set, namely the maximum evapotranspiration scenario and the minimum evapotranspiration scenario.

[0023] Substituting the two types of water constraint scenarios into the canopy energy balance model, the energy balance equations are solved hourly or daily on the unified time axis of S3 to obtain the canopy temperature series for the corresponding time steps, which are then used as T. c ,wet and T c The dynamic sequence of dry; and the consistency constraints and rationality checks of the upper and lower temperature limits are performed to ensure T c dry≥T c The final output is the dynamic upper and lower limits of canopy temperature, which correspond one-to-one with the plot identifier and time axis.

[0024] Furthermore, S5 calculates the temperature-normalized crop water stress index (CWSI): it retrieves thermal infrared remote sensing image data and its metadata corresponding to the target plot from the unified data base, performs quality control and effective pixel screening on the thermal infrared remote sensing images, converts the thermal infrared image radiance into brightness temperature or land surface temperature, and extracts the canopy temperature observation value T based on the spatial range of the plot. c ,obs;

[0025] Tc ,obs and T c ,wet and T c The dry function matches plot identifiers and time identifiers, interpolates on the time scale or aggregates them by time window to make the three elements fit on the same spatiotemporal scale.

[0026] The Crop Water Stress Index (CWSI) is calculated based on the upper and lower temperature limits, and the results are subject to boundary constraints to ensure that 0 ≤ 0 ≤ 1. Abnormal cases are marked, corrected, or removed. Finally, the CWSI results corresponding to the plot identifier, time dimension, and spatial location are output.

[0027] Furthermore, the method for calculating the crop nitrogen nutrient index (NNI) in S6 is as follows: Hyperspectral remote sensing image data and its metadata corresponding to the target plot are retrieved from the unified data base in step S3. Quality control and effective pixel screening are performed on the hyperspectral image, and the canopy reflectance spectrum at the plot scale is extracted to construct spectral features for model input. The spectral features are then input into the initial prediction model in S2 to obtain CNC and DM estimates corresponding to the plot identifier, time dimension, and spatial location.

[0028] Calculation of critical nitrogen concentration N based on DM estimation value c NNI = CNC / N c The nitrogen nutrient index (NNI) was obtained.

[0029] Furthermore, in the S7 multi-round incremental iteration and version update mechanism: newly added training samples are combined with representative historical samples extracted from the replay sample library for training; at the same time, a frozen validation set is maintained to evaluate the performance and verify the consistency of the model before and after the update. When the updated model meets the preset performance constraints, it is released as a new version model for subsequent diagnosis; otherwise, the original version is maintained and the update results are recorded.

[0030] Furthermore, the historical playback sample library is maintained and updated. Representative samples from newly added measured data are included in the playback sample library according to dimensions such as crop type, regional conditions and growth stage. When the library capacity is limited, sample replacement or stratified retention strategies are implemented to ensure long-term coverage of key scenarios.

[0031] Compared with the prior art, the present invention can achieve at least the following beneficial effects:

[0032] (1) Low data dependence and rapid availability: Based on the physical model, simulation samples are generated and an initial diagnostic model is constructed, reducing the dependence on a large number of previous field trials and measured data, so that water-nitrogen stress diagnosis can still be carried out under the condition of scarce data.

[0033] (2) Clear physical mechanism and strong interpretability: The diagnostic process is based on the mechanism of radiation transmission and energy balance, and the results have clear physical meaning, which improves the consistency and interpretability of diagnosis.

[0034] (3) Multi-source fusion and better adaptability: It integrates information from multiple sources such as remote sensing, meteorology, soil and growth period, and improves data consistency through unified cleaning and spatiotemporal alignment processes, thereby enhancing its adaptability to complex environmental conditions.

[0035] (4) Dynamic boundary reduces threshold deviation: The upper and lower limits of canopy temperature are calculated by energy balance model and water stress index is calculated by thermal infrared observation, which reduces the deviation of fixed threshold when applied across crops, regions and growth stages.

[0036] (5) Water-nitrogen synergistic output facilitates management: outputting water stress and nitrogen nutrition diagnosis results within the same framework facilitates coordinated decision-making on irrigation and nitrogen application.

[0037] (6) Continuous updates and gradual specialization: Incremental learning mechanism is introduced to continuously correct the model using subsequent measured data, so as to achieve adaptive optimization for specific crop and regional conditions and gradually improve diagnostic accuracy. Attached Figure Description

[0038] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0039] Figure 1 This is a flowchart of a crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion in an embodiment of the present invention.

[0040] Figure 2 This diagram illustrates the steps of a multi-source data fusion method using UAV remote sensing images, meteorological data, soil data, crop data, and other data in an embodiment of the present invention.

[0041] Figure 3 This is a diagram illustrating the water diagnostic steps in an embodiment of the present invention.

[0042] Figure 4 This is a diagram illustrating the nitrogen diagnosis steps in an embodiment of the present invention.

[0043] Figure 5 This is a flowchart illustrating the water nitrogen diagnosis and parameter adaptive correction method in an embodiment of the present invention. Detailed Implementation

[0044] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as a limitation of the present invention, but rather as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0045] It should be understood that the terminology used in this invention is merely for describing particular embodiments and is not intended to limit the invention. Furthermore, with respect to numerical ranges in this invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Every smaller range between any stated value or intermediate value within a stated range, and any other stated value or intermediate value within said range, is also included in this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0046] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. While only preferred methods and materials have been described herein, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this invention. All references to this specification are incorporated by way of citation to disclose and describe methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0047] Various modifications and variations can be made to the specific embodiments described in this specification without departing from the scope or spirit of the invention, as will be apparent to those skilled in the art. Other embodiments derived from this specification will also be obvious to those skilled in the art. This application specification and embodiments are merely exemplary.

[0048] The terms “include,” “including,” “have,” “contain,” etc., used in this article are all open-ended terms, meaning that they include but are not limited to.

[0049] This invention provides a method for diagnosing and adaptively correcting crop water-nitrogen stress based on a physical model-driven approach and multi-source data fusion. The overall process includes: first, generating simulated spectral-canopy parameter samples based on a radiative transfer physical model and training an initial prediction model for spectral-canopy parameters; then, collecting and constructing a unified data foundation of multi-source observation data for the target plot; based on this, calculating the dynamic upper and lower limits of canopy temperature based on the energy balance mechanism and calculating the water stress index (CWSI) using thermal infrared remote sensing; simultaneously, using hyperspectral remote sensing combined with the initial prediction model to estimate canopy nitrogen concentration and aboveground dry matter and calculate the nitrogen nutrient index (NNI) to achieve joint diagnosis of water-nitrogen stress; during system operation, continuously integrating newly measured data, and iteratively correcting and updating the prediction model and diagnostic-related key parameters through incremental learning, thereby achieving adaptive specialization and improved diagnostic accuracy for different crops, regions, and growth stages.

[0050] To achieve the above objectives, the method steps of this invention are as follows.

[0051] S1: A physical model of the leaf-canopy radiative transfer mechanism is used to generate simulated canopy reflectance spectra and corresponding canopy parameter samples. These canopy parameter samples include parameters related to crop canopy growth status, leaf biochemical composition, and canopy structure from the model input parameters, as well as target diagnostic parameters calculated from these parameters. The model input parameters may include leaf structure parameters, chlorophyll content, leaf water content, protein content, carbon-based composition, leaf area index, average leaf tilt angle, soil brightness, and observed geometric parameters, as listed in Table 1 below. The generated simulated spectral-canopy parameter sample set serves as the initial training sample for subsequently constructing a spectral-canopy parameter prediction model.

[0052] In this embodiment, the physical model of the radiative transfer mechanism is preferably implemented using the PROSAIL-PRO model (see Table 1), which is used to generate canopy reflectance spectra covering different crop types, different growth stages and different canopy conditions within a preset parameter space; however, the present invention is not limited to this model, and any equivalent physical model that can realize leaf-canopy radiative transfer simulation can be used instead.

[0053] More preferably, to ensure that the simulated spectrum is consistent with the data acquired by the UAV hyperspectral remote sensing equipment in terms of band settings, the simulated spectrum needs to be processed. Specific operations include: firstly, cropping the simulated spectrum according to the band range of the target equipment; then, based on the center wavelength and bandwidth information of each band of the target equipment, resampling and band-aligning the cropped simulated spectrum using a band response uniformity method. The band response uniformity method can be implemented, for example, using a Gaussian weighted approach: using the center wavelength of the target band as the weight center, determining the standard deviation based on the band bandwidth or preset parameters to construct a weight function, and then performing a weighted summation and normalization of the simulated spectrum within the corresponding band range to obtain a simulated reflectance that corresponds one-to-one with the measured band.

[0054] Furthermore, to simulate the effects of sensor noise, atmospheric disturbances, and stray light in actual observations, noise perturbations (such as Gaussian white noise or other random perturbations) can be superimposed on the resampled simulated spectrum to obtain simulated training samples that are closer to the actual collection distribution.

[0055] Table 1. PROSAIL-PRO Input Parameters

[0056]

[0057] Note: The parameters listed in Table 1 are example parameters. The actual parameter items and units can be adjusted as the model is implemented.

[0058] S2: Based on the simulated spectral-canopy parameter sample set generated in step S1, an initial prediction model for spectral-canopy parameters is trained. The initial prediction model can be used to achieve an initial estimate of crop canopy parameters under the condition of insufficient measured samples.

[0059] Specifically, step S1 constructs a simulation training set covering different crop types, different growth stages, and different canopy states. The simulated spectral features after band response uniformization and observational perturbation simulation processing in step S1 are used as model inputs, and the canopy parameters corresponding one-to-one with the simulated spectra are used as model outputs. Based on this, an initial prediction model for predicting crop canopy nitrogen concentration (CNC) and crop aboveground dry matter (DM) and their related intermediate variables is trained.

[0060] The initial prediction model can be implemented using a machine learning model or a deep learning model, and can employ a multi-task learning approach to achieve joint prediction of multiple parameters. DM can be the direct output of the model, or it can be obtained from intermediate variables output by the model combined with a preset conversion relationship. This initial prediction model serves as the foundation for subsequent incremental learning and continuous correction using newly added measured data.

[0061] S3: Acquire multi-source observation data related to the target crop plots and construct a unified data foundation.

[0062] S3.1: Acquire multi-source observation data (see Table 2). The multi-source observation data includes at least remote sensing image data, meteorological data, soil data, and crop growth period information.

[0063] S3.2: Add metadata to multi-source observation data. The metadata includes at least the data source identifier, acquisition time identifier, spatial reference information, and plot identifier or geographical location identifier.

[0064] S3.3: Perform an integrity check on the multi-source observation data, identify missing fields, abnormal records or invalid data, and perform validity screening according to preset quality rules, removing or marking useless data. The preset quality rules include field integrity rules, numerical range rules, temporal validity rules, spatial validity rules, remote sensing image quality rules, meteorological and soil data quality rules, and duplicate record consistency rules. Specifically, field integrity rules determine whether the acquisition time, plot identifier, spatial coordinates, crop type, growth stage, and key observation variables are missing; numerical range rules determine whether reflectance, canopy temperature, meteorological factors, soil moisture, and crop parameters are within a preset reasonable physical range; temporal validity rules determine whether the observation data is within the target growth stage or a preset time window; spatial validity rules determine whether the observation point or image data falls within the target plot area and has a valid spatial reference; remote sensing image quality rules remove cloud shadows, shadows, overexposure, underexposure, splicing misalignment, non-vegetation backgrounds, and areas with severe noise; meteorological and soil data quality rules identify abnormal records such as sensor offline, continuous missing measurements, abnormal mutations, or long-term constant values; and duplicate record consistency rules identify and merge duplicate or conflicting records from the same plot, the same time, and the same data source.

[0065] S3.4: Perform cleaning and quality control processing on the selected available datasets, including outlier identification and processing, missing value completion or imputation, noise suppression, and duplicate record merging; among them, remote sensing image data can be further processed by radiometric homogenization and quality masking, and meteorological and soil time series data can be processed by abrupt change point identification and smoothing.

[0066] S3.5: Spatial Matching. After completing quality control, spatial correspondence between data from different sources and target plots is established based on geographic location identifiers and spatial reference information.

[0067] S3.6: Time Alignment. Based on time identifiers, various types of data are aligned to construct a unified time axis or time window, making remote sensing observation times comparable and fusionable with meteorological and soil time series at the same time scale.

[0068] S3.7: Perform data format unification, unit consistency and standardization processing on the spatiotemporally aligned data, including field naming and data type unification, numerical unit and dimension conversion consistency; if necessary, perform normalization or standardization on the input features to meet the input requirements of subsequent models.

[0069] S3.8: Organize and store the processed multi-source data according to plot identifier, time dimension and crop type to form a unified data base; wherein, the data base supports indexing and querying by plot, time and crop category, and supports data calls for subsequent diagnostic calculations and incremental model updates.

[0070] Table 2. Data Acquisition and Corresponding Platforms

[0071]

[0072] S4: Based on the physical model of the canopy energy balance mechanism, and supported by the unified data base formed in step S3, the dynamic upper and lower limits of canopy temperature of the target crop plot on the corresponding time axis are calculated. The canopy energy balance model is based on the energy budget relationship between net radiation, sensible heat flux, latent heat flux and soil heat flux, and its basic relationship is shown in equation (6) below. In this embodiment, the canopy energy balance model can preferably be implemented using the publicly available SCOPE model. The SCOPE model is a soil-canopy integrated model that couples canopy radiative transfer, photosynthesis, energy balance and thermal infrared radiation processes. It can simulate canopy temperature and energy flux under given meteorological driving, canopy structure parameters and soil moisture status. However, this invention is not limited to the SCOPE model. Other equivalent models that can realize canopy energy balance solution and canopy temperature simulation can also be used as alternatives.

[0073] Specifically, based on the unified data base, input data related to energy balance solution is called according to plot identification and time axis. The input data includes at least meteorological driving quantities, soil moisture state quantities and crop growth period information, and canopy structure parameters used for energy balance calculation are determined. The canopy structure parameters are kept consistent during the calculation of upper and lower limits to reduce the impact of canopy structure differences on temperature boundary estimation.

[0074] Subsequently, while keeping the meteorological driving force and canopy structure conditions unchanged, at least two types of water constraint scenarios were set up: the maximum evapotranspiration scenario and the minimum evapotranspiration scenario. Among them, the maximum evapotranspiration scenario was used to characterize the condition of sufficient water supply. By setting the water constraint factor to a state of high evapotranspiration and low resistance, the latent heat flux was made relatively high, thereby obtaining the lower limit T of the canopy temperature. c The minimum transpiration scenario is used to characterize water-constrained conditions. By setting the water constraint factor to a low transpiration and high drag state, the latent heat flux is relatively low, thereby obtaining the upper limit of canopy temperature T. c The water constraint factor can be determined jointly by soil moisture state parameters and crop growth period information, and can be set using threshold, interval, or continuous function methods.

[0075] Substituting the two types of water constraint scenarios into the canopy energy balance model, the canopy temperature is solved hourly or daily on the unified time axis in step S3 to obtain the canopy temperature sequence for the corresponding time step, which is then used as T. c ,wet and T c The dynamic sequence of dry; and the consistency constraints and rationality checks of the upper and lower temperature limits are performed to ensure T c dry≥T c The final output corresponds one-to-one with the plot identifier and time axis to the dynamic upper and lower limits of canopy temperature, which is used for subsequent calculation and diagnosis of the water stress index.

[0076] Among them, the moisture constraint parameters and resistance-related parameters used for temperature upper and lower limit calculations, including but not limited to aerodynamic resistance-related parameters, blade boundary layer resistance-related parameters, soil evaporation resistance-related parameters, and moisture constraint mapping parameters, can be used as correctable parameters and updated and optimized based on newly added measured data during subsequent incremental learning.

[0077] S5: Obtaining crop canopy observation temperature T based on thermal infrared remote sensing images c The temperature-normalized crop water stress index (CWSI) is calculated by combining the dynamic upper and lower limits of canopy temperature obtained in step S4. Specifically, thermal infrared remote sensing image data and its metadata corresponding to the target plot are retrieved from the unified data base in step S3. Quality control and effective pixel screening are performed on the thermal infrared remote sensing images, including invalid pixel removal, masking of occlusion / abnormal areas, radiometric calibration, and temperature inversion consistency processing. The thermal infrared image radiance is converted into brightness temperature or land surface temperature, and the observed canopy temperature T is extracted based on the spatial range of the plot. c ,obs, where T c ,obs can be a statistical measure or pixel-level temperature value of the effective pixels of the canopy within the plot, and can be used to remove or correct mixed pixels of the soil background based on vegetation masking or threshold rules.

[0078] Then, T c ,obs and T output in step S4 c ,wet and T c The dry index is matched according to the plot identifier and time identifier. If necessary, interpolation is performed on the time scale or aggregation is performed according to the time window so that the three are on the same spatiotemporal scale. The crop water stress index (CWSI) is calculated based on the above and below temperature limits, see formula (7) below, and the results are subject to boundary constraints to ensure that 0 ≤ T ≤ 1; for abnormal situations (including but not limited to T c ,obs exceeds the upper and lower limits or the upper and lower limits do not meet T c dry≥T cThe data (including wet) is marked, corrected, or removed. The final output is the CWSI result corresponding to the plot identifier, time dimension, and spatial location, which is used to identify the degree of crop water stress and water shortage status, and to provide a basis for subsequent visualization and management decisions.

[0079] S6: Extract the spectral features of the crop canopy based on the hyperspectral remote sensing image, and input the initial prediction model of the spectral-canopy parameters in step S2 to obtain the estimated values ​​of crop CNC and DM. Then, combine the critical nitrogen dilution curve (see formula (8) below) to calculate the crop nitrogen nutrition index NNI, which is used to identify the degree of crop nitrogen stress and nitrogen deficiency status.

[0080] Specifically, the hyperspectral remote sensing image data and its metadata corresponding to the target plot are retrieved from the unified data base in step S3. Quality control and effective pixel screening are performed on the hyperspectral image, and the canopy reflectance spectrum at the plot scale is extracted to construct spectral features for model input. These spectral features can be full-spectrum reflectance features, selected sensitive band features, or spectral index features calculated from the reflectance spectrum. Band response consistency and numerical standardization are then performed on the spectral features to ensure consistency with the model input in step S2. The processed spectral features are then input into the initial prediction model in step S2 to obtain CNC and DM estimates corresponding to the plot identifier, time dimension, and spatial location. DM can be directly output by the model or obtained from intermediate variables output by the model combined with conversion relationships.

[0081] Subsequently, the critical nitrogen dilution curve was determined based on crop type and growth stage information, and the critical nitrogen concentration N was calculated based on the DM estimate. c NNI = CNC / N c The nitrogen nutrient index (NNI) was obtained, and the CNC, DM, and N were compared. c Reasonable range constraints and anomaly handling are implemented with NNI to improve computational stability.

[0082] The critical nitrogen dilution curve can be selected from publicly available literature or existing databases as a priori curve in the early stage of model deployment. During system operation, the curve parameters are specifically corrected and iteratively updated based on continuously accumulated measured samples to improve the applicability and stability under different crops, different regions, and cross-year conditions.

[0083] S7: During agricultural production and experimentation, new measured data are continuously acquired and introduced into the incremental learning mechanism as new samples. The spectral-canopy parameter prediction model in step S2 and the key parameters related to diagnosis in steps S4 to S6 are continuously corrected and optimized to achieve adaptive specialization of the model for different crops, different regions and different growth stages.

[0084] Specifically, the system receives new measured data in an irregular and variable manner, and performs cleaning, quality control and spatiotemporal alignment processing on the new measured data in the same way as in step S3 to form a new training sample set. When updating the model, iterative updates are preferred to the parameters of the correction layer or output layer of the initial prediction model. If necessary, small updates are made to some internal parameters of the model under stability constraints.

[0085] To mitigate the risk of catastrophic forgetting caused by continuous updates, the system maintains a historical replay sample library and uses newly added training samples and representative historical samples extracted from the replay sample library for training during each iteration update. At the same time, a frozen validation set is maintained to evaluate the performance and verify the consistency of the model before and after the update. When the updated model meets the preset performance constraints, it is released as a new version model for subsequent diagnosis; otherwise, the original version is maintained and the update results are recorded.

[0086] After the model version is released, the system maintains and updates the historical playback sample library, and incorporates representative samples from the newly added measured data into the playback sample library according to dimensions such as crop type, regional conditions and growth stage. When the library capacity is limited, the system implements sample replacement or stratified retention strategies to ensure long-term coverage of key scenarios.

[0087] Through the aforementioned multi-round incremental iteration and version update mechanism, the model can be gradually adapted to specific crop types, specific regional climate and soil conditions, and differences in different growth stages, thereby continuously improving the accuracy, stability, and generalizability of water-nitrogen stress diagnosis.

[0088] To facilitate understanding of the implementation process of this invention, some calculation formulas involved in this embodiment are given below; it should be understood that the formulas are illustrative and can be adjusted or equivalently replaced according to crop type, sensor characteristics and application scenario without departing from the technical concept of this invention.

[0089] Leaf dry mass (LMA) is estimated from leaf components:

[0090] (1)

[0091] Where LMA is the leaf dry matter mass; C prot Leaf protein content; C c This refers to the content of carbon-based components.

[0092] Relationship for estimating canopy nitrogen concentration (CNC):

[0093] (2)

[0094] Wherein, CNC is the canopy nitrogen concentration; LCC is the leaf chlorophyll content; LAI is the leaf area index; and 𝑘 PN ,NPN These are the model coefficients.

[0095] DM conversion relationships:

[0096] (3)

[0097] Wherein, DM represents the aboveground dry matter mass of the crop; f leaf ∈(0,1] is the proportion coefficient of leaf dry matter in the aboveground dry matter, which can be set according to crop type and growth period or obtained by calibration.

[0098] The weights of each band in the simulated data are calculated using formulas 4 and 5. Here, σ is the Gaussian kernel standard deviation, λ0 is each measured spectral band, λ is each simulated spectral band, and g(λ) is the weighting function for each spectral band. The simulated data is then weighted and summed using normalized weights to ultimately align the simulated and measured spectra.

[0099] (4)

[0100] (5)

[0101] The energy balance relationship upon which the dynamic upper and lower limits of canopy temperature are calculated can be expressed as:

[0102] (6)

[0103] Among them, R n λ represents net radiation, H represents sensible heat flux, λE represents latent heat flux, and G represents soil heat flux.

[0104] The temperature-normalized water stress index (CWSI) formula is used, assuming the canopy observation temperature obtained from thermal infrared remote sensing is T. c The lower limit of the maximum evaporation scenario obtained from the energy balance model is T. c The upper temperature limit T obtained from the minimum evaporation scenario is... c ,dry.

[0105] (7)

[0106] Critical nitrogen dilution curve:

[0107] (8)

[0108] Where, N c denoted as critical nitrogen concentration; DM as aboveground dry matter; a and b are curve parameters, which can be given prior from the literature and corrected and updated in subsequent incremental learning.

[0109] Nitrogen Nutritional Index (NNI):

[0110] (9)

[0111] NNI is used to characterize the degree of nitrogen stress.

[0112] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for diagnosing and adaptively correcting crop water-nitrogen stress based on a physical model-driven approach and multi-source data fusion, characterized in that, Includes the following steps: S1: Simulate and generate a simulated spectral-canopy parameter sample set covering different crop types, different growth stages and different canopy conditions. The simulated spectral-canopy parameter sample set includes simulated canopy reflectance spectra and corresponding canopy parameter samples. S2: Using the simulated spectrum-canopy parameter sample set, an initial prediction model for the spectrum-canopy parameters is trained. The simulated spectrum features are used as the model input, and the canopy parameters that correspond one-to-one with the simulated spectrum are used as the model output to predict the crop canopy nitrogen concentration (CNC) and the crop aboveground dry matter (DM). S3: Acquire multi-source observation data related to the target crop plot and construct a unified data base. The multi-source observation data includes at least remote sensing image data, meteorological data, soil data, and crop growth period information. The multi-source observation data needs to be processed, including but not limited to additional metadata, cleaning, spatiotemporal alignment, standardization, and indexing. The processed multi-source data is organized and stored according to plot identification, time dimension, and crop type to form a unified data base. S4: Based on the physical model of canopy energy balance mechanism, and with the support of a unified data base, calculate the dynamic upper and lower limits of canopy temperature of the target crop plot on the corresponding time axis. S5: Obtain crop canopy observation temperature T based on thermal infrared remote sensing image c , obs, and combine the upper and lower limits of canopy temperature dynamic to calculate the temperature normalized crop water stress index CWSI; S6: Extract crop canopy spectral features based on hyperspectral remote sensing images, input them into the initial prediction model of spectral-canopy parameters, obtain the estimated values ​​of crop CNC and DM, and calculate the crop nitrogen nutrient index NNI by combining the critical nitrogen dilution curve, which is used to identify the degree of crop nitrogen stress and nitrogen deficiency status. S7: Continuously acquire new measured data and introduce them as new samples into the incremental learning mechanism to continuously correct and optimize the spectral-canopy parameter prediction model of S2 and the key parameters related to diagnosis in S4 to S6, so as to achieve adaptive specialization of the model for different crops, different regions and different growth stages.

2. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion as described in claim 1, characterized in that, S1 uses a physical model of the leaf-canopy radiative transfer mechanism for simulation, generating canopy reflectance spectra covering different crop types, different growth stages, and different canopy conditions within a preset parameter space.

3. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 2, characterized in that, The simulated spectrum is processed to ensure that its band settings are consistent with the data acquired by the UAV hyperspectral remote sensing equipment. This includes: firstly, cropping the simulated spectrum according to the band range of the target equipment; then, based on the center wavelength and bandwidth information of each band of the target equipment, resampling and band-aligning the cropped simulated spectrum is performed using a band response consistency method.

4. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 2, characterized in that, Noise perturbation is superimposed on the simulated spectrum after resampling.

5. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 1, characterized in that, The initial prediction model employs a machine learning model or a deep learning model.

6. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 1, characterized in that, S4 calculates the dynamic upper and lower limits of canopy temperature for the target crop plot on the corresponding time axis: Based on the unified data base, input data related to energy balance solution is called according to plot identification and time axis. The input data includes at least meteorological driving quantities, soil moisture state quantities and crop growth period information, and the canopy structure parameters used for energy balance calculation are determined. Under the premise of keeping the meteorological driving force and canopy structure conditions unchanged, at least two types of water constraint scenarios are set, namely the maximum evapotranspiration scenario and the minimum evapotranspiration scenario. Substituting the two types of water constraint scenarios into the canopy energy balance model, the energy balance equations are solved hourly or daily on the unified time axis of S3 to obtain the canopy temperature series for the corresponding time steps, which are then used as T. c ,wet and T c The dynamic sequence of dry; and the consistency constraints and rationality checks of the upper and lower temperature limits are performed to ensure T c dry≥T c The final output is the dynamic upper and lower limits of canopy temperature, which correspond one-to-one with the plot identifier and time axis.

7. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 1, characterized in that, S5 calculates the Temperature Normalized Crop Water Stress Index (CWSI): It retrieves thermal infrared remote sensing image data and metadata corresponding to the target plot from a unified data base, performs quality control and effective pixel screening on the thermal infrared remote sensing images, converts thermal infrared image radiance into brightness temperature or land surface temperature, and extracts canopy temperature observations (T) based on the spatial extent of the plot. c ,obs; T c ,obs and T c ,wet and T c The dry function matches plot identifiers and time identifiers, interpolates on the time scale or aggregates them by time window to make the three elements fit on the same spatiotemporal scale. The Crop Water Stress Index (CWSI) is calculated based on the upper and lower temperature limits, and the results are subject to boundary constraints to ensure that 0 ≤ 0 ≤ 1. Abnormal cases are marked, corrected, or removed. Finally, the CWSI results corresponding to the plot identifier, time dimension, and spatial location are output.

8. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 1, characterized in that, The method for calculating the crop nitrogen nutrient index (NNI) in S6 is as follows: Hyperspectral remote sensing image data and its metadata corresponding to the target plot are retrieved from the unified data base in step S3. Quality control and effective pixel screening are performed on the hyperspectral image, and the canopy reflectance spectrum at the plot scale is extracted to construct spectral features for model input. The spectral features are then input into the initial prediction model in S2 to obtain CNC and DM estimates corresponding to the plot identifier, time dimension, and spatial location. Calculation of critical nitrogen concentration N based on DM estimation value c NNI = CNC / N c The nitrogen nutrient index (NNI) was obtained.

9. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 1, characterized in that, The S7 multi-round incremental iteration and version update mechanism combines newly added training samples with representative historical samples extracted from the replay sample library for training; at the same time, it maintains a frozen validation set to evaluate the performance and verify the consistency of the model before and after the update. When the updated model meets the preset performance constraints, it is released as a new version model for subsequent diagnosis; otherwise, the original version is maintained and the update results are recorded.

10. The crop water-nitrogen stress diagnosis and adaptive correction method based on physical model-driven and multi-source data fusion according to claim 1, characterized in that, The historical playback sample library is maintained and updated. Representative samples from newly added measured data are included in the playback sample library according to dimensions such as crop type, regional conditions and growth stage. When the library capacity is limited, sample replacement or stratified retention strategies are implemented to ensure long-term coverage of key scenarios.

Images