Winter wheat quality monitoring method and device for coupling integrated learning rib shape correction and data assimilation

Abstract

The invention discloses a winter wheat quality monitoring method and device for coupling integrated learning rib shape correction and data assimilation, and relates to the field of crop quality monitoring. The method comprises the steps of preprocessing acquired information data, determining a winter wheat planting area in a target monitoring area based on a classification model, and dividing the winter wheat planting area into different rib type variety areas; performing inversion and extraction of quality-related crop parameters by adopting a machine learning regression model to obtain a crop parameter time sequence; determining a simulation state quantity time sequence of the rib type variety area based on a crop growth model; and performing multivariable collaborative data assimilation processing according to the crop parameter time sequence and the simulation state quantity time sequence, driving the crop growth model to the winter wheat mature harvest time according to the assimilated data, and performing GPC correction prediction on the winter wheat in the tendon type variety area based on the correction factor to represent the quality of the winter wheat grains. The invention aims to improve the quality monitoring precision of regional winter wheat.

Description

Technical Field

[0001] This application relates to the field of crop quality monitoring, and particularly to a winter wheat quality monitoring method and device that couples integrated learning rib type correction and data assimilation. Background Art

[0002] As a major global food crop, the yield and quality of wheat are of great significance for national food security and diverse human consumption needs. Among them, wheat grain quality, especially the grain protein content (GPC), is not only a core indicator for evaluating the nutritional value and processing characteristics of wheat, but also a key parameter for distinguishing weak gluten (GPC ≤ 12%), medium gluten (12% < GPC ≤ 14%), secondary strong gluten (14% < GPC ≤ 15%), and first-class strong gluten (GPC < 15%) wheat. With the rapid development of social economy and the continuous improvement of residents' consumption levels, the market demand for high-quality special wheat is increasing day by day. Therefore, accurate and rapid wheat quality monitoring has become an urgent need for the development of modern agriculture.

[0003] The early determination of wheat GPC mainly relied on field destructive sampling and laboratory chemical analysis (such as the Kjeldahl method). Although this method has high accuracy, it has inherent defects such as low efficiency, high cost, and insufficient spatial representativeness, and it is difficult to meet the requirements of large-scale and rapid dynamic monitoring. Remote sensing technology, with its advantages of macroscopy, rapidity, and non-destructive detection, provides a new technical means for crop growth monitoring at the regional scale. By constructing empirical or semi-empirical semi-mechanistic models between remote sensing information at key growth stages and GPC, or by combining machine learning to construct data-driven models, researchers can invert the GPC of wheat at the regional scale. Although such methods have simple processes and are easy to implement, they generally have problems of insufficient mechanism and poor spatio-temporal universality. In recent years, data assimilation technology has provided a new idea for solving the above problems. Crop growth models have strong mechanism and time dynamic simulation capabilities, and can simulate the whole process of crop growth from sowing to maturity at the plot or farm scale. By integrating the spatial information observed by remote sensing and the temporal dynamic information of crop models, data assimilation technology can effectively optimize the state variables or key parameters of the models, and significantly improve the accuracy and reliability of crop models in simulating crop growth and yield at the regional scale.

[0004] However, research on applying data assimilation methods to wheat quality (especially GPC) monitoring at the regional scale is still in the exploratory stage. On the one hand, grain protein formation involves the comprehensive results of multiple complex physiological and ecological processes, such as nitrogen absorption, nitrogen storage in multiple organs, and the redistribution and translocation of nitrogen to the grain during growth and reproduction. Compared to simulating overall yield, accurately simulating these mechanistic processes in crop models and selecting key state variables or parameters that play a decisive role in grain protein formation and can be effectively assimilated is significantly more difficult. On the other hand, current mainstream remote sensing observation variables used for assimilation (such as LAI) reflect more the overall canopy structure and biomass accumulation of crops, and their correlation with the intrinsic physiological processes of final GPC formation is relatively indirect. Relying solely on such single remote sensing observation variables often makes it difficult to directly and sensitively constrain model parameters related to quality formation, ultimately resulting in limited accuracy in regional quality monitoring. Therefore, there is an urgent need to develop a regional winter wheat quality monitoring method and system that can combine multi-source remote sensing observation information, accurately simulate crop nitrogen cycling and distribution processes, apply advanced assimilation algorithms, and integrate gluten type and climate information for correction, in order to fill the limitations of existing technologies and better meet the urgent needs of precision agriculture for high-quality wheat production. Summary of the Invention

[0005] The purpose of this application is to provide a method and apparatus for monitoring the quality of winter wheat by coupling integrated learning of gluten type correction and data assimilation, which can improve the accuracy of regional winter wheat quality monitoring.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] Firstly, this application provides a method for monitoring the quality of winter wheat by coupling integrated learning of gluten type correction and data assimilation, including:

[0008] Acquire information data; the information data includes remote sensing satellite data covering the target monitoring area and the key growth and development stages of winter wheat, remote sensing reanalysis product data, and necessary ground measurement data;

[0009] The information data is preprocessed, and the winter wheat planting area within the target monitoring area is determined based on the classification model, and the winter wheat planting area is divided into different gluten type variety areas; the classification model is obtained by supervising classification training using an ensemble learning method based on the information data corresponding to the sample points of the known winter wheat planting area.

[0010] For different rib-type variety areas, machine learning regression models were used to invert and extract quality-related crop parameters to obtain time series of crop parameters; the crop parameters include: leaf area index, aboveground biomass, aboveground nitrogen accumulation, and soil moisture.

[0011] The time series of simulated state quantities for the gluten-type variety region were determined based on a crop growth model; the crop growth model is a mechanistic model used to simulate the growth and development, carbon-nitrogen coupling cycle, and grain quality formation process of winter wheat.

[0012] Based on the crop parameter time series and the simulated state quantity time series, multivariate collaborative data assimilation processing is performed to obtain assimilated data;

[0013] The crop growth model is driven by the assimilated data until the winter wheat matures and is harvested. Based on the correction factor, the GPC correction prediction is performed on the winter wheat in the gluten variety area to obtain the GPC prediction value. The GPC prediction value is used to characterize the quality of winter wheat grains.

[0014] Secondly, this application provides a winter wheat quality monitoring device that couples integrated learning gluten type correction and data assimilation, comprising:

[0015] The information data acquisition module is used to acquire information data, which includes remote sensing satellite data covering the target monitoring area and the key growth and development period of winter wheat, remote sensing reanalysis product data, and necessary ground measurement data.

[0016] The classification module is used to preprocess the information data, determine the winter wheat planting area within the target monitoring area based on the classification model, and divide the winter wheat planting area into different gluten type variety areas; the classification model is obtained by supervising classification training using an ensemble learning method based on the information data corresponding to the sample points of the known winter wheat planting area.

[0017] The inversion and extraction module is used to invert and extract quality-related crop parameters for different rib-type variety areas using a machine learning regression model, thereby obtaining a time series of crop parameters. The crop parameters include: leaf area index, aboveground biomass, aboveground nitrogen accumulation, and soil moisture.

[0018] The simulated state quantity time series determination module is used to determine the simulated state quantity time series of the gluten-type variety area based on the crop growth model; the crop growth model is a mechanism model used to simulate the growth and development, carbon-nitrogen coupling cycle and grain quality formation process of winter wheat.

[0019] The assimilation processing module is used to perform multivariate collaborative data assimilation processing based on the crop parameter time series and the simulated state quantity time series to obtain assimilated data;

[0020] The correction prediction module is used to drive the crop growth model to the winter wheat maturity and harvest period based on the assimilated data, and to perform GPC correction prediction on winter wheat in the gluten variety area based on the correction factor to obtain the GPC prediction value; the GPC prediction value is used to characterize the quality of winter wheat grains.

[0021] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0022] This application provides a method and apparatus for monitoring the quality of winter wheat by coupling integrated learning gluten type correction and data assimilation. The method involves preprocessing the acquired information data and determining the winter wheat planting area within the target monitoring region based on a classification model, dividing the winter wheat planting area into different gluten type variety zones. A machine learning regression model is used to invert and extract quality-related crop parameters, obtaining a time series of crop parameters. A simulated state quantity time series for the gluten type variety zones is determined based on a crop growth model. Multivariate collaborative data assimilation is performed based on the crop parameter time series and the simulated state quantity time series. The assimilated data drives the crop growth model to the winter wheat maturity and harvest period. Based on a correction factor, GPC (Gross Cycle Performance) correction prediction is performed on winter wheat in the gluten type variety zones to characterize the quality of winter wheat grains. This application, by using a machine learning regression model to invert and extract crop parameters and performing multivariate collaborative data assimilation based on the crop parameter time series and the simulated state quantity time series, can more accurately constrain the key physiological parameters controlling nitrogen cycling and allocation in the crop model, thereby significantly improving the monitoring accuracy of winter wheat GPC at the regional scale. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a winter wheat quality monitoring method that couples integrated learning gluten type correction and data assimilation.

[0025] Figure 2 A flowchart illustrating a regional winter wheat quality monitoring method that couples integrated learning gluten type correction and data assimilation in practical applications;

[0026] Figure 3 A structural block diagram of a regional winter wheat quality monitoring system that couples and integrates learning gluten type correction and data assimilation in practical applications;

[0027] Figure 4A structural diagram of a winter wheat quality monitoring device that couples integrated learning gluten type correction and data assimilation. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] In one exemplary embodiment, such as Figure 1 As shown, a method for monitoring the quality of winter wheat that couples gluten type correction and data assimilation is provided, including:

[0031] Step 100: Obtain information data.

[0032] Step 200: Preprocess the information data and determine the winter wheat planting areas within the target monitoring area based on the classification model, and divide the winter wheat planting areas into different gluten type variety areas. The classification model is obtained by supervising the training of the model using ensemble learning methods, based on the information data corresponding to the sample points of the known winter wheat planting areas.

[0033] Specifically, information data is masked, and information data from different sources and at different resolutions is cropped, reprojected, and time-series synthesized to ensure data consistency in space and time.

[0034] Step 300: For different rib-type variety areas, a machine learning regression model is used to invert and extract quality-related crop parameters to obtain the time series of crop parameters.

[0035] The machine learning regression model adopts the random forest regression model. The machine learning regression model is based on machine learning methods and is trained on a feature vector set of known crop parameters. The feature vector set includes: normalized vegetation index, enhanced vegetation index, soil-regulated vegetation index, normalized differential water index, green normalized differential vegetation data, green leaf green index, optimized soil-regulated vegetation index, green red-edge vegetation data, radar backscattering coefficient, radar vegetation index, microwave backscattering cross-polarization ratio, and polarization decomposition parameters.

[0036] Step 400: Determine the time series of simulated state variables for the gluten-type variety region based on the crop growth model. The crop growth model is a mechanistic model used to simulate the growth and development, carbon-nitrogen coupling cycle, and grain quality formation process of winter wheat.

[0037] The crop growth model used was the APSIM crop model. The crop growth model was obtained by performing a global sensitivity analysis of the model parameters using the Extended Fourier Transform (EFT) amplitude sensitivity test to determine key parameters. These key parameters were then calibrated and optimized using the Markov chain Monte Carlo method, combined with ground-based measured data from sample points in the study area. The ground-based measured data included meteorological data, soil data, field management data, and crop parameters. The field management data included sowing time, planting density, fertilizer application rate, and irrigation amount.

[0038] The key parameters were determined by calculating the first-order sensitivity index and the global sensitivity index.

[0039] The expression for the first-order sensitivity index is:

[0040]

[0041] The expression for the global sensitivity index is:

[0042]

[0043] in, It is a first-order sensitivity index; Let be the variance of the i-th input factor; x represents the expected value of all input factors except the i-th input factor; i Let be the i-th input factor; y be the objective function value; V(y) be the variance of the objective function value; x i-1 All input factors except the i-th input factor; Let be the expected value of the i-th input factor; Let be the variance of all input factors except the i-th input factor; This is a global sensitivity index.

[0044] Step 500: Based on the time series of crop parameters and the time series of simulated state variables, perform multivariate collaborative data assimilation processing to obtain assimilated data.

[0045] In one embodiment, multivariate collaborative data assimilation processing is performed based on the time series of crop parameters and the time series of simulated state variables to obtain assimilated data, specifically including:

[0046] Gaussian perturbations were added to the time series of crop parameters and the time series of simulated state variables, and the ensemble Kalman filter algorithm was used for multivariate collaborative data assimilation to obtain assimilated data.

[0047] The expression corresponding to the Gaussian perturbation is:

[0048] S = s + aε.

[0049] Where S is a multivariate vector after adding Gaussian perturbation; s is the input variable data; a is a constant; and ε is a random number vector.

[0050] As an optional implementation, an ensemble Kalman filter algorithm is used for multivariate collaborative data assimilation processing, and the corresponding mathematical expression is:

[0051] B t =HA t +v t .

[0052]

[0053] Among them, B t Let H be a set of observation data at time t; H be the observation operator; A ... t Let v be the set of state variables at time t; t For measuring noise; M represents the predicted set of state variables; M represents the state transformation equation. The set of optimal estimates of state variables at time t-1; w t This is process error; K represents the set of optimal estimates of the state variables at time t. t For Kalman gain.

[0054] Step 600: Drive the crop growth model to the winter wheat maturity and harvest period based on the assimilated data, and perform GPC correction prediction on winter wheat in the gluten-type variety area based on the correction factor to obtain the GPC predicted value. The GPC predicted value is used to characterize the quality of winter wheat grains.

[0055] The correction factors include: rib type correction factor and climate factors; the climate factors include: temperature influence index and moisture influence index; the expression for the GPC prediction value is:

[0056] GPC=γ·[AGB*AGN+f·(LAI)]+FT(t)+FW(SM).

[0057] Wherein, GPC is the predicted GPC value; γ is the rib type correction factor; FT(t) is the temperature influence index at time t; FW is the water influence index; AGB is the aboveground biomass; AGN is the aboveground nitrogen accumulation; LAI is the leaf area index; f is the empirical parameter obtained from least squares regression analysis; SM is soil moisture; and FW(SM) is the water influence index corresponding to soil moisture.

[0058] As an optional implementation method, the winter wheat quality monitoring method based on crop models and data assimilation also includes: outputting and displaying GPC prediction values ​​in the form of maps, reports or data services.

[0059] This application integrates multi-source remote sensing data to generate remote sensing feature data related to winter wheat growth, and uses ensemble learning methods to identify different winter wheat cultivar categories with different gluten types. It extracts time-series LAI, AGB, AGN, and soil moisture remote sensing observation data using multi-parameter remote sensing inversion, and co-assimilates the multivariate remote sensing observation set into a crop model, dynamically constraining and optimizing the model's state variables and key parameters. The optimized crop model is run to generate a set of key parameters, and cultivar gluten type, climate, and phenological corrections are introduced for regional winter wheat GPC monitoring. By co-assimilating multiple remote sensing observation variables and introducing multi-factor correction factors, the accuracy of regional-scale winter wheat GPC monitoring can be effectively improved.

[0060] In practical applications, the method mentioned in this application can also be implemented using the following steps.

[0061] S1. Acquire multi-source data covering the target monitoring area and the key growth and development period of winter wheat. The multi-source data shall include at least remote sensing satellite data, remote sensing reanalysis product data and ground measurement data, and preprocess the multi-source data.

[0062] Remote sensing satellite data includes MODIS remote sensing imagery, Landsat remote sensing imagery, and Sentinel-1 dual-polarization SAR data; remote sensing reanalysis product data includes MODIS LAI products, AgERA5 reanalysis meteorological datasets, and CSDLv2 soil property datasets; necessary ground-based measured data includes winter wheat and non-winter wheat sample data, LAI, AGB, AGN, and SM data of winter wheat samples during key growth stages, and agricultural management data of the plots where the winter wheat samples are located. Cloud / snow / masking processing was performed on the acquired remote sensing satellite data; data from different sources and at different resolutions were cropped, reprojected, and time-series synthesized to ensure spatial and temporal consistency.

[0063] The AgERA5 reanalysis meteorological dataset includes daily wind speed at 10m near the ground, daily maximum temperature at 2m near the ground, daily minimum temperature at 2m near the ground, daily precipitation flux, daily solar radiation flux, and daily vapor pressure data for the study area; the CSDLv2 soil property dataset in step S1 includes soil texture, bulk density, organic matter, and pH data for the study area; the key growth stages of winter wheat in step S1 are the greening stage, jointing stage, and heading stage; the agricultural management data of the winter wheat plots in step S1 includes sowing time, planting density, fertilizer application, and irrigation amount.

[0064] S2. Based on preprocessed hyperspectral and SAR remote sensing satellite data, the winter wheat planting area within the target monitoring area is extracted by fusing spectral and structural features, and the winter wheat planting area is divided into different gluten type variety areas.

[0065] Specifically, the classification of winter wheat gluten varieties involved training a machine learning classification model for supervised classification using winter wheat and non-winter wheat sample points within the study area, along with their corresponding remote sensing image reflectance and key features. The machine learning model was a random forest classification model, generating a spatial distribution map of winter wheat planting at a certain resolution (e.g., 30m) for the study area. Then, spectral and structural features were integrated to extract winter wheat planting areas within the target monitoring area, which were then divided into different gluten variety zones. The greening, jointing, and heading stages were selected as key growth stages for winter wheat. Structural sensitivity parameters were constructed using Sentinel-1 data, and optical vegetation indices were constructed using Landsat-8 and Landsat-9 data. A machine learning-based classification model for gluten varieties at key growth stages was built. Combining data from the three growth stages, three machine learning models were used as base models, and a logistic regression model was used as the meta-model to construct an ensemble learning model to obtain classification information for different gluten varieties in the winter wheat planting area.

[0066] Among them, the spectral and structural features used for classifying winter wheat gluten varieties in step S2 include at least one or more of the following: Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Normalized Difference Water Index (NDWI), Chlorophyll Index (GCVI), Normalized Difference Yellowness Index (NDYI), Radar Vegetation Index (RVI), Microwave Backscattering Cross-Polarization Ratio (Rc), and Polarization Decomposition Parameters (Scattering Entropy, Anti-Entropy, and Average Scattering Angle Alpha); The calculation formulas for these features are shown in Formulas (1) to (7), respectively.

[0067]

[0068]

[0069] Where, ρ NIR ρ RED ρ BLUE ρ SWIR1 ρ GREEN These represent the reflectance of optical remote sensing images in the near-infrared, red, blue, short-wave infrared, and green bands, respectively. and These are the backscattering coefficients of microwave remote sensing data with VH and VV polarizations, respectively. RVI VV This represents the radar vegetation index after VV polarization.

[0070] S3. Based on preprocessed remote sensing satellite data and remote sensing reanalysis product data, invert and extract time series (crop parameter time series) of at least multiple remote sensing observation variables, including leaf area index (LAI), aboveground biomass (AGB), aboveground nitrogen accumulation (AGN), and soil moisture (SM), for the target operational unit.

[0071] Based on the preprocessed time-series remote sensing images from step S1, feature parameters sensitive to LAI, AGB, AGN, and SM were calculated from images of key growth stages of winter wheat, and feature vector sets were constructed. Combined with machine learning methods, inversion models for LAI, AGB, AGN, and SM were trained respectively to obtain predicted data for LAI, AGB, AGN, and SM during key growth stages of winter wheat. Feature parameters sensitive to LAI, AGB, AGN, and SM include one or more of the following: NDVI, EVI, Soil-Adjusted Vegetation Index (SAVI), NDWI, Green Normalized Differential Vegetation Data (GNDVI), Green Leaf Greenness Index (GCVI), Optimized Soil-Adjusted Vegetation Index (OSAVI), Green-Red Edge Vegetation Data (GRVI), Radar Backscattering Coefficient, Radar Vegetation Index (RVI), Microwave Backscattering Cross-Polarization Ratio, and polarization decomposition parameters (Scattering Entropy, Anisotropy, and Mean Scattering Angle Alpha). The machine learning regression model was a random forest regression model.

[0072] The calculation formulas for the vegetation indices SAVI and GNDVI that are sensitive to AGB and the vegetation indices GCVI, OSAVI and GRVI that are sensitive to AGN in step S3 are shown in formulas (8) to (12).

[0073]

[0074]

[0075] S4. Collect soil, meteorological and crop parameters in the study area, perform scale variation and location matching processing on the data, and use them to calibrate a mechanistic crop growth model that can simulate winter wheat growth and grain protein content (GPC) formation. Then, use the model to simulate the time series of state variables of the target work unit.

[0076] Crop model construction and parameter optimization: The APSIM crop model, a mechanistic crop growth model capable of accurately simulating the growth and development, carbon-nitrogen coupling cycle, and grain quality formation process of winter wheat, was selected. Global sensitivity analysis was performed on the crop model to identify key model parameters that significantly affect the simulated LAI, AGB, AGN, and SM outputs and are closely related to the GPC formation process. Combined with meteorological, soil, field management measures, and crop parameters in the study area, the crop model was calibrated and preliminarily optimized. For each target grid, the optimized crop model was run to generate an unassimilated time series set of simulated state variables.

[0077] In step S4, the crop model is the APSIM crop model. The extended Fourier amplitude sensitivity test (EFAST) is used to perform global sensitivity analysis on the crop parameters of the APSIM model. The sensitivity index and global sensitivity index are solved according to formula (13) to formula (14).

[0078]

[0079] In step S4, the calibration method for sensitive parameters is the Markov Chain Monte Carlo (MCMC) method, while the default values ​​of the model are used directly for insensitive parameters.

[0080] S5. Using a sequence data assimilation algorithm, the time series of multiple remote sensing observation variables obtained in step S3 are co-assimilated with the time series of model simulation state variables obtained in step S4 in order to optimize the model's state variables and key model parameters that are related to GPC.

[0081] Multivariate Collaborative Data Assimilation: Construct an assimilation framework based on sequence data assimilation algorithms; add Gaussian perturbations to the time series of various remote sensing observation variables (e.g., LAI, AGB, AGN, SM) output in step S3 to generate multivariate extended state vectors and observation member sets; add Gaussian perturbations to the corresponding state variables output by the crop model simulation in step S4 to generate model simulation member sets; input the observation member sets and model simulation members into the data assimilation algorithm to constrain the internal state of the model, dynamically adjust and optimize the model's state variables and key parameters closely related to GPC formation.

[0082] In the Ensemble Kalman Filter (EnKF) algorithm, the state vector is a multivariable extended state vector X:

[0083] X = [LAI1…LAI] n AGB1…AGB n ,AGN…AGN n ,SM1…SM n ].

[0084] In step S5, a Gaussian perturbation is added, and the calculation is as shown in formula (13):

[0085] S=s+aε (13)

[0086] Where S is the multivariate vector after adding Gaussian perturbation; s is the input variable data, i.e., the initial observations or model state variables; a is a constant; ε is a random number vector, i.e., ε represents a random number vector with the same dimension as S, whose members conform to a normal distribution, i.e., ε i ~N(0,1).

[0087] In step S5, the ensemble Kalman filter algorithm is used to assimilate the remote sensing observation variables and APSIM simulation variables, and the calculations are performed using formulas (14) to (16):

[0088] B t =HA t +v t (14)

[0089]

[0090] The mean of is the optimal estimate of the state at that moment, H is the observation operator, M is the state transformation equation, i.e., the APSIM model, v t It measures noise, w t It is process error, K t It is the Kalman gain, representing the weight of the observed data.

[0091] S6. Using the model state variables and key parameters assimilated in step S5, drive the crop model to run, introduce multi-factor correction factors to predict and output the final GPC value of the target work unit.

[0092] Using the model parameter set and final state variables obtained after data assimilation in step S5, the crop model is driven to the winter wheat maturity and harvest period; for each target grid cell determined in step S2, the final predicted LAI, AGB, AGN, and SM values ​​are output; rib type correction factors and climate factors are introduced to construct corrected GPC prediction values:

[0093] GPC=γ·[AGB*AGN+f·(LAI)]+FT(t)+FW(SM).

[0094] The final prediction results are verified using ground-based measured data. At the same time, spatial distribution maps of indicators such as yield and biomass can be output simultaneously according to application needs. The prediction results are output in the form of maps, reports or data services to provide scientific decision support for regional wheat quality evaluation, precise harvesting and storage and processing.

[0095] This application is used to perform the steps of the regional winter wheat quality monitoring method based on crop model and multi-source remote sensing data assimilation and rib type correction.

[0096] This application introduces and co-assimilates multi-source remote sensing observation variables such as LAI, AGB, AGN, and SM, and focuses on optimizing model parameters directly related to nitrogen cycle and quality formation, achieving the following beneficial effects:

[0097] Effective improvement in regional GPC monitoring accuracy: By adding AGB and AGN, which are more sensitive to nitrogen accumulation in plants, as key remote sensing observation variables for multivariate collaborative data assimilation, compared with existing technologies that rely on single observation variables (such as LAI), it can more accurately constrain the key physiological parameters controlling nitrogen cycling and distribution in crop models, thereby significantly improving the monitoring accuracy of winter wheat GPC at the regional scale.

[0098] It enhances the mechanistic and universality of regional GPC monitoring: by dynamically optimizing key physiological parameters in crop models through data assimilation algorithms, the models can better adapt to nitrogen dynamics under different environmental conditions, and improve the stability of cross-regional and cross-year applications.

[0099] It achieves collaborative and non-destructive monitoring of multiple objective variables: it can quickly, non-destructively, and synchronously acquire multi-dimensional information such as GPC, LAI, AGB, and SM of winter wheat over a large area, providing more comprehensive information support for precision management.

[0100] Taking a certain city as an example, the city has a relatively flat terrain and the soil type is mainly alluvial soil, which is suitable for the growth of winter wheat. However, it also faces challenges such as large interannual variations in meteorological conditions and uneven spatial and temporal distribution of water resources, which affect the yield and quality of winter wheat. Therefore, carrying out precise quality monitoring in this area has important practical significance.

[0101] Data source introduction:

[0102] Remote sensing satellite data mainly includes NASA's MODIS ground reflectance products (spatial resolution 500m, temporal resolution 8 days), Sentinel-1 dual-polarization SAR data, and Landsat 8 / 9 surface reflectance products (spatial resolution 30m, temporal resolution 16 days).

[0103] Remote sensing reanalysis product data include MODIS LAI (spatial resolution 500m, temporal resolution 8 days, providing global leaf area index time series), AgERA5 reanalysis meteorological dataset released by the European Centre for Medium-Range Weather Forecasts (spatial resolution 0.1°, providing daily meteorological elements), and CSDLv2 China Soil Properties dataset released by the School of Atmospheric Sciences, Sun Yat-sen University (spatial resolution 1km, providing key soil physicochemical properties).

[0104] Ground-based measured data include GPS location information of winter wheat and non-winter wheat sample points obtained through field surveys within the city, LAI, AGB, and AGN data obtained from destructive sampling of winter wheat sample points during key growth stages (jointing, heading, and elongation stages), agricultural management information of the plots where the winter wheat sample points are located (such as sowing, fertilization, and irrigation records), and measured GPC values ​​of winter wheat obtained from laboratory analysis after final harvest.

[0105] Reference Figure 2 The specific process is as follows:

[0106] Step S1, Data Acquisition and Preprocessing. Acquire remote sensing satellite data and remote sensing reanalysis product data covering the entire winter wheat growing season from October 2023 to June 2024 for the city. The remote sensing satellite data requires preprocessing steps including cloud / snow / shade masking, cropping, missing value imputation, resampling, and time-series image synthesis to generate a 30m spatial resolution time-series optical remote sensing image set. The remote sensing reanalysis product data needs to be unified to the WGS84 / UTM Zone 50N coordinate system, spatially cropped to match the study area of ​​the city, and resampled to a uniform spatial resolution. Specifically, extract daily near-surface 10m wind speed, 2m maximum / minimum temperature, precipitation flux, solar radiation flux, and vapor pressure data for the study area from the AgERA5 dataset, and extract soil texture (percentage of sand, silt, and clay), bulk density, organic matter content, and pH value from the CSDLv2 dataset.

[0107] Step S2: Classification of winter wheat gluten-type varieties. Based on the preprocessed time-series optical remote sensing image set, vegetation indices such as NDVI, EVI, NDWI, NDYI, RVI, Rc, and polarization decomposition parameters are calculated. Remote sensing images of key growth stages of winter wheat and corresponding calculated features are selected. An ensemble learning classifier is trained by combining ground sample points. The trained classification model is then applied to remote sensing images of the entire study area to generate a 30m resolution distribution map of winter wheat gluten-type varieties.

[0108] Step S3: Remote sensing observation variable inversion and extraction. Based on the preprocessed time-series optical remote sensing image set, calculate the sensitive vegetation indices; combine the acquired ground-measured AGN, LAI, AGB, and SM data and their corresponding pixel vegetation index values, train random forest inversion models respectively, and apply the trained inversion models to remote sensing images of key growth stages of winter wheat to generate time-series sets of spatial distribution data of AGN, LAI, AGB, and SM for winter wheat during the greening, jointing, and heading stages, for subsequent assimilation.

[0109] Step S4: Crop Model Construction and Parameter Optimization. To identify model parameters that significantly impact key outputs of the model simulation (such as LAI, AGB, AGN, and final GPC) and are closely related to quality formation, EFAST is used to perform global sensitivity analysis on potential key crop parameters (such as photoperiod sensitivity, leaf-thermal spacing, nitrogen uptake efficiency, etc.). Based on this, for the key parameters selected by the sensitivity analysis, meteorological and soil data of each target grid obtained in step S1 of this embodiment, as well as representative field management data (sowing date, density, fertilization and irrigation information, etc.) are used as model input drivers. Combined with ground-measured data (LAI, AGB, AGN, and GPC observations of winter wheat samples), the MCMC method is used for localized calibration and optimization of the parameters. For parameters with low sensitivity, the model default values ​​or regional recommended values ​​are directly adopted. Finally, after completing the parameter optimization, the APSIM model is run using the parameter set obtained from the optimization of each target grid and the corresponding driving data to simulate and generate the time series set of key state variables such as LAI, AGB, and AGN of the grid under unassimilated conditions, providing model prior information for the multivariate collaborative data assimilation in the subsequent step S5.

[0110] Step S5, Multivariate Collaborative Data Assimilation. A sequence data assimilation framework based on Ensemble Kalman Filter (EnKF) is constructed. Gaussian perturbations are added to the AGN, LAI, AGB, and SM observation time series generated in Step S3 based on their uncertainties (e.g., inverted RMSE values) to generate an observation member set. Gaussian perturbations are added to the model simulation state variables (including AGN, LAI, AGB, and SM) and key parameters to be optimized generated in Step S4 based on model uncertainties or parameter prior ranges to generate a model simulation member set. The EnKF algorithm is run at time nodes with remote sensing observations, updating the model prediction set using the observation set to obtain the analysis set. This process is iterative, using the four observation variables (AGN, LAI, AGB, and SM) to collaboratively constrain the model state and key parameters.

[0111] Step S6: Generation and output of quality monitoring results. Based on the model state variables and parameter set finally optimized in step S5, the APSIM model is driven to run completely until the winter wheat maturity period; the relevant parameters of the GPC value for each target grid are output; the rib type correction factor and climate factor are introduced to construct the corrected GPC prediction value; the predicted GPC value is compared with the measured GPC value of the independent validation samples obtained from field sampling, and a scatter plot is drawn.

[0112] In one exemplary embodiment, such as Figure 4 As shown, a winter wheat quality monitoring device that couples integrated learning gluten type correction and data assimilation is provided, comprising:

[0113] The information data acquisition module is used to acquire information data.

[0114] The classification module is used to preprocess information data and determine the winter wheat planting area within the target monitoring area based on the classification model, and to divide the winter wheat planting area into different gluten type variety areas. The classification model is obtained by supervising classification training using an ensemble learning method based on the information data corresponding to the sample points of the known winter wheat planting area.

[0115] The inversion and extraction module is used to invert and extract quality-related crop parameters for different rib-type variety areas using machine learning regression models, thereby obtaining the time series of crop parameters.

[0116] The simulated state quantity time series determination module is used to determine the simulated state quantity time series of the gluten-type variety area based on the crop growth model; the crop growth model is a mechanism model used to simulate the growth and development, carbon-nitrogen coupling cycle and grain quality formation process of winter wheat.

[0117] The assimilation module is used to perform multivariate collaborative data assimilation processing based on the time series of crop parameters and the time series of simulated state variables to obtain assimilated data.

[0118] The modified prediction module is used to drive the crop growth model to the maturity and harvest period of winter wheat based on the assimilated data, and to perform GPC modified prediction on winter wheat in the gluten variety area based on the modification factor to obtain the GPC prediction value; the GPC prediction value is used to characterize the quality of winter wheat grains.

[0119] In practical applications, regional winter wheat quality monitoring systems based on crop models and multi-source remote sensing data assimilation can be implemented through software programming. For example... Figure 3As shown, the system includes: a data acquisition and preprocessing module (201), a winter wheat gluten type variety classification module (202), a remote sensing observation variable inversion and extraction module (203), a crop model construction and optimization module (204), a multivariate co-assimilation module (205), and a result generation and output module (206). These modules can be implemented using Python, libraries such as GDAL, Rasterio, Scikit-learn, NumPy, and SciPy, combined with the APSIM model's calling interface. Each module is scheduled through a preset workflow or script and can run on a high-performance computing cluster or cloud platform to achieve automated processing and monitoring.

[0120] The data acquisition and preprocessing module (201) is used to acquire multi-source data covering the target monitoring area and the key growth and development period of winter wheat. The multi-source data includes at least remote sensing satellite data, remote sensing reanalysis product data and ground measurement data, and preprocesses the multi-source data.

[0121] The winter wheat gluten type variety classification module (202) is used to extract the winter wheat planting area within the target monitoring area based on the preprocessed multi-source data, and divide the winter wheat planting area into different gluten type variety areas.

[0122] The remote sensing observation variable inversion and extraction module (203) is used to invert or extract the time series of at least multiple remote sensing observation variables, including LAI, AGB, AGN, and SM, of the target operational unit based on the preprocessed remote sensing satellite data and / or the remote sensing reanalysis product data.

[0123] The crop model construction and optimization module (204) is used to construct and optimize a mechanistic crop growth model that can simulate the growth of winter wheat and the formation of GPC, and to use the model to simulate the time series of state variables of the target work unit.

[0124] The multivariate co-assimilation module (205) is used to co-assimilate multiple remote sensing observation variable time series obtained by the remote sensing observation variable extraction module with the model simulation state variable time series obtained by the crop model and optimization module using a sequence data assimilation algorithm, so as to optimize the model's state variables and key parameters related to GPC formation.

[0125] The results generation and output module (206) is used to drive the crop model operation by using the model state variables and / or key parameters assimilated by the data assimilation module, and to introduce multi-factor correction factors to predict and output the final GPC value of the target work unit.

[0126] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor; the electronic device is characterized in that the processor executes the program to implement the steps of the above-described method.

[0127] A non-transitory computer-readable storage medium storing a computer program; when executed by a processor, the computer program implements the steps of the above method.

[0128] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the scope of protection of this application. For example, the study area can be extended to other winter wheat growing areas; the data source can be replaced with Sentinel series data or Gaofen series data; the crop model can be a mechanistic model such as DSSAT; and the data assimilation algorithm can also adopt particle filtering (PF), variational assimilation (4DVar), etc. These changes do not depart from the core idea of ​​this application.

[0129] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for monitoring the quality of winter wheat by coupling integrated learning gluten type correction and data assimilation, characterized in that, include: Acquire information and data; The information data is preprocessed, and the winter wheat planting area within the target monitoring area is determined based on the classification model, and the winter wheat planting area is divided into different gluten type variety areas; the classification model is obtained by supervising classification training using an ensemble learning method based on the information data corresponding to the sample points of the known winter wheat planting area. For different rib-type variety areas, machine learning regression models are used to invert and extract quality-related crop parameters to obtain crop parameter time series. The time series of simulated state quantities for the gluten-type variety region were determined based on a crop growth model; the crop growth model is a mechanistic model used to simulate the growth and development, carbon-nitrogen coupling cycle, and grain quality formation process of winter wheat. Based on the crop parameter time series and the simulated state quantity time series, multivariate collaborative data assimilation processing is performed to obtain assimilated data; The crop growth model is driven to the winter wheat maturity and harvest period based on the assimilated data, and the GPC correction prediction is performed on the winter wheat in the gluten variety area based on the correction factor to obtain the GPC prediction value. The GPC prediction values ​​are used to characterize the quality of winter wheat grains.

2. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 1, characterized in that, The information data is masked, and information data from different sources and at different resolutions is cropped, reprojected, and time-series synthesized to ensure the consistency of the data in space and time.

3. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 1, characterized in that, The machine learning regression model adopts a random forest regression model; the machine learning regression model is trained based on machine learning methods and is obtained by training on a feature vector set of known crop parameters; the feature vector set includes: normalized vegetation index, enhanced vegetation index, soil-regulated vegetation index, normalized differential water index, green normalized differential vegetation data, green leaf green index, optimized soil-regulated vegetation index, green red-edge vegetation data, radar backscattering coefficient, radar vegetation index, microwave backscattering cross-polarization ratio, and polarization decomposition parameters.

4. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 1, characterized in that, The crop growth model is the APSIM crop model. The crop growth model is obtained by performing a global sensitivity analysis on the model parameters using the extended Fourier amplitude sensitivity test to determine the key parameters, and by using the Markov chain Monte Carlo method in combination with ground-measured data from sample points in the study area to calibrate and optimize the key parameters. The ground-based measured data includes: meteorological data, soil data, field management data, and crop parameters; the field management data includes: sowing time, planting density, fertilizer application rate, and irrigation amount.

5. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 4, characterized in that, The key parameters were determined by calculating the first-order sensitivity index and the global sensitivity index. The expression for the first-order sensitivity index is: The expression for the global sensitivity index is: in, It is a first-order sensitivity index; Let be the variance of the i-th input factor; x represents the expected value of all input factors except the i-th input factor; i Let be the i-th input factor; y be the objective function value; V(y) be the variance of the objective function value; x i-1 All input factors except the i-th input factor; Let be the expected value of the i-th input factor; Let be the variance of all input factors except the i-th input factor; This is a global sensitivity index.

6. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 1, characterized in that, Based on the crop parameter time series and the simulated state quantity time series, multivariate collaborative data assimilation processing is performed to obtain assimilated data, specifically including: Gaussian perturbations are added to the time series of crop parameters and the time series of simulated state variables, and multivariate collaborative data assimilation processing is performed using an ensemble Kalman filter algorithm to obtain assimilated data. The expression corresponding to the Gaussian perturbation is: S = s + aε; Where S is a multivariate vector after adding Gaussian perturbation; s is the input variable data; a is a constant; and ε is a random number vector.

7. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 6, characterized in that, The ensemble Kalman filter algorithm is used for multivariate collaborative data assimilation processing. The corresponding mathematical expression is: B t =HA t +v t ; Among them, B t Let H be a set of observation data at time t; H be the observation operator; A ... t Let v be the set of state variables at time t; t For measuring noise; M represents the predicted set of state variables; M represents the state transformation equation. The set of optimal estimates of state variables at time t-1; w t This is process error; K represents the set of optimal estimates of the state variables at time t. t This is the Kalman gain.

8. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 1, characterized in that, The correction factors include: a ridge type correction factor and a climate factor; the climate factor includes: a temperature influence index and a moisture influence index; the expression for the GPC prediction value is: GPC=γ·[AGB*AGN+f·(LAI)]+FT(t)+FW(SM); Wherein, GPC is the predicted GPC value; γ is the rib type correction factor; FT(t) is the temperature influence index at time t; FW is the water influence index; AGB is the aboveground biomass; AGN is the aboveground nitrogen accumulation; LAI is the leaf area index; f is the empirical parameter obtained from least squares regression analysis; SM is soil moisture; and FW(SM) is the water influence index corresponding to soil moisture.

9. The method for monitoring the quality of winter wheat by coupled integrated learning gluten type correction and data assimilation according to claim 1, characterized in that, The winter wheat quality monitoring method based on crop models and data assimilation also includes: The GPC prediction values ​​are output and displayed in the form of maps, reports, or data services.

10. A winter wheat quality monitoring device that couples integrated learning gluten type correction and data assimilation, characterized in that, include: Information data acquisition module, used to acquire information data; The classification module is used to preprocess the information data, determine the winter wheat planting area within the target monitoring area based on the classification model, and divide the winter wheat planting area into different gluten type variety areas; the classification model is obtained by supervising classification training using an ensemble learning method based on the information data corresponding to the sample points of the known winter wheat planting area. The inversion and extraction module is used to invert and extract quality-related crop parameters for different rib-type variety areas using a machine learning regression model, thereby obtaining a time series of crop parameters. The simulated state quantity time series determination module is used to determine the simulated state quantity time series of the gluten-type variety area based on the crop growth model; the crop growth model is a mechanism model used to simulate the growth and development, carbon-nitrogen coupling cycle and grain quality formation process of winter wheat. The assimilation processing module is used to perform multivariate collaborative data assimilation processing based on the crop parameter time series and the simulated state quantity time series to obtain assimilated data; The correction prediction module is used to drive the crop growth model to the winter wheat maturity and harvest period based on the assimilated data, and to perform GPC correction prediction on winter wheat in the gluten variety area based on the correction factor to obtain the GPC prediction value. The GPC prediction values ​​are used to characterize the quality of winter wheat grains.

Images