Corn yield remote sensing estimation method and system based on multispectral data coupling radiation transmission and crop growth model

Abstract

The invention belongs to the field of crop yield estimation, and discloses a corn yield remote sensing estimation method and system based on multispectral data coupling radiation transmission and a crop growth model, and the method comprises the steps: obtaining and preprocessing a Sentinel-2 multispectral remote sensing image, and obtaining the reflectivity data of a corn planting region; constructing a PROSAIL forward simulation spectrum library, and performing domain correction on the simulation spectrum by using an auto-encoder and a residual error alignment network; establishing a machine learning inversion model based on the corrected simulated spectrum and the actually measured spectrum, and generating regional leaf area index (LAI) distribution; key agronomic parameters are obtained, a localized WOFOST crop growth model is constructed, the LAI state of the model is assimilated by adopting ensemble Kalman filtering at a remote sensing observation moment, and the crop growth process is dynamically corrected; and advancing the model to a mature period, outputting the dry weight of the corn kernels, and realizing remote sensing estimation of the regional corn yield. The method can effectively improve the LAI inversion precision and yield prediction reliability, and is suitable for the fields of agricultural monitoring, grain evaluation, agricultural condition management and the like.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of crop yield prediction technology, and particularly relates to a remote sensing prediction method and system for maize yield based on multispectral data coupled with radiative transfer and crop growth models. Background Technology

[0002] Crop yield estimation is a crucial aspect of agricultural production management, playing a vital role in ensuring food security, optimizing resource allocation, and formulating agricultural policies. Accurate yield forecasts can help governments plan grain reserves and allocate resources in advance, and provide farmers with market information, thereby optimizing planting decisions.

[0003] Traditional methods for estimating corn yield mainly rely on sampling surveys and field observations, which have several drawbacks. First, these methods are time-consuming and labor-intensive, requiring significant human and material resources, and are relatively inefficient. Second, the sample size is limited, making it difficult to cover a large area, resulting in insufficient accuracy and representativeness in yield estimation.

[0004] With the development of remote sensing technology, dynamic monitoring of crop growth and yield over large areas using satellite remote sensing images has become a research hotspot. Remote sensing has advantages such as wide spatial coverage, high data acquisition frequency, and non-contact operation. Multispectral remote sensing data can be used to extract key agronomic parameters such as vegetation indices and leaf area indices, thereby indirectly reflecting the growth status and yield potential of crops.

[0005] Furthermore, crop growth models can simulate the growth process of crops under different meteorological, soil, and management conditions, making them an important tool for yield prediction. However, using crop models alone often relies on high-quality input parameters, lacks spatial resolution, and struggles to accurately describe the crop growth status at the field level. Radiative transfer models can establish a link between remote sensing inversion and physiological variables, serving as a crucial bridge for introducing remote sensing information into crop models.

[0006] Based on the above analysis, the urgent technical problems that need to be solved in the existing technology are:

[0007] There is an urgent need for a novel yield estimation method that integrates multispectral remote sensing observations, radiative transfer models, and crop growth models to couple spatial scales with physiological mechanisms and improve the accuracy and adaptability of maize yield prediction. Summary of the Invention

[0008] To address the problems existing in the prior art, this invention provides a remote sensing prediction method and system for maize yield based on multispectral data coupled with radiative transfer and crop growth models.

[0009] This invention is implemented as follows: a remote sensing prediction method for maize yield based on multispectral data coupled with radiative transfer and a crop growth model, characterized in that the method specifically includes:

[0010] S1: Remote Sensing Data Acquisition and Preprocessing: Remote sensing image data of the target area's maize planting zone is acquired using multispectral remote sensing satellites. Based on phenological information, remote sensing images of the target area's maize planting zone are selected and subjected to radiometric, atmospheric, and orthorectification corrections to generate high-precision remote sensing image data. Then, the remote sensing image data is cropped and mosaicked according to vector boundaries.

[0011] S2: Acquisition of key agronomic parameters: Leaf area index (LAI) sampling was performed on the target area on the observation date; based on the obtained maize remote sensing images, the LAI was retrieved using machine learning algorithms based on the Prosail radiative transfer model to obtain the regional maize LAI.

[0012] S3: Crop growth model input data acquisition: Acquire maize crop data, soil data, farmland management data, and meteorological data for the target area;

[0013] S4: Yield Prediction: The crop growth model is localized, and the obtained maize leaf area index is used as an assimilation variable. The leaf area index output by the crop growth model is assimilated by remote sensing based on ensemble Kalman filtering. The deviation between the model simulation value and the remote sensing observation value is dynamically adjusted, and finally the maize yield is predicted.

[0014] Furthermore, the remote sensing data acquisition and preprocessing steps are specifically as follows:

[0015] (1) The multispectral remote sensing data came from Seninel2, and the input features included: blue band, green band, red band, and near-infrared band;

[0016] (2) Perform radiometric correction, atmospheric correction, orthorectification, and resampling operations on the obtained multi-scene multispectral remote sensing data;

[0017] (3) Perform registration processing;

[0018] (4) Obtain remote sensing images of corn based on corn crop mask.

[0019] Furthermore, the specific steps for obtaining the key agronomic parameters are as follows:

[0020] (1) Leaf area index sampling was performed on the target area on the observation date;

[0021] (2) Construct the Prosail forward simulation library and set the range of variables such as leaf area index (LAI), chlorophyll content, and leaf tilt angle distribution;

[0022] (3) Map the simulated spectrum according to the Sentinel 2 spectral response function;

[0023] (4) Using an autoencoder AE coupled with AlignNet, the simulated spectrum is mapped to a domain consistent with the measured spectrum to obtain the corrected simulated spectrum;

[0024] The autoencoder-alignment network (AE-AlignNet) coupled architecture includes an input layer, an encoding layer, an alignment network layer, and a decoding layer;

[0025] The output of the input layer is connected to the input of the coding layer, the output of the coding layer is connected to the input of the alignment network layer, and the output of the alignment network layer is connected to the input of the decoding layer.

[0026] in,

[0027] The input layer is used to receive simulated or measured multispectral remote sensing reflectance data;

[0028] The coding layer is used to perform feature compression and nonlinear mapping on the input multispectral reflectance data to extract low-dimensional latent feature representations.

[0029] The alignment network layer is used to map and correct the latent features of the encoded analog spectrum. By learning the feature offset relationship between the analog spectrum and the measured spectrum in the latent feature space, the distribution of the corrected latent features is made closer to the distribution of the measured spectral features.

[0030] The decoding layer is used to inverse decode the aligned and corrected latent features to generate a corrected spectrum consistent with the measured spectral distribution, thereby achieving domain alignment between the simulated spectrum and the measured spectrum. Furthermore, the encoding layer adopts a multi-layer fully connected neural network structure, including at least two hidden layers. Each hidden layer uses the ReLU activation function, and after encoding, forms a low-dimensional latent feature representation used to characterize the high-level semantic information of the spectrum.

[0031] The decoding layer adopts a multi-layer fully connected neural network structure that is symmetrical to the encoding layer structure. Its output dimension is consistent with the dimension of the input multispectral reflectance data, which is used to realize spectral reconstruction and maintain the physical consistency of the spectrum.

[0032] Furthermore, the alignment network layer is a residual alignment network AlignNet, which includes at least two fully connected neural network layers and adopts a residual connection structure to correct and map the encoded analog spectral latent features. The mapping relationship satisfies:

[0033] ,

[0034] Where zsim represents the latent eigenvector corresponding to the simulated spectrum, and zcorr represents the latent eigenvector after alignment correction. This represents the nonlinear mapping function learned by the alignment network.

[0035] (5) Machine learning modeling is performed based on the corrected spectrum to retrieve the LAI value;

[0036] (6) Output the spatial distribution map of LAI in the corn region.

[0037] Furthermore, the specific steps for acquiring input data for the crop growth model are as follows:

[0038] Data on corn crops, soil, farmland management, and meteorological conditions were obtained through field measurements or by local agricultural management departments. The crop and farmland management data were in YAML format, the soil data was in JSON format, and the meteorological data was in CSV format.

[0039] Furthermore, the production forecasting step specifically includes:

[0040] (1) Localization of crop growth models;

[0041] (2) Run the crop growth model according to the daily step size;

[0042] (3) Use ensemble Kalman filtering to assimilate LAI at the remote sensing observation time points and update the model state;

[0043] (4) Output yield of crop growth model.

[0044] Another objective of this invention is to provide a remote sensing prediction system for maize yield based on multispectral data coupled with radiative transfer and crop growth models. This system specifically includes:

[0045] The data acquisition and preprocessing module is used to acquire remote sensing image data of the corn planting area in the target region and perform preprocessing such as radiometric correction, atmospheric correction and orthorectification.

[0046] The key agronomic parameter acquisition module is used to obtain the regional maize leaf area index;

[0047] The crop growth model input data acquisition module acquires corn crop data, soil data, farmland management data, and meteorological data for the target area;

[0048] The yield forecasting module is used to forecast corn yield.

[0049] This method uses the coefficient of determination and mean square error as indicators for evaluating the accuracy of LAI inversion:

[0050] ,

[0051] in, and These are the measured value and the predicted value, respectively. The mean of the measured values ​​is n, where n is the sample size.

[0052] ,

[0053] in, and These are the measured value and the predicted value, respectively, and n is the sample size.

[0054] This method uses the average relative error as the yield verification indicator:

[0055] ,

[0056] in, and These are the measured value and the predicted value, respectively, and n is the sample size.

[0057] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0058] This invention significantly reduces the difference between simulated and measured spectral domains by correcting the spectrum using AE-AlignNet; it significantly improves the accuracy of LAI inversion and enhances the WOFOST assimilation effect; the remote sensing model and the mechanistic model are deeply coupled, and the yield prediction has interpretability, dynamism and spatial continuity; it is applicable to scenarios such as regional management, disaster reduction assessment and food monitoring.

[0059] First, at the level of remote sensing inversion mechanism, this invention introduces an autoencoded feature mapping and residual distribution alignment mechanism to systematically correct the statistical distribution differences between simulated spectra generated based on the radiative transfer model and actual remote sensing observation spectra. This effectively overcomes the long-standing "domain inconsistency" problem between simulated and measured spectra in traditional physical models. Through this correction mechanism, the statistical characteristics of the simulated spectrum can approximate those of actual remote sensing observations while maintaining physically interpretable constraints. This significantly reduces the systematic errors caused by spectral distribution shifts during the inversion process, resulting in higher stability and accuracy of leaf area index inversion results across different growth stages and spatial locations.

[0060] Secondly, regarding model coupling and information utilization, this invention does not simply substitute remote sensing inversion results as static input into the crop growth model. Instead, it uses an assimilation mechanism to apply the inverted leaf area index as an observation constraint, guiding the dynamic correction of the internal state of the crop growth model. This closed-loop structure of "inversion-assimilation-evolution" allows remote sensing information to continuously participate in model state updates over time, avoiding the fragmented and phased utilization of remote sensing information in traditional methods. This significantly improves the consistency between simulated growth processes and actual crop physiological changes, thereby enhancing the reliability of yield prediction results.

[0061] Furthermore, regarding the scientific validity and interpretability of yield prediction results, this invention embeds remote sensing observation information into the crop physiological evolution path through mechanistic model-driven growth process simulation. This ensures that the final yield output not only has numerical predictive significance but also allows for tracing key physiological state changes during its formation. This approach overcomes the "black box" limitations of purely data-driven models, providing a clear physical and physiological basis for the prediction results and offering more credible support for agricultural management and decision-making.

[0062] Finally, in practical applications, this invention achieves an organic unity between spatial continuous inversion of leaf area index at the regional scale and dynamic simulation of crop growth status, making it suitable for monitoring maize yield in different regions, years, and under different management conditions. This technical solution has good adaptability and promotional value in scenarios such as regional precision agricultural management, agricultural disaster assessment, food security monitoring, and macro-agricultural decision support, and can provide stable, high-precision, and interpretable technical support for the development of agricultural informatization and smart agriculture.

[0063] The technical solution of this invention overcomes technical biases: For a long time, those skilled in the art have generally held the following technical cognitive biases: On the one hand, they believe that radiative transfer models have clear physical mechanisms, but due to their complex parameters and numerous idealized assumptions, there are unavoidable systematic deviations between them and actual remote sensing observations, making them difficult to directly use for high-precision inversion; on the other hand, they believe that although data-driven models can fit remote sensing observation data, they lack physical constraints, have poor generalization ability, and are difficult to apply stably in different regions, years, or crop conditions. Therefore, existing technologies usually use physical models and data-driven models separately, or only perform simple fusion at a shallow level, generally believing that the two are difficult to achieve deep synergy.

[0064] This invention overcomes the aforementioned technical biases and proposes for the first time a technical approach that systematically couples radiative transfer models, deep feature learning, and crop growth models. By introducing an autoencoder and a residual alignment network, the distribution alignment between simulated and measured spectra is achieved in the latent feature space, thus effectively unifying the physical model output with remote sensing observation data.

[0065] Therefore, this invention breaks through the long-standing technical bias in the field of remote sensing inversion and crop model coupling from a methodological perspective, and provides a new and scalable technical paradigm for remote sensing estimation of crop yield. Attached Figure Description

[0066] Figure 1 This is a flowchart of a remote sensing prediction method for maize yield based on multispectral data coupled with radiative transfer and crop growth model provided in an embodiment of the present invention.

[0067] Figure 2 This is a flowchart of the production forecasting process provided in an embodiment of the present invention;

[0068] Figure 3 This is a module diagram of a remote sensing prediction system for maize yield based on multispectral data coupled radiative transfer and crop growth model provided in an embodiment of the present invention.

[0069] Figure 4 This is a comparison chart of the average reflectance of the simulated spectrum, the measured spectrum, and the spectrum corrected by the method of this invention in each band, provided in the embodiments of this invention.

[0070] Figure 5 This is a machine learning regression modeling diagram of the corrected spectrum provided in an embodiment of the present invention. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0072] like Figure 1 As shown in the figure, this invention provides a remote sensing prediction method for maize yield based on multispectral data coupled with radiative transfer and crop growth models. The method specifically includes:

[0073] S101, acquire multispectral remote sensing reflectance data of the target area;

[0074] S102, based on the radiative transfer mechanism, constructs a positive spectral simulation space that includes leaf area index, chlorophyll content and leaf structure parameters;

[0075] S103, map the positive analog spectrum to a multispectral reflectance space consistent with that of the remote sensing sensor;

[0076] S104, The distribution consistency between the simulated spectrum and the measured remote sensing spectrum is corrected by a domain alignment mechanism;

[0077] S105, after completing the distribution consistency correction, constructs the inversion model and outputs the spatial distribution results of the crop leaf area index.

[0078] The specific method is as follows:

[0079] (1) Remote sensing data processing.

[0080] The remote sensing data specifically refers to L2A level data of the target area in Seninel2A / B. The input features include four bands: blue light, green light, red light, and near-infrared light.

[0081] The center wavelength of the blue light band is 490nm, the bandwidth is 65nm, and the spatial resolution is 10m.

[0082] The center wavelength of the green light band is 560nm, the bandwidth is 35nm, and the spatial resolution is 10m.

[0083] The center wavelength of the red light band is 665nm, the bandwidth is 30nm, and the spatial resolution is 10m;

[0084] The center wavelength of the near-infrared band is 842nm, the bandwidth is 115nm, and the spatial resolution is 10m.

[0085] After obtaining the corn crop mask of the target area, the remote sensing data is resampled, registered, and the mask is extracted. The crop planting area within the target area is retained to obtain the remote sensing data of the corn crop.

[0086] (2) Steps for obtaining key agronomic parameters.

[0087] 1) Construct the Prosail forward simulation database:

[0088] Set the value ranges for leaf area index (LAI), chlorophyll content (Cab), leaf structure parameter (N), etc., and randomly select 10,000 combinations from them according to a certain distribution to input into the Prosail model to construct a simulation database.

[0089] The simulated spectrum was convolved using Sentinel-2 SRF (spectral response function) to ensure band uniformity:

[0090] ,

[0091] in, Simulated Sentinel-2A equivalent remote sensing reflectance; : High spectral reflectance simulated by PROSAIL; : The spectral response function of the sensor in the target band.

[0092] 2) Using an autoencoder (AE) coupled with AlignNet, the simulated spectrum is mapped to a domain consistent with the measured spectrum to obtain the corrected simulated spectrum. The specific steps are as follows:

[0093] Nearest neighbor pairing based on Mahalanobis distance: Pair each measured spectrum with the closest spectrum in the simulated spectral library to obtain... ;

[0094] ,

[0095] in, Sentinel-2 measured spectral data; PROSAIL simulated spectral data; : The covariance matrix of the simulated spectrum set; : Covariance inverse matrix; Mahalanobis distance.

[0096] Build a shared encoder With dual decoders :

[0097] ,

[0098] in, Encoder network; : Low-dimensional latent vector.

[0099] ,

[0100] in, : Reconstructing the simulated spectrum; : Reconstruct the measured spectrum.

[0101] Reconstruction losses:

[0102] ,

[0103] ResidualAlignNet domain alignment module:

[0104] To further reduce the difference in latent distribution between the simulated and measured domains, a residual alignment network is introduced:

[0105] ,

[0106] in,

[0107] : Residual alignment network composed of multilayer sensing mechanism.

[0108] Domain alignment loss is calculated using maximum mean difference (MMD):

[0109] ,

[0110] ,

[0111] ,

[0112] in,

[0113] : Core bandwidth, ranging from 0.1 to 2.0, determined through cross-validation.

[0114] The final total loss is:

[0115] ,

[0116] in,

[0117] Weighting coefficient, taken as 0.5 here.

[0118] Generate corrected simulated spectra:

[0119] After training, the simulated spectrum is input into AE-AlignNet to obtain the corrected spectrum mapped to the "measured domain".

[0120] ,

[0121] 3) Construct an LAI inversion model based on the corrected spectrum.

[0122]

[0123] in,

[0124] : A dataset consisting of simulated and measured spectra for calibration.

[0125] The final inversion model is obtained by using a machine learning model for regression training.

[0126] 4) Based on the constructed model, output the spatial distribution map of LAI in the corn region.

[0127] (3) Steps for obtaining input data for crop growth model.

[0128] Meteorological data was obtained by downloading ERA5 data from the European Space Agency, maize crop data was set up based on previous studies in the same region, and soil and farmland management data for the target area were obtained through field measurements.

[0129] The crop data includes the effective accumulated temperature of the crop, etc.

[0130] The soil data includes wilting moisture, field water holding capacity, and saturated moisture content.

[0131] The farmland management data includes fertilization, irrigation operations, crop sowing dates, and crop harvest dates.

[0132] The meteorological data includes daily minimum temperature, daily maximum temperature, daily average vapor pressure, daily average wind speed, daily total precipitation, total shortwave radiation, altitude, and latitude and longitude.

[0133] (4) Production forecasting steps.

[0134] Production forecasting process as follows Figure 2 As shown.

[0135] A crop growth model was established based on the WOFOST model using the acquired data.

[0136] The obtained maize leaf area index (LAI) was used as an assimilation variable to perform ensemble Kalman filtering remote sensing assimilation on the leaf area index output by the crop growth model. The specific steps are as follows:

[0137] Initialization set: Based on the localized crop growth model, an initial set of state variables is constructed in the form of perturbations.

[0138] ,

[0139] Each of them Includes LAI state; N is the number of set members, ranging from 50 to 100; the initial set constructs the uncertainty of the initial state of the simulation model by introducing a normal distribution perturbation to the initial state.

[0140] Prediction Phase: Before each remote sensing observation date t, the crop growth model is used to advance the set from the previous time step to the current time step, resulting in a predicted state set:

[0141] ,

[0142] in,

[0143] : The predicted state of the i-th member;

[0144] Localized WOFOST crop growth model;

[0145] : Inputs to the driving model (data such as meteorological, soil, and management measures);

[0146] : Prediction process noise.

[0147] Extract the LAI value corresponding to each set member to form a predicted LAI set:

[0148] LAI observations obtained through inversion of key agronomic parameters Assuming the observation error follows a Gaussian distribution, the error covariance matrix is ​​defined. :

[0149] ,

[0150] in The inversion standard deviation of the remote sensing LAI is 0.2.

[0151] The observed disturbance value is:

[0152] ,

[0153] Analysis and Update Phase: Based on the deviation between the predicted state and the perturbation observation, update the LAI state value for each ensemble member:

[0154] Calculate the mean of the prediction set Covariance :

[0155] ,

[0156] Calculate Kalman gain :

[0157] ,

[0158] Where H is the observation matrix; T represents the matrix transpose.

[0159] Update the LAI state of the collection members:

[0160] ,

[0161] The updated LAI will replace the LAI component in the model state and serve as the basis for the next step of model advancement.

[0162] Yield simulation phase: After the final remote sensing LAI assimilation is completed, the crop growth model is used to advance the state of each ensemble member to maturity, the grain dry weight is output, and the average of the simulated yields of all members is taken to obtain the final yield estimate.

[0163] ,

[0164] like Figure 3 As shown in the figure, an embodiment of the present invention provides a remote sensing prediction system for maize yield based on multispectral data coupled with radiative transfer and crop growth models, specifically including:

[0165] The data acquisition and preprocessing module is used to acquire remote sensing image data of the corn planting area in the target region and perform preprocessing such as radiometric correction, atmospheric correction and orthorectification.

[0166] The key agronomic parameter acquisition module is used to obtain the regional maize leaf area index;

[0167] The crop growth model input data acquisition module acquires corn crop data, soil data, farmland management data, and meteorological data for the target area;

[0168] The yield forecasting module is used to forecast corn yield.

[0169] Example 1: Example of Maize Leaf Area Index Inversion Based on Radiative Transfer Mechanism and Domain Alignment

[0170] This embodiment takes a typical maize-producing area as the research object and acquires multi-temporal, multispectral remote sensing reflectance data during the maize growth period. First, the remote sensing images undergo radiometric correction, atmospheric correction, orthorectification, resampling, and spatial registration. Then, a maize planting area reflectance dataset is generated by combining this data with plot boundary information. Simultaneously, field surveys are conducted on the observation dates to obtain leaf area index samples at different growth stages for subsequent inversion model construction and validation.

[0171] Building upon this foundation, a forward spectral simulation space covering the variation ranges of leaf area index, chlorophyll content, and leaf structural parameters is constructed based on the radiative transfer mechanism. The simulated spectra are then converted into reflectance forms consistent with multispectral remote sensing. By mapping the simulated and measured remote sensing spectra to a unified latent feature space and minimizing the statistical distribution differences between the two types of samples in this space, the simulated spectral distribution is corrected to the measured distribution. Finally, an inversion model is jointly constructed using the corrected spectral samples and the measured samples, outputting the spatial distribution results of leaf area index at the regional scale.

[0172] Example 2: Stable generation of multi-temporal leaf area index inversion results

[0173] Building upon Example 1, this example further focuses on continuous inversion of multi-temporal remote sensing data. Multiple remote sensing images covering the seedling, jointing, tasseling, and grain-filling stages of maize are selected. Preprocessing and spectral extraction are performed on each image to generate time-series multispectral reflectance data. A unified simulated spectrum generation rule and distribution correction mechanism ensure the consistency of statistical characteristics of the inversion input data across different observation phases.

[0174] In the continuous inversion process, the same inversion model structure was used to process remote sensing data from different observation dates, outputting multi-temporal leaf area index spatial distribution results. Temporal consistency verification confirmed that the inverted leaf area index exhibits continuous variation characteristics consistent with maize growth patterns during the growing season, avoiding the stage-specific biases that easily occur in single-temporal inversions, and providing stable and reliable observational input for subsequent dynamic coupling with crop growth models.

[0175] Example 3: Dynamic Update of Crop Growth Status Based on Leaf Area Index Assimilation

[0176] This embodiment constructs a crop growth simulation process based on the multi-temporal leaf area index inversion. First, crop variety information, soil physicochemical properties, field management practices, and meteorological data for the target area are collected and organized into a data format readable by the crop growth model. Starting from the seedling emergence period, the crop growth model is driven to run at daily steps to obtain simulation results of the crop growth process under unassimilated conditions.

[0177] On the remote sensing observation dates, the retrieved leaf area index is incorporated as observational information into the model's operation. Ensemble statistical methods are used to dynamically correct the model's internal growth state. After each assimilation, the model state adjusts towards the constraints of the remote sensing observations and continues to evolve in the next time step. After multiple observation assimilations, the crop growth trajectory gradually approximates the actual growth process over time, providing a physically consistent and observationally constrained state basis for the final yield estimation.

[0178] Example 4: Regional Localized Parameter Calibration and Maize Yield Output Example

[0179] This embodiment further verifies the adaptability of model parameters under different regional conditions. Several representative maize planting plots were selected, and corresponding crop, soil, and management parameters were obtained, thus constructing a model parameter search space. Key model parameters were adjusted using optimization methods to ensure that the simulated leaf area index and biomass accumulation characteristics were consistent with regional observation results, thereby achieving regional adaptive determination of model parameters.

[0180] After parameter calibration, the assimilated model state is used to advance the crop growth process to the maturity stage, outputting the maize yield results for the corresponding plot. Comparing the simulated yield with measured statistical data verifies that this method has good stability and generalization ability under different plots and years, and can effectively couple remote sensing observation, crop growth mechanism, and yield output.

[0181] Spectral correction based on AE-AlignNet: Figure 4The comparison results of average reflectance in each band are shown for the simulated spectrum, the measured spectrum, and the spectrum corrected by the method of this invention. It can be seen that there are significant deviations between the original PROSAIL simulated spectrum and the measured spectrum in the visible and near-infrared bands, especially in the red and near-infrared bands, where there is a systematic overestimation, affecting the accuracy of subsequent parameter inversion. After correction by the autoencoder-alignment network (AE-AlignNet), the reflectance amplitude and variation trend of the corrected spectrum in each band are highly consistent with the measured spectrum, significantly reducing the distribution difference between the simulated and measured domains. Particularly in the vegetation-sensitive red and near-infrared bands, it can more accurately reflect the characteristics of canopy structure and leaf area changes.

[0182] Figure 5 Machine learning regression modeling is performed on the corrected spectrum, R 2 The range is between 0.59 and 0.71.

[0183] Using this method to predict regional maize yield, the results show that the average estimated yield of the fields in the study area is only 2.5% lower than the statistical data.

[0184] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0185] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A remote sensing prediction method for maize yield based on multispectral data coupled with radiative transfer and crop growth models, characterized in that, Includes the following steps: Step 1: Obtain multispectral remote sensing reflectance data of the target area; Step 2: Construct a forward spectral simulation space based on the radiative transfer mechanism, which includes leaf area index, chlorophyll content, and leaf structure parameters; Step 3: Map the positive simulated spectrum to a multispectral reflectance space consistent with that of the remote sensing sensor; Step 4: Correct the distribution consistency between the simulated spectrum and the measured remote sensing spectrum using a domain alignment mechanism; Step 5: After completing the distribution consistency correction, construct the inversion model and output the spatial distribution results of the crop leaf area index.

2. The method as described in claim 1, characterized in that, The positive spectral simulation space is established through a radiative transfer model, which simultaneously characterizes the coupled influence of canopy structure parameters and leaf optical properties on reflectivity.

3. The method as described in claim 1, characterized in that, The multispectral reflectance mapping process is achieved by weighted integration of the hyperspectral simulated reflectance according to the sensor's spectral response characteristics.

4. The method as described in claim 1, characterized in that, The distribution consistency correction includes: Map simulated and measured spectra to a shared latent feature space; Alignment of simulated spectra with measured spectral distributions is achieved by minimizing the difference in the distribution of the two types of samples in the latent feature space.

5. A crop yield prediction method based on the coupling of remote sensing inversion results and a crop growth model, implementing the remote sensing prediction method for maize yield based on multispectral data coupled with a crop growth model as described in any one of claims 1-4, characterized in that, Includes the following steps: Construct a crop growth model and set the initial state of the model; The leaf area index obtained by the method described in any one of claims 1 to 4 is used as the observation variable; During crop growth, the relative model state is dynamically updated at multiple observation times. After completing all observation assimilation, the crop growth model is advanced to the maturity stage and the crop yield is output.

6. The method as described in claim 5, characterized in that, The model state update is achieved through ensemble statistical methods, using observed variables to correct the internal physiological state of the model.

7. The method as described in claim 5, characterized in that, The model states include at least growth state variables and photosynthetic accumulation state variables related to leaf area index.

8. An adaptive calibration method for crop growth model parameters oriented towards a target region, implementing the remote sensing prediction method for maize yield based on multispectral data coupled with radiative transfer and crop growth models as described in any one of claims 1-4, characterized in that, include: Obtain crop parameters, soil parameters, management parameters, and meteorological parameters for the target area; Constructing the parameter search space for crop growth models; The model parameters were adjusted by optimization methods to make the model simulation results consistent with the regional observation characteristics. The calibrated model parameters will be used for subsequent growth simulations and yield predictions.

9. The method as described in claim 8, characterized in that, The optimization method achieves adaptive determination of regional parameters by minimizing the difference between the model output and the observed data.

10. The method as described in claim 8, characterized in that, The calibrated model parameters are used as fixed parameters input into the crop yield prediction method of claim 5.

Images