A crop model online calibration method and system based on large sprinkler image feedback

Abstract

The application discloses a crop model online calibration method and system based on image feedback of a large-scale sprinkler. The application converts real-time high spatial resolution image data obtained by the large-scale sprinkler into key parameters that can be used for assimilation of a crop growth model, and dynamically corrects a WOFOST model (World Food Studies) by using a data assimilation technology. The application effectively overcomes the shortcomings of static parameters and spatial homogenization of a traditional model, and enables the model to perceive and adapt to spatial variation in a field. The application realizes closed-loop optimization of crop growth monitoring and model prediction, significantly improves the growth simulation and yield prediction accuracy of the WOFOST model at a specific field scale, and provides more reliable data support for precise irrigation decision-making based on the model.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture and agricultural meteorological model technology, specifically to an online calibration method and system for crop models based on image feedback from large sprinkler irrigation machines. Background Technology

[0002] Crop growth models such as WOFOST (World Food Studies) are core to precision irrigation decision-making, but their predictive accuracy heavily relies on accurate initial parameters and conditions (such as leaf area index (LAI), biomass, and crop growth stage). Traditional parameter acquisition methods have significant shortcomings: relying on historical data or default parameters cannot reflect actual field differences; manual field measurements, while accurate, are inefficient, costly, and fail to characterize spatial heterogeneity in large fields.

[0003] Satellite and UAV remote sensing offer the possibility of large-scale monitoring, but they face limitations in practical applications: high-resolution satellites have long revisit cycles, are susceptible to cloud interference, and have insufficient data timeliness; high-frequency satellites have too low spatial resolution to match the scale of sprinkler irrigation operations. UAVs have high operating costs, making it difficult to achieve routine monitoring synchronized with daily irrigation. Existing remote sensing data streams are often disconnected from the irrigation decision-making chain, making it difficult to form a real-time closed loop.

[0004] Large sprinkler irrigation machines (center-pivot type and translational type) are the backbone equipment of modern field irrigation, and their mobility and full-coverage characteristics make them naturally suitable as field monitoring platforms. However, at present, sprinkler irrigation machines are mainly limited to performing irrigation operations, and the sensors they carry are mostly used to monitor engineering parameters. They have not yet been effectively developed into real-time acquisition terminals for crop growth information, let alone integrated online with crop growth models.

[0005] Therefore, how to use sprinkler irrigation platforms to collect key crop growth parameters (such as LAI) at low cost and high frequency, and use this to dynamically correct the parameters of the mechanism model to generate real-time and accurate irrigation decisions, has become a technical problem that urgently needs to be solved in the field of smart agriculture. Summary of the Invention

[0006] To address the problems mentioned above, this invention provides a method for dynamically correcting WOFOST model parameters based on image data from a large sprinkler irrigation machine camera. The core of this method lies in using leaf area index (LAI) observations retrieved from the images to perform online assimilation and correction of the model.

[0007] To achieve the above objectives, this invention provides an online calibration method for crop models based on image feedback from large sprinkler irrigation machines, comprising the following steps: S1. Obtain crop canopy coverage based on the color images of the crop canopy captured by the visible light camera mounted on the truss of the large sprinkler irrigation machine; S2. Obtain leaf area index observations based on crop canopy coverage; S3. Based on the observed leaf area index, the ensemble Kalman filter algorithm is used to assimilate and correct the WOFOST model online, and the corrected model parameters are obtained. S4. Restart the WOFOST model based on the corrected model parameters and obtain the crop growth prediction results.

[0008] Preferably, S1 includes: The crop canopy color image is binarized based on the supergreen index to obtain green vegetation pixels; The crop canopy coverage is obtained by the ratio of green vegetation pixels to the total number of pixels in the image.

[0009] Preferably, S2 includes: Based on a pre-established empirical model of leaf area index and crop canopy coverage, the estimated value of leaf area index is obtained. Based on the leaf area index estimates from several images, the average leaf area index of the field was obtained.

[0010] Preferably, based on the leaf area index observation, S3 includes: The model set is obtained by applying random perturbations to the initial state and specific leaf area parameter table of the WOFOST model; Run the model ensemble to the current observation time point and obtain the simulated leaf area index for each ensemble member; The observation operator is obtained based on the ratio of effective leaf area to land area; The Kalman gain is calculated based on the observed leaf area index, the simulated leaf area index for each set member, and the observed operator. The state vector and parameter vector of the model set are updated based on the Kalman gain to obtain the updated set; Based on the mean of the updated set, obtain the corrected specific leaf area parameter table.

[0011] Preferably, the state vector includes: developmental stage, total aboveground dry weight, effective leaf area, stem dry weight, storage organ dry weight, root dry weight, and values ​​for each developmental stage in the specific leaf area parameter table.

[0012] Preferably, the state vector and parameter vector of the model set are updated according to the Kalman gain, including: The update amounts of state and parameters are obtained based on the difference between the observed and simulated leaf area index, as well as the Kalman gain. The state vector and parameter vector of each member in the model set are updated synchronously according to the update amount.

[0013] Preferably, S4 includes: Restart the WOFOST model based on the revised leaf area parameter table and the latest model status to obtain future leaf area index dynamics and biomass accumulation prediction results. Based on the future dynamics of leaf area index and the prediction results of biomass accumulation, crop water requirement forecasts are obtained; Based on crop water requirement predictions, a variable irrigation prescription map is generated.

[0014] This invention also provides an online calibration system for crop models based on image feedback from large sprinkler irrigation machines. The system is used to implement the above method and includes: The acquisition module is a visible light camera mounted on the truss of a large sprinkler irrigation machine, used to acquire color images of the crop canopy and obtain the crop canopy coverage. The acquisition module is used to obtain leaf area index observations based on crop canopy coverage. The correction module is used to perform online assimilation and correction of the WOFOST model based on the leaf area index observations and the ensemble Kalman filter algorithm to obtain the corrected model parameters. The prediction module is used to restart the WOFOST model based on the corrected model parameters and obtain crop growth prediction results. Compared with the prior art, the beneficial effects of this invention are as follows: This invention transforms real-time, high spatial resolution image data acquired by a large sprinkler irrigation machine—a mobile platform—into key parameters that can be assimilated into crop growth models, and then uses data assimilation technology to dynamically correct the WOFOST model. This effectively overcomes the shortcomings of traditional model parameters being static and spatially homogenized, enabling the model to perceive and adapt to spatial variations within the field. This invention achieves closed-loop optimization of crop growth monitoring and model prediction, significantly improving the accuracy of WOFOST model growth simulation and yield prediction at the specific field scale, and providing more reliable data support for model-based precision irrigation decisions. Attached Figure Description

[0015] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the method flow in this embodiment; Figure 2 This is a schematic diagram of the system structure in this embodiment. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] Example 1 like Figure 1 As shown in this embodiment, an online calibration method for crop models based on image feedback from a large sprinkler irrigation machine is provided. The steps include: S1. Obtain crop canopy coverage based on the color images of the crop canopy captured by the visible light camera mounted on the truss of the large sprinkler irrigation machine.

[0020] During the operation of large sprinkler irrigation machines (central pivot type or translation type), visible light cameras mounted on their trusses continuously collect color images of the crop canopy below; simultaneously recording the geographical location, time, and environmental data corresponding to each image.

[0021] S2. Obtain leaf area index observations based on crop canopy coverage.

[0022] The image acquired by S1 is processed to extract crop canopy information. First, a color threshold-based segmentation algorithm is used to separate the green vegetation region from the image. The algorithm is implemented as follows: the preprocessed RGB image is converted to the HSV color space, and the H (hue) and S (saturation) channels are extracted; based on the distribution characteristics of green vegetation in the HSV space, a fixed threshold range H∈[35°, 85°] and S∈[0.2, 1.0] are set, and the image is binarized. Pixels that meet the threshold conditions are marked as green vegetation, and the rest are marked as soil background; morphological opening operations (3×3 rectangular kernel) are performed on the segmented binary image to remove isolated noise points, resulting in the final crop canopy binary mask. Then, the canopy coverage and canopy texture features of the region are calculated. Using a pre-established LAI inversion model for the target crop, the canopy coverage, texture features, and solar altitude angle at the time of image acquisition are used as inputs to output the estimated crop canopy LAI value for the corresponding region of the image. The inversion model was obtained by regression training using ground-measured LAI data and synchronously acquired features from sprinkler camera images.

[0023] S3. Based on the observed leaf area index, the ensemble Kalman filter algorithm is used to assimilate and correct the WOFOST model online, and the corrected model parameters are obtained.

[0024] The core of this step is to optimize the key parameters controlling leaf growth in the WOFOST model using LAI observation sequences.

[0025] S3.1: Model State and Parameter Definition: In the WOFOST model, the key state variable directly related to LAI dynamics is defined as effective leaf area (LV), and the key process parameter is specific leaf area (SLATB). SLATB is a parameter table that varies with developmental stage, determining the leaf area that can be formed per unit dry matter, and is the core of controlling the LAI growth rate.

[0026] S3.2: Data Assimilation Execution: An assimilation framework is constructed using the Ensemble Kalman Filter (EnKF) algorithm. The specific process is as follows: Set initialization: Generate a WOFOST model set. Set diversity is achieved by applying random perturbations consistent with prior knowledge to the initial state of the model and the values ​​of the key parameter table SLATB.

[0027] Model prediction step: Starting from the last assimilation time point, run the model for each ensemble member up to the current observation time point to obtain the simulated LAI value (LAI_sim_i) and other states (such as biomass TAGP) for each member.

[0028] Application of the observation operator: The WOFOST model directly outputs the effective leaf area LV (㎡), which can be converted into the simulated LAI using the formula LAI = LV / land area. This is the observation operator H, which maps the model state space to the observation space (LAI).

[0029] Assimilation Update Step: The current-time field-average LAI observation value (LAI_obs) and its observation error obtained from the inversion in step S2 are introduced. The EnKF algorithm calculates the ensemble statistics (mean, variance) of the simulated LAI, as well as the covariance between the simulated LAI and all model state variables and parameters (including the values ​​of each developmental stage of SLATB). Subsequently, the Kalman gain is calculated, which determines the weight of the observation information for each state variable and parameter update. Finally, using the difference between LAI_obs and each LAI_sim_i, the entire state vector (including LV, TAGP, etc.) and parameter vector (mainly the SLATB table) of all members are synchronously updated according to the Kalman gain.

[0030] S3.3: Correction Results Output and Model Restart: After assimilation and update, the ensemble mean of all member states and parameters is the optimal estimate after correction based on the current LAI observations. This correction directly optimizes the physiological parameter SLATB, which controls leaf growth, to better reflect the actual growth characteristics of the crop in the current field. Using this optimal estimate as the new initial condition, the WOFOST model is restarted.

[0031] S4: Restart the WOFOST model based on the corrected model parameters and obtain crop growth prediction results.

[0032] By using the revised model for rolling forecasts, more accurate predictions of future LAI dynamics, biomass accumulation, and derived crop water requirements are obtained. Based on this, variable irrigation prescription maps are generated to guide sprinkler irrigation.

[0033] Example 2 This embodiment uses a wheat field planted on a farm in a plain in my country as an example to illustrate the implementation process of the present invention.

[0034] Step S1: Obtain the crop canopy coverage based on the color image of the crop canopy captured by the visible light camera mounted on the truss of the large sprinkler irrigation machine. Four high-definition visible light cameras are evenly spaced on the truss of the center-supported sprinkler irrigation machine, ensuring their field of view covers the entire operating circle. The sprinkler control cabinet integrates a GPS module and a 4G communication module. The sprinkler operates one revolution along a fixed path every day at 10:00 AM. During operation, the cameras capture images of the crops below at one frame per second, each image automatically appended with a timestamp and location coordinates generated by the GPS module. Simultaneously, a small automatic weather station installed on the central tower of the sprinkler records temperature, humidity, and solar radiation data. All images, location, and environmental data are uploaded in real-time to a central processing unit on a cloud server via a 4G network.

[0035] Step S2: Obtain leaf area index observations based on crop canopy coverage.

[0036] Image preprocessing and segmentation: Color correction is performed on the image using an ultragreen index ExG=2. GRB is binarized to segment out green vegetation pixels, and the canopy coverage (CC) of a single image is calculated as: number of vegetation pixels / total number of pixels.

[0037] LAI Inversion Model: The relationship between CC and LAI was established through ground experiments. At different growth stages of wheat, LAI was measured at multiple sampling points in the field using an LAI-2200 canopy analyzer. Simultaneously, sprinkler irrigation equipment was used to capture images from above the corresponding locations, and CC was calculated. Eighty sets of data were collected to establish an empirical LAI-CC model for this wheat variety: LAI_est = -(1 / (2 k)) ln(1-CC). Where k is the extinction coefficient, determined through fitting (k≈0.65 in this example). For each image, substitute its CC value into this formula to obtain the estimated LAI value for that point. After processing the images for the entire day, a field LAI distribution map can be generated, and the average LAI observation value (LAI_obs) for the current day can be calculated.

[0038] Step S3: Based on the observed leaf area index, the ensemble Kalman filter algorithm is used to assimilate and correct the WOFOST model online, and the corrected model parameters are obtained.

[0039] Objective: To correct the key parameter – leaf area SLATB.

[0040] Assimilation system settings: State vector: x=[DVS,TAGP,LV,WST,WSO,WRT,...,SLATB_1,SLATB_2,...], where DVS is the developmental stage, TAGP is the total dry weight of the aboveground parts, LV is the effective leaf area, and SLATB_1, SLATB_2,... represent the values ​​of SLATB at different developmental stages.

[0041] Observation vector: y = [LAI].

[0042] The observation operator H is: H(x) = LV / (land area), which means that the LAI is calculated from the model state LV.

[0043] Assimilation process: Forecast: Starting from the state on day 55, run a model ensemble of 50 members until day 60. Each member will simulate a different LAI growth trajectory due to a different initial SLATB parameter table, resulting in 50 LAI_sim_i models.

[0044] renew: a. Obtain the average LAI_obs = 3.8 obtained by inverting the sprinkler machine image on day 60, and set the observation error.

[0045] b. Calculating the Kalman gain: The core of the EnKF algorithm is calculating the gain matrix K. K is proportional to the covariance of the state / parameter and the simulated LAI, and inversely proportional to the sum of the variance of the simulated LAI and the observation error. This means that if a parameter is highly correlated with the changes in the simulated LAI (large covariance), and the current model set has a large uncertainty in predicting the LAI (large variance), then the correction magnitude (gain) for that parameter in the current observed LAI will be larger.

[0046] c. State and Parameter Updates: For each member i, the update amount for its state and parameters is: Δx_i = K (LAI_obs - LAI_sim_i). Regarding the SLATB parameter, if the model generally overestimates LAI (mean LAI_sim_i > 3.8), then (LAI_obs - LAI_sim_i) is negative, resulting in a negative SLATB update amount Δ(SLATB). This leads to a reduction in the SLATB value after the update, making the model's subsequent leaf growth rate predictions more conservative and closer to reality. Conversely, if the model underestimates LAI, then SLATB is increased.

[0047] d. After all members have completed their updates, the set mean is taken as the optimal estimate. At this point, the model not only corrects the current state (such as LV, TAGP), but more importantly, it obtains a set of SLATB parameter tables that are calibrated with the current growth data and are more consistent with the characteristics of the crops in this field.

[0048] Step S4: Restart the WOFOST model based on the corrected model parameters and obtain the crop growth prediction results.

[0049] Using assimilated and corrected SLATB parameters and the latest status, the model predicts the peak LAI and biomass growth for the next 10 days. Combined with weather forecasts, the model's predicted LAI dynamics are used to calculate a more accurate daily crop water requirement. The system identifies areas of rapid LAI growth and high water demand, generating high-resolution variable irrigation maps to guide sprinkler irrigation systems in increasing irrigation volume in these areas during the next cycle.

[0050] Example 3 This embodiment provides an online calibration system for crop models based on image feedback from a large sprinkler irrigation machine, such as... Figure 2 As shown, it includes: a data acquisition module, which is a visible light camera mounted on the truss of a large sprinkler irrigation machine, used to acquire color images of the crop canopy and obtain the crop canopy coverage; an acquisition module, used to obtain leaf area index observations based on the crop canopy coverage; a correction module, used to perform online assimilation and correction of the WOFOST model using an ensemble Kalman filter algorithm based on the leaf area index observations, and obtain the corrected model parameters; and a prediction module, used to restart the WOFOST model based on the corrected model parameters and obtain crop growth prediction results.

[0051] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for online calibration of crop models based on image feedback from large sprinkler irrigation machines, characterized in that, Includes the following steps: S1. Obtain crop canopy coverage based on the color images of the crop canopy captured by the visible light camera mounted on the truss of the large sprinkler irrigation machine; S2. Obtain leaf area index observations based on crop canopy coverage; S3. Based on the observed leaf area index, the ensemble Kalman filter algorithm is used to assimilate and correct the WOFOST model online, and the corrected model parameters are obtained. S4. Restart the WOFOST model based on the corrected model parameters and obtain the crop growth prediction results.

2. The online calibration method for crop models based on image feedback from large sprinkler irrigation machines according to claim 1, characterized in that, S1 includes: The crop canopy color image is binarized based on the supergreen index to obtain green vegetation pixels; The crop canopy coverage is obtained by the ratio of green vegetation pixels to the total number of pixels in the image.

3. The online calibration method for crop models based on image feedback from a large sprinkler irrigation machine according to claim 1, characterized in that, S2 includes: Based on a pre-established empirical model of leaf area index and crop canopy coverage, the estimated value of leaf area index is obtained. Based on the leaf area index estimates from several images, the average leaf area index of the field was obtained.

4. The online calibration method for crop models based on image feedback from a large sprinkler irrigation machine according to claim 1, characterized in that, Based on the observed leaf area index, S3 includes: The model set is obtained by applying random perturbations to the initial state and specific leaf area parameter table of the WOFOST model; Run the model ensemble to the current observation time point and obtain the simulated leaf area index for each ensemble member; The observation operator is obtained based on the ratio of effective leaf area to land area; The Kalman gain is calculated based on the observed leaf area index, the simulated leaf area index for each set member, and the observed operator. The state vector and parameter vector of the model set are updated based on the Kalman gain to obtain the updated set; Based on the mean of the updated set, obtain the corrected specific leaf area parameter table.

5. The online calibration method for crop models based on image feedback from a large sprinkler irrigation machine according to claim 4, characterized in that, The state vector includes: developmental stage, total aboveground dry weight, effective leaf area, stem dry weight, storage organ dry weight, root dry weight, and values ​​for each developmental stage in the specific leaf area parameter table.

6. The online calibration method for crop models based on image feedback from a large sprinkler irrigation machine according to claim 4, characterized in that, The state vector and parameter vector of the model set are updated based on the Kalman gain, including: The update amounts of state and parameters are obtained based on the difference between the observed and simulated leaf area index, as well as the Kalman gain. The state vector and parameter vector of each member in the model set are updated synchronously according to the update amount.

7. The online calibration method for crop models based on image feedback from a large sprinkler irrigation machine according to claim 1, characterized in that, S4 includes: Restart the WOFOST model based on the revised leaf area parameter table and the latest model status to obtain future leaf area index dynamics and biomass accumulation prediction results. Based on the future dynamics of leaf area index and the prediction results of biomass accumulation, crop water requirement forecasts are obtained; Based on crop water requirement predictions, a variable irrigation prescription map is generated.

8. An online calibration system for crop models based on image feedback from a large sprinkler irrigation machine, the system being used to implement the method described in any one of claims 1-7, characterized in that, include: The acquisition module is a visible light camera mounted on the truss of a large sprinkler irrigation machine, used to acquire color images of the crop canopy and obtain the crop canopy coverage. The acquisition module is used to obtain leaf area index observations based on crop canopy coverage. The correction module is used to perform online assimilation and correction of the WOFOST model based on the leaf area index observations and the ensemble Kalman filter algorithm to obtain the corrected model parameters. The prediction module is used to restart the WOFOST model based on the corrected model parameters and obtain crop growth prediction results.

Images