Deep learning-based irrigation area rice yield estimation method and system

Abstract

The invention provides an irrigation area rice yield estimation method and system based on deep learning, and the method comprises the steps: collecting a satellite remote sensing image, an unmanned plane multispectral remote sensing image and ground meteorological station data of a target irrigation area, and generating a crop information image data set; performing remote sensing semantic segmentation on the crop information image data set and processing based on a deep learning network model, and extracting rice physiological indexes; an AquaCrop-EnKF data assimilation method is adopted, and on the basis of the rice physiological indexes, an EnKF method is adopted to carry out parameter correction on an AquaCrop model; and performing regional scale yield estimation on the target irrigation area, and combining with a growth simulation result of the AquaCrop model to obtain the final yield of the rice in the target irrigation area range. The method overcomes the limitations of a traditional crop yield estimation method in the aspects of long data acquisition period, high labor cost and the like, and meanwhile avoids the defects that a traditional statistical model and an empirical formula are insufficient in precision and poor in environmental adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural remote sensing, and particularly relates to a method and system for estimating rice yield in irrigation areas based on deep learning. Background Art

[0002] Accurate prediction of crop yields is of great significance for regional agricultural management and food security assurance. Current crop growth simulations mostly rely on physical models (such as AquaCrop). However, the parameterization process of these models is complex, and they rely highly on the quality and timeliness of input data. Although the crop growth simulation method combined with remote sensing data has improved the possibility of regional estimation, it still faces challenges in practical applications due to the spatio-temporal resolution of remote sensing images and data uncertainty.

[0003] Due to its powerful feature extraction ability, deep learning has gradually been applied to the processing of agricultural remote sensing data. Its hybrid method combined with physical models can improve the simulation accuracy while reducing the dependence on parameterization. In addition, the Ensemble Kalman Filter (EnKF) algorithm has been widely used in updating crop growth states due to its dynamic estimation ability for high-dimensional nonlinear systems. However, the existing research has insufficient timeliness processing of phenological data, which limits the yield estimation accuracy of the model.

[0004] Therefore, it is necessary to propose a crop yield estimation method combining deep learning and AquaCrop-EnKF, which can significantly improve the accuracy of crop growth simulation and yield prediction at the irrigation area scale. Summary of the Invention

[0005] The present invention provides a method and system for estimating rice yield in irrigation areas based on deep learning to solve the defects existing in the prior art, realize the dynamic simulation of various crop growth states, support model migration between different crops, improve the generality of the method. This method can not only predict the growth process of regional crops in real time, but also improve the estimation accuracy of the final yield of crops in irrigation areas through model optimization.

[0006] In a first aspect, the present invention provides a method for estimating rice yield in irrigation areas based on deep learning, including: Collect satellite remote sensing images, unmanned aerial vehicle multispectral remote sensing images, and ground meteorological station data of the target irrigation area to generate a crop information image dataset; Perform remote sensing semantic segmentation and processing based on a deep learning network model on the crop information image dataset to extract rice physiological indicators; Adopt the AquaCrop-EnKF data assimilation method, and based on the rice physiological indicators, use the EnKF method to correct the parameters of the AquaCrop model; Estimate the yield at the regional scale for the target irrigation area, and combine the growth simulation results of the AquaCrop model to obtain the final rice yield within the target irrigation area.

[0007] According to a method for estimating rice yield in an irrigation area based on deep learning provided by the present invention, it further includes: Construct a visualization platform based on WebGIS to display the crop growth status and the final rice yield within the target irrigation area in real time.

[0008] According to a method for estimating rice yield in an irrigation area based on deep learning provided by the present invention, collect satellite remote sensing images, unmanned aerial vehicle (UAV) multi-spectral remote sensing images, and ground meteorological station data of the target irrigation area, and generate a crop information image dataset, including: Collect high-resolution remote sensing images at different time points during the crop growth season using a UAV. The high-resolution remote sensing images include crop canopy coverage, leaf area index, and above-ground biomass; Collect daily meteorological data using a ground meteorological station. The meteorological data includes air temperature, precipitation, and radiation; Preprocess and standardize the high-resolution remote sensing images and the meteorological data to generate the crop information image dataset.

[0009] According to a method for estimating rice yield in an irrigation area based on deep learning provided by the present invention, perform remote sensing semantic segmentation and processing based on a deep learning network model on the crop information image dataset to extract rice physiological indicators, including: Use the DeepLab v3 architecture and adopt ResNet-101 as the backbone to perform semantic segmentation on the crop information image dataset, and use the intersection over union (IOU) and accuracy (ACC) as evaluation indicators; Estimate the leaf area index (LAI) and above-ground biomass (AGB) in the rice physiological indicators using VPANet, and determine the spatio-temporal loss function , so as to improve the correlation between the observed and estimated values at time and reduce the angle between the linear equation and the 1:1 straight line

[0010]

[0011]

[0012]

[0013]

[0014]

[0015] Among them, and are the -th i observed value and estimated value at the observation time respectively, represents the degree of closeness between the slope of the linear equation and 1. The smaller its value, the smaller the included angle is. is and the truncated slope of the linear equation. represents and the coefficient of determination. , are loss scaling coefficients. is the time loss, and its function is to optimize the accuracy of the estimated mean value in time. is the spatial loss, and its optimization goal is to reduce the included angle at the moment and improve the correlation between the estimated value and the true value at this time. Estimate the BBCH, the plant phenological development stage coding system in rice physiological indicators, using the ResNet-50 network, and use the mean square error MSE as the loss function for the regression task.

[0016] According to a method for estimating rice yield in an irrigation area based on deep learning provided by the present invention, the AquaCrop-EnKF data assimilation method is adopted. Based on the rice physiological indicators, the EnKF method is used to correct the parameters of the AquaCrop model, including: Determine that the predicted output of the AquaCrop model for the ensemble sample at the moment is , and the calculation of the ensemble mean vector and the covariance matrix is as follows:

[0017]

[0018] Among them, the superscript f represents the prior estimate, and Ne is the number of ensemble samples; The calculation formula of the Kalman gain K is as follows:

[0019]

[0020]

[0021]

[0022] wherein, in the formula is the transformation matrix for transforming the state space to the observation space, is the observation error covariance matrix. Let the observation vector at time be the updated state the mean vector and the error covariance matrix is the identity matrix, and the superscript represents the posterior estimate; Determine the uncertainty parameters of the data assimilation coefficient based on the global sensitivity analysis of the Aquacrop model; Optimize the parameters of the AquaCrop model through the data assimilation algorithm, reduce the prediction error, and adjust the key parameters of crop growth in real time.

[0023] According to a method for estimating rice yield in an irrigation area based on deep learning provided by the present invention, the final rice yield within the target irrigation area is evaluated by the determination coefficient R2 and the root mean square error RMSE.

[0024] In a second aspect, the present invention further provides a system for estimating rice yield in an irrigation area based on deep learning, including: A collection module, configured to collect satellite remote sensing images, unmanned aerial vehicle multispectral remote sensing images, and ground meteorological station data of the target irrigation area, and generate a crop information image data set; An extraction module, configured to perform remote sensing semantic segmentation on the crop information image data set and process it based on a deep learning network model to extract rice physiological indicators; A calibration module, configured to use the AquaCrop-EnKF data assimilation method to calibrate the parameters of the AquaCrop model by the EnKF method based on the rice physiological indicators; An estimation module, configured to perform regional scale yield estimation on the target irrigation area, and combine the growth simulation results of the AquaCrop model to obtain the final rice yield within the target irrigation area.

[0025] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the method for estimating rice yield in an irrigation area based on deep learning as described in any one of the above when executing the program.

[0026] Fourthly, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for estimating the rice yield in an irrigation area based on deep learning as described in any one of the above.

[0027] Fifthly, the present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for estimating the rice yield in an irrigation area based on deep learning as described in any one of the above.

[0028] Compared with the prior art, the beneficial effects of the present application are as follows: By combining a deep learning model to extract high-dimensional spatial features and temporal changes in remote sensing images, the present invention makes up for the deficiencies of traditional physical models in the quality of input data, thereby improving the accuracy and efficiency of crop growth simulation. Through this combination, the model can not only dynamically respond to environmental changes, but also update growth parameters and prediction results in real time; The present invention uses the Ensemble Kalman Filter (EnKF) algorithm for data assimilation. Through the dynamic fusion of remote sensing observation data and model prediction data, the growth state and key parameters of crops are updated in real time, ensuring the accuracy and timeliness of crop growth simulation in the time dimension. Compared with traditional static models, the dynamic optimization ability of the present invention makes the yield estimation results more real-time and adaptable; The present invention combines data assimilation technology and uses the Ensemble Kalman Filter (EnKF) to dynamically update the crop model, which can track and adjust the changes in crop growth in real time. Compared with the static processing method of traditional models, this dynamic optimization greatly improves the timeliness and adaptability of the model. By continuously integrating the latest remote sensing observation data, the model can continuously optimize parameters, thereby reducing prediction errors and improving the ability to simulate the crop growth process. Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0030] Figure 1 is one of the flow schematic diagrams of the method for estimating the rice yield in an irrigation area based on deep learning provided by the present invention; Figure 2 is another flow schematic diagram of the method for estimating the rice yield in an irrigation area based on deep learning provided by the present invention; Figure 3It is a schematic diagram of the main physiological processes of the Aquacrop model simulating the SPAC system provided by the present invention; Figure 4 It is a schematic diagram of the network architecture of the VPANet, a deep learning crop parameter estimation model provided by the present invention; Figure 5 It is a comparison chart of the impact of assimilating different types of observations on rice yield estimation provided by the present invention; Figure 6 It is a schematic diagram of the strategy for inferring rice physiological indicators based on grid images provided by the present invention; Figure 7 It is a schematic diagram of the structure of the deep learning-based irrigation area rice yield estimation system provided by the present invention; Figure 8 It is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed implementation manners

[0031] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] Figure 1 It is one of the schematic diagrams of the process of the deep learning-based irrigation area rice yield estimation method provided by the embodiments of the present invention. As Figure 1 shown, it includes: Step 100: Collect satellite remote sensing images, unmanned aerial vehicle multispectral remote sensing images and ground meteorological station data of the target irrigation area to generate a crop information image dataset; Step 200: Perform remote sensing semantic segmentation and processing based on a deep learning network model on the crop information image dataset to extract rice physiological indicators; Step 300: Adopt the AquaCrop-EnKF data assimilation method, and based on the rice physiological indicators, use the EnKF method to correct the parameters of the AquaCrop model; Step 400: Perform regional scale yield estimation on the target irrigation area, and combine the growth simulation results of the AquaCrop model to obtain the final rice yield within the target irrigation area.

[0033] Specifically, as Figure 2 shown, the embodiments of the present invention adopt a crop growth and yield estimation method based on deep learning observations and AquaCrop-EnKF, aiming to provide high-precision real-time yield estimation predictions for crops in irrigation areas. The specific implementation includes: Step 1: Obtain multi-spectral remote sensing images, RGB images of the UAV, and ground crop growth parameters, and perform environmental modeling in combination with meteorological data; Step 2: Semantic segmentation of remote sensing images and estimation of rice physiological indicators based on CNN or Transformer models; Step 3: Adopt the AquaCrop-EnKF data assimilation method. Based on the rice physiological indicators estimated in Step 2, use the EnKF method to correct the parameters of the AquaCrop model; Step 4: Conduct regional-scale yield estimation. Combine the growth simulation results of the AquaCrop model to predict the final yield of rice within the irrigation area; Step 5: System implementation. Build a visualization platform based on WebGIS to display the crop growth status and yield estimation results in real time.

[0034] Compared with traditional methods, the present invention overcomes the limitations of traditional crop yield estimation methods in terms of long data acquisition cycle and high labor cost, etc., and at the same time avoids the disadvantages of insufficient accuracy of traditional statistical models and empirical formulas and poor environmental adaptability. In addition, the present invention can dynamically optimize crop growth simulation parameters, improve water and nutrient use efficiency, and achieve cross-regional and cross-species applicability through deep learning and data assimilation methods, further verifying the robustness, generalization ability and prediction reliability of the model.

[0035] In one embodiment, the multi-source data collection and preprocessing in Step 1 include: In this embodiment, the data collection comes from satellite remote sensing images, multi-spectral cameras carried by UAVs, and ground meteorological stations, as Figure 3 shown. The specific steps include: Use the UAV to collect high-resolution remote sensing images at different time points during the crop growth season. The image data contains information such as crop canopy coverage, leaf area index, above-ground biomass, etc.

[0036] Use the ground meteorological station to collect daily meteorological data, including temperature, precipitation, radiation, etc.

[0037] These data are processed and standardized for input into the deep learning model. First, preprocess the remote sensing images, remove background noise, retain the crop area, and generate an image dataset containing crop growth information.

[0038] In one embodiment, in Step 2, for semantic segmentation of the regional rice fields from the remote sensing images using the DeepLab v3 architecture with ResNet-101 as the backbone, the following formula can be used as an evaluation index:

[0039]

[0040] In the formula, TP (true positive) represents the intersection of predicted positive examples and true positive examples, FP (false positive) represents the intersection of predicted positive examples and true negative examples, FN (false negative) represents the intersection of predicted negative examples and true positive examples, and TN (true negative) represents the intersection of predicted negative examples and true negative examples.

[0041] Among the physiological indicators of rice, the leaf area index (LAI) and above-ground biomass are estimated using VPANet. As Figure 4 shown, a spatio-temporal loss function (ST-loss) is proposed as the loss function. In addition to optimizing the MSE, the optimization objectives are to improve the correlation between the observed and estimated values at time t and to reduce the angle θ between the linear equation and the 1:1 straight line. The proposed ST-loss is shown in the following formula:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047] Among them, and are the th observed value and estimated value at the observation time i respectively. characterizes and the degree of closeness of the slope of the linear equation to 1. The smaller its value, the smaller the angle . is and the truncated slope of the linear equation. represents and the coefficient of determination. , are loss scaling coefficients. is the time loss, and its function is to optimize the accuracy of the estimated mean in time. is the spatial loss, and it is used to reduce Included angle at a moment And taking improving the correlation between the estimated value and the true value at this time as the optimization goal.

[0048] BBCH uses the ResNet-50 network for estimation and uses the mean square error MSE as the loss function for the regression task, which can be expressed by the following formula:

[0049] Among them, is the total number of samples, is the true value of the th sample, is the predicted value of the th sample, is the square of the prediction error of the

[0050] In one embodiment, in step 3, based on the crop physiological parameters estimated in step 2, the EnKF method is used to correct the parameters of the AquaCrop model to improve the model simulation accuracy. Specifically, it includes the following 4 key steps: (1) Model prediction The predicted output of the model for the ensemble samples at time t is , and the calculation of the ensemble mean vector and the covariance matrix is as follows:

[0051]

[0052] Among them, the superscript f represents the prior estimate, and Ne is the number of ensemble samples; (2) State update The calculation formula of the Kalman gain K is as follows:

[0053]

[0054]

[0055]

[0056] Among them, in the formula is the transformation matrix from the state space to the observation space, is the observation error covariance matrix, and let the observation vector at time be , mean vector and error covariance matrix , is the identity matrix, and the superscript represents the posterior estimate; (3) Determine the uncertainty parameters of the data assimilation coefficient based on the global sensitivity analysis of the Aquacrop model.

[0057] (4) While updating the state, several model sensitive parameters are also updated. The data assimilation strategy adopted is shown in Table 1. When there is no phenological observation, GDD and Tbase are updated simultaneously. When there is phenological observation, Tbase is not updated.

[0058] Table 1

[0059] Note: Pc is the canopy growth related parameter, including CDC, CGC, CCx In one embodiment, in step 4, under the model state after data assimilation, the biomass and final yield of the crop are calculated. The system generates a dynamic change map of regional crop growth and a final yield prediction based on the estimated yield results, and provides them to farmers or agricultural management departments for decision-making support.

[0060] Under different crop varieties and regional environmental conditions, the model is adaptively adjusted through transfer learning. The specific steps include: Use the trained model to test different crops (such as rice, corn), and adjust the model parameters according to the test results, as Figure 5 shown.

[0061] According to the growth characteristics of different crop varieties, transfer the model parameters to make it adapt to different environmental conditions, and ensure the stability and robustness of the model among multiple regions and multiple crops, as Figure 6 shown.

[0062] Furthermore, the output yield estimation results are evaluated by the coefficient of determination R2 and the root mean square error RMSE. The calculation formulas are as follows:

[0063]

[0064] Among them, is the total number of samples, is the true value of the th sample, is the predicted value of the th sample, is the average value of the samples.

[0065] Verify the advantages of the method of the present invention in terms of accuracy and stability by comparing it with traditional yield estimation methods.

[0066] In one embodiment, in step 5, the embodiment of the present invention adopts a computing framework accelerated by GPU to improve the efficiency of large-scale data processing and real-time render high-resolution crop growth maps on the WebGIS platform.

[0067] The following describes the irrigation area rice yield estimation system based on deep learning provided by the present invention. The irrigation area rice yield estimation system based on deep learning described below can be correspondingly referred to the irrigation area rice yield estimation method based on deep learning described above.

[0068] Figure 7 is a schematic structural diagram of the irrigation area rice yield estimation system based on deep learning provided by the embodiment of the present invention, as Figure 7 shown, including: a collection module 71, an extraction module 72, a calibration module 73, and an estimation module 74, wherein: The collection module 71 is used to collect satellite remote sensing images, unmanned aerial vehicle multispectral remote sensing images, and ground meteorological station data of the target irrigation area to generate a crop information image data set; the extraction module 72 is used to perform remote sensing semantic segmentation and deep learning network model-based processing on the crop information image data set to extract rice physiological indicators; the calibration module 73 is used to adopt the AquaCrop-EnKF data assimilation method and use the EnKF method to calibrate the parameters of the AquaCrop model based on the rice physiological indicators; the estimation module 74 is used to perform regional-scale yield estimation on the target irrigation area and combine the growth simulation results of the AquaCrop model to obtain the final rice yield within the target irrigation area.

[0069] Figure 8 Illustrates a schematic structural diagram of an electronic device, as Figure 8As shown in the figure, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the method for estimating the rice yield in the irrigation area based on deep learning. The method includes: collecting satellite remote sensing images, unmanned aerial vehicle multispectral remote sensing images, and ground meteorological station data of the target irrigation area to generate a crop information image dataset; performing remote sensing semantic segmentation and processing based on a deep learning network model on the crop information image dataset to extract rice physiological indicators; using the AquaCrop-EnKF data assimilation method, based on the rice physiological indicators, using the EnKF method to correct the parameters of the AquaCrop model; performing regional scale yield estimation on the target irrigation area, and combining the growth simulation results of the AquaCrop model to obtain the final rice yield within the target irrigation area.

[0070] In addition, when the logical instructions in the above-mentioned memory 830 are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, and other various media that can store program codes.

[0071] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the deep learning-based irrigation area rice yield estimation method provided by each of the above methods. The method includes: collecting satellite remote sensing images, unmanned aerial vehicle multispectral remote sensing images and ground meteorological station data of the target irrigation area to generate a crop information image dataset; performing remote sensing semantic segmentation and processing based on a deep learning network model on the crop information image dataset to extract rice physiological indicators; using the AquaCrop-EnKF data assimilation method, and based on the rice physiological indicators, using the EnKF method to correct the parameters of the AquaCrop model; performing regional-scale yield estimation on the target irrigation area, and combining the growth simulation results of the AquaCrop model to obtain the final rice yield within the target irrigation area.

[0072] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the deep learning-based irrigation area rice yield estimation method provided by each of the above methods. The method includes: collecting satellite remote sensing images, unmanned aerial vehicle multispectral remote sensing images and ground meteorological station data of the target irrigation area to generate a crop information image dataset; performing remote sensing semantic segmentation and processing based on a deep learning network model on the crop information image dataset to extract rice physiological indicators; using the AquaCrop-EnKF data assimilation method, and based on the rice physiological indicators, using the EnKF method to correct the parameters of the AquaCrop model; performing regional-scale yield estimation on the target irrigation area, and combining the growth simulation results of the AquaCrop model to obtain the final rice yield within the target irrigation area.

[0073] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0074] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or equivalently replace some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for estimating rice yield in irrigation areas based on deep learning, characterized in that: include: Collect satellite remote sensing images, drone multispectral remote sensing images and ground meteorological station data of the target irrigation area to generate crop information image datasets; Performing remote sensing semantic segmentation on the crop information image dataset and processing based on a deep learning network model to extract rice physiological indicators; Using the AquaCrop-EnKF data assimilation method, based on the rice physiological indicators, the EnKF method is used to calibrate the parameters of the AquaCrop model; The yield of the target irrigation area was estimated at a regional scale, and combined with the growth simulation results of the AquaCrop model, the final rice yield within the target irrigation area was obtained.

2. The method for estimating rice yield in irrigation areas based on deep learning according to claim 1, characterized in that: Also includes: A visualization platform was built based on WebGIS to display the crop growth status and the final rice yield within the target irrigation area in real time.

3. The method for estimating rice yield in irrigation areas based on deep learning according to claim 1, characterized in that: Collect satellite remote sensing images, drone multispectral remote sensing images and ground weather station data of the target irrigation area to generate crop information image datasets, including: Using drones to collect high-resolution remote sensing images at different time points during the crop growing season, the high-resolution remote sensing images include crop canopy coverage, leaf area index, and aboveground biomass; Using ground weather stations to collect daily meteorological data, the meteorological data includes temperature, precipitation and radiation; The high-resolution remote sensing image and the meteorological data are preprocessed and standardized to generate the crop information image data set.

4. The method for estimating rice yield in irrigation areas based on deep learning according to claim 1, characterized in that: The crop information image dataset is subjected to remote sensing semantic segmentation and deep learning network model processing to extract rice physiological indicators, including: Using the DeepLab v3 architecture, ResNet-101 is used as the skeleton to perform semantic segmentation on the crop information image dataset, and the intersection-over-union (IOU) and accuracy (ACC) are used as evaluation indicators; VPANet was used to estimate the leaf area index LAI and aboveground biomass AGB in rice physiological indicators, and the spatiotemporal loss function was determined. , to improve The correlation between the observed and estimated values ​​at the time, reducing the angle between the linear equation and the 1:1 straight line To optimize the goal: in, and The observation time The next i observations and estimates, Characterization and The closer the slope of the linear equation is to 1, the smaller the value is, indicating that the angle The smaller, for and The intercept slope of the linear equation, represent and The coefficient of determination, , is the loss scaling factor, is the time loss, which is used to optimize the accuracy of the estimated mean in time. is the space loss, which is reduced Angle at moment And improving the correlation between the estimated value and the true value at this time is the optimization goal; The plant phenological development stage coding system BBCH in rice physiological indicators was estimated using the ResNet-50 network, and the mean square error MSE was used as the loss function of the regression task.

5. The method for estimating rice yield in irrigation areas based on deep learning according to claim 1, characterized in that: The AquaCrop-EnKF data assimilation method was used, and based on the rice physiological indicators, the EnKF method was used to calibrate the parameters of the AquaCrop model, including: Determine the AquaCrop model in The predicted output for the set sample at time is , the set mean vector and the covariance matrix The calculation of is as follows: Among them, the superscript f represents the prior estimate, Ne is the number of samples in the set; The calculation formula of Kalman gain K is as follows: Among them, is the transformation matrix from state space to observation space, is the observation error covariance matrix, let The observation vector at time , the updated status , mean vector and the error covariance matrix , is the identity matrix, the superscript represents the posterior estimate; The uncertainty parameters of the data assimilation coefficients were determined based on a global sensitivity analysis of the Aquacrop model; Through data assimilation algorithms, the parameters of the AquaCrop model are optimized, prediction errors are reduced, and key parameters of crop growth are adjusted in real time.

6. The method for estimating rice yield in irrigation areas based on deep learning according to claim 1, characterized in that: The final rice yield within the target irrigation area is evaluated using the determination coefficient R2 and the root mean square error RMSE as evaluation indicators.

7. A rice yield estimation system for irrigation areas based on deep learning, characterized in that: include: The acquisition module is used to collect satellite remote sensing images of the target irrigation area, multispectral remote sensing images of unmanned aerial vehicles, and ground meteorological station data to generate crop information image datasets; An extraction module, used to perform remote sensing semantic segmentation on the crop information image dataset and process it based on a deep learning network model to extract rice physiological indicators; A correction module, for using the AquaCrop-EnKF data assimilation method to perform parameter correction on the AquaCrop model using the EnKF method based on the rice physiological indexes; The estimation module is used to estimate the yield of the target irrigation area at the regional scale and obtain the final rice yield within the target irrigation area by combining the growth simulation results of the AquaCrop model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the irrigation area rice yield estimation method based on deep learning as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for estimating rice yield in an irrigation area based on deep learning as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the method for estimating rice yield in irrigation areas based on deep learning as described in any one of claims 1 to 6.

Images