Crop height inversion method and device based on compact polarimetric SAR (Synthetic Aperture Radar) data

Abstract

The invention discloses a crop height inversion method and device based on compact polarimetric SAR data, and relates to the technical field of polarimetric radar remote sensing quantitative inversion, and the crop height inversion method based on compact polarimetric SAR data mainly comprises the steps: carrying out the data preprocessing of an original polarimetric SAR image, and obtaining compact polarimetric data; a multi-dimensional feature set is obtained according to polarization parameters, a machine learning algorithm model is trained and verified by combining a field actually-measured crop height data set to obtain a height prediction model, and crop height estimation data is obtained accordingly; and feature subset optimization is carried out by using a forward feature selection method, the machine learning algorithm model is trained and verified again based on the optimized feature subset to obtain a height prediction model, and the optimized feature subset is predicted to obtain a higher-precision crop height inversion result. According to the crop height inversion method and device based on the compact polarimetric SAR data, the inversion efficiency and precision can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of polarimetric radar remote sensing quantitative inversion technology, and more specifically, to a method and device for inverting crop height based on simplified polarimetric SAR data. Background Technology

[0002] Crop height is a key indicator characterizing crop growth dynamics and is of significant value in precision agriculture management, disaster monitoring, and yield prediction. Traditional manual measurement is time-consuming and labor-intensive, and has limited spatiotemporal resolution. Synthetic Aperture Radar (SAR), with its all-weather, all-day observation capabilities and penetration through vegetation, has become an important tool for modern agricultural monitoring. Currently, SAR-based crop height retrieval mainly falls into two categories: Polarimetric Interferometric SAR (PolInSAR) and Polarimetric SAR (PolSAR). PolInSAR technology is limited by factors such as low crop height and rapid growth, and conventional interferometry struggles to meet the spatial baseline requirements for height measurement and overcome temporal decorrelation. In contrast, PolSAR technology, utilizing its sensitivity to the dielectric properties and geometry of vegetation, is not constrained by spatiotemporal baselines, providing a more feasible approach for high-frequency dynamic monitoring of crop height at the regional scale.

[0003] Regarding inversion models, while physical models have broad applicability, they are computationally complex and prone to ill-posed problems; semi-empirical models rely on numerous parameters and are difficult to directly correlate with crop growth parameters. Data-driven machine learning methods, however, utilize prior knowledge to establish empirical relationships, circumventing complex physical modeling. They can effectively capture nonlinear relationships, handle multi-source features, and possess adaptive capabilities, making them more suitable for large-scale monitoring. It is worth noting that the choice of data source significantly impacts inversion efficiency. While fully polarized systems provide complete information, they are limited by narrow swath width and large data volume. In contrast, reduced polarization, as a special dual-channel polarization mode, such as the configuration used in the RADARSAT constellation mission and the RISAT-1 satellite, retains richer polarization information than dual polarization while overcoming the limitations of fully polarized systems in swath width and revisit period, resulting in higher data acquisition efficiency and lower costs.

[0004] However, current research on crop height inversion using simplified polarimetric data combined with advanced machine learning algorithms and feature optimization still needs improvement. How to select effective parameters sensitive to plant height from high-dimensional CP features and construct a high-precision inversion model is a key problem that urgently needs to be solved in current polarimetric SAR agricultural applications. Summary of the Invention

[0005] The purpose of this invention is to provide a method and equipment for crop height inversion based on simplified polarimetric SAR data, which can effectively improve the efficiency and accuracy of the inversion.

[0006] This invention provides a method for crop height inversion based on reduced polarimetric SAR data, comprising the following steps: S1: Acquire raw polarimetric SAR images during the crop growth cycle, perform data preprocessing on the raw polarimetric SAR images, and obtain simulated simplified polarimetric data. S2: Extract the polarization parameters from the simplified polarization data, and obtain a multidimensional feature set based on the polarization parameters; S3: Obtain a training set and a validation set based on the field measured crop height dataset; use the training set and validation set to train and validate the machine learning algorithm model to obtain a height prediction model; use the height prediction model to predict the multidimensional feature set to obtain crop height estimation data; S4: Based on the multidimensional feature set, optimize using the forward feature selection method to obtain the optimal feature subset; retrain and validate the machine learning algorithm model based on the optimal feature subset to obtain the height prediction model; use the height prediction model to predict the optimal feature subset to obtain a more accurate crop height inversion result.

[0007] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described crop height inversion method based on reduced polarimetric SAR data.

[0008] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described crop height inversion method based on reduced polarimetric SAR data.

[0009] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for crop height inversion based on reduced polarimetric SAR data.

[0010] The method and equipment for crop height inversion based on simplified polarimetric SAR data provided by this invention have the following beneficial effects: This invention first acquires raw polarimetric SAR images during the crop growth cycle, preprocesses them, and simulates to generate simplified polarimetric data. Different types of polarimetric parameters are extracted to fully utilize the data information, obtaining a multi-dimensional feature set. Four machine learning algorithms—random forest regression, bagged decision tree, extreme gradient, and Gaussian process regression—are used to construct a crop height inversion model, obtaining crop height estimation data. A forward feature selection algorithm is used to optimize the feature subset to reduce redundancy, and the optimized subset is input to obtain higher-precision plant height inversion results. This invention, based on a crop height inversion framework combining polarimetric SAR data and machine learning, fully utilizes simplified polarimetric data information, avoids complex physical modeling and mathematical derivation, and effectively improves the efficiency and accuracy of the inversion. Attached Figure Description

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of the crop height inversion method based on simplified polarimetric SAR data provided by the present invention; Figure 2 This is a schematic diagram illustrating the operation steps of the crop height inversion method based on simplified polarimetric SAR data provided by the present invention.

[0012] Figure 3 This is a location map of the research area provided by the present invention.

[0013] Figure 4 This is a scatter plot of the optimal results of the optimal model inversion provided by this invention.

[0014] Figure 5 This is a schematic diagram of the feature optimization process provided by the present invention.

[0015] Figure 6 This is a structural block diagram of the computer device provided by the present invention. Detailed Implementation

[0016] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] Figure 1 A schematic diagram of the crop height inversion method based on reduced polarimetric SAR data in this embodiment is shown. In this embodiment, the crop height inversion method based on reduced polarimetric SAR data includes the following steps: S1: Acquire raw polarimetric SAR images during the crop growth cycle, perform data preprocessing on the raw polarimetric SAR images, and obtain simulated simplified polarimetric data.

[0018] In one exemplary embodiment, the data preprocessing of the original polarimetric SAR image includes: performing single-view complex transformation, coherence matrix extraction, polarimetric filtering, geocoding, and data cropping on the original polarimetric SAR image, and simulating it using a right-hand circular hybrid mode.

[0019] S2: Extract the polarization parameters from the simplified polarization data, and obtain a multidimensional feature set based on the polarization parameters.

[0020] In one exemplary embodiment, the polarization parameters include basic backscattering coefficient characteristics, Stokes vector characteristics, polarization target decomposition characteristics, polarization state and statistical characteristics, inter-channel relationships, and vegetation index characteristics.

[0021] In one exemplary embodiment, the basic backscattering coefficient characteristics include the backscattering coefficients of horizontally polarized reception under right-hand circular polarization emission, the backscattering coefficients of vertically polarized reception, the backscattering coefficients of right-hand circular polarization reception, and the backscattering coefficients of left-hand circular polarization reception.

[0022] In one exemplary embodiment, the Stokes vector characteristics include the total power intensity of the scattered wave, the power difference between the horizontal and vertical linear polarization components, the power difference between the 45-degree and 135-degree linear polarization components, and the power difference between the left-hand and right-hand circular polarization components.

[0023] In one exemplary embodiment, the polarization target decomposition features include those based on reduced polarization. Decompose parameters Decompose parameters Decomposition parameters, model-free three-component decomposition parameters, and random body scattering and surface scattering decomposition parameters.

[0024] In one exemplary embodiment, the reduced polarization-based The decomposition parameters include polarization entropy, average backscattering angle, and antipolarizability; The The decomposition parameters are based on polarizability. and ellipticity angle The power values ​​corresponding to the surface scattering component, dihedral scattering component, and volume scattering component extracted by decomposition; The The decomposition parameters are based on polarizability. and relative phase The power values ​​corresponding to the surface scattering component, dihedral scattering component, and volume scattering component extracted by decomposition; The model-free three-component decomposition parameters are the corresponding pseudo-surface scattering, pseudo-dihedral scattering, and pseudo-volume scattering power components extracted without relying on a specific physical model. The decomposition parameters of random body scattering and surface scattering are the extinction coefficient and the surface scattering amplitude ratio obtained by inversion based on the RVOG model.

[0025] In one exemplary embodiment, step S2 further includes: normalizing the multidimensional feature set to obtain a normalized multidimensional feature set.

[0026] S3: Obtain a training set and a validation set based on the field measured crop height dataset; use the training set and validation set to train and validate the machine learning algorithm model to obtain a height prediction model; use the height prediction model to predict the multidimensional feature set to obtain crop height estimation data.

[0027] In one exemplary embodiment, the machine learning algorithm model includes a random forest regression algorithm model, a bagged decision tree algorithm model, an extreme gradient boosting algorithm model, and a Gaussian process regression algorithm model.

[0028] In an exemplary embodiment, the process of obtaining the training set and validation set based on the field measured crop height dataset includes: using a stratified random sampling method, dividing the height distribution in the field measured crop height dataset into six smaller subgroups, randomly assigning each subgroup, and dividing them according to a preset ratio to obtain the training set and validation set.

[0029] In one exemplary embodiment, the preset ratio is: .

[0030] In one exemplary embodiment, the process of obtaining the training set and validation set based on the field measured crop height dataset further includes: performing 10 independent samplings on the field measured crop height dataset to obtain different training sets and validation sets.

[0031] In one exemplary embodiment, the verification process of training and validating the machine learning algorithm model using the training set and validation set involves using root mean square error and correlation coefficient as evaluation metrics to assess the prediction accuracy of the highly predictive model.

[0032] S4: Based on the multidimensional feature set, optimize using the forward feature selection method to obtain the optimal feature subset; retrain and validate the machine learning algorithm model based on the optimal feature subset to obtain the height prediction model; use the height prediction model to predict the optimal feature subset to obtain a more accurate crop height inversion result.

[0033] In one exemplary embodiment, the specific process of optimizing using the forward feature selection method is as follows: Candidate feature sets are obtained based on the multidimensional feature set. Starting from an empty feature set, candidate feature sets are added to the current feature set in sequence. Specifically, this includes: obtaining a training set and a validation set based on the current feature set and a field measured crop height dataset; training a machine learning algorithm model based on the training set; calculating the root mean square error of the test set; and selecting the candidate feature set that minimizes the root mean square error to add to the current feature set. The current feature set is used as the optimal feature subset until the number of features in the current feature set reaches a preset value or the root mean square error converges.

[0034] In one exemplary embodiment, the predicted value is 18.

[0035] In some embodiments, the above-described crop height inversion method based on reduced polarimetric SAR data can also be implemented in the following ways.

[0036] In this embodiment, the crop height inversion method based on simplified polarimetric SAR data includes the following steps: Step 1: Acquire raw polarimetric SAR images during the crop growth cycle, perform data preprocessing on the raw images, and obtain simulated simplified polarimetric data. In one exemplary embodiment, in step 1, a study area is selected, and raw polarimetric SAR images covering the crop growth cycle of the study area in the time dimension are obtained. The raw data is subjected to single-view complex transformation, coherence matrix extraction, polarimetric filtering, geocoding and data cropping, and simplified polarimetric data is obtained by simulating the right-hand circular hybrid mode.

[0037] Step 2: Extract different types of polarization parameters (including backscattering coefficient, polarization decomposition parameters, and radar vegetation index, etc.) to fully utilize the data information and obtain a multi-dimensional feature set; In one exemplary embodiment, step 2 incorporates backscattering coefficients, model decomposition parameters, polarization entropy / scattering angle, and radar vegetation index under different reduced polarization modes into a feature subset to systematically evaluate the contribution of multidimensional polarization information to crop height retrieval performance. Specifically, this includes 37 polarization feature parameters: backscattering coefficients under right-hand circular polarization transmission (horizontal, vertical, left-hand, or right-hand circular polarization reception); Stokes vector components (SV0, SV1, SV2, SV3); and parameters based on reduced polarization. Decomposition parameters The surface scattering, dihedral scattering, and volume scattering components obtained from the decomposition; The surface scattering, dihedral scattering, and volume scattering components obtained from the decomposition; Shannon entropy decomposition parameters (SE_Int, SE_Pol); polarizability (DoP), depolarizability (DoD), circular polarizability (DoC), and ellipticity (DoE); uniformity coefficient ( and related to the ellipticity of the reduced scattered wave Parameters; phase difference and correlation coefficient between RH and RV channels; linear polarization ratio and circular polarization ratio; reduced polarization radar vegetation index (CpRVI); model-free three-component decomposition parameters ( , , , ); Random scattering and surface scattering (RVOG) decomposition parameters.

[0038] In one exemplary embodiment, in step 2, considering the wide range of variation of the feature values ​​and the non-Gaussian distribution characteristics, it is proposed to normalize the feature set to ensure that all values ​​remain within the limit range of 0-1.

[0039] Step 3: Use four machine learning algorithms—random forest regression, bagged decision tree, extreme gradient, and Gaussian process regression—to construct a crop height inversion model and obtain crop height estimation data; In one exemplary embodiment, step 3 employs four algorithms—random forest, gradient boosting tree, extreme gradient regression, and Gaussian process regression—to predict the height of corn, wheat, and soybean. A stratified random sampling method is used for each crop, dividing the height distribution based on a five-year measured dataset into six smaller subgroups to ensure broad distribution. Each subgroup is randomly assigned, with 80% used as the training set and 20% for accuracy validation. To enhance the model's generalization and robustness, ten independent samplings are performed to create different datasets, used for training and validation of the four machine learning models across the three crop scenarios.

[0040] In one exemplary embodiment, in step 3, the root mean square error and correlation coefficient are used as evaluation indicators to assess the model's prediction accuracy.

[0041] Step 4: Optimize the feature subset using the forward feature selection algorithm to reduce redundancy, and input the optimized subset to obtain a more accurate plant height inversion result.

[0042] In an exemplary embodiment, the specific process of optimizing using the forward feature selection algorithm in step 4 is as follows: starting from an empty feature set, candidate features are added to the current feature set sequentially; the model is trained based on the training set, and the root mean square error of the test set is calculated; the feature that minimizes the root mean square error is selected and added to the feature set; the above steps are repeated until the number of features reaches a preset value or the root mean square error no longer decreases significantly, thereby determining the optimal feature subset.

[0043] In one exemplary embodiment, in step 4, the best feature subset is used as the model input to train and validate four machine learning models again, resulting in a more accurate high-inversion result.

[0044] In some embodiments, the above-described crop height inversion method based on reduced polarimetric SAR data can also be implemented in the following ways.

[0045] Please refer to Figure 2 To avoid complex modeling processes and mathematical derivations, this invention fully utilizes simplified polarimetric data information and provides a method for crop height inversion based on simplified polarimetric SAR data and machine learning. This embodiment selects two adjacent agricultural areas in a certain region as the study area. Figure 3 The location of the study area is shown. Thirty-seven C-band RADARSAT scenes were selected throughout the growth period of maize, wheat, and soybeans. 2. Fully Polarimetric SAR Imagery. Ground-based data acquisition was conducted in close coordination with the acquisition dates of the aforementioned images, with a maximum time difference of no more than 3 days, to verify the inversion results. The specific implementation steps are as follows: Step 1: Polarimetric SAR Image Preprocessing The study area was selected, and full polarimetric synthetic aperture radar images covering the crop growth cycle of the study area in the time dimension were acquired. The raw data were preprocessed by single-view complex transformation, coherence matrix extraction, polarimetric filtering, geocoding, data cropping, etc., and simplified polarimetric data were obtained by simulating the right-hand circular hybrid mode.

[0046] Step 2: Extract polarization parameters for different types of polarization To fully exploit the rich ground object scattering information contained in abbreviated polarimetric SAR data, a high-dimensional feature space was constructed. The aim is to improve the accuracy and robustness of crop height retrieval in complex farmland scenarios by systematically evaluating multi-dimensional polarimetric information. In this step, a total of 37 polarimetric feature parameters were extracted from the preprocessed abbreviated polarimetric data, forming a feature subset for subsequent model training. These feature parameters cover multiple dimensions, including backscattering intensity, polarimetric state descriptor, target decomposition components, and radar vegetation index. The specific composition and extraction methods of these 37 polarimetric feature parameters are detailed below: 1. Basic Backscattering Coefficient Characteristics. This embodiment employs a right-hand circular polarization emission mode. Under this emission mode, the backscattering coefficients (in dB) for different receiving polarizations are extracted, directly reflecting the scattering intensity of the target to different polarized waves: RH: Backscattering coefficient for right-hand circularly polarized emission and horizontally polarized reception; RV: Backscattering coefficient for right-hand circularly polarized emission and vertically polarized reception; RR: Backscattering coefficient of right-hand circularly polarized emission and right-hand circularly polarized reception; RL: Backscattering coefficient for right-hand circularly polarized emission and left-hand circularly polarized reception.

[0047] 2. Stokes Vector Characteristics. To fully describe the polarization state of the scattered wave, the four components of the Stokes vector were calculated based on reduced polarization data. The Stokes vector comprehensively characterizes the intensity, polarizability, shape, and orientation of the polarization ellipse of the electromagnetic wave: SV0: Represents the total power intensity of the scattered wave; SV1: Represents the power difference between the horizontal and vertical linear polarization components; SV2: Represents the power difference between the 45-degree linear polarization component and the 135-degree linear polarization component; SV3: Represents the power difference between the left-hand circular polarization component and the right-hand circular polarization component.

[0048] 3. Polarization Target Decomposition Characteristics. To separate the different scattering mechanisms of crop canopy, stems, and underlying soil, this invention introduces several advanced simplified polarization target decomposition methods, extracting the following key parameters: Based on simplified polarization Decomposition parameters include polarization entropy, which describes the randomness of the scattering process; average backscattering angle, which identifies the main scattering mechanisms (such as surface, dipole, or dihedral scattering); and antipolarizability, which describes the relative importance of secondary scattering mechanisms.

[0049] Decomposition parameters: using polarizability and ellipticity angle The decomposition process extracts the power values ​​corresponding to the surface scattering component, dihedral scattering component, and volume scattering component.

[0050] Decomposition parameters: using polarizability and relative phase The decomposition process also extracts the power values ​​corresponding to the surface scattering component, dihedral scattering component, and volume scattering component.

[0051] Model-free three-component decomposition parameters: without relying on the assumptions of a specific physical model, extract the dynamic degree parameters, as well as the corresponding pseudo-surface scattering, pseudo-dihedral scattering, and pseudo-volume scattering power components.

[0052] Random volume scattering and surface scattering decomposition parameters: These are parameters obtained from the RVOG model inversion and are used to remove the influence of vegetation layer volume scattering and surface scattering. They typically include model-related parameters such as extinction coefficient and surface scattering amplitude ratio.

[0053] 4. Polarization State and Statistical Characteristics. Parameters used to describe the polarization purity, disorder, and geometry of the scattered wave: Polarimetric parameters include total polarimetric, depolarization, circular polarimetric, and ellipticity. These parameters measure the proportion and shape of the fully polarized portion of the scattered wave.

[0054] Shannon entropy decomposition parameters: Based on information theory, Shannon entropy decomposition includes intensity entropy, which reflects the uncertainty of total power; and polarization entropy, which reflects the uncertainty of polarization state.

[0055] Geometric parameters: consistency coefficients, used to describe the degree of consistency between the scattered wave and a specific reference polarization state.

[0056] 5. Inter-channel relationships and vegetation index characteristics. Features constructed using the phase and amplitude relationships between different polarization channels: Inter-channel parameters: The phase difference between the RH channel and the RV channel, as well as the magnitude of the complex correlation coefficient, reflect the coherence between different polarization channels.

[0057] Polarization ratio parameters: including linear polarization ratio and circular polarization ratio, used to enhance sensitivity to specific scattering mechanisms.

[0058] Radar Vegetation Index: Reduced Polarimetric Radar Vegetation Index is a comprehensive index specifically designed for reduced polarimetric data and sensitive to vegetation biomass and structure.

[0059] Considering the wide variation range of eigenvalues ​​and their non-Gaussian distribution, the feature set is to be normalized to ensure that all values ​​remain within the 0-1 range. This embodiment of the invention successfully extracted 37 feature parameters covering multiple dimensions such as scattering intensity, phase information, polarization state, and scattering mechanism, laying a solid data foundation for the subsequent establishment of a high-precision crop height inversion model.

[0060] Step 3: Constructing an inversion model using four machine learning algorithms This embodiment selects corn, wheat, and soybean as typical crop research objects, and constructs a height prediction model for each crop. In order to deal with the complex nonlinear relationship between SAR data and crop parameters, this invention uses four advanced machine learning algorithms as regressors: Random Forest Regression (RFR), Bagged Decision Tree (BAGTREE), Extreme Gradient Boosting (XGB), and Gaussian Process Regression (GPR).

[0061] By leveraging the powerful learning capabilities of these algorithms, a mapping relationship between the input SAR feature set and the measured crop height is established.

[0062] To ensure the sufficiency of model training and the reliability of validation results, this embodiment employs a rigorous hierarchical random sampling strategy. Specifically, based on a five-year field-measured crop height dataset, the data is divided into six smaller subgroups according to the height distribution characteristics throughout the crop growth cycle. This division ensures that the samples have broad and balanced representativeness across all stages of the crop's growing season, avoiding model bias caused by too many or too few samples at a particular growth stage. Within each subgroup, random allocation is performed, with 80% of the sample data designated as the training set for model learning and parameter optimization; the remaining 20% ​​is designated as the validation set for evaluating the model's generalization ability. To further enhance the robustness of the evaluation results and reduce the random errors introduced by a single random sampling, this embodiment implements the above sampling process 10 times independently. This means that for the four machine learning models in three crop scenarios, 10 distinct combinations of training and validation datasets were created for repeated experiments.

[0063] In this embodiment, the number of random forest decision trees is set to 200, with all other parameters remaining at their default settings. Similarly, the number of bagged decision trees is set to 200, with all other parameters remaining at their default settings. Furthermore, for the extreme gradient boosting model, the number of decision trees is specified as 600, the maximum depth as 2, the subsampling rate as 0.2, and the learning rate as 0.01, with all other parameters remaining at their default settings. For the Gaussian regression model, three key parameters are calculated: the RBF kernel width, the scaling factor used to determine the kernel signal, and the noise standard deviation.

[0064] During the model evaluation phase, root mean square error and correlation coefficient are used as quantitative evaluation indicators to measure the difference and correlation between the model predictions and the measured values.

[0065] The inversion results were compared with the actual crop heights measured at sampling points in the experimental area. Figure 4 Scatter plots of the inversion results using the random forest model for the three crops are shown in Table 1. Table 1 illustrates that the random forest model achieved the best inversion accuracy for all three crops. The optimal mean root mean square error (RMSE) for maize reached 58.76 cm, with a correlation coefficient of 0.84. For wheat, the optimal RMSE reached 17.32 cm, with a correlation coefficient of 0.64. For soybean, the optimal RMSE reached 25.10 cm, with a correlation coefficient of 0.58. Overall, wheat showed the best inversion performance, and all models achieved highly efficient and accurate vegetation inversion.

[0066] Table 1: Comparison of Inversion Results

[0067] Step 4: Optimize the feature subset using the forward feature selection algorithm. In the high-resolution inversion process, not all input features contribute positively to the inversion accuracy. Instead, the feature data may contain redundant information or noise, thus reducing model performance. To address this technical problem, this embodiment employs a forward feature selection method to process the input data. Forward feature selection is an encapsulated feature selection method that, compared to traditional filtering-based feature selection methods, fully considers the interaction effects between features. By eliminating irrelevant features and noise, this method effectively improves computational efficiency, reduces the model's generalization error, and ultimately enhances the final high-resolution inversion accuracy.

[0068] First, initialize the feature set, setting it to an empty set initially, and establish a candidate feature set. In each iteration, select a feature from the candidate feature set that is not yet included in the current feature set and add it to the current feature set. Retrain the prediction model based on the updated feature set.

[0069] The root mean square error (RMSE) between the measured and predicted altitude data is calculated to evaluate the model's performance. In a single iteration, the feature that minimizes the RMSE is selected from all candidate combinations and formally added to the feature set.

[0070] The above steps are repeated until a predetermined stopping criterion is met. This stopping criterion may include, but is not limited to, reaching a predetermined number of selected features, or the root mean square error no longer significantly decreasing. Finally, the algorithm converges and outputs the optimal feature set with the lowest root mean square error among all feature interaction combinations.

[0071] Figure 5 The process of feature selection using a random forest model for the three crops shows that 18 features were selected in the corn scenario, 15 features were selected in the wheat scenario, and 13 features were selected in the soybean scenario. All three scenarios selected model-free three-component decomposition parameters and RVOG decomposition parameters, which demonstrates that these parameters are highly sensitive to the crops selected in this embodiment of the invention.

[0072] The inversion results were compared with the actual crop heights at the measured sampling points in the experimental area. Table 1 shows that the three crops still achieved the best inversion accuracy under the random forest model. The optimal mean root mean square error (RMSE) for maize reached 57.04 cm, with a correlation coefficient of 0.85, representing a 3.02% improvement in accuracy. For wheat, the optimal RMSE reached 16.97 cm, with a correlation coefficient of 0.65, representing a 2.02% improvement in accuracy. For soybean, the optimal RMSE reached 24.43 cm, with a correlation coefficient of 0.54, representing a 2.67% improvement in accuracy. All models showed varying degrees of improvement in inversion performance, achieving more accurate height inversion.

[0073] As can be seen, the crop height inversion framework based on polarimetric SAR data combined with machine learning in this invention makes full use of simplified polarimetric data information, avoids complex physical modeling and mathematical derivation, and effectively improves the efficiency and accuracy of inversion.

[0074] This embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of the crop height inversion method based on reduced polarimetric SAR data described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0075] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-described method for crop height inversion based on reduced polarimetric SAR data.

[0076] like Figure 6As shown, the computer device 120 may include: at least one processor 121, such as a central processing unit (CPU), at least one communication interface 123, memory 124, and at least one communication bus 122. The communication bus 122 is used to enable communication between these components. The communication interface 123 may include a display screen and a keyboard; optionally, the communication interface 123 may also include a standard wired interface or a wireless interface. The memory 124 may be high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory 124 may also be at least one storage device located remotely from the aforementioned processor 121. The memory 124 stores application programs, and the processor 121 calls the program code stored in the memory 124 to execute any of the aforementioned method steps. The communication bus 122 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 122 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6The term 124 is represented by a single line, but this does not imply a single bus or a single type of bus. The memory 124 may include volatile memory, such as random-access memory (RAM); it may also include non-volatile memory, such as flash memory, hard disk drive (HDD), or solid-state drive (SSD); or it may include combinations of the above types of memory. The processor 121 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. The processor 121 may further include hardware chips. These hardware chips may be application-specific integrated circuits (ASICs), programmable logic devices (PLDs), or combinations thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. Optionally, the memory 124 is also used to store program instructions. The processor 121 can call the program instructions to implement the crop height inversion method based on reduced polarimetric SAR data as described in this embodiment.

[0077] This embodiment provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for crop height inversion based on reduced polarimetric SAR data.

[0078] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for inverting crop height based on reduced polarimetric SAR data, characterized in that, Includes the following steps: S1: Acquire raw polarimetric SAR images during the crop growth cycle, perform data preprocessing on the raw polarimetric SAR images, and obtain simulated simplified polarimetric data. S2: Extract the polarization parameters from the simplified polarization data, and obtain a multidimensional feature set based on the polarization parameters; S3: Obtain a training set and a validation set based on the field measured crop height dataset; use the training set and validation set to train and validate the machine learning algorithm model to obtain a height prediction model; use the height prediction model to predict the multidimensional feature set to obtain crop height estimation data; S4: Based on the multidimensional feature set, optimize using the forward feature selection method to obtain the optimal feature subset; retrain and validate the machine learning algorithm model based on the optimal feature subset to obtain the height prediction model; use the height prediction model to predict the optimal feature subset to obtain a more accurate crop height inversion result.

2. The crop height inversion method based on reduced polarimetric SAR data according to claim 1, characterized in that, The data preprocessing of the original polarimetric SAR image includes: performing single-view complex transformation, coherence matrix extraction, polarimetric filtering, geocoding, and data cropping on the original polarimetric SAR image, and simulating it using a right-hand circular hybrid mode.

3. The method for crop height inversion based on reduced polarimetric SAR data according to claim 1, characterized in that, The polarization parameters include basic backscattering coefficient characteristics, Stokes vector characteristics, polarization target decomposition characteristics, polarization state and statistical characteristics, inter-channel relationships and vegetation index characteristics.

4. The method for crop height inversion based on reduced polarimetric SAR data according to claim 1, characterized in that, The machine learning algorithm models include random forest regression algorithm model, bagged decision tree algorithm model, extreme gradient boosting algorithm model, and Gaussian process regression algorithm model.

5. The method for crop height inversion based on reduced polarimetric SAR data according to claim 1, characterized in that, The process of obtaining the training set and validation set based on the field measured crop height dataset includes: using a stratified random sampling method, the height distribution in the field measured crop height dataset is divided into six smaller subgroups, each subgroup is randomly assigned, and the subgroups are divided according to a preset ratio to obtain the training set and validation set.

6. The method for crop height inversion based on reduced polarimetric SAR data according to claim 1, characterized in that, The verification process for training and validating the machine learning algorithm model using the training set and validation set involves using root mean square error and correlation coefficient as evaluation metrics to assess the prediction accuracy of the highly predictive model.

7. The method for crop height inversion based on reduced polarimetric SAR data according to claim 1, characterized in that, The specific process of optimization using the forward feature selection method is as follows: Candidate feature sets are obtained based on the multidimensional feature set. Starting from an empty feature set, candidate feature sets are added to the current feature set in sequence. Specifically, this includes: obtaining a training set and a validation set based on the current feature set and a field measured crop height dataset; training a machine learning algorithm model based on the training set; calculating the root mean square error of the test set; and selecting the candidate feature set that minimizes the root mean square error to add to the current feature set. The current feature set is used as the optimal feature subset until the number of features in the current feature set reaches a preset value or the root mean square error converges.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the crop height inversion method based on simplified polarimetric SAR data as described in any one of claims 1-7.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the crop height inversion method based on simplified polarimetric SAR data as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the crop height inversion method based on simplified polarimetric SAR data as described in any one of claims 1-7.

Images