A method and apparatus for surface parameter inversion

Abstract

This invention discloses a method and apparatus for inverting surface parameters. The method includes: dynamically selecting or weightedly fusing an AIEM model and a KA model based on the surface roughness parameter ks to construct an AIEM-KA hybrid model to adapt to surface roughness at different scales; generating a simulation dataset based on the AIEM-KA hybrid model, the simulation dataset including simulated backscattering coefficient data under different combinations of surface parameters and radar observation parameters; establishing a mapping model from backscattering coefficients to surface parameters based on the simulation dataset and Gaussian process regression; and determining the target surface parameters based on the measured backscattering coefficients using the mapping model. Thus, by fusing the physical model with Gaussian process regression, efficient inversion is achieved while maintaining the correctness of the electromagnetic scattering physical mechanism, effectively improving the accuracy and efficiency of surface parameter inversion.

Description

Technical Field

[0001] This invention relates to the field of microwave remote sensing and SAR image simulation technology, and in particular to a method and apparatus for inverting surface parameters. Background Technology

[0002] The dielectric constant and roughness of the Earth's surface are key characteristic parameters describing its physical and chemical properties. Radar backscattering coefficient, as the core link between surface parameters and microwave remote sensing observation information, together constitute the basic input data for quantitative soil moisture retrieval, numerical simulation of surface microwave scattering, and agricultural remote sensing soil moisture monitoring. Leveraging the all-weather, 24 / 7 observation advantages of microwave remote sensing, the backscattering coefficient, as a direct observation output of radar remote sensing, directly determines the overall level of surface parameter retrieval and quantitative applications based on its accuracy and reliability.

[0003] Currently, microwave inversion technology for surface parameters is mainly divided into two categories: physical mechanism-driven inversion and data-driven machine learning inversion. Regarding physical models, the improved integral equation model is suitable for small-scale rough surfaces and is currently the most accurate classical forward model, while the Kirchhoff approximation is suitable for large-scale undulating rough surfaces. However, traditional physical models involve multiple nested integrals and cumbersome iterative solutions, resulting in extremely low computational efficiency and making it difficult to meet the needs of large-scale real-time inversion.

[0004] Therefore, there is an urgent need for an inversion correction method that integrates the physical mechanism of electromagnetic scattering with efficient regression modeling, so as to improve the inference efficiency while ensuring the correctness of the physical mechanism, thereby improving the accuracy and efficiency of surface parameter inversion. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method and apparatus for inverting surface parameters to solve the problems of low accuracy and poor efficiency in surface parameter inversion in the prior art.

[0006] In a first aspect, embodiments of the present invention provide a method for retrieving surface parameters, including: The AIEM-KA hybrid model is obtained by dynamically selecting or weighting and fusing the AIEM and KA models based on the surface roughness parameter ks. A simulation dataset was generated based on the AIEM-KA hybrid model. The simulation dataset includes backscattering coefficient simulation data under different combinations of surface parameters and radar observation parameters. A mapping model from backscattering coefficients to surface parameters was established based on simulation datasets and Gaussian process regression. Based on the mapping model, the target surface parameters are determined according to the measured backscattering coefficient. The target surface parameters are key physical property parameters used for surface microwave scattering modeling.

[0007] Secondly, embodiments of the present invention provide a surface parameter inversion device, comprising: The hybrid model determination unit is used to dynamically select or weightedly fuse the AIEM model and the KA model to obtain the AIEM-KA hybrid model based on the surface roughness parameter ks. The simulation dataset determination unit is used to generate a simulation dataset based on the AIEM-KA hybrid model. The simulation dataset includes backscattering coefficient simulation data under different combinations of surface parameters and radar observation parameters. The mapping model building unit is used to build a mapping model from backscattering coefficients to surface parameters based on simulation datasets and Gaussian process regression. The target determination unit is used to determine the target surface parameters based on the measured backscattering coefficient according to the mapping model. The target surface parameters are key physical property parameters used for surface microwave scattering modeling.

[0008] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in the embodiments of the present invention.

[0009] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in the embodiments of the present invention.

[0010] The technical solution of this invention constructs an AIEM-KA hybrid model by dynamically selecting or weighting and fusing the AIEM and KA models based on the surface roughness parameter ks. This model adapts to surface roughness at different scales. A simulation dataset is generated based on the AIEM-KA hybrid model, including simulated backscattering coefficient data under different combinations of surface parameters and radar observation parameters. A mapping model from backscattering coefficients to surface parameters is established based on the simulation dataset and Gaussian process regression. The target surface parameters are then determined based on the measured backscattering coefficients using this mapping model. Thus, by fusing the physical model with Gaussian process regression, efficient inversion is achieved while maintaining the correctness of the electromagnetic scattering physical mechanism, effectively improving the accuracy and efficiency of surface parameter inversion. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of the surface parameter inversion method provided in this embodiment of the invention; Figure 2 This is a hardware architecture diagram of the electronic device provided in an embodiment of the present invention; Figure 3 This is a structural diagram of the surface parameter inversion device provided in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0014] Figure 1 This is a flowchart of the surface parameter inversion method provided in this embodiment of the invention, as follows: Figure 1 As shown, the method includes: Step 100: Dynamically select or weightedly fuse the AIEM model and the KA model based on the surface roughness parameter ks to obtain the AIEM-KA hybrid model.

[0015] Step 200: Generate a simulation dataset based on the AIEM-KA hybrid model. The simulation dataset includes backscattering coefficient simulation data under different combinations of surface parameters and radar observation parameters.

[0016] Step 300: Establish a mapping model from backscattering coefficients to surface parameters based on the simulation dataset and Gaussian process regression.

[0017] Step 400: Based on the mapping model, determine the target surface parameters according to the measured backscattering coefficient. The target surface parameters are key physical property parameters used for surface microwave scattering modeling.

[0018] The specific implementation methods of each step are described below. Before introducing each step, the background of this invention embodiment should be explained first. In the field of microwave remote sensing, the surface backscattering coefficient is the core link connecting surface physical parameters and radar observation signals. The simulation accuracy and correction level of the backscattering coefficient directly determine the overall level of surface parameter inversion and quantitative remote sensing applications. Among them, surface parameters include dielectric constant, root mean square height, and correlation length.

[0019] The AIEM (Advanced Integral Equation Model) model, based on the physical laws of electromagnetic wave scattering, accurately calculates the backscattering coefficient by summing Kirchhoff terms, complementary terms, and cross terms. It boasts advantages such as a complete physical mechanism and high simulation accuracy, making it particularly suitable for small-scale rough surfaces. Specifically, AIEM is an improved version of the integral equation model, capable of calculating radar backscattering coefficients for given surface and sensor parameters. Where p and q represent the polarization mode (V or H). In AIEM, given the polarization combination pq, the backscattering coefficient is obtained by summing the Kirchhoff terms, complementary terms, and cross terms, specifically through the following formula: In the formula, The backscattering coefficient is... For the incident wave number, , s is the root-mean-square height of the Earth's surface, and W is the surface power spectrum. For the complex weighting coefficients, the Fourier transform of the correlation function of the surface power spectrum W is expressed by the following formula: In the formula, W(K) is the surface power spectrum, K is the space wavenumber, ρ(ξ) is the surface height autocorrelation function, and ξ is the horizontal interval distance.

[0020] The Kirchhoff Approximation (KA) model, based on the tangent plane approximation, treats the scattered field at any point on a rough surface as equivalent to the scattered field produced by an infinitely large tangent plane at that point under the same incident wave. This model is suitable for large-scale undulating rough surfaces with gentle slopes and slow changes. Specifically, for Gaussian rough surfaces, the incoherent scattering coefficient derived by KA... for: In the formula, f is the incoherent scattering coefficient. pq v is the Fresnel scattering coefficient. x and v y Let ρ(x, y) be the scattering vector component, ρ(x, y) be the surface height autocorrelation function, k be the wavenumber, and θ be the scattering vector component. i Let θ be the angle of incidence. s The scattering angle is... Let x be the azimuth angle, and y be the spatial coordinates of two orthogonal directions within the horizontal plane of the rough surface.

[0021] However, a single model is insufficient to cover the full range of surface types from smooth to rough. At low roughness, AIEM is computationally expensive while KA lacks accuracy, while at high roughness, AIEM is highly complex while KA is more efficient. Therefore, a hybrid model is needed that can adaptively switch between or fuse the two models based on surface roughness characteristics, thus balancing computational accuracy and efficiency. It should be understood that the surface parameters mentioned below refer to surface physical parameters.

[0022] Regarding step 100, the surface roughness parameter is ks, which is the dimensionless ratio of the surface undulation to the radar detection wavelength. In this embodiment of the invention, the surface roughness parameter serves as a dimensionless index for judging the surface undulation relative to the radar wavelength, in order to determine which physical model to use to calculate the backscattering coefficient under given surface parameters and radar observation conditions. The AIEM model is an improved integral equation model, which starts from the physical laws of electromagnetic wave scattering and accurately establishes a quantitative mapping relationship between surface roughness, dielectric constant, and radar backscattering coefficient. It is currently the mainstream method for analyzing the variation law of backscattering and conducting scattering mechanism research, and is suitable for small-scale rough surfaces. The KA model is the Kirchhoff approximation model, which is a classic analytical method for analyzing the scattering of electromagnetic waves or light waves on rough surfaces. In the fields of electromagnetic scattering and optics, it is also often referred to as the geometric optics approximation or tangent plane approximation. Its core idea is that the scattering field of any point P on the rough surface is equivalent to the scattering field produced by an infinitely large tangent plane at that point under the same incident wave. It is suitable for large-scale undulating rough surfaces with small slopes and slow changes.

[0023] In one alternative implementation, dynamically selecting or weighting and fusing the AIEM model and the KA model based on the surface roughness parameter ks includes: When ks < 0.3, the AIEM model is selected to calculate the backscattering coefficient; When ks>3, the KA model is selected to calculate the backscattering coefficient; When 0.3 ≤ ks ≤ 3, the backscattering coefficient is calculated using the following weighted fusion formula: In the formula, The backscattering coefficients are output by the mixture model. The backscattering coefficients are output by the AIEM model. Here, α represents the backscattering coefficients output by the KA model, and α is the weighting coefficient. The weighting coefficients are obtained by fitting the surface slope and curvature parameters.

[0024] When ks < 0.3, the surface undulations are relatively gentle relative to the radar wavelength. At this time, electromagnetic scattering from the surface is dominated by coherent scattering. The AIEM model, based on rigorous integral equations, has accurate analytical capabilities for this type of scattering mechanism. Therefore, using the AIEM model alone to calculate the backscattering coefficient can ensure the highest simulation accuracy. When ks > 3, the surface undulations are very severe relative to the wavelength. At this time, the scattering field is dominated by incoherent scattering. The KA model, based on the tangent plane approximation, can efficiently describe this type of scattering characteristic, and its computational efficiency is much higher than that of the AIEM model. Therefore, using the KA model alone to calculate the backscattering coefficient can significantly improve computational efficiency while ensuring sufficient accuracy. When 0.3 ≤ ks ≤ 3, the surface is in the medium roughness range, and both coherent and incoherent scattering mechanisms coexist. It is difficult to obtain the best results by using either model alone. By using a weighted fusion method, the advantages of both models can be combined to more accurately reflect the actual surface scattering response characteristics to electromagnetic waves. Furthermore, within the transition region (0.3≤ks≤3), the applicability of the AIEM and KA models continuously varies with surface micro-topographic features (such as local slope and curvature). By introducing geometric parameters such as surface slope and curvature, a functional relationship between the weighting coefficient α and these geometric parameters is established using numerical fitting methods. This allows the hybrid model to adaptively adjust the contribution ratio of the AIEM and KA models as the surface morphology changes, thereby ensuring the continuity and smoothness of the backscattering coefficient calculation within the transition region.

[0025] Step 200: Generate a simulation dataset based on the AIEM-KA hybrid model. The simulation dataset includes backscattering coefficient simulation data under different combinations of surface parameters and radar observation parameters.

[0026] Regarding step 200, after obtaining the AIEM-KA hybrid model through step 100, a simulation dataset needs to be generated based on the AIEM-KA hybrid model. This simulation dataset is a set of training sample data covering a wide parameter space, generated by the forward modeling of the AIEM-KA hybrid model. In this embodiment of the invention, the simulation dataset serves to provide training samples with clear physical meaning, no manual annotation errors, and sufficient parameter coverage for the subsequent training of the Gaussian process regression model. This compensates for the limitations of the limited bands and angles in the raw radar backscattering data obtained from field measurements, which restricts the improvement of the accuracy of microwave remote sensing quantitative inversion from the data source level. Generating the simulation dataset through a physical model ensures the physical correctness and completeness of the training samples.

[0027] In one alternative implementation, the simulation dataset is represented as: in, , ε, s, l, θ, and f represent the dielectric constant, root mean square height, correlation length, incident angle, and frequency, respectively. These represent the backscattering coefficients of VV polarization and HH polarization, respectively. Specifically, the range of input parameters can be set according to the actual surface type and radar system parameters. For example, the dielectric constant ε can be set to 3~30 to cover the common range from dry to wet soil, the root mean square height s can be set to 0.1~5cm to cover typical surface morphologies from smooth to rough, the correlation length l can be set to 1~30cm to cover surface texture features at different scales, the incident angle θ can be set to 10°~70° to cover common observation geometries of spaceborne and airborne SAR systems, and the frequency f can be set to the corresponding center frequency value according to the radar operating band.

[0028] For each combination of input parameters, the corresponding VV and HH polarization backscattering coefficients are calculated using the AIEM-KA hybrid model determined in step 100, and all samples constitute a complete simulation dataset.

[0029] Regarding step 300, after determining the simulation dataset through step 200, it is necessary to establish a mapping model based on the simulation dataset and Gaussian process regression. The mapping model is used to map the measured radar backscattering coefficients to surface parameters, including the dielectric constant ε, root mean square height s, and correlation length l.

[0030] Gaussian Process Regression (GPR) is a nonparametric regression method based on a Bayesian framework. Its core idea is to assume that the function to be regressed follows a Gaussian process prior distribution, meaning that the function values ​​corresponding to any finite number of input points follow a joint Gaussian distribution. Unlike parametric regression methods, GPR does not require a predefined fixed function form. It can flexibly characterize the similarity relationships in the input sample space through kernel functions and provide predicted values ​​and their uncertainty estimates through Bayesian inference. Therefore, it has significant advantages in handling small-sample, high-dimensional nonlinear regression problems. More importantly, GPR naturally supports the confidence assessment of inversion results, providing a confidence interval for each inversion output, thus overcoming the shortcomings of existing algorithms in quantifying the uncertainty of inversion results.

[0031] Traditional modeling methods are divorced from the physical mechanism of electromagnetic scattering and rely solely on sample data to fit numerical relationships. As a result, the model has poor generalization performance and is prone to overfitting. In contrast, this invention trains Gaussian process regression based on simulation data generated by the physical model. This not only retains the mechanistic constraints of the physical model but also leverages the advantages of efficient data-driven model inference, achieving a deep integration of physical mechanisms and data-driven approaches.

[0032] In one optional implementation, establishing a mapping model from backscattering coefficients to surface parameters based on a simulation dataset and Gaussian process regression includes: establishing independent Gaussian process regression (GPR) models for different polarization modes; constructing the covariance matrix of each GPR model using the Matern kernel function; optimizing the hyperparameters of each GPR model using the simulation dataset as training samples through maximum likelihood estimation; and enabling each GPR model to learn the nonlinear mapping relationship from backscattering coefficients to surface parameters to obtain the trained mapping model.

[0033] Specifically, independent Gaussian process regression (GPR) models are established for different polarization modes. Two independent GPR models are built for the VV-polarization backscattering coefficient and the HH-polarization backscattering coefficient, respectively, with the corresponding backscattering coefficient values ​​as inputs and surface parameter vectors as outputs. Since different polarization modes have varying sensitivities to surface parameters—for example, HH polarization is more sensitive to horizontal surface structure information, while VV polarization is more sensitive to vertical structure information—separate modeling allows for more precise capture of complementary surface information contained in each polarization channel, thereby improving inversion accuracy. The covariance matrix of each GPR model is constructed using the Matern kernel function. The covariance matrix quantifies the correlation between output values ​​of different training samples and is a core component of Gaussian process regression. Using the simulation dataset as training samples, the hyperparameters of each GPR model are optimized through maximum likelihood estimation. Specifically, maximum likelihood estimation optimizes the scale hyperparameter l and other relevant hyperparameters in the kernel function by maximizing the logarithmic marginal likelihood function, maximizing the likelihood probability of the model on the training data, thus obtaining the optimal model configuration. Finally, each GPR model is trained to learn the nonlinear mapping relationship from backscattering coefficients to surface parameters, resulting in a fully trained mapping model.

[0034] Specifically, during training, the GPR model establishes a high-dimensional nonlinear mapping between the input backscattering coefficient space and the output surface parameter space through the covariance matrix, and achieves model convergence through hyperparameter iterative optimization. After training, given a new measured backscattering coefficient input, the GPR model can predict the most likely surface parameter values ​​based on the covariance structure learned during training, and can simultaneously output the prediction variance as an uncertainty evaluation index of the inversion results, providing a quantitative basis for assessing the reliability of the inversion results.

[0035] In one alternative implementation, the Matern kernel expression is: = In the formula, ( ) is the input sample vector. Used to represent input samples and The covariance between them, d is and The Euclidean distance between them, where l is a size hyperparameter used to control the smoothness of the kernel function.

[0036] The hyperparameters are optimized during model training using maximum likelihood estimation. This kernel function possesses first-order mean-square differentiability, corresponding to finite-order smoothness, which, while preserving sufficient smoothness, can effectively fit the slight non-stationary characteristics that may exist in the surface scattering response. Existing Gaussian kernels are infinitely differentiable, resulting in a rapid decrease in correlation, which does not conform to the patterns of exponential spectrum simulation data, weakening the sensitivity of the scattering coefficient to parameter changes. The Matern kernel (ν=3 / 2), however, supports finite-order smoothness, compensating for the Gaussian kernel's inability to fit slight non-stationary characteristics. Compared to the infinitely differentiable RBF kernel, the Matern kernel (ν=3 / 2) has stronger adaptability in regions of abrupt parameter changes, effectively avoiding spurious smoothing. Furthermore, the Matern kernel (ν=5 / 4) has excessively strong second-order differentiability, which can also produce spurious smoothing in regions of abrupt parameter changes, making it unsuitable for the local nonlinear characteristics of rough surface scattering. Therefore, this invention preferably uses the Matern kernel (ν=3 / 2).

[0037] Regarding step 400, after obtaining the mapping model through step 300, the target surface parameters are determined based on the measured backscattering coefficient according to the mapping model. The obtained target surface parameters are the key physical property parameters used for surface microwave scattering modeling.

[0038] Specifically, the GPR model has been trained using a simulation dataset, establishing a high-dimensional nonlinear mapping relationship from surface parameters to backscattering coefficients. After training, the measured backscattering coefficients are used as input observations y. GPR calculates their posterior distribution through Bayesian inference, that is, given the observed data and the covariance structure of the kernel function, it solves for the combination of surface parameters x that maximizes the posterior probability. This process is equivalent to searching in the parameter space for the physical parameter solution that best matches the measured scattering coefficients.

[0039] Meanwhile, as a nonparametric probabilistic model, GPR can simultaneously provide the posterior variance (i.e., uncertainty interval) of the inversion results. This variance reflects both the data sparsity of the model in the input region and the uniqueness and reliability of the parameter solutions. In practice, optimization algorithms such as conjugate gradient or L-BFGS are typically used to maximize the posterior probability density, obtaining point estimates as the final inversion results. The obtained uncertainty information can serve as a confidence evaluation index for the inversion results. For example, excessively large variance indicates low reliability of the inversion values ​​or the need for additional observation constraints. It can also be used in subsequent scattering coefficient correction steps to design weighted fusion strategies, such as assigning higher weights to high-confidence samples to improve the overall robustness of the correction, thereby ensuring that the input parameters of the physical model have optimal physical consistency and statistical reliability during forward modeling.

[0040] In one alternative implementation, after obtaining the target surface parameters, the following steps are also included: substituting the target surface parameters into the AIEM-KA hybrid model to obtain the target backscattering coefficient, which is the corrected backscattering coefficient.

[0041] The target backscattering coefficient is the backscattering coefficient value obtained by substituting the retrieved, self-consistent surface parameters into the AIEM-KA hybrid model for forward modeling. Specifically, since raw radar backscattering data obtained from field measurements is typically affected by a combination of factors, including system thermal noise, calibration errors, propagation path attenuation, and non-ideal surface conditions (such as inhomogeneous surface dielectric properties and inaccurate roughness parameter measurements), directly using the measured backscattering coefficient for subsequent quantitative applications may introduce non-negligible biases. By substituting the high-precision surface parameters obtained from inversion into the AIEM-KA hybrid model to calculate the backscattering coefficient, the resulting target backscattering coefficient maintains inherent consistency with the surface parameters in terms of physical mechanism. This effectively eliminates non-physical biases in the measured data and achieves systematic correction of the measured backscattering coefficient.

[0042] Through steps 100-400, surface parameter inversion and scattering coefficient correction based on the AIEM-KA hybrid model and Matern kernel Gaussian process regression were achieved. The technical solution of this invention integrates the electromagnetic scattering mechanism analysis capability of the AIEM-KA hybrid model with the efficient data inference capability of Gaussian process regression. It overcomes the shortcomings of pure physical models, such as low computational efficiency, cumbersome internal nested integrals and iterative solution logic, and difficulty in direct application for parameter inversion. It also compensates for the weaknesses of pure data-driven models, such as being detached from the constraints of electromagnetic scattering physical mechanisms, insufficient generalization ability, and susceptibility to overfitting. The Gaussian process regression framework naturally supports the quantitative assessment of prediction uncertainty, providing a theoretical basis and quantitative indicators for the credibility assessment of inversion results, and solving the problem that traditional methods cannot conduct credibility assessments of surface parameter inversion results obtained from backscattering data. Finally, by substituting the inverted surface parameters back into the forward model for scattering coefficient correction, a self-consistent agreement between the measured backscattering coefficient and the surface physical parameters was achieved. This effectively eliminated non-physical biases in the measured backscattering coefficient, improved the accuracy and reliability of quantitative processing of microwave remote sensing data, and can be widely applied to the quantitative processing of radar surface scattering observation data and SAR image simulation.

[0043] like Figure 2 , Figure 3 As shown, this embodiment of the invention provides a surface parameter inversion device. The device embodiment can be implemented through software, hardware, or a combination of both. From a hardware perspective, as... Figure 2 The diagram shown is a hardware architecture diagram of an electronic device used for a surface parameter inversion device according to an embodiment of the present invention. (Except for...) Figure 2 In addition to the processor, memory, mesh interface, and non-volatile memory shown, the electronic device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing packets. Taking software implementation as an example, such as... Figure 3 As shown, a device in a logical sense is formed by the CPU of the electronic device in which it is located reading the corresponding computer program from the non-volatile memory into the memory for execution.

[0044] This embodiment provides a surface parameter inversion device, comprising: The hybrid model determination unit 31 is used to dynamically select or weightedly fuse the AIEM model and the KA model to obtain the AIEM-KA hybrid model based on the surface roughness parameter ks. Simulation dataset determination unit 32 is used to generate simulation datasets based on the AIEM-KA hybrid model. The simulation datasets include backscattering coefficient simulation data under different combinations of surface parameters and radar observation parameters. The mapping model building unit 33 is used to build a mapping model from backscattering coefficients to surface parameters based on the simulation dataset and Gaussian process regression. The target determination unit 34 is used to determine the target surface parameters based on the measured backscattering coefficient according to the mapping model. The target surface parameters are key physical property parameters used for surface microwave scattering modeling.

[0045] In some embodiments, the device is further configured to: Substitute the target surface parameters into the AIEM-KA hybrid model to obtain the target backscattering coefficient, which is the corrected backscattering coefficient.

[0046] In some embodiments, dynamically selecting or weighting and fusing the AIEM model and the KA model based on the surface roughness parameter ks includes: When ks < 0.3, the AIEM model is selected to calculate the backscattering coefficient; When ks>3, the KA model is selected to calculate the backscattering coefficient; When 0.3 ≤ ks ≤ 3, the backscattering coefficient is calculated using the following weighted fusion formula: In the formula, The backscattering coefficients are output by the mixture model. The backscattering coefficients are output by the AIEM model. α represents the backscattering coefficients output by the KA model, and α is the weighting coefficient.

[0047] In some embodiments, the weighting coefficients are obtained by fitting surface slope and curvature parameters.

[0048] In some embodiments, establishing a mapping model from backscattering coefficients to surface parameters based on a simulation dataset and Gaussian process regression includes: Independent Gaussian process regression (GPR) models were established for different polarization modes. The covariance matrix of each GPR model is constructed using the Matern kernel function; Using the simulation dataset as training samples, the hyperparameters of each GPR model are optimized through maximum likelihood estimation; Each GPR model learns the nonlinear mapping relationship from backscattering coefficients to surface parameters, resulting in a trained mapping model.

[0049] In some embodiments, the Matern kernel expression is: = In the formula, ( ) is the input sample vector. Used to represent input samples and The covariance between them, d is and The Euclidean distance between them, where l is a size hyperparameter used to control the smoothness of the kernel function.

[0050] In some embodiments, the simulation dataset is represented as: in, , ; ε, s, l, θ, and f represent the dielectric constant, root mean square height, correlation length, angle of incidence, and frequency, respectively. These represent the backscattering coefficients for VV polarization and HH polarization, respectively.

[0051] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on a surface parameter inversion device. In other embodiments of the present invention, a surface parameter inversion device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0052] The information interaction and execution process between the modules in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0053] This invention also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements a surface parameter inversion method according to any embodiment of this invention.

[0054] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program causes the processor to perform a surface parameter inversion method according to any embodiment of this invention.

[0055] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.

[0056] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0057] Storage media embodiments for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0058] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0059] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion module connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion module execute some and all of the actual operations, thereby realizing the function of any of the above embodiments.

[0060] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or electronic device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or electronic device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or electronic device that includes that element.

[0061] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various storage media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions 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 inverting surface parameters, characterized in that, include: The AIEM-KA hybrid model is obtained by dynamically selecting or weighting and fusing the AIEM and KA models based on the surface roughness parameter ks. A simulation dataset was generated based on the AIEM-KA hybrid model. The simulation dataset includes backscattering coefficient simulation data under different combinations of surface parameters and radar observation parameters. A mapping model from backscattering coefficients to surface parameters was established based on simulation datasets and Gaussian process regression. Based on the mapping model, the target surface parameters are determined according to the measured backscattering coefficient. The target surface parameters are key physical property parameters used for surface microwave scattering modeling.

2. The method according to claim 1, characterized in that, Also includes: Substitute the target surface parameters into the AIEM-KA hybrid model to obtain the target backscattering coefficient, which is the corrected backscattering coefficient.

3. The method according to claim 1, characterized in that, Based on the surface roughness parameter ks, AIEM and KA models are dynamically selected or weighted and fused, including: When ks < 0.3, the AIEM model is selected to calculate the backscattering coefficient; When ks>3, the KA model is selected to calculate the backscattering coefficient; When 0.3 ≤ ks ≤ 3, the backscattering coefficient is calculated using the following weighted fusion formula: In the formula, The backscattering coefficients are output by the mixture model. The backscattering coefficients are output by the AIEM model. α represents the backscattering coefficients output by the KA model, and α is the weighting coefficient.

4. The method according to claim 3, characterized in that, The weighting coefficients are obtained by fitting the surface slope and curvature parameters.

5. The method according to claim 1, characterized in that, The mapping model from backscattering coefficients to surface parameters, established based on simulation datasets and Gaussian process regression, includes: Independent Gaussian process regression (GPR) models were established for different polarization modes. The covariance matrix of each GPR model is constructed using the Matern kernel function; Using the simulation dataset as training samples, the hyperparameters of each GPR model are optimized through maximum likelihood estimation; Each GPR model learns the nonlinear mapping relationship from backscattering coefficients to surface parameters, resulting in a trained mapping model.

6. The method according to claim 1, characterized in that, The Matern kernel expression is: = In the formula, ( ) is the input sample vector. Used to represent input samples and The covariance between them, d is and The Euclidean distance between them, where l is a size hyperparameter used to control the smoothness of the kernel function.

7. The method according to claim 1, characterized in that, The simulation dataset is represented as follows: in, , ; ε, s, l, θ, and f represent the dielectric constant, root mean square height, correlation length, angle of incidence, and frequency, respectively. These represent the backscattering coefficients for VV polarization and HH polarization, respectively.

8. A surface parameter inversion device, characterized in that, include: The hybrid model determination unit is used to dynamically select or weightedly fuse the AIEM model and the KA model to obtain the AIEM-KA hybrid model based on the surface roughness parameter ks. The simulation dataset determination unit is used to generate a simulation dataset based on the AIEM-KA hybrid model. The simulation dataset includes backscattering coefficient simulation data under different combinations of surface parameters and radar observation parameters. The mapping model building unit is used to build a mapping model from backscattering coefficients to surface parameters based on simulation datasets and Gaussian process regression. The target determination unit is used to determine the target surface parameters based on the measured backscattering coefficient according to the mapping model. The target surface parameters are key physical property parameters used for surface microwave scattering modeling.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-7.

Images