SAR image vegetation coverage inversion method and system based on multi-dimensional feature optimization and model optimization, storage medium and electronic equipment

Abstract

The invention provides an SAR image vegetation coverage inversion method and system based on multi-dimensional feature optimization and model optimization, a storage medium and electronic equipment, and the method comprises the following steps: carrying out the preprocessing of a Sentinel-2 image, calculating a normalized difference vegetation index (NDVI) through a wave band calculation formula, and carrying out the calculation of the NDVI; a reference vegetation coverage (FVC) of a research area is obtained by combining a pixel bipartite model (DPM), and then an optimal SAR feature combination is selected by applying an improved genetic algorithm ET-GA, using XGBoost as an adaptability evaluation function and using a root-mean-square error (RMSE) as an evaluation index. According to the method, SAR features are extracted through multiple feature decomposition methods, multi-dimensional feature optimization, intelligent algorithm optimization and a deep learning model are combined, the core problems of feature redundancy, insufficient precision, low efficiency and the like of SAR data in vegetation parameter inversion are solved, and an innovative technical scheme is provided for the field of remote sensing quantitative inversion.

Description

Technical Field

[0001] The present application relates to the field of quantitative remote sensing, and in particular to a SAR image vegetation cover inversion method, system, storage medium and electronic equipment based on multi-dimensional feature selection and model optimization. Background Art

[0002] Vegetation is the natural "link" connecting the atmosphere, hydrosphere, and pedosphere, playing a crucial role in ecosystems, directly impacting processes such as regional energy balance, hydrological cycles, and soil erosion control. In agriculture, especially in research areas dominated by farmland, vegetation cover is a key factor influencing crop growth, yield estimates, and farmland ecological health. Accurately obtaining information on farmland vegetation cover helps optimize agricultural management decisions, improve resource utilization efficiency, and provide data support for precision agriculture. Traditional vegetation cover measurements rely primarily on ground surveys. While they can provide relatively accurate data in some cases, they are often time-consuming and labor-intensive, and difficult to implement over large areas. Advances in remote sensing technology have made it possible to rapidly and accurately obtain vegetation cover over large areas.

[0003] In recent years, remote sensing technology has been mainly divided into optical remote sensing satellites and microwave remote sensing satellites. Optical remote sensing, with its high temporal and spatial resolution, provides a reliable data basis for vegetation cover inversion. However, optical remote sensing is easily affected by clouds, aerosols and lighting conditions. It is often difficult to obtain continuously available observation data during the critical growth period of crops (such as the rainy season or cloudy areas), which limits its application in real-time monitoring of large areas. In contrast, synthetic aperture radar (SAR) has all-day and all-weather observation capabilities due to its active microwave imaging characteristics, providing a stable data source for vegetation cover monitoring. Polarimetric SAR is sensitive to vegetation structure, water content and surface roughness through multi-channel backscatter information and polarization characteristics, providing richer polarization information for vegetation parameter inversion, giving it a unique advantage in vegetation parameter inversion.

[0004] Methods for inverting FVC from SAR images are primarily categorized into pixel decomposition, regression analysis, and machine learning. Pixel decomposition was originally proposed for optical data. Optical vegetation indices have a good linear correlation with surface vegetation conditions, but are difficult to effectively represent using linear mixed models in SAR data. Regression analysis estimates FVC by building a regression model between measured and remotely sensed data. However, this method fails to fully account for the complex scattering mechanisms and nonlinear relationships between variables in SAR data, resulting in limited accuracy in the inversion results and an inability to meet the needs of refined monitoring. Machine learning has gradually become the primary method for inverting FVC. In recent years, numerous researchers have attempted to apply SAR imagery to machine learning, improving FVC inversion accuracy. However, its application in deep learning remains limited. The KAN model, due to its advantages in nonlinear modeling and automatic feature learning, has been widely used in remote sensing. Furthermore, machine learning models are highly sensitive to hyperparameter settings, and proper hyperparameter configuration can improve model prediction accuracy. While excessive features increase information content, they can also introduce noise and redundancy, increasing model complexity. Therefore, optimizing multiple features is an urgent technical challenge. Summary of the Invention

[0005] The purpose of this application is to provide a SAR image vegetation cover inversion method, system, storage medium and electronic device based on multi-dimensional feature optimization and model optimization to solve or alleviate the problems existing in the above-mentioned prior art.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] The present application provides a SAR image vegetation cover inversion method based on multidimensional feature optimization and model optimization, comprising: step S101, obtaining a fully polarized SAR remote sensing image of a target study area and preprocessing it to obtain backscattering coefficients VV, HH and VH; step S102, applying multiple polarization decomposition techniques to extract SAR feature parameters to construct a multidimensional scattering feature set; step S103, obtaining a Sentinel-2 optical image of the target study area, calculating the normalized difference vegetation index NDVI, and calculating a reference FVC in combination with a pixel binary model DPM; step S104, applying a remote sensing land cover classification product to mask non-vegetation feature types and creating a vegetation-non-vegetation mask file to reduce interference from non-vegetation information; step S105, applying the ET-GA algorithm to perform feature screening on the SAR multidimensional feature set to obtain the optimal feature combination to reduce feature redundancy; step S106, applying the BKA hyperparameter optimization algorithm to obtain the optimal hyperparameters of the model, and using the optimal model BKA-KAN to estimate vegetation cover.

[0008] Preferably, in step S101, a fully polarized SAR remote sensing image of the target study area is acquired and preprocessed to obtain backscatter coefficients VV, HH, and VH, specifically:

[0009] A series of preprocessing operations on the Radarsat-2 fully polarimetric data from the target study area yields backscatter coefficients VV, HH, and VH. The preprocessing process primarily includes radiometric calibration, deburst processing, multi-look processing, polarimetric filtering, and terrain correction. Radiometric calibration converts raw digital quantized values ​​into physically meaningful radar backscatter coefficients, ensuring quantitative comparability of data across different time phases. Taking advantage of the strip-wise acquisition characteristics of SLC data, deburst processing eliminates inter-strip discontinuities and generates seamless, continuous images. To suppress coherent speckle noise, a multi-look processing scheme with one look in range and two looks in azimuth is employed, effectively improving the signal-to-noise ratio while maintaining resolution. Furthermore, a 7×7 windowed Refined Lee filter is used for polarimetric filtering, reducing speckle noise while effectively preserving feature edges and textures. Finally, terrain correction is performed using a range-Doppler model, combined with high-precision DEM data, to eliminate geometric distortion caused by topographical fluctuations. After this series of rigorous processing steps, fully polarized SAR data with accurate geometric position and reliable radiation characteristics were finally obtained, including radar reflection information under vertical-vertical (VV), horizontal-horizontal (HH) and vertical-horizontal (VH) polarization modes, which also provided a high-quality data foundation for subsequent polarization decomposition, radar vegetation index calculation and vegetation cover inversion.

[0010] Preferably, in step S102, multiple polarization decomposition techniques are applied to extract SAR characteristic parameters to construct a multi-dimensional scattering feature set, specifically:

[0011] First, the polarization ratio parameter is calculated using the SAR backscatter coefficient to enhance the feature expression and differentiation effect. The specific formula is as follows:

[0012]

[0013] Among them, σ 0 HH represents the backscatter coefficient of the horizontal transmission-horizontal reception (HH) channel, σ 0 HV It represents the backscatter coefficient of the horizontal transmit-vertical receive (HV) path.

[0014] Pauli decomposition is performed on the SAR scattering matrix to extract the components corresponding to different scattering mechanisms by converting the scattering matrix into the form of Pauli basis vectors k. The specific formula is as follows:

[0015] Paulir =|S HH -S VV | 2

[0016] Pauli b =|S HH +S VV | 2

[0017] Pauli g =|2S HV | 2

[0018] Among them Pauli r 、Pauli b 、Pauli g represent even scattering, surface scattering, and volume scattering, respectively.

[0019] Then, the polarization coherence matrix [T] is constructed using the Pauli basis vectors. The specific formula is as follows:

[0020]

[0021] Where superscripts * and T denote conjugate and transpose, respectively, and <·> denotes spatial or temporal averaging. The total scattered power provides a global measure of scattered energy and is expressed as follows:

[0022] Span=T 11 +T 22 +T 33

[0023] Then, various polarization decomposition techniques are applied, including Cloude-Pottier decomposition, Van Zyl decomposition, Freeman three-component decomposition, and Yamaguchi four-component decomposition.

[0024] By applying Cloude-Pottier to perform eigendecomposition on [T], we can obtain three variables: polarization scattering entropy (Entropy, H), average scattering angle (Alpha Angle, α), and anisotropy (Anisotropy, A). Among them, H represents the randomness of scattering, α represents the type of dominant scattering mechanism, and A represents the relative importance of the secondary dominant scattering mechanism. The specific formula is as follows:

[0025]

[0026] Among them, λ1, λ2, and λ3 are the eigenvalues ​​of the polarization coherence matrix, and e1, e2, and e3 are the corresponding eigenvectors. The combination of H and A can better distinguish different ground types. The combination formula is as follows:

[0027] 1mH1mA=(1-H0(1-A)

[0028] 1mHA=(1-H)A

[0029] H1mA=H(1-A)

[0030] HA=HA

[0031] Van Zyl decomposition considers azimuthally symmetric targets, that is, the correlation coefficient between the co-polarization and cross-polarization ratios is 0. General natural targets such as vegetation and soil meet this condition. Van Zyl decomposition is based on the Pauli basis to decompose the polarization scattering matrix into three scattering mechanisms, namely surface scattering (VZ surf ), even scattering (VZ dbl ) and volume scattering (VZ vol ).

[0032] Freeman three-component decomposition. Based on the work of Van Zyl, Freeman et al. first proposed to use the polarization coherence matrix [T] to establish three scattering mechanism models: surface scattering (Free surf ), Free dbl ), Volume Scattering (Free vol ).

[0033] Yamaguchi four-component decomposition. Using the same even-order scattering model as Freeman decomposition (Yamaguchi four-component decomposition) dbl ) and surface scattering models (Yama surf ) based on the non-reflection symmetry, the helicoid scattering component (Yama helix ) and modify the volume scattering component of vegetation by changing the probability density function of the relevant azimuth angle (Yama vol ).

[0034] Finally, there are three radar vegetation indices as the main features, namely, radar vegetation index (RVI) based on backscatter intensity kim ), Radar Vegetation Index (RVI) based on eigenvalues ​​calculated based on Cloude-Pottier decomposition van ) and based on Freeman decomposition, by calculating surface scattering (Free surf ), Free dbl ) and volume scattering (Free vol Radar vegetation index (RVI) free ), the specific details of the three radar vegetation indices are shown in the following formulas:

[0035]

[0036] In summary, this paper extracts a total of 32 SAR features, including backscattering coefficient, polarization decomposition parameter, polarization coherence matrix and radar vegetation index.

[0037] Preferably, in step S103, a Sentinel-2 optical image of the target study area is obtained, the Normalized Difference Vegetation Index (NDVI) is calculated, and the reference FVC is calculated in combination with the pixel dichotomy model (DPM), specifically:

[0038] First, Sentinel-2 optical images of the target study area, quasi-synchronized with Radarsat-2 SAR images, were acquired from the Copernicus Open Data Center (https: / / dataspace.copernicus.eu / ). Both data sets were resampled to the same resolution and georeferenced to ensure consistency in spatial extent. Next, based on vegetation spectral sensitivity analysis, the Normalized Difference Vegetation Index (NDVI) was constructed using the red band (B4, central wavelength 665nm) and the near-infrared band (B8, central wavelength 842nm). The reference vegetation cover was then calculated using the NDVI-DPM hybrid model. The calculation formula is:

[0039]

[0040] Where ρ NIR represents the near-infrared band (B8), ρ RED Represents the red band (B4), NDVI soil represents the NDVI value with a 5% confidence interval, NDVI veg represents the NDVI value with a 95% confidence interval, and FVC refers to the fractional vegetation cover.

[0041] Preferably, in step S104, the remote sensing land cover classification product is used to mask the non-vegetation feature type and create a vegetation-non-vegetation mask file to reduce the interference of non-vegetation information, specifically:

[0042] The European Space Agency (ESA) used the 2020 10-meter resolution global land cover classification product (https: / / viewer.esa-worldcover.org / worldcover / ), which provides detailed and reliable data support for land cover classification worldwide. To improve the accuracy and reliability of vegetation cover inversion within the study area, this paper precisely aligned this land cover data with Radarsat-2 and Sentinel-2 imagery acquired within the study area. The classification results were used to effectively remove non-vegetated areas such as buildings and water bodies, generating a vegetation-non-vegetation mask file. This process significantly improved image quality and analysis accuracy.

[0043] Preferably, in step S105, the ET-GA algorithm is applied to perform feature screening on the SAR multidimensional feature set to obtain the optimal feature combination to reduce feature redundancy, specifically:

[0044] Considering that directly using all features for inversion modeling not only increases computational costs but also may introduce noise and redundant information, thereby reducing the model's predictive performance, a genetic algorithm (ET-GA) combining an elite retention strategy with a tournament selection method is applied to optimize SAR features, thereby balancing the algorithm's global search capability and local convergence ability.

[0045] This algorithm is a global optimization algorithm based on the principles of natural selection and genetics. It is widely used in high-dimensional feature selection problems. It gradually optimizes feature subsets by simulating operations such as selection, crossover, and mutation in the biological evolution process. The main steps of applying the ET-GA algorithm are as follows:

[0046] Encoding: Encode the feature set into a binary string, with each bit corresponding to one feature;

[0047] Generate initial population: Randomly generate 100 binary-coded individuals as the initial population, each individual represents a feature subset.

[0048] Fitness evaluation: For each individual, the XGBoost model is used as the fitness function, the root mean square error (RMSE) is used as the evaluation indicator, and the fitness value is calculated using 5-fold cross-validation.

[0049] Selection: During the selection process, the elite retention strategy is used to retain the top 5% individuals with the highest fitness values ​​in each generation to directly enter the next generation, and the remaining individuals are selected using the tournament selection method.

[0050] Crossover and mutation operations: Set a certain crossover probability to perform a two-point crossover operation on the selected individuals, so that the parent individuals produce new individuals; and reverse each bit of the individual code with a certain probability, thereby introducing a new feature combination.

[0051] Population update and iteration: Merge the elite individuals with the newly generated individuals to form a new population and enter the next iteration. The algorithm sets a maximum number of iterations of 30 and terminates early when the fitness value remains unchanged for 10 consecutive generations.

[0052] Preferably, in step S106, the BKA hyperparameter optimization algorithm is applied to obtain the optimal hyperparameters of the model, and the optimal model BKA-KAN is used to estimate vegetation coverage, specifically:

[0053] Hyperparameter optimization (HPO) is a critical step in ensuring model performance. Properly selecting hyperparameters plays a crucial role in the model's convergence speed, generalization ability, and prediction accuracy. The Black-winged Kite Algorithm (BKA) is used to automatically optimize the hyperparameter configurations of multiple network models.

[0054] The Black Kite Algorithm (BKA) is a newly proposed metaheuristic optimization algorithm inspired by the aggressive behavior of black kites during predation and their migratory behavior in response to environmental changes. Unlike traditional optimization methods, the BKA simulates the predation and migration mechanisms of black kites, achieving an efficient combination of global search and local exploitation. In the initial phase of the algorithm, individual positions are initialized by randomly generating a population within the search space. The root mean squared error (RMSE) of a five-fold cross-validation is used as the fitness function to evaluate the performance of each individual. In the attack phase, the algorithm updates individual positions through adaptive perturbations and random search mechanisms, enhancing global search capabilities and exploring potential optimal solutions. In the migration phase, the algorithm introduces a leader strategy and Cauchy mutation strategies, leveraging information from the best-performing individuals in the current population to guide other individuals toward the optimal solution, thereby achieving more refined local exploitation. Throughout the optimization process, the algorithm continuously updates the global optimal solution based on fitness, repeating the attack and migration steps until the maximum number of iterations is reached.

[0055] BKA optimizes the model's hyperparameters and improves the model's robustness. It achieves a balance between breadth search and depth development through attack and migration operations, solving the problem that traditional grid search and random search methods are inefficient in high-dimensional parameter space and are prone to falling into local optimality.

[0056] Kolmogorov-Arnold Networks is a new type of neural network architecture designed based on the Kolmogorov-Arnold Representation Theorem (KART). According to the KART theorem, any continuous multivariate function can be represented as a combination of a series of unary functions, and its mathematical expression is as follows:

[0057]

[0058] Among them, Φ q and is a continuous univariate function. This theory provides a mathematical basis for decomposing high-dimensional complex functions into simple univariate functions, which can effectively capture the nonlinear relationship between input features.

[0059] The core idea of ​​KAN is to replace the fixed activation function in traditional networks with learnable B-splines to significantly improve the model's expressiveness and generalization performance. Its mathematical expression is:

[0060]

[0061] Among them, B i (x) represents the B-spline basis function, c i are the coefficients optimized during the training process. The spline function dynamically adjusts the parameters by minimizing the loss function to achieve accurate fitting of the data and effective modeling.

[0062] Unlike traditional neural networks, KAN introduces a learnable activation function in the hidden layer. This design can more efficiently fit complex nonlinear relationships and significantly enhance the flexibility and interpretability of the model.

[0063] Finally, the feature combinations selected by the ET-GA algorithm were input into the model optimized by the BKA algorithm for training. Through comparative experiments, it was found that the BKA-KAN model achieved highly accurate inversion of FVC by learning the complex nonlinear relationship between features and target variables.

[0064] The present application also provides a SAR image vegetation cover inversion system based on multi-dimensional feature optimization and model optimization, including:

[0065] An image preprocessing unit is configured to perform image preprocessing operations on fully polarimetric Radarsat-2 images using snap software to obtain backscatter features VV, HH, and VH. For Sentinel-2 optical images, it is resampled to the same resolution as the Radarsat-2 images and accurately registered;

[0066] a characteristic parameter extraction unit configured to use the scattering matrix to perform Pauli decomposition, [T] matrix construction, Cloude-Pottier decomposition, Van Zyl decomposition, Freeman three-component decomposition, and Yamaguchi four-component decomposition to obtain characteristic parameters such as SAR polarization parameters, polarization coherence matrix, and radar vegetation index;

[0067] A reference FVC calculation unit is configured to calculate vegetation coverage from the pre-processed optical image using the NDVI-DPM model as a reference;

[0068] The land cover classification unit is configured to apply remote sensing land cover classification products to mask non-vegetation feature types and produce vegetation-non-vegetation files to reduce the interference of non-vegetation information.

[0069] The feature parameter optimization unit is configured to apply the ET-GA method to perform feature optimization to obtain the optimal feature parameter combination with the minimum relative error, thereby improving the efficiency of the model;

[0070] Model hyperparameter optimization unit, configured to apply the Black Kite BKA algorithm to optimize the model's hyperparameters to improve the model's robustness and generalization ability;

[0071] The vegetation cover inversion unit is configured to apply the optimal model BAK-KAN for vegetation cover estimation.

[0072] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the program is any of the above-described methods for inverting vegetation cover from SAR images based on multi-dimensional feature optimization and model optimization.

[0073] An embodiment of the present application also provides an electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for inverting vegetation coverage of SAR images based on multidimensional feature optimization and model optimization as described above is implemented.

[0074] Beneficial effects:

[0075] In the SAR image vegetation cover inversion method based on multi-dimensional feature selection and model optimization provided in this application, the fully polarized Radarsat-2 (Single Look Complex) data is first preprocessed, and a variety of SAR feature parameters are extracted through different polarization decomposition methods, including backscatter coefficients VV and VH, polarization coherence matrix, polarization decomposition parameters, radar vegetation index, etc. The Sentinel-2 (Level-2A) image is preprocessed, and the normalized difference vegetation index NDVI (Normalized Difference Vegetation Index) is calculated using the band calculation formula. The reference vegetation cover FVC (Fractional Vegetation Cover) of the study area is obtained by combining the pixel dichotomy model DPM (Dimidiate Pixel Model). To remove interference from non-vegetation information, the European Space Agency's land use product was used to mask non-vegetation features, creating a vegetation-non-vegetation mask file. An improved genetic algorithm (ET-GA) was then applied using XGBoost as the fitness evaluation function, with the root mean square error (RMSE) used as the evaluation metric to select the optimal SAR feature combination. Model training was performed based on this optimal feature set. To improve model robustness, the Black Kite Algorithm (BKA) was used to optimize the model's hyperparameters. Through comparative experiments, the BKA-KAN model was used to estimate the FVC of the study area. This method, by combining multiple feature decomposition methods to extract SAR features, multidimensional feature selection, intelligent algorithm optimization, and deep learning models, addresses core issues in vegetation parameter inversion from SAR data, such as feature redundancy, insufficient accuracy, and low efficiency. This method provides an innovative technical solution for the field of quantitative remote sensing inversion. This method can be applied to dynamic monitoring of vegetation cover in farmland, grassland, and other areas, providing technical support for precision agriculture, ecological monitoring, disaster early warning, and other fields, possessing significant practical significance and promising industrial applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The drawings and descriptions that constitute part of this application are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. Among them:

[0077] Figure 1 A schematic diagram of a process according to the present invention;

[0078] Figure 2 A diagram showing the unit configuration according to the present invention;

[0079] Figure 3 A schematic diagram of the ET-GA algorithm flow according to an embodiment of the present invention;

[0080] Figure 4 A schematic diagram comparing the prediction accuracy of the BKA-KAN model provided by an embodiment of the present invention and other machine learning models;

[0081] Figure 5 This is a schematic diagram of the vegetation coverage inversion results of the study area using the BKA-KAN model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0082] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments. Each example is provided by way of explanation of the present application and does not limit the present application. In fact, it will be clear to those skilled in the art that modifications and variations can be made in the present application without departing from the scope or spirit of the present application. For example, a feature shown or described as part of one embodiment can be used in another embodiment to produce yet another embodiment. Therefore, it is expected that the present application includes such modifications and variations within the scope of the appended claims and their equivalents.

[0083] Exemplary Methods

[0084] like Figure 1 As shown in FIG, the SAR image vegetation cover inversion method based on multi-dimensional feature optimization and model optimization includes:

[0085] In step S101, a fully polarimetric SAR remote sensing image of the target study area is acquired and preprocessed to obtain backscatter coefficients VV, HH, and VH;

[0086] Specifically, a series of preprocessing operations are performed on the Radarsat-2 fully polarimetric data in the target study area to obtain the backscatter coefficients VV, HH, and VH. The preprocessing process primarily includes radiometric calibration, deburst processing, multi-look processing, polarimetric filtering, and terrain correction. Radiometric calibration converts raw digital quantized values ​​into physically meaningful radar backscatter coefficients, ensuring quantitative comparability of data across different time phases. Taking advantage of the strip-wise acquisition characteristics of SLC data, deburst processing eliminates discontinuities between strips, generating seamless, continuous images. To suppress coherent speckle noise, a multi-look processing scheme with one look in range and two looks in azimuth is employed, effectively improving the signal-to-noise ratio while maintaining resolution. Furthermore, a 7×7 windowed Refined Lee filter is used for polarimetric filtering, reducing speckle noise while effectively preserving feature edges and textures. Finally, combined with high-precision DEM data, terrain correction is performed using a range-Doppler model to eliminate geometric distortion caused by topographical fluctuations. After this series of rigorous processing steps, fully polarized SAR data with accurate geometric position and reliable radiation characteristics were finally obtained, including radar reflection information under vertical-vertical (VV), horizontal-horizontal (HH) and vertical-horizontal (VH) polarization modes, which also provided a high-quality data foundation for subsequent polarization decomposition, radar vegetation index calculation and vegetation cover inversion.

[0087] Figure 3 The target study area is shown, which is located in Kaifeng City, Henan Province. The vegetation in this area is mainly composed of winter wheat and garlic.

[0088] In step S102, a plurality of polarization decomposition techniques are applied to extract SAR characteristic parameters to construct a multi-dimensional scattering characteristic set;

[0089] Specifically: First, the polarization ratio parameter is calculated using the SAR backscatter coefficient to enhance the feature expression ability and differentiation effect. The specific formula is as follows:

[0090]

[0091] Among them, σ 0 HH represents the backscatter coefficient of the horizontal transmission-horizontal reception (HH) channel, σ 0 HV It represents the backscatter coefficient of the horizontal transmit-vertical receive (HV) path.

[0092] Pauli decomposition is performed on the SAR scattering matrix to extract the components corresponding to different scattering mechanisms by converting the scattering matrix into the form of Pauli basis vectors k. The specific formula is as follows:

[0093] Paulir =|S HH -S VV | 2

[0094] Pauli b =|S HH +S VV | 2

[0095] Pauli g =|2S HV | 2

[0096] Among them Pauli r 、Pauli b 、Pauli g represent even scattering, surface scattering, and volume scattering, respectively.

[0097] Then, the polarization coherence matrix [T] is constructed using the Pauli basis vectors. The specific formula is as follows:

[0098]

[0099] Where superscripts * and T denote conjugate and transpose, respectively, and <·> denotes spatial or temporal averaging. The total scattered power provides a global measure of scattered energy and is expressed as follows:

[0100] Span=T 11 +T 22 +T 33

[0101] Then, various polarization decomposition techniques are applied, including Cloude-Pottier decomposition, Van Zyl decomposition, Freeman three-component decomposition, and Yamaguchi four-component decomposition.

[0102] By applying Cloude-Pottier to perform eigendecomposition on [T], we can obtain three variables: polarization scattering entropy (Entropy, H), average scattering angle (Alpha Angle, α), and anisotropy (Anisotropy, A). Among them, H represents the randomness of scattering, α represents the type of dominant scattering mechanism, and A represents the relative importance of the secondary dominant scattering mechanism. The specific formula is as follows:

[0103]

[0104] Among them, λ1, λ2, and λ3 are the eigenvalues ​​of the polarization coherence matrix, and e1, e2, and e3 are the corresponding eigenvectors. The combination of H and A can better distinguish different ground types. The combination formula is as follows:

[0105] 1mH1mA=(1-H)(1-A)

[0106] 1mHA=(1-H)A

[0107] H1mA=H(1-A)

[0108] HA=HA

[0109] Van Zyl decomposition considers azimuthally symmetric targets, that is, the correlation coefficient between the co-polarization and cross-polarization ratios is 0. General natural targets such as vegetation and soil meet this condition. Van Zyl decomposition is based on the Pauli basis to decompose the polarization scattering matrix into three scattering mechanisms, namely surface scattering (VZ surf ), even scattering (VZ dbl ) and volume scattering (VZ vol ).

[0110] Freeman three-component decomposition. Based on the work of Van Zyl, Freeman et al. first proposed to use the polarization coherence matrix [T] to establish three scattering mechanism models: surface scattering (Free surf ), Free dbl ), Volume Scattering (Free vol ).

[0111] Yamaguchi four-component decomposition. Using the same even-order scattering model as Freeman decomposition (Yamaguchi four-component decomposition) dbl ) and surface scattering models (Yama surf ) based on the non-reflection symmetry, the helicoid scattering component (Yama helix ) and modify the volume scattering component of vegetation by changing the probability density function of the relevant azimuth angle (Yama vol ).

[0112] Finally, there are three radar vegetation indices as the main features, namely, radar vegetation index (RVI) based on backscatter intensity kim ), Radar Vegetation Index (RVI) based on eigenvalues ​​calculated based on Cloude-Pottier decomposition van ) and based on Freeman decomposition, by calculating surface scattering (Free surf ), Free dbl ) and volume scattering (Free vol Radar vegetation index (RVI) free ), the specific details of the three radar vegetation indices are shown in the following formulas:

[0113]

[0114] In summary, this paper extracts 32 SAR features, including σ 0 HH ,σ 0 VV ,σ 0 HV ,σ 0 HH / σ 0 VV ,σ 0 HH / σ 0 HV ,Pauli r ,Pauli b ,Pauli g ,λ1,λ2,λ3,Span,α,H,A,1mH1mA,1mHA,H1mA,HA,VZ surf ,VZ dbl ,VZ vol ,Free surf ,Free dbl ,Free vol ,Yama surf ,Yama dbl ,Yama vol ,Yama helix ,RVI kim ,RVI van ,RVI free .

[0115] In step S103, a Sentinel-2 optical image of the target study area is obtained, the Normalized Difference Vegetation Index (NDVI) is calculated, and a reference FVC is calculated in combination with the pixel dichotomy model (DPM);

[0116] Specifically: First, Sentinel-2 optical images of the target study area were obtained from the Copernicus Open Data Center (https: / / dataspace.copernicus.eu / ) in quasi-synchronous fashion with Radarsat-2 SAR images. The two data sets were resampled to the same resolution and georeferenced to ensure consistency in spatial range. Then, based on vegetation spectral sensitivity analysis, the red band (B4, central wavelength 665nm) and the near-infrared band (B8, central wavelength 842nm) were selected to construct the Normalized Difference Vegetation Index (NDVI). The reference vegetation cover was then calculated using the NDVI-DPM hybrid model. The calculation formula is:

[0117]

[0118] Where ρ NIR represents the near-infrared band (B8), ρ REDRepresents the red band (B4), NDVI soil represents the NDVI value with a 5% confidence interval, NDVI veg represents the NDVI value with a 95% confidence interval, and FVC refers to the fractional vegetation cover.

[0119] In step S104, the remote sensing land cover classification product is used to mask the non-vegetation feature type and generate a vegetation-non-vegetation mask file to reduce the interference of non-vegetation information.

[0120] Specifically, we used the 2020 10-meter resolution global land cover classification product (https: / / viewer.esa-worldcover.org / worldcover / ) released by the European Space Agency (ESA). This product provides detailed and reliable data support for land cover type classification worldwide. To improve the accuracy and reliability of vegetation cover inversion within the study area, this paper accurately aligned this land cover data with Radarsat-2 and Sentinel-2 imagery acquired within the study area. The classification results were used to effectively remove non-vegetated areas such as buildings and water bodies, generating a vegetation-non-vegetation mask file. This process significantly improved image quality and analysis accuracy.

[0121] In step S105, the ET-GA algorithm is applied to perform feature screening on the SAR multi-dimensional feature set to obtain the optimal feature combination to reduce feature redundancy;

[0122] Specifically, considering that directly using all features for inversion modeling not only increases computational costs but also may introduce noise and redundant information, thereby reducing the model's predictive performance, a genetic algorithm (ET-GA) combining an elite retention strategy with a tournament selection method is applied to optimize SAR features, thereby balancing the algorithm's global search capability and local convergence ability.

[0123] This algorithm is a global optimization algorithm based on the principles of natural selection and genetics. It is widely used in high-dimensional feature selection problems. It gradually optimizes feature subsets by simulating operations such as selection, crossover, and mutation in the biological evolution process. The main steps of applying the ET-GA algorithm are as follows:

[0124] Encoding: Encode the feature set into a binary string, with each bit corresponding to one feature;

[0125] Generate initial population: Randomly generate 100 binary-coded individuals as the initial population, each individual represents a feature subset.

[0126] Fitness evaluation: For each individual, the XGBoost model is used as the fitness function, the root mean square error (RMSE) is used as the evaluation indicator, and the fitness value is calculated using 5-fold cross-validation.

[0127] Selection: During the selection process, the elite retention strategy is used to retain the top 5% individuals with the highest fitness values ​​in each generation to directly enter the next generation, and the remaining individuals are selected using the tournament selection method.

[0128] Crossover and mutation operations: Set a certain crossover probability to perform a two-point crossover operation on the selected individuals, so that the parent individuals produce new individuals; and reverse each bit of the individual code with a certain probability, thereby introducing a new feature combination.

[0129] Population update and iteration: Merge the elite individuals with the newly generated individuals to form a new population and enter the next iteration. The algorithm sets a maximum number of iterations of 30 and terminates early when the fitness value remains unchanged for 10 consecutive generations.

[0130] Figure 4 Flowchart of the ET-GA algorithm implementation.

[0131] In step S106, the BKA hyperparameter optimization algorithm is applied to obtain the optimal hyperparameters of the model, and the optimal model BKA-KAN is used to estimate vegetation coverage;

[0132] Specifically: Hyperparameter optimization (HPO) is a critical step in ensuring model performance. Properly selecting hyperparameters plays a crucial role in the model's convergence speed, generalization ability, and prediction accuracy. The Black-winged Kite Algorithm (BKA) is used to automatically optimize the hyperparameter configurations of multiple network models.

[0133] The Black Kite Algorithm (BKA) is a newly proposed metaheuristic optimization algorithm inspired by the aggressive behavior of black kites during predation and their migratory behavior in response to environmental changes. Unlike traditional optimization methods, the BKA simulates the predation and migration mechanisms of black kites, achieving an efficient combination of global search and local exploitation. In the initial phase of the algorithm, individual positions are initialized by randomly generating a population within the search space. The root mean squared error (RMSE) of a five-fold cross-validation is used as the fitness function to evaluate the performance of each individual. In the attack phase, the algorithm updates individual positions through adaptive perturbations and random search mechanisms, enhancing global search capabilities and exploring potential optimal solutions. In the migration phase, the algorithm introduces a leader strategy and Cauchy mutation strategies, leveraging information from the best-performing individuals in the current population to guide other individuals toward the optimal solution, thereby achieving more refined local exploitation. Throughout the optimization process, the algorithm continuously updates the global optimal solution based on fitness, repeating the attack and migration steps until the maximum number of iterations is reached.

[0134] BKA optimizes the model's hyperparameters and improves the model's robustness. It achieves a balance between breadth search and depth development through attack and migration operations, solving the problem that traditional grid search and random search methods are inefficient in high-dimensional parameter space and are prone to falling into local optimality.

[0135] Kolmogorov-Arnold Networks is a new type of neural network architecture designed based on the Kolmogorov-Arnold Representation Theorem (KART). According to the KART theorem, any continuous multivariate function can be represented as a combination of a series of unary functions, and its mathematical expression is as follows:

[0136]

[0137] Among them, Φ q and is a continuous univariate function. This theory provides a mathematical basis for decomposing high-dimensional complex functions into simple univariate functions, which can effectively capture the nonlinear relationship between input features.

[0138] The core idea of ​​KAN is to replace the fixed activation function in traditional networks with learnable B-splines to significantly improve the model's expressiveness and generalization performance. Its mathematical expression is:

[0139]

[0140] Among them, B i (x) represents the B-spline basis function, ci are the coefficients optimized during the training process. The spline function dynamically adjusts the parameters by minimizing the loss function to achieve accurate fitting of the data and effective modeling.

[0141] Unlike traditional neural networks, KAN introduces a learnable activation function in the hidden layer. This design can more efficiently fit complex nonlinear relationships and significantly enhance the flexibility and interpretability of the model.

[0142] Finally, the feature combinations selected by the ET-GA algorithm were input into the model optimized by the BKA algorithm for training. Through comparative experiments, it was found that the BKA-KAN model achieved highly accurate inversion of FVC by learning the complex nonlinear relationship between features and target variables.

[0143] Using R 2 , RMSE and MAE were used as evaluation indicators to compare the BKA-KAN model with other optimized models to evaluate the performance of the model in predicting FVC. Figure 5 The scatter plots of all model test results are shown, and Figure 6 shows the vegetation cover inversion results of the target study area using the optimal model BKA-KAN.

[0144] This application innovatively uses the ET-GA algorithm to select the optimal feature combination from a high-dimensional SAR feature set. Compared to traditional feature selection methods, the ET-GA algorithm has a powerful global search capability. Simultaneously, the BKA hyperparameter optimization algorithm simulates natural phenomena. By optimizing the model with these methods, the accuracy of the model's FVC inversion is improved, providing a new approach for high-precision remote sensing monitoring of vegetation parameters.

[0145] In summary, the application of remote sensing technology and machine learning in vegetation cover inversion marks a major advancement in vegetation monitoring technology. This not only improves the accuracy of FVC inversion, but is also crucial for ecological and environmental protection, resource management, agricultural planning, urban construction, and climate change response. It has far-reaching social and economic value and broad application prospects.

[0146] Exemplary Systems

[0147] Figure 2This is a schematic diagram of a SAR image vegetation cover inversion system based on multidimensional feature optimization and model optimization, according to an embodiment of the present application. The system includes an image preprocessing unit configured to perform image preprocessing operations on fully polarized Radarsat-2 images using snap software to obtain backscatter features VV, HH, and VH. For Sentinel-2 optical images, they are resampled to the same resolution as Radarsat-2 images and accurately aligned; the feature parameter extraction unit is configured to use the scattering matrix to perform Pauli decomposition, [T] matrix construction, Cloude-Pottier decomposition, Van Zyl decomposition, Freeman three-component decomposition, and Yamaguchi four-component decomposition to obtain feature parameters such as SAR polarization parameters, polarization coherence matrix, and radar vegetation index; the reference FVC calculation unit is configured to apply the NDVI-DPM model to calculate vegetation cover from the preprocessed optical image as a reference; the land cover classification unit is configured to use remote sensing land cover classification products to mask non-vegetation land feature types and create vegetation-non-vegetation files to reduce interference from non-vegetation information. The feature parameter optimization unit is configured to apply the ET-GA method for feature optimization to obtain the optimal feature parameter combination with the minimum relative error, thereby improving the efficiency of the model; the model hyperparameter optimization unit is configured to apply the Black Kite BKA algorithm to optimize the model's hyperparameters to improve the model's robustness and generalization ability; the vegetation cover inversion unit is configured to apply the optimal model BAK-KAN for vegetation cover estimation.

[0148] The SAR image vegetation cover inversion system based on multi-dimensional feature optimization and model optimization provided in the embodiment of the present application can implement the steps and processes of any of the above-mentioned SAR image vegetation cover inversion method embodiments based on multi-dimensional feature optimization and model optimization, and achieve the same technical effects, which will not be repeated here one by one.

[0149] Exemplary devices

[0150] The present application provides an electronic device, including a storage device and a processor, the processor being suitable for executing various programs; and a memory being used to store multiple programs. The electronic device is characterized in that when the memory executes the program on the processor, the method for inverting SAR image vegetation cover based on multidimensional feature optimization and model optimization as described in any one of claims 1 to 7 is implemented.

[0151] Since the steps of SAR image vegetation coverage inversion based on multi-dimensional feature selection and model optimization have been introduced in detail in the specific implementation method example, they will not be described in detail here.

[0152] The processor includes a central processing unit (CPU), a network processor (NP), etc., and may also be a digital signal processor, an application-specific integrated circuit, an off-the-shelf programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The methods, steps, and logic block diagrams disclosed in the embodiments of this application can be implemented or executed. A general-purpose processor may be a microprocessor or any conventional processor.

[0153] The processor can be specifically configured as follows: obtaining fully polarimetric SAR remote sensing images of the target study area and preprocessing them to obtain backscatter coefficients VV, HH and VH; applying a variety of polarimetric decomposition techniques to extract SAR feature parameters to construct a multidimensional scattering feature set; obtaining Sentinel-2 optical images of the target study area, calculating the normalized difference vegetation index NDVI, and calculating the reference FVC in combination with the pixel dichotomy model DPM; applying remote sensing land cover classification products to mask non-vegetation landforms and create vegetation-non-vegetation mask files to reduce interference from non-vegetation information; applying the ET-GA algorithm to perform feature screening on the SAR multidimensional feature set to obtain the optimal feature combination to reduce feature redundancy; applying the BKA hyperparameter optimization algorithm to obtain the optimal hyperparameters of the model, and using the optimal model BKA-KAN to estimate vegetation cover.

[0154] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present application.

[0155] The above-mentioned method according to the embodiment of the present application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or as computer code originally stored in a remote recording medium or a non-temporary machine storage medium downloaded via a network and stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor or programmable or dedicated hardware (such as an ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the single-track synthetic aperture radar interference source location method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown herein.

[0156] Those skilled in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of this application.

[0157] It should be noted that the various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences from the other embodiments. In particular, the device and system embodiments are described briefly because they are generally similar to the method embodiments. For relevant parts, refer to the description of the method embodiments.

[0158] The device and system embodiments described above are merely illustrative. Units not described as separate may or may not be physically separate, and units not described as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiments. Persons of ordinary skill in the art will be able to understand and implement the present embodiments without inventive effort.

[0159] The foregoing description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A SAR image vegetation cover inversion method based on multi-dimensional feature selection and model optimization is characterized by: The steps include: Step S101: Acquire a fully polarimetric SAR remote sensing image of the target study area and perform preprocessing to obtain backscatter coefficients VV, HH, and VH; Step S102, applying multiple polarization decomposition techniques to extract SAR characteristic parameters to construct a multi-dimensional scattering feature set; Step S103, obtaining Sentinel-2 optical images of the target study area, calculating the Normalized Difference Vegetation Index (NDVI), and calculating the reference FVC in combination with the pixel dichotomy model (DPM); Step S104: applying the remote sensing land cover classification product to mask the non-vegetation feature type and creating a vegetation-non-vegetation mask file to reduce the interference of non-vegetation information; Step S105, applying the ET-GA algorithm to perform feature screening on the SAR multi-dimensional feature set to obtain the optimal feature combination to reduce feature redundancy; Step S106: Apply the BKA hyperparameter optimization algorithm to obtain the optimal hyperparameters of the model, and use the optimal model BKA-KAN to estimate vegetation coverage.

2. The SAR image vegetation cover inversion method based on multidimensional feature optimization and model optimization according to claim 1 is characterized in that: In step S101, the pre-processing process includes radiometric calibration, deburst processing, multi-look processing, polarization filtering and terrain correction in sequence; The radiometric calibration converts the original digital quantization value into a radar backscatter coefficient with physical meaning, ensuring the quantitative comparability of data at different time phases; The deburst process is used to eliminate the discontinuity of the splicing between strips and generate a seamless continuous image; Multi-look processing with one look in range and two looks in azimuth is used to suppress speckle noise, effectively improving the signal-to-noise ratio while maintaining resolution. By using a 7×7 window Refined Lee filter for polarization filtering, the edge and texture features of the ground objects are well preserved while reducing speckle noise. Finally, combined with high-precision DEM data, terrain correction is performed using the range-Doppler model to eliminate geometric distortion caused by terrain undulations.

3. The SAR image vegetation cover inversion method based on multidimensional feature optimization and model optimization according to claim 1 is characterized in that: The step S102 specifically includes the following steps: first, the polarization ratio parameter is calculated using the SAR backscatter coefficient to enhance the feature expression capability and the discrimination effect; the specific formula is as follows: Among them, σ 0 HH represents the backscatter coefficient of the horizontal transmission-horizontal reception HH channel, σ 0 HV represents the backscatter coefficient of the horizontal transmission-vertical reception HV channel; The SAR scattering matrix is ​​subjected to Pauli decomposition to extract the components corresponding to different scattering mechanisms by converting the scattering matrix into the form of Pauli basis vectors k. The specific formula is as follows: Pauli r =|S HH -S VV | 2 Pauli b =|S HH +S VV | 2 Pauli g =|2S HV | 2 Among them Pauli r 、Pauli b 、Pauli g represent even scattering, surface scattering, and volume scattering, respectively; Then, the polarization coherence matrix [T] is constructed using the Pauli basis vectors. The specific formula is as follows: Where superscripts * and T denote conjugation and transpose, respectively, and <·> denotes spatial or temporal averaging. The total scattered power provides a global measure of scattered energy and is expressed as follows: Span=T 11 +T 22 +T 32 Then, various polarization decomposition techniques are applied, including Cloude-Pottier decomposition, Van Zyl decomposition, Freeman three-component decomposition, and Yamaguchi four-component decomposition; By applying Cloude-Pottier to perform eigendecomposition on [T], we can obtain three variables: polarization scattering entropy (Entropy, H), average scattering angle (Alpha Angle, α), and anisotropy (Anisotropy, A). Among them, H represents the randomness of scattering, α represents the type of dominant scattering mechanism, and A represents the relative importance of the secondary dominant scattering mechanism. The specific formula is as follows: Among them, λ1, λ2, and λ3 are the eigenvalues ​​of the polarization coherence matrix, and e1, e2, and e3 are the corresponding eigenvectors. The combination of H and A can better distinguish different ground types. The combination formula is as follows: 1mH1mA=(1-H)(1-A) 1mHA=(1-H)A H1mA=H(1-A) HA=HA Van Zyl decomposition considers azimuthally symmetric targets, that is, the correlation coefficient between the co-polarization and cross-polarization ratios is 0. General natural targets such as vegetation and soil meet this condition. Van Zyl decomposition is based on the Pauli basis to decompose the polarization scattering matrix into three scattering mechanisms, namely surface scattering VZ surf , even scattering VZ dbl and volume scattering VZ vol ; Freeman three-component decomposition; using the polarization coherence matrix [T] to establish three scattering mechanism models: surface scattering Free surf 、Even scatteringFree dbl 、Volume ScatteringFree vol ; Yamaguchi four-component decomposition; using the same even-order scattering model as Freeman decomposition dbl and surface scattering model Yama surf On the basis of the non-reflection symmetry, the helical scattering component Yama is introduced. helix , and modify the volume scattering component of vegetation by changing the probability density function of the relevant azimuth angle vol ; Finally, there are three radar vegetation indices as the main features, namely, radar vegetation index (RVI) calculated based on backscatter intensity kim , Radar Vegetation Index (RVI) based on eigenvalues ​​calculated based on Cloude-Pottier decomposition van Based on Freeman decomposition, by calculating the surface scattering Free surf 、Even scatteringFree dbl and volume scattering Free vol Radar Vegetation Index (RVI) free , the specific details of the three radar vegetation indices are shown in the following formulas:

4. The SAR image vegetation cover inversion method based on multidimensional feature optimization and model optimization according to claim 1 is characterized in that: Step S103 specifically includes the following steps: based on the vegetation spectral sensitivity analysis, the red band and the near infrared band are selected to construct the Normalized Difference Vegetation Index NDVI, and then the NDVI-DPM hybrid model is used to calculate the reference vegetation coverage; the calculation formula is: Where ρ NIR represents the near-infrared band, ρ RED Represents the red band, NDVI soil Represents the NDVI value with a 5% confidence interval, NDVI veg represents the NDVI value with a 95% confidence interval, and FVC refers to the fractional vegetation cover.

5. The SAR image vegetation cover inversion method based on multidimensional feature optimization and model optimization according to claim 1 is characterized in that: In step S104, the 2020 10-meter resolution global land cover classification product released by the European Space Agency is used; the land cover data is accurately aligned with the Radarsat-2 and Sentinel-2 images obtained in the study area, and the classification results are used to effectively remove non-vegetation areas such as buildings and water bodies to obtain a vegetation-non-vegetation mask file.

6. The SAR image vegetation cover inversion method based on multidimensional feature optimization and model optimization according to claim 1 is characterized in that: Step S105 specifically includes the following steps: Encoding: Encode the feature set into a binary string, with each bit corresponding to one feature; Generate initial population: Randomly generate 100 binary-coded individuals as the initial population, each individual represents a feature subset; Fitness evaluation: For each individual, the XGBoost model is used as the fitness function, the root mean square error (RMSE) is used as the evaluation indicator, and the fitness value is calculated using 5-fold cross-validation; Selection: During the selection process, the elite retention strategy is used to retain the top 5% individuals with the highest fitness values ​​in each generation to directly enter the next generation, and the remaining individuals are selected using the tournament selection method; Crossover and mutation operations: Set a certain crossover probability to perform a two-point crossover operation on the selected individuals, so that the parent individuals produce new individuals; and reverse each bit of the individual code with a certain probability, thereby introducing a new feature combination; Population update and iteration: Merge the elite individuals and the newly generated individuals into a new population and enter the next iteration; the algorithm sets the maximum number of iterations to 30, and terminates the algorithm early when the fitness value remains unchanged for 10 consecutive generations.

7. The SAR image vegetation cover inversion method based on multidimensional feature optimization and model optimization according to claim 1 is characterized in that: In step S106, the BKA hyperparameter optimization algorithm specifically includes the following steps: In the initial stage, the individual positions are initialized by randomly generating a population in the search space, and the root mean square error of the five-fold cross validation is used as the fitness function value to evaluate the performance of each individual; In the attack phase, the algorithm updates individual positions through adaptive perturbations and random search mechanisms, thereby enhancing global search capabilities and exploring potential optimal solution areas; In the migration phase, the algorithm introduces the leader strategy and the Cauchy mutation strategy, utilizing the information of the best performing individuals in the current group to guide other individuals toward the optimal solution, thereby achieving more refined local development. During the entire optimization process, the algorithm continuously updates the global optimal solution based on the fitness value and repeats the attack and migration steps until the maximum number of iterations is reached.

8. A SAR image vegetation cover inversion system based on multi-dimensional feature selection and model optimization, characterized in that: include: An image preprocessing unit is configured to perform image preprocessing operations on fully polarimetric Radarsat-2 images using snap software to obtain backscatter features VV, HH, and VH; and to resample Sentinel-2 optical images to the same resolution as Radarsat-2 images and perform precise registration; a characteristic parameter extraction unit configured to use the scattering matrix to perform Pauli decomposition, [T] matrix construction, Cloude-Pottier decomposition, Van Zyl decomposition, Freeman three-component decomposition, and Yamaguchi four-component decomposition to obtain characteristic parameters such as SAR polarization parameters, polarization coherence matrix, and radar vegetation index; A reference FVC calculation unit is configured to calculate vegetation coverage from the pre-processed optical image using the NDVI-DPM model as a reference; The land cover classification unit is configured to apply remote sensing land cover classification products to mask non-vegetation features and create vegetation-non-vegetation files to reduce the interference of non-vegetation information; The feature parameter optimization unit is configured to apply the ET-GA method to perform feature optimization to obtain the optimal feature parameter combination with the minimum relative error, thereby improving the efficiency of the model; Model hyperparameter optimization unit, configured to apply the Black Kite BKA algorithm to optimize the model's hyperparameters to improve the model's robustness and generalization ability; The vegetation cover inversion unit is configured to apply the optimal model BAK-KAN for vegetation cover estimation.

9. A storage device storing a plurality of programs, characterized in that: The program application is loaded and executed by a processor to implement the SAR image vegetation cover inversion method based on multi-dimensional feature optimization and model optimization as described in any one of claims 1-7.

10. An electronic device comprising a storage device and a processor; the processor being adapted to execute various programs; and the memory being adapted to store a plurality of programs; characterized in that: When the memory executes the program on the processor, the SAR image vegetation cover inversion method based on multi-dimensional feature optimization and model optimization as described in any one of claims 1 to 7 is implemented.

Images