Fast inversion method of significant wave height from sar range-compressed data

Abstract

The present application belongs to the technical field of marine microwave remote sensing and environmental monitoring, and particularly relates to a method for quickly retrieving sea wave significant wave height based on SAR range-compressed data. The imaging phase in SLC data is inversely compensated by inverse Chirp Scaling processing, and range-direction pulse compression is performed to generate RC data which is only focused in the range direction and unfocused in the azimuth direction. The normalized variance of RC intensity, the Doppler power spectrum statistical parameters extracted along the azimuth direction, the two-dimensional image spectrum orthogonal projection coefficients and the radar incidence angle are extracted. An XGBoost machine learning model is used to establish a nonlinear mapping relationship between the RC domain feature parameters and the significant wave height. Compared with the prior art, the present application can significantly reduce the calculation complexity while realizing accurate retrieval of the significant wave height by constructing a multi-dimensional feature parameter system in the range-compressed domain and combining a machine learning model, and can meet the application requirements of low power consumption and fast response.

Description

Technical Field

[0001] This invention belongs to the field of marine microwave remote sensing and environmental monitoring technology, specifically relating to a method for rapid inversion of the significant wave height of ocean waves based on SAR range compression data. Background Technology

[0002] Significant wave height (SWH) is an important parameter characterizing wavefield energy levels and sea state intensity. Synthetic Aperture Radar (SAR), with its high spatial resolution and wide mapping band, has become a crucial tool for observing sea surface wave structures. Existing SAR-based wave parameter inversion methods, such as wavenumber domain physical inversion, image domain empirical models, and deep learning methods, all rely on standard SAR image products that have already undergone full-focus processing. However, traditional processing methods suffer from high computational and storage resource consumption in practical applications: the complete two-dimensional matched filtering imaging process requires substantial computational resources. In near-real-time operational monitoring scenarios with extremely high timeliness requirements, such as typhoon warnings, the traditional full-focus imaging processing chain is too long, making it difficult to meet the application demands for low power consumption and rapid response. Summary of the Invention

[0003] To address the shortcomings of existing wave inversion techniques that rely excessively on two-dimensional images from fully focused synthetic aperture radar (SAR), resulting in high computational complexity and poor processing timeliness, this invention proposes a wave significant wave height inversion method based on SAR range compression data. By constructing a multi-dimensional feature parameter system in the range compression domain and combining it with a machine learning model, the method achieves accurate inversion of significant wave height while significantly reducing computational complexity.

[0004] The technical solution adopted by this invention to solve its technical problem is as follows: a rapid inversion method for significant wave height of ocean waves based on SAR range compression data, comprising the following steps: S1. SAR Range Compression Data Reconstruction: Acquire SAR single-look complex data, abbreviated as SLC data, and perform quality screening. Perform inverse chirp scaling to compensate the imaging phase in the SLC data and perform range pulse compression to generate RC data that is focused only in the range direction and not in the azimuth direction. S2. Sample set construction: Simultaneously acquire the true effective wave height from the marine environment reanalysis data within the spatiotemporal range of radar observation, and perform spatiotemporal matching of the true effective wave height with the RC data through spatiotemporal interpolation to construct a training set and a validation set; S3. Multidimensional feature parameter extraction: From the RC data, extract the RC intensity normalized variance used to quantify the signal dispersion caused by ocean waves, the Doppler power spectrum statistical parameters extracted along the azimuth direction used to capture the dynamic kinematic features of the sea surface, the two-dimensional image spectrum orthogonal projection coefficient used to characterize the spatial spectral structure information in the range direction, and the radar incident angle as the observation geometric feature. S4. Construction of nonlinear inversion model: Combine the multidimensional feature parameters extracted in step S3 into an input feature vector, take the true value of effective wave height as the target variable, and use the XGBoost machine learning model for training to establish a nonlinear mapping relationship between RC domain feature parameters and effective wave height. S5. Inversion Result Output: Input the multi-dimensional feature parameters corresponding to the RC data of the sea area to be measured into the trained inversion model, and output the effective wave height prediction result.

[0005] Preferably, the inverse Chirp Scaling algorithm in step S1 includes: S101. Perform azimuth FFT transformation on the SLC data to the range-Doppler domain. Multiply the SLC data with the inverse function of the residual phase function to compensate for the azimuth phase shift caused by the range change, and obtain the data with the residual phase removed. S102. Perform range FFT on the data with the remaining phase removed to enter the two-dimensional frequency domain. Multiply the data with the inverse function of the phase function corresponding to range compression and range migration correction to remove the range frequency modulation and migration correction introduced in the imaging process, and obtain the data with range compression and RCMC phase removed. S103. The data with removed range compression and RCMC phase is returned to the range-Doppler domain via range-to-IFFT. By multiplying it with the inverse function of the Chirp Scaling phase function, the range curvature introduced by Chirp Scaling is corrected to obtain the data with removed Chirp Scaling phase. S104. The data with the Chirp Scaling phase removed is processed by azimuth IFFT to obtain the reconstructed SAR raw echo data. S105. Perform range pulse compression processing on the reconstructed SAR raw echo data to generate RC data that is focused only in the range direction.

[0006] Preferably, the spatiotemporal interpolation in step S2 includes: using the imaging time and geographic coordinates of the RC data as a reference, performing spatiotemporal bilinear interpolation on the effective wave height true value of the marine environmental reanalysis data, and combining the spatial distance weights of grid points around the image center to achieve parameter correlation.

[0007] Preferably, the extraction of orthogonal projection coefficients of the two-dimensional image spectrum in step S3 includes: performing a two-dimensional Fourier transform on the RC intensity data to obtain a two-dimensional image spectrum, and projecting the two-dimensional image spectrum onto a preset two-dimensional orthogonal basis function within the effective wavenumber range to obtain 20 orthogonal projection coefficients.

[0008] Preferably, in step S3, the statistical parameters of the Doppler power spectrum extracted along the azimuth direction include the Doppler spectrum center, spectral width, skewness, and kurtosis.

[0009] Preferably, in step S4, a 26-dimensional comprehensive feature vector is obtained through combination for input into the subsequent machine learning model. Its form is expressed as: ; in, The radar incident angle; Normalize the variance of the RC data; Centered on the Doppler spectrum; For Doppler spectral width; Doppler spectral skewness; Doppler spectrum peaks; Represents the spectrum of two-dimensional images The 20 extracted orthogonal projection coefficients.

[0010] Preferably, the XGBoost machine learning model is a regression model based on the limiting gradient boosting tree, employing a squared error loss function, and accumulating... The output of each regression tree yields the final predicted value.

[0011] Preferably, the XGBoost machine learning model further includes: a second-order Taylor expansion of the objective function, using the first-order gradient and the second-order Hessian matrix to guide tree growth; built-in L1 / L2 regularization terms to control model complexity and prevent overfitting; and a histogram-based tree growth algorithm to optimize memory consumption and computational efficiency.

[0012] Preferably, the training and validation of the XGBoost machine learning model adopts a stratified sampling strategy.

[0013] Preferably, the correlation coefficient R between the inversion result of step S5 and the ERA5 reference true value is ≥0.895.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention adopts an inversion path based on unfocused data in the range compression domain, eliminates the azimuth focusing phase through the inverse Chirp Scaling algorithm, skips the complete imaging process, has a short processing link, and can improve data processing speed and computational efficiency.

[0015] 2. This invention constructs a feature system that includes normalized variance, Doppler power spectrum statistical parameters, orthogonal projection coefficients of two-dimensional image spectrum, and radar incident angle. It integrates signal strength, kinematics, spatial spectrum and geometric multi-dimensional information, which can effectively and accurately characterize the state of ocean waves and achieve high-precision effective wave height inversion without the need for a fully focused image.

[0016] 3. This invention uses an XGBoost regression model based on feature engineering to establish a nonlinear mapping from the high-dimensional feature space of the distance compression domain to the effective wave height of the sea surface. The inversion is fast and it is suitable for near real-time operational monitoring scenarios with extremely high timeliness requirements, such as typhoon warnings.

[0017] In summary, this invention, by constructing a multi-dimensional feature parameter system in the distance compression domain and combining it with a machine learning model, achieves accurate inversion of effective wave height while significantly reducing computational complexity, thus meeting the application requirements of low power consumption and fast response. Attached Figure Description

[0018] Figure 1 This is a flowchart of the process of generating RC data from SLC data inversion in this invention.

[0019] Figure 2 This is a schematic diagram illustrating the effective wave height inversion principle based on the XGBoost ensemble learning framework.

[0020] Figure 3 This is a scatter plot of the effective wave height inversion results based on multi-dimensional feature fusion.

[0021] Figure 4 This is a deviation diagram of model inversion under different significant wave heights.

[0022] Figure 5 This is a scatter plot comparing the effective wave height inversion results based on multi-dimensional feature fusion with altimeter data.

[0023] Figure 6 This is a scatter plot comparing the significant wave height retrieved from the CWAVE model with the ERA5 reference value. Detailed Implementation

[0024] To facilitate understanding of the present invention, it will be described in more detail below with reference to the accompanying drawings and specific embodiments. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in this specification. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention.

[0025] The specific technical solutions for the effective wave height inversion method based on SAR range compression data include: 1. Reconstruction and preprocessing of SAR range compressed data (RC data): Acquire SAR single-look complex (SLC) data and perform quality screening. Then, use an inverse reconstruction algorithm (such as the inverse chirp scaling algorithm) to eliminate the azimuth focusing phase. That is, perform inverse chirp scaling processing to inversely compensate the imaging phase in the SLC data and re-execute range pulse compression to generate RC data that is focused only in the range direction and not in the azimuth direction.

[0026] 2. Multi-source data spatiotemporal matching to construct sample sets: Simultaneously acquire marine environmental reanalysis data (such as ERA5 significant wave height) within the spatiotemporal range of radar observation, and construct training and validation sets by spatiotemporal matching with RC data through spatiotemporal interpolation.

[0027] 3. Extract multidimensional feature parameters from RC data: (1) Calculate the normalized variance (NV) of the RC intensity to quantify the signal dispersion caused by ocean waves; (2) Extract the statistical parameters of the Doppler power spectrum (including Doppler spectrum center, spectrum width, skewness and kurtosis) along the azimuth direction to capture the dynamic kinematic characteristics of the sea surface; (3) Extract the orthogonal projection coefficients of the two-dimensional image spectrum to characterize the spatial spectral structure information in the range direction; (4) Extract the radar incident angle as the observation geometric feature.

[0028] 4. Construct a nonlinear effective wave height inversion model: Combine the extracted multidimensional features into an input feature vector, use the true value of the effective wave height as the target variable, and train it using the machine learning model XGBoost to establish a nonlinear mapping relationship between the RC domain feature parameters and the effective wave height.

[0029] 5. Model Validation and Operational Application: Input the RC feature parameters of the sea area to be measured into the trained inversion model to quickly output the effective wave height prediction results.

[0030] The implementation process of this invention mainly includes three core steps: constructing an experimental dataset, extracting features in the distance compression (RC) domain and training the model, and evaluating the model performance.

[0031] I. Constructing the experimental dataset: This invention utilizes spaceborne SAR data and multi-source validation data to construct a high-dimensional sample library, providing physical constraints and truth references for the model.

[0032] (1) Selection of raw data. The experimental data used in this invention mainly consists of three parts: SAR observation data, environmental reference data, and independent validation data. For SAR observation, global ocean wave mode (WV) single-look complex (SLC) data from the European Space Agency's Sentinel-1A / B satellites were selected. This data covers two observation geometries: WV1 (incident angle of approximately 23°) and WV2 (incident angle of approximately 36°). Moreover, the vertical transmission and vertical reception VV polarization mode of the radar waves used in this data has high sensitivity to wave tilt modulation. For environmental reference, ERA5 global reanalysis data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) was used. Effective wave height (SWH) parameters matching the SAR observations were extracted through spatiotemporal interpolation and used as true references for model training and parameter optimization. In addition, to achieve cross-sensor performance evaluation, this invention also introduces the along-track observation data of Jason-2, Jason-3 (Ku band) and Saral (Ka band) radar altimeters from the same period as an independent validation set. The stability and universality of the inversion model in the real ocean environment are cross-validated by high-precision observations at the nadir point.

[0033] (2) Data Quality Screening. Before performing reverse reconstruction, this invention first performs quality control and interference removal on the original samples based on geographical location and statistical indicators. Geographically, data with an absolute latitude greater than 60° are removed using geographical location information to effectively avoid the nonlinear interference caused by sea ice covering high-latitude sea areas year-round on the backscattering characteristics of the sea surface, ensuring that the dataset only contains samples from open sea areas. Statistically, this invention introduces Normalized Variance (NV) as a core indicator for measuring the homogeneity and signal-to-noise ratio of SLC images. By calculating the ratio of the standard deviation to the mean of image intensity, differentiated screening thresholds are set for different observation geometries: for the WV1 mode with an incident angle of approximately 23°, the screening range is set to... For the WV2 mode with an incident angle of approximately 36°, due to the weakened tilt modulation effect and susceptibility to background noise, the filtering range is set to... .

[0034] (3) Spatiotemporal matching. To form a usable matching dataset, spatiotemporal registration of the reconstructed RC data and auxiliary reference data is required. For the gridded ERA5 reanalysis data, a spatiotemporal bilinear interpolation method is used to lock adjacent nodes before and after the imaging time within a 1-hour time window, and parameter correlation is achieved by combining the spatial distance weights of grid points around the image center. For the satellite altimeter data distributed along the orbit, a nearest neighbor matching method is used, with the matching window set to 1 hour in time and 50 km in space. The observation point with the smallest geographical distance from the RC data center is selected, and its measured effective wave height is used as the independent verification true value of the sample.

[0035] (4) RC data reconstruction based on the Inverse Chirp Scaling (ICS) algorithm. To obtain RC data, this invention uses the ICS algorithm to reconstruct Sentinel-1 wave mode SLC data. The Inverse Chirp Scaling algorithm refers to the "reverse operation" used for echo data simulation or interference signal generation based on the linear frequency modulated scaling (LFM) CS imaging algorithm. Specific steps are combined with… Figure 1 understand.

[0036] First, the SLC data Transform to the range-Doppler domain, by using the remaining phase function The reverse Multiplication compensates for the azimuth phase shift caused by changes in range, and the corresponding phase function is shown below: (1); in, For azimuth frequency, For distance to fast time, The imaginary unit; This represents the range-direction frequency modulation slope; At the speed of light, This is the reference slope distance. It is a scaling factor that varies with azimuth frequency, defined as ,in For radar wavelength, This is the satellite's equivalent velocity.

[0037] In elimination Subsequently, the data undergoes a range-wise FFT to enter the two-dimensional frequency domain, and is then processed by the phase function corresponding to range compression and range migration correction (RCMC). The reverse Multiplying to remove range-direction frequency modulation and migration correction introduced during imaging processing, the phase function is as follows: (2); in For distance frequency.

[0038] Subsequently, the data is fed back to the range-Doppler domain via the IFFT, and then compared with the Chirp Scaling phase function. The reverse Multiplication corrects for the distance curvature introduced by Chirp Scaling, and its phase function is shown below: (3).

[0039] Through the above steps and the final azimuth IFFT, the reconstructed SAR raw echo data can be obtained. .

[0040] Based on the reconstructed original echo, this invention further implements standard range-direction pulse compression processing to generate RC data focused only in the range direction. This step is implemented in the range frequency domain using a matched filter. Let the linear frequency modulated signal of the transmitted pulse be... The frequency domain response of the matched filter is Then the distance to the compressed signal It can be calculated using equation (4): (4); in For direction, slow time, and These represent the Fourier transform and inverse Fourier transform of the range direction, respectively. It is the complex conjugate of the matched filter.

[0041] II. Extraction of wave feature parameters in the RC domain: This invention constructs 26-dimensional feature parameters by deeply mining the scattering energy distribution, kinematic Doppler information, and spatial spectral structure in RC data. The specific feature extraction steps are as follows: (1) Normalized variance (NV) of RC data RC data undergoes pulse compression and energy convergence in the range direction, while retaining its original phase history in the azimuth direction. The tilt modulation and hydrodynamic modulation of long-wave sea surface waves cause periodic variations in echo amplitude in the range direction. To quantify this signal dispersion caused by sea waves, this invention calculates the RC intensity map. Normalized variance: (5); in, This represents the variance of RC strength data in the spatial domain; This represents the mean of the RC strength data in the spatial domain.

[0042] (2) Statistical parameters of Doppler power spectrum RC data retains complete azimuth Doppler phase information, which is related to the orbital velocity field of the sea surface scattering cells. This invention performs FFT on the RC data along the azimuth direction and averages it along the range direction to obtain the average Doppler power spectrum. Treating it as a frequency domain energy distribution function, the following statistical characteristics are extracted: Doppler Spectrum Center The centroid frequency shift reflecting the echo energy is calculated from the first-order moment at the origin. (6).

[0043] Doppler spectral width The extent to which the quantized energy broadens in the frequency domain (positively correlated with the intensity of random sea surface motion) is calculated using the second-order central moments. (7); Doppler spectral skewness Calculated using normalized third-order central moments, this characteristic characterizes the asymmetry of the Doppler spectrum distribution, specifically the spectral tailing feature caused by the nonlinear motion of ocean waves. The calculation formula is as follows: (8).

[0044] Doppler spectral peak Calculated from the normalized fourth-order central moment, it describes the concentration of energy distribution in the Doppler spectrum, i.e., the sharpness of the spectral peaks and the tail distribution characteristics. The calculation formula is: (9).

[0045] (3) Orthogonal projection coefficients of two-dimensional image spectrum To characterize the spatial spectral structure in RC data, a two-dimensional Fourier transform is first performed on the RC intensity data to obtain a two-dimensional image spectrum. ,in and These are the range and azimuth wavenumbers, respectively. Then, Projecting over the effective wavenumber range onto an orthogonal basis consisting of Gegenbauer polynomials and trigonometric harmonic functions. Above, calculate the projection coefficients: (10).

[0046] Finally, 20 orthogonal projection coefficients were extracted (denoted as...). This is to comprehensively reflect the spatial structure information of ocean waves at different scales.

[0047] (4) Radar incident angle ( ) The sea surface scattering mechanism is closely related to observation geometry; the same sea state may produce different echo characteristics at different incident angles. Therefore, the radar incident angle... Introduced into the model as a priori geometric parameter to mitigate systematic bias caused by changes in the observation angle.

[0048] III. Construction of an effective wave height inversion model based on XGBoost: This invention establishes a nonlinear mapping from the high-dimensional feature space of the distance compression domain to the sea surface significant wave height (SWH) by constructing a regression model based on the extreme gradient boosting (XGBoost) tree.

[0049] Compared to the traditional gradient boosting decision tree (GBDT), the core mechanism of XGBoost in this invention includes: a second-order Taylor expansion of the objective function, using the first-order gradient and the second-order Hessian matrix to guide tree growth; built-in L1 / L2 regularization terms to control model complexity and prevent overfitting; and a histogram-based tree growth algorithm to optimize memory consumption and computational efficiency.

[0050] The specific modeling and training steps are as follows: (1) Constructing high-dimensional feature vectors The extracted independent features are combined to construct a 26-dimensional comprehensive feature vector for each RC sample, which is then used as input to the subsequent machine learning model. Its form is expressed as: ; in, The radar incident angle; Normalize the variance of the RC data; Centered on the Doppler spectrum; For Doppler spectral width; Doppler spectral skewness; Doppler spectrum peaks; Represents the spectrum of two-dimensional images The 20 extracted orthogonal projection coefficients.

[0051] The target variable of the model This is the true value of the effective wave height of ERA5 after spatiotemporal matching at the corresponding location.

[0052] (2) Integrating learning prediction logic, combined with Figure 2 understand.

[0053] The model uses the squared error loss function to optimize the objective function. In the feature-engineered XGBoost regression model, the first tree serves as the base learner, the second tree fits the first / second derivative information, the Kth tree iteratively optimizes, and finally, the results are accumulated... The output of the regression tree yields the final predicted value. : (11); in, Let k be the function space of the regression tree; k represents the ordinal number of the regression tree, ranging from 1 to K. These are the RC data feature parameters, also known as the 26-dimensional comprehensive feature vector constructed in this invention. During model training, the following regularization objective function is minimized. To improve generalization ability: (12); In the formula, The regularization term controls the complexity of the tree and is used to suppress overfitting while fitting complex nonlinear wavefield features.

[0054] (3) Stratified sampling of dataset and parameter tuning Stratified Sampling Strategy: Addressing the long-tailed distribution of SWH samples (mainly concentrated in the 1.5–4.0m range), this invention employs stratified sampling. SWH samples are binned at 0.5m intervals, and within each bin, the training and validation sets are divided in a 3:1 ratio, ultimately yielding approximately 80,000 training samples and 27,000 validation samples, ensuring sample representativeness across each wave height range.

[0055] Hyperparameter configuration: Randomized search combined with K-fold cross-validation was used for optimization. The optimal parameters were determined as follows: maximum tree depth (max_depth) of 6, row sampling rate (subsample) and column sampling rate (colsample_bytree) of 0.8, learning rate (learning_rate) of 0.05, and number of regression trees (n_estimators) of 3000.

[0056] IV. Model Performance Evaluation and Applicability Analysis: (1) Verification of model inversion accuracy The multi-dimensional feature fusion inversion method proposed in this invention (i.e., fusing 26 dimensions of features including Doppler kinematics features, image texture features, and spatial spectral structure features) demonstrates excellent inversion performance on the validation set. The inversion results of this invention's model are compared with the ERA5 reference ground truth, and the results are shown in [the table below]. Figure 3 As can be seen, its correlation coefficient R reaches 0.903, the root mean square error RMSE is as low as 0.551m, and the scattering index SI is 20.5%. The above evaluation indicators fully demonstrate that the multidimensional physical features extracted by the present invention in the range-compressed RC domain can effectively and accurately characterize the wave state, and achieve high-precision effective wave height inversion without the need for a fully focused image.

[0057] (2) Systematic deviation analysis Analysis of the variation trend of the inversion residual with wave height revealed that the average deviation remained stable within ±0.3m in the SWH range of 2.0–6.0m. However, it showed a weak positive deviation in SWH < 2.0m and turned into a negative deviation in SWH > 8.0m. Figure 4 As shown in the figure. This phenomenon reveals that the data-driven model has a mean regression effect in the sample-scarce interval (at the end of the long-tail distribution), but it also highlights the excellent adaptability of the present invention to the most common sea conditions in the world.

[0058] (3) Independent verification across sensors Cross-validation was performed using Jason-2 / 3 (Ku-band) and Saral (Ka-band) altimeter data on 2351 matched samples. The results are shown in [Figure number missing]. Figure 5 As can be seen, the correlation coefficient R remains above 0.895, and the RMSE is 0.662m. This verifies the robustness of the inversion model under different radar bands and different observation geometries.

[0059] (4) Comparison with the classic CWAVE algorithm in the image domain The method of this invention is compared with the CWAVE empirical model based on full-focus image features, and the results are shown in [the table below]. Figure 6 Experimental results show that while the CWAVE model has slightly higher accuracy on the validation set (R=0.916), it relies on complex azimuth focusing processing. This invention, while completely avoiding the enormous computational burden of full-focus imaging, achieves inversion performance comparable to CWAVE (R=0.903), with both methods exhibiting similar overall inversion performance. The difference lies in the fact that CWAVE-like methods rely on full-focus SAR images, typically requiring steps such as raw echo data downlink, ground reception, Level-0 data decompression, range and azimuth focusing processing, image product generation, product distribution, and secondary inversion at the user end. Typical near-real-time remote sensing product release times are usually 1–3 hours after observation, with some products reaching several hours to 24 hours. In contrast, this invention has lower computational requirements, supports direct on-board processing, and can directly extract feature parameters and complete significant wave height inversion on-board based on range-compressed RC data, only downlinking small amounts of result information such as significant wave height, latitude and longitude, imaging time, and quality indicators. Referring to existing on-board processing and rapid alarm systems, on-board result generation and distribution can be achieved within minutes or even tens of seconds. Therefore, while maintaining similar inversion accuracy, this invention simplifies the traditional hour-level chain of "complete image product generation - user-end inversion" into a minute-level chain of "on-board inversion - result download - server-end distribution," thus demonstrating the advantage of balancing accuracy and business timeliness in inverting significant wave height based on RC data feature parameters.

[0060] Those skilled in the art should recognize that the above embodiments are only used to illustrate this application and are not intended to limit this application. Any appropriate changes and variations made to the above embodiments within the essential spirit and scope of this application fall within the scope of protection claimed in this application.

Claims

1. A method for rapid inversion of significant wave height based on SAR range compression data, characterized in that, Includes the following steps: S1. SAR Range Compression Data Reconstruction: Acquire SAR single-look complex data, abbreviated as SLC data, and perform quality screening. Perform inverse chirp scaling to compensate the imaging phase in the SLC data and perform range pulse compression to generate RC data that is focused only in the range direction and not in the azimuth direction. S2. Sample set construction: Simultaneously acquire the true effective wave height from the marine environment reanalysis data within the spatiotemporal range of radar observation, and perform spatiotemporal matching of the true effective wave height with the RC data through spatiotemporal interpolation to construct a training set and a validation set; S3. Multidimensional feature parameter extraction: From the RC data, extract the RC intensity normalized variance used to quantify the signal dispersion caused by ocean waves, the Doppler power spectrum statistical parameters extracted along the azimuth direction used to capture the dynamic kinematic features of the sea surface, the two-dimensional image spectrum orthogonal projection coefficient used to characterize the spatial spectral structure information in the range direction, and the radar incident angle as the observation geometric feature. S4. Construction of nonlinear inversion model: Combine the multidimensional feature parameters extracted in step S3 into an input feature vector, take the true value of effective wave height as the target variable, and use the XGBoost machine learning model for training to establish a nonlinear mapping relationship between RC domain feature parameters and effective wave height. S5. Inversion Result Output: Input the multi-dimensional feature parameters corresponding to the RC data of the sea area to be measured into the trained inversion model, and output the effective wave height prediction result.

2. The method for rapid inversion of ocean wave significant wave height based on SAR range compression data according to claim 1, characterized in that, The inverse Chirp Scaling algorithm in step S1 includes: S101. Perform azimuth FFT transformation on the SLC data to the range-Doppler domain. Multiply the SLC data with the inverse function of the residual phase function to compensate for the azimuth phase shift caused by the range change, and obtain the data with the residual phase removed. S102. Perform range FFT on the data with the remaining phase removed to enter the two-dimensional frequency domain. Multiply the data with the inverse function of the phase function corresponding to range compression and range migration correction to remove the range frequency modulation and migration correction introduced in the imaging process, and obtain the data with range compression and RCMC phase removed. S103. The data with removed range compression and RCMC phase is returned to the range-Doppler domain via range-to-IFFT. By multiplying it with the inverse function of the Chirp Scaling phase function, the range curvature introduced by Chirp Scaling is corrected to obtain the data with removed Chirp Scaling phase. S104. The data with the Chirp Scaling phase removed is processed by azimuth IFFT to obtain the reconstructed SAR raw echo data. S105. Perform range pulse compression processing on the reconstructed SAR raw echo data to generate RC data that is focused only in the range direction.

3. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, The spatiotemporal interpolation in step S2 includes: using the imaging time and geographic coordinates of the RC data as a reference, performing spatiotemporal bilinear interpolation on the effective wave height true value of the marine environment reanalysis data, and combining the spatial distance weights of grid points around the image center to achieve parameter correlation.

4. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, The extraction of orthogonal projection coefficients of the two-dimensional image spectrum in step S3 includes: performing a two-dimensional Fourier transform on the RC intensity data to obtain a two-dimensional image spectrum, and projecting the two-dimensional image spectrum onto a preset two-dimensional orthogonal basis function within the effective wavenumber range to obtain 20 orthogonal projection coefficients.

5. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, In step S3, the statistical parameters of the Doppler power spectrum extracted along the azimuth direction include the Doppler spectrum center, spectral width, skewness, and kurtosis.

6. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, In step S4, a 26-dimensional comprehensive feature vector is obtained through combination for input into the subsequent machine learning model. Its form is expressed as: ； in, The radar incident angle; Normalize the variance of the RC data; Centered on the Doppler spectrum; For Doppler spectral width; Doppler spectral skewness; Doppler spectrum peaks; Represents the spectrum of two-dimensional images The 20 extracted orthogonal projection coefficients.

7. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, The XGBoost machine learning model is a regression model based on the limit gradient boosting tree, employing a squared error loss function, and accumulating... The output of each regression tree yields the final predicted value.

8. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, The XGBoost machine learning model also includes: a second-order Taylor expansion of the objective function, using the first-order gradient and the second-order Hessian matrix to guide tree growth; built-in L1 / L2 regularization terms to control model complexity and prevent overfitting; and a histogram-based tree growth algorithm to optimize memory consumption and computational efficiency.

9. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, The training and validation of XGBoost machine learning models employ a stratified sampling strategy.

10. The method for rapid inversion of significant wave height based on SAR range compression data according to claim 1, characterized in that, The correlation coefficient R between the inversion result of step S5 and the ERA5 reference true value is ≥0.895.

Images