Satellite-borne microwave scatterometer data enhancement method based on improved adaptive bilateral total variation regularization algorithm and multi-track coverage strategy

Abstract

The invention provides a satellite-borne microwave scatterometer data enhancement method based on an improved self-adaptive bilateral total variation regularization algorithm and a multi-track coverage strategy, belongs to the technical field of satellite remote sensing image processing and application, and aims at overcoming the defects of an existing resolution enhancement processing method. On the basis of satellite-borne microwave scatterometer degradation mechanism modeling, on the basis of a Lorentz norm-based adaptive bilateral total variation (LABTV) regularization algorithm, an adaptive adjustment step length is introduced in a steepest descent method solution, computing resources are saved, adaptive adjustment regularization term parameters are introduced in an iteration process, and the calculation efficiency is improved. The improved adaptive bilateral total variation regularization (LABTV +) algorithm is provided, and the method is also combined with a multi-track synthesis strategy, so that a satellite-borne microwave scatterometer image with lower spatial resolution can be reconstructed into an image with higher spatial resolution, and the image can be better applied to ocean and land quantitative remote sensing.

Description

Technical Field

[0002] The present invention belongs to the technical field of satellite remote sensing image processing and applications. Background Art

[0003] Spaceborne microwave scatterometers are active microwave sensors that can be used to accurately measure the normalized backscattering coefficient of the Earth's surface ( ), and their initial design goal was to measure large-scale ocean surface winds. Currently, spaceborne microwave scatterometers can be mainly divided into three types: fixed fan operation mechanism, rotating pencil operation mechanism, and rotating fan operation mechanism. For spaceborne microwave scatterometers with fixed fan and rotating fan operation mechanisms, wide beams are used to achieve global observations, and for spaceborne microwave scatterometers with rotating pencil and rotating fan operation mechanisms, rotating scans are used to achieve global observations. Due to the advantages of wide observation swath, fast global coverage speed, high measurement accuracy, etc., spaceborne microwave scatterometers play an important role in the field of remote sensing applications. In the field of meteorological forecasting, they are widely used in meteorological research, especially in the monitoring and forecasting of wind fields and disastrous sea conditions, providing real-time sea surface wind field information; in the field of environmental monitoring, they are used for polar sea ice mapping, monitoring, classification, etc.

[0004] The spatial resolution of spaceborne microwave scatterometers depends on factors such as the distance between the satellite and the Earth, microwave frequency, antenna size, and processing of measurement results. For traditional scatterometers, the resolution in the range dimension is processed to be several kilometers using range gates, while the resolution in the azimuth dimension is still determined by the antenna beam width of approximately several tens of kilometers. In order to obtain the required high resolution and use scatterometer data with a regular grid, in 1992, Long et al. proposed the Additive Algebraic Reconstruction Technique (AART), the Multiplicative Algebraic Reconstruction Technique (MART), and the Scatterometer Image Reconstruction (SIR) (see D. G. Long, P. J. Hardin, and P. T. Whiting, “Resolution enhancement of spaceborne scatterometer data,” IEEE IEEE Transactions on Geoscience and Remote Sensing, vol. 31, no. 3, pp. 700–715, May 1993.), and applied these three reconstruction algorithms to a one-dimensional narrow sinc function. The aperture function was selected as a rectangular function, and it was found that all three algorithms could basically reconstruct the attenuated signal components. Moreover, in the presence of noise, the SIR algorithm was more robust than the MART algorithm and the AART algorithm. Based on the SIR algorithm, Long et al. further proposed the SIR with Filter (SIRF) algorithm by combining a hybrid median linear filter, and reconstructed the 3-month SASS (Seasat - A Satellite Scatterometer, SASS) data with a resolution of 0.5°×0.5° to 1 / 30°×1 / 30°, thus obtaining a reconstruction result with less noise. Subsequently, the SIR reconstruction algorithm was also applied to ERS-2 (ERS-2, European Remote - Sensing Satellite-2 Scatterometer) and QuickSCAT (QuickSCAT, Quick Scatterometer) data. For the ERS-2, the aperture function had a Hamming window function with a 3dB width of 50 km, and 30-day (JD 280-310, 1996) ERS-2 scatterometer data was used, thus achieving a resolution-enhanced reconstruction of a 2.225 km high-resolution grid; for the QuickSCAT data, 8-day data (JD 230-237, 2000) was selected, and on the same study area, it was reconstructed from an effective resolution of the aperture function of 7 km×30 km to a 2.225 km high-resolution grid. In 1997, Quinn P. et al. applied the SIRF algorithm (see Sensing, M. E. R., Remund, Q. P., & Long, D.G., “Optimization of SIRF for NSCAT”, 1997.) to the reconstruction of NSCAT (NSCAT, NASA Scatterometer). Using 6-day NSCAT data (JD 276-281, 1996) for the Amazon Basin, a grid reconstruction process of 4.45 km was achieved, and the SIRF reconstructed images all had a high correlation coefficient and a low RMS error, indicating that the addition of the hybrid median linear filter made the trade-off between resolution and noise more obvious.In 2011, Williams et al. (see B. A. Williams and D. G. Long, “Reconstruction from aperture filtered samples with application to scatterometer image reconstruction,” IEEE IEEE Transactions on Geoscience and Remote Sensing , vol. 49, no. 5, pp. 1663–1676, May 2011.) developed a maximum a posteriori (MAP) estimator that considered scatterometer noise for analysis. The MAP and SIR reconstruction algorithms were applied to the Advanced Scatterometer (ASCAT) and SeaWinds scatterometers. The ASCAT slice footprint size was 4.2 km × (20 - 35 km), and the SeaWinds slice footprint size was 6 km × 25 km. Using 4 days of data (JD 201 - 204, 2008), high-resolution grid reconstruction processing of 2.225 km was achieved for two regions, the Amazon region and the Weddell Sea. The results showed that MAP could obtain results consistent with the mature SIR algorithm, but significantly improved the resolution at the cost of noise amplification. In 2019, Long et al. also applied the Backus–Gilbert Inversion (BGI) (see D. G. Long, “Scatterometer backscatter imaging using Backus–Gilbert inversion,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no.6, 1010 pp. 3179–3190, Jun. 2019.) was applied to the resolution enhancement reconstruction of the QuickSCAT scatterometer. In the study, 4-day QuickSCAT data (JD 354-257, 1999) were used, where the swath footprint size of QuickSCAT was 6 km × 25 km. By extracting the actual position, geometry, and SRF of the swath measurements from the Level-1B (L1B) data of QuickSCAT, the reconstruction of a 2.225-km resolution grid in the study area was achieved. Although the signal resolution enhancement of BGI is visually similar to that of SIR, SIR can more effectively suppress noise at low signal-to-noise ratios. Wang et al. (see Q. Wang, T. Yun, and X. Dong, “Super-resolution scatterometer image reconstruction using total variation regularization method,” in Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS) , pp. 3065–3068, Jul. 2014.) developed a total variational (TV) regularization reconstruction (RR) algorithm for HY2-SCAT data with a footprint size of 25 km × 38 km and obtained a reconstructed scatterometer image with a resolution of 0.01° × 0.01°. In addition, Liu et al. (see L. Liu, X. Dong, J. Zhu, et al., “An improved adaptive regularization method for forward-looking azimuth super-resolution of a dual-frequency polarized scatterometer,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 9, no. 6, pp. 2145–2159, Mar. 2016.) The improved adaptive regularization deconvolution (IARD) algorithm and the gradient histogram preservation (GHP) constraint are used to reconstruct the dual-frequency polarized scatterometer (DFPSCAT) with a resolution of about 15 km, and an enhanced image with a resolution of about 2 - 5 km is obtained. However, the current reconstruction method still improves the resolution at the cost of noise amplification. In our previous work (see L. Li, L. Gu, X. Li, T. Jiang and X. Fan, "Adaptive Bilateral-Total-Variation Regularization Algorithm for Enhancing HY2-SCAT Data," IEEE IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1-14, 2024, Art no. 5108514.) The LABTV regularization reconstruction algorithm (LABTV, adaptive bilateral-total-variation regularization algorithm with Lorentzian norm) is used to quadruple the reconstruction of different orbit HY2-SCAT data with a spatial resolution pixel size of 25 km to a spatial resolution pixel size of 6.25 km, effectively removing noise while retaining the texture details of the image. However, since HY-2SCAT is a rotating pencil-beam scatterometer, the sampling density of the data is low, and a single orbit of data cannot achieve higher-resolution enhanced reconstruction. Therefore, while effectively reducing noise amplification and maintaining the detailed information of the image, achieving higher-resolution enhanced reconstruction processing remains a challenge in the resolution enhancement reconstruction of scatterometers.

[0005] So far, many methods for enhancing the resolution of spaceborne microwave scatterometer data have been proposed by scholars at home and abroad. However, there are still some obvious deficiencies: (1) How to balance noise and maintain the detailed information of the image in the reconstruction algorithm; (2) The sampling density of spaceborne microwave scatterometer data cannot achieve higher-resolution enhanced reconstruction processing. Summary of the Invention

[0006] To solve the above problems, the present invention discloses a method for enhancing spaceborne microwave scatterometer data based on an improved adaptive bilateral total variation regularization algorithm and a multi-track coverage strategy, realizing the enhancement processing of the spatial resolution of remote sensing images of spaceborne microwave scatterometers with a rotating fan operation mechanism.

[0007] This method is based on the modeling of the degradation mechanism of spaceborne microwave scatterometers. On the basis of the LABTV (LABTV, adaptive bilateral-total-variation regularization algorithm with Lorentzian norm) regularization algorithm, an adaptively adjusted step size is introduced in the steepest descent method to save computing resources. At the same time, an adaptively adjusted regularization term parameter is introduced during the iteration process to further improve the performance and robustness of the reconstruction algorithm. The proposed improved adaptive bilateral total variation regularization (LABTV+) algorithm also combines a multi-track synthesis strategy, which can reconstruct spaceborne microwave scatterometer images with a lower spatial resolution into images with a higher spatial resolution, making it better applied to ocean and land quantitative remote sensing. The technical solution adopted by the present invention is as follows: The method for enhancing spaceborne microwave scatterometer data based on an improved adaptive bilateral total variation regularization algorithm and a multi-track coverage strategy specifically includes the following steps: Step 1. Data preprocessing: Obtain the data of the spaceborne microwave scatterometer with a rotating fan operation mechanism, and successively perform incident angle normalization and region of interest (ROI) selection processing to obtain the scatterometer data to be processed.

[0008] (a) Method for incident angle normalization: Select the angle within the incident angle range of the spaceborne microwave scatterometer as the normalized incident angle. According to the approximate linear dependence relationship between the backscattering coefficient value and the incident angle, as shown in Equation (1), (1) In the formula, is the backscattering coefficient value, is the incident angle, A is the backscattering coefficient value after normalizing the incident angle to , and B represents the dependence on .

[0009] Determine the estimate of the parameter through the linear regression of the backscattering coefficient according to Equation (1), and then determine the estimated value of according to Equation (2); (2) The backscattering coefficient value after normalizing the incident angle is obtained hereby.

[0010] (b) Select the region of interest according to the latitude and longitude range.

[0011] Step 2: Combine the multi-track synthesis strategy to synthesize and process the data of multiple-track scatterometers: Perform data connection processing on the preprocessed data of different-track scatterometers, that is , so as to improve the sampling density of the data and achieve enhanced reconstruction processing with higher resolution. represents the connection operation, represents the data of different-track scatterometers, amount The value is guaranteed to ensure that at least one data is included in all high-resolution image pixel grids, so as to meet the requirements of the Gröchenig criterion.

[0012] Step 3: Simulate and calculate the spatial response function of the spaceborne microwave scatterometer with a rotating fan-shaped operation mechanism, and obtain the high-resolution image data X; According to the degradation mechanism of the spaceborne microwave scatterometer, there is: (3) In formula (4), Z is the low-resolution observation data, with a size of 1× M , X is the high-resolution image data, with a size of N × N , noise is the Gaussian additive noise, H is the spatial response function matrix; In order to obtain the spatial response function matrix, first calculate the normalized radar cross section (NRCS, Normalized Radar Cross Section), and the calculation formula is (4) In formula (4), and respectively represent the power transmitted and received by the spaceborne microwave scatterometer, is the wavelength, is the antenna pattern, is the scanning area of the spaceborne microwave scatterometer, A represents the unit area scanned by the spaceborne scatterometer, is the loss factor caused by the antenna standing wave, is the slant range, and are the transmission and reception loss factors, is the attenuation of the electromagnetic wave of the spaceborne microwave scatterometer by the atmosphere; Spaceborne scatterometers with a rotating fan-shaped operation mechanism often adopt a pulsed radar system, which emits linearly frequency-modulated pulse signals. At the receiving end, techniques such as pulse compression are used to divide the received signals into multiple range gates. The wide-pulse received signals are compressed into narrow-pulse signals with shorter time and wider bandwidth through a matched filter, that is, slice processing is performed on the corresponding orbital fan-shaped footprints in the range direction through FFT processing.

[0013] The obtained backscattering coefficient value corresponding to each slice is (5) In equation (5), is the gain of the slice filter, h is the spatial response function of each slice. The spatial response function within the slice is quantized into a constant, that is, the spatial response function within the slice range is set to 1, and the SRF outside the range is set to 0; Thus, according to the slice center position and rotation angle in each slice data of the measurement data, the slice position is obtained, and the position relationship corresponding to the slice and the high-resolution image grid is determined. According to whether the pixel falls within the slice range, the spatial response function matrix of the high-resolution image is obtained H , the element of the spatial response function matrix is (6) where .

[0014] Thus, the slice center position, the corresponding rotation angle of the slice, and the spatial response function matrix of the spaceborne microwave scatterometer are obtained. Combining with the backscattering coefficient value corresponding to the slice, high-resolution image data X is obtained according to the above data.

[0015] Step 4: Resolution enhancement processing of multi-track synthetic data of spaceborne microwave scatterometer: An improved adaptive bilateral total variation regularization algorithm is used for resolution enhancement reconstruction processing of multi-track synthetic data of spaceborne microwave scatterometer; the formula of this algorithm is as follows: (7) where is the objective function, is the reconstructed scatterometer image, is the Lorentzian norm function, is the image after multi-track synthesis, is the adaptively adjusted regularization parameter, and the matrix represents moving the high-resolution image X horizontally and vertically by l and m pixels respectively; If the algorithm uses the steepest descent method for solution, then the k -th iteration solution is expressed as: (8) where is the step size adaptively adjusted at the k -th iteration of the steepest descent method, is the descent direction, and the formula is as follows: (9) where is the first-order derivative function of the Lorentzian norm function; The step size (10) where the two step sizes and are calculated according to the following formula: (11) (12) In formulas (11) and (12), and , represents the first-order derivative function of the objective function.

[0016] represents the truncation parameter, and there is (13) In formula (13), (14) (15) In the k -th iteration, the regularization parameter adaptively adjusted is In formula (16), is the regularization term parameter set for the initial iteration, and the adaptive adjustment of the regularization term parameter is achieved by comparing the gap between the regularization term parameters of each iteration and the first iteration.

[0017] Advantages of the present invention: 1. Considering that single-track data cannot achieve higher-resolution enhanced reconstruction processing, the present invention proposes a multi-track synthesis strategy. On the premise of time invariance, multiple track data are combined to improve the sampling density of scatterometer data and achieve higher-resolution enhanced reconstruction processing; 2. Based on the LABTV regularization reconstruction algorithm, the present invention proposes the LABTV+ regularization reconstruction algorithm. First, the step size is adaptively adjusted. In each iteration, the step size can be adaptively adjusted to select the optimal step size to quickly approach the optimal solution, thereby saving more computing resources. And an adaptively adjusted regularization term parameter is introduced to further improve the performance and robustness of the LABTV+ regularization reconstruction algorithm. When the LABTV+ regularization reconstruction algorithm performs resolution enhancement reconstruction processing on the data of the spaceborne microwave scatterometer with a rotating fan operation mechanism, an effective noise suppression effect is achieved, and the texture detail information of the spaceborne microwave scatterometer data can be effectively maintained. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] FIG. 1 is a schematic diagram of the operation mechanism of the spaceborne microwave scatterometer used in the present invention.

[0019] FIG. 2 is a flowchart of the spaceborne microwave scatterometer data enhancement method based on the improved adaptive bilateral total variation regularization algorithm and the multi-track coverage strategy of the present invention.

[0020] FIG. 3 shows the resolution enhancement reconstruction results of the single-track data of the spaceborne microwave scatterometer of the present invention: (a) original measurement data, (b) 25 km single-track grid image, (c) 3.125 km bicubic interpolation image, (d) 3.125 km LABTV regularization reconstruction image, and (e) 3.125 km LABTV+ regularization reconstruction image.

[0021] FIG. 4 shows the resolution enhancement reconstruction results of the multi-track composite data of the spaceborne microwave scatterometer of the present invention: (a) original measurement data, (b) 25 km multi-track composite grid image, (c) 1.5625 km bicubic interpolation image, (d) 1.5625 km LABTV regularization reconstruction image, and (e) 1.5625 km LABTV+ regularization reconstruction image DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The technical solution of the present invention will be further explained in the form of specific embodiments below. Embodiment

[0023] Obtain the FY3E WindRAD data from December 10th to 11th, 2023 ( Figure 1)Using the C-band HH polarization ascending and descending orbits with a total of 4 orbits of data as the data source, the relevant parameters of the FY3E WindRAD C-band HH polarization data are shown in Table 1. After preprocessing the WindRAD data, combined with the multi-orbit synthesis strategy, the data of multiple orbits are synthesized, and the spatial response function matrix corresponding to the WindRAD data is calculated. Finally, the LABTV+ regularization reconstruction algorithm is used for resolution enhancement reconstruction to obtain high-resolution WindRAD image data. The selected study area is located in Hainan Island, China, with diverse landforms and natural features, covering mountains, oceans, islands, plains, etc., which is conducive to verifying the effectiveness of the FY3E WindRAD resolution enhancement reconstruction results. The overall process of this embodiment is referred to Figure 2 。

[0024] Table 1 Step 1. Data preprocessing: For the obtained 4-orbit WindRAD data, incidence angle normalization and ROI selection are performed respectively to obtain the WindRAD data to be processed.

[0025] (a) Incidence angle normalization: As a scatterometer with a rotating fan operation mechanism, the C-band incidence angle range of FY3E WindRAD is 36.1° - 43.4°. Since the backscatter coefficient value has an approximately linear dependence on the incidence angle, the expression is (1) In the formula, is the backscatter coefficient value, is the incidence angle, A is the backscatter coefficient value after normalizing the incidence angle to , and B is the parameter representing the dependence of on . Considering the rotating fan operation mechanism of the spaceborne microwave scatterometer, which has a large incidence angle range and affects the backscatter coefficient value, the backscatter coefficient in the WindRAD data is normalized by the incidence angle. Select , and according to the backscatter coefficient value, through linear regression, the estimation of parameter is determined , so the estimated value of parameter is calculated , and the formula is (2) Therefore, the backscatter coefficient value after incidence angle normalization is obtained.

[0026] (b)ROI Selection: The selected study area is Hainan Island, China and its adjacent areas, with a latitude and longitude range of 18°N - 20.5°N, 108°E - 111.5°E. The study area has diverse geomorphologies and natural features, covering mountains, oceans, islands, plains, etc., which is conducive to verifying the effectiveness of the FY3E WindRAD resolution enhancement and reconstruction results.

[0027] Step 2: Perform multi-track synthesis processing on WindRAD data by combining the multi-track synthesis strategy: (a)Gröchenig Criterion: Considering the Nyquist sampling theorem, when the sampling rate exceeds twice the Nyquist sampling rate of the maximum frequency in the signal, a band-limited function can be completely reconstructed from regularly spaced samples. When the sample spacing is irregular or the aperture function varies with different observations, according to Gröchenig's research, the resolution limit of two-dimensional reconstruction with irregular sampling is based on Gröchenig's - density theory, that is, for an arbitrary irregular grid, it can be parameterized by which describes the maximum rectangular box containing the center position of a single sample. The Gröchenig criterion states that to recover a signal with a wavenumber (spatial frequency) of , the sampling must satisfy (3) That is, the minimum sampling density is higher than the Nyquist sampling density of a given value, or 1 / ln2 times the Nyquist rate of uniformly spaced samples. This theorem shows that in this way, a function can be completely reconstructed from irregular samples.

[0028] (b) Multi-track synthesis strategy: For observation areas where the backscattering characteristics do not change significantly within the time range, such as in studies with relatively slow changes like land and ice evolution processes, according to the Gröchenig criterion, the sampling density of single-track data can no longer support higher-resolution enhancement and reconstruction. Therefore, it is possible to consider combining data from multiple orbital scatterometers to increase the sampling density of the data. For the preprocessed data of different orbital scatterometers, consider the multi-track synthesis strategy to achieve data combination.

[0029] After preprocessing the WindRAD data of 4 orbits obtained, combine the multi-track synthesis strategy and perform joint processing on it, that is , increase the data sampling density. For a high-resolution grid with a pixel spacing of 1.5625 km, basically all grid pixels contain at least one WindRAD data, that is, it satisfies the Gröchenig criterion, and high-resolution enhancement and reconstruction processing with a pixel spacing of 1.5625 km can be achieved.

[0030] Step 3: Simulate and analyze the spatial response function of FY3E WindRAD According to the degradation mechanism of the spaceborne microwave scatterometer, we have: (4) In Equation (4), Z is the low-resolution observation data, with a size of 1× M , X is the high-resolution image data, with a size of N × N , noise is the Gaussian additive noise, H is the spatial response function matrix; Each measurement of the spaceborne microwave scatterometer has a corresponding spatial response function (SRF). The SRF is the contribution of each point on the surface to the entire backscatter measurement and is the combined effect of the antenna gain pattern and the radar ambiguity function. The SRF varies with the antenna azimuth and orbital position. The SRF is affected by multiple factors, mainly including the antenna beam response, the frequency response of the onboard processing based on the fast Fourier transform (FFT), etc. The normalized radar cross-section (NRCS) can be expressed as (5) In Equation (5), and respectively represent the power transmitted and received by the spaceborne microwave scatterometer, is the wavelength, is the antenna pattern, is the scanning area of the spaceborne microwave scatterometer, A represents the unit area scanned by the spaceborne scatterometer, is the loss factor caused by the antenna standing wave, is the slant range, and are the transmission and reception loss factors, is the attenuation of the electromagnetic wave of the spaceborne microwave scatterometer by the atmosphere.

[0031] FY3E WindRAD adopts a pulsed radar system (see J. Shang et al., "Preliminary Performance of the WindRAD Scatterometer Onboard the FY-3E Meteorological Satellite," in IEEE Transactions on Geoscience and Remote Sensing, vol.62, pp. 1-13, 2024, Art no. 5100813.), emits a chirp pulse signal, and at the receiving end, the received signal is de-chirped through pulse compression technology, and the received signal is compressed by a matched filter to obtain a pulse slice signal, which is manifested as performing FFT processing on the received signal, realizing the slicing processing of the fan footprint, and obtaining the backscattering coefficient value corresponding to each slice as (6) In equation (6), is the gain of the slice filter, h is the spatial response function of each slice, and its formula is (7) For WindRAD data, the spatial resolution in the C band is 25 km × 0.25 km. After slicing processing, the fan footprint is very narrow in the range direction; at the same time, the corresponding antenna beam width is very narrow, and the antenna response remains basically unchanged. Quantifying the SRF as a constant, that is, setting the SRF within the slice range to 1 and the SRF outside the range to 0 can significantly reduce the computational complexity, but cause very little loss to the quality of the reconstructed image.

[0032] According to the slice center position and rotation angle in each slice data of the measurement data, the slice position is obtained, and the position relationship corresponding to the slice and the high-resolution image grid is determined. According to whether the pixel points fall within the slice range, the spatial response function matrix of the high-resolution image is obtained H The element of the spatial response function matrix is (8) where .

[0033] According to the slice center longitude and latitude data, the corresponding azimuth data and the spatial response function matrix provided in the FY3E WindRAD data, combined with the slice backscattering coefficient value data of FY3E WindRAD, the high-resolution image data X is reconstructed.

[0034] Step 4. Resolution enhancement reconstruction processing of FY3E WindRAD multi-track synthetic data: According to the on-board microwave scatterometer degradation model, the LABTV+ regularization reconstruction algorithm is used to reconstruct the FY3E WindRAD C-band HH polarization data synthesized from multiple orbits from 25 km × 0.25 km to 1.5625 km, realizing the resolution enhancement reconstruction processing of FY3E WindRAD data.

[0035] To improve the performance and robustness of the LABTV RR algorithm, the LABTV RR algorithm is improved to obtain the LABTV+RR algorithm. The formula is (9) where is the reconstructed scatterometer image, is the Lorentzian norm function, is the image after multi-track synthesis, is the adaptively adjusted regularization parameter, and the matrix represents moving the high-resolution image X by l and m pixels in the horizontal and vertical directions respectively.

[0036] In the present invention, the steepest descent method is used for solution, and the k -th iteration solution of the obtained LABTV+RR algorithm is: (10) In the formula, is the step size adaptively adjusted at the k -th iteration, is the descent direction, and the formula is as follows: (11) (a) Step size adaptive adjustment: In the regularization reconstruction algorithm, considering that the step size determines the distance moved towards the optimal solution in each iteration, an appropriate step size can accelerate the convergence speed, while too large or too small a step size may lead to a slowdown in the convergence speed. Using a fixed step size can quickly approach the optimal solution, but near the optimal solution, it will also be unable to reach the optimal solution due to too large a step size, increasing the computational complexity. To reduce the computational complexity, the BB step size method is considered to adaptively adjust the step size . The BB step size uses the information of the previous step to estimate the current step size, enabling the new step size to better adapt to the curvature of the objective function. Referring to the idea of iterative approximation in the quasi-Newton method, the step size needs to satisfy the requirement (12) (13) In the formula, and .

[0037] Thus, two step sizes are derived, and there are (14) (15) The introduction of the BB step size enables the algorithm to select the optimal step size in each iteration, thus achieving the rapid solution of the objective function and further improving the robustness of the algorithm. To prevent falling into the local optimum, a truncation parameter is introduced, and there are (16) (17) (18) Therefore, the formula for adaptively adjusting the step size is obtained (19) (a) To further improve the algorithm performance, the regularization term parameter is adaptively adjusted. By dynamically adjusting the regularization term parameter, the stability and convergence efficiency of the numerical solution are improved. In each iteration, the regularization term parameter is dynamically adjusted according to the previous regularization term parameter and the iteration step size. The regularization term parameter is adaptively adjusted with the step size, and their relationship formula is: (20) where is the regularization term parameter set for the initial iteration, and the adaptive adjustment of the regularization term parameter is realized by comparing the gap between the regularization term parameter of each iteration and that of the first iteration.

[0038] In the present invention, the up-track data of one track on December 10, 2023 and the combined data of 4 tracks of up and down tracks from December 10 to 11, 2023 are respectively processed by the LABTV regularization reconstruction algorithm and the LABTV+ regularization reconstruction algorithm for resolution enhancement reconstruction. The final reconstruction results of the single-track WindRAD data include the original measurement data, the 25 km single-track grid image, the 3.125 km bicubic interpolation image, the 3.125 km LABTV regularization reconstruction image, and the 3.125 km LABTV+ regularization reconstruction image ( Figure 3), the final reconstruction results of multi-track synthetic WindRAD data include the original measurement data, 25 km multi-track synthetic grid images, 1.5625 km bicubic interpolation images, 1.5625 km LABTV-regularized reconstruction images, and 1.5625 km LABTV+-regularized reconstruction images ( Figure 4 ).

[0039] Step 5. Accuracy evaluation: Since there is no corresponding true-value image for comparative analysis, only the and MG of the results are calculated to evaluate the performance of the reconstruction algorithm. The formulas are as follows: (21) (22) In the formula, the size of image X is M×N, is the pixel in the image, and LL is the linear filter.

[0040] Tables 2 and 3 are the evaluation results of the resolution enhancement reconstruction of single-track and multi-track WindRAD data using different regularized reconstruction algorithms respectively. Visually, the regularized reconstruction results of multi-track synthetic data have a clearer coastline and remove more systematic noise. For single-track WindRAD data, the of the LABTV-regularized reconstruction image is 12.0445 and the MG is 4.6059 dB, while the of the LABTV+-regularized reconstruction image is 12.0721 and the MG is 3.8503 dB. The LABTV+-regularized reconstruction algorithm can maintain the edges and detailed information of the image while enhancing the image resolution and filter out more noise; for multi-track synthetic WindRAD data, the of the LABTV-regularized reconstruction image is 15.3820 and the MG is 4.0454 dB, while the of the LABTV+-regularized reconstruction image is 15.3793 and the MG is 2.5424 dB. The LABTV+-regularized reconstruction algorithm removes more noise components while enhancing the image resolution, and the retained texture information is also more consistent with the original measurement data.

[0041] Table 2 <![CDATA[σ n > MG (dB) Original measurement data 12.3499 3.0695 25 km grid image 10.4705 5.7986 Bicubic interpolation image 12.4672 4.4584 LABTV reconstructed image 12.0445 4.6059 LABTV+ reconstructed image 12.0721 3.8503 Table 3 <![CDATA[σ n > MG (dB) Original measurement data 12.1879 3.0283 25 km grid image 10.7310 4.9996 Bicubic interpolation image 15.6976 4.1259 LABTV reconstructed image 15.3820 4.0454 LABTV+ reconstructed image 15.3793 2.5424 Traditional spaceborne microwave scatterometers have relatively low spatial resolution, generally 25 - 50 km, which cannot meet the detection of some phenomena at the kilometer scale such as mesoscale wind field characteristics, storms, and coastal areas. At the same time, with the increasing demand for land detection, high-spatial-resolution spaceborne microwave scatterometers are needed for the monitoring and detection research of snow cover, glaciers, etc. Developing high-resolution (generally required to be below 10 km) data processing technology based on wide-swath microwave scatterometers has become an important scientific and application requirement. In order to improve the data resolution of spaceborne microwave scatterometers, the present invention proposes a new and effective algorithm for enhancing and reconstructing the spatial resolution of spaceborne microwave scatterometer images. Based on the LABTV regularization reconstruction algorithm, the LABTV+ regularization reconstruction algorithm is proposed, which can achieve the adaptive adjustment of the step size and the regularization term parameter, further improving the performance and robustness of the regularization reconstruction algorithm. In addition, based on the Gröchenig criterion, a multi-track synthesis strategy is proposed to combine WindRAD data from multiple orbits to increase the sampling density of the data and achieve higher-resolution enhanced reconstruction processing of spaceborne microwave scatterometers.

Claims

1. A method for enhancing satellite-borne microwave scatterometer data based on an improved adaptive bilateral total variation regularization algorithm and a multi-track coverage strategy, characterized in that: The steps of this method are as follows: Step 1: Data preprocessing: Obtain the data of the rotating sector operation mechanism satellite-borne microwave scatterometer, perform incident angle normalization and ROI selection processing in sequence, and obtain the scatterometer data to be processed; Step 2: Combine the multi-track synthesis strategy to synthesize the data of multiple track scatterometers: perform data connection processing on the pre-processed data of different track scatterometers, that is, , thereby increasing the sampling density of the data and achieving enhanced reconstruction processing with higher resolution. Indicates a connection operation. represents the scatterometer data of different orbits, amount The value is selected to ensure that all high-resolution image pixel grids contain at least one data, so that it meets the Gröchenig criterion requirements; Step 3, simulating and calculating the corresponding spatial response function of the rotating sector operation mechanism spaceborne microwave scatterometer; According to the degradation mechanism of satellite-borne microwave scatterometer, we have: (3) In formula (3), Z is low-resolution observation data, size 1× M , X is the high-resolution image data, size is N × N , noise is Gaussian additive noise, H is the spatial response function matrix; In order to obtain the spatial response function matrix, the normalized radar backscatter cross section is first calculated, and the calculation formula is: (4) In formula (4), and denote the power transmitted and received by the satellite-borne microwave scatterometer, is the wavelength, is the antenna pattern, is the scanning area of ​​the satellite-borne microwave scatterometer, A represents the unit area scanned by the spaceborne microwave scatterometer, is the loss factor caused by the antenna standing wave, is the slope distance, and are the transmission and reception loss factors, is the atmospheric attenuation of electromagnetic waves from satellite-borne microwave scatterometers; The received wide pulse signal of the satellite-borne microwave scatterometer is compressed by a matched filter to obtain a narrow pulse signal with shorter time and wider bandwidth, which is manifested by slicing the corresponding orbital fan-shaped footprint in the range direction through fast Fourier transform processing; The backscatter coefficient value corresponding to each slice is (5) In formula (5), is the gain of the slice filter, h is the spatial response function of each slice, and the spatial response function within the slice is quantized to a constant, that is, the spatial response function within the slice range is set to 1, and the spatial response function outside the range is set to 0; The slice position is obtained according to the slice center position and rotation angle of each slice data in the measurement data, and the position relationship between the slice and the high-resolution image grid is determined. The spatial response function matrix of the high-resolution image is obtained according to whether the pixel point falls within the slice range. H , the elements of the spatial response function matrix for (6) in, ; Thus, the center position of the slice of the satellite-borne microwave scatterometer, the rotation angle corresponding to the slice, and the spatial response function matrix are obtained, and combined with the backscattering coefficient value corresponding to the slice, the high-resolution image data X is reconstructed according to the above data; Step 4: Resolution enhancement processing of multi-track synthetic data of satellite-borne microwave scatterometer: The improved adaptive bilateral total variation regularization algorithm is used to process the resolution enhancement reconstruction of multi-track synthetic data of spaceborne microwave scatterometer; the formula of the algorithm is as follows: (7) in, is the objective function, is the reconstructed scatterometer image, is the Lorentzian norm function, It is a multi-track composite image. is the adaptive regularization parameter, the matrix Indicates that the high-resolution image is moved horizontally and vertically X move l and m pixels; This algorithm uses the steepest descent method to solve the problem. k The iterative solution is expressed as: (8) in, The steepest descent method k The step size of adaptive adjustment at the iteration, is the descending direction, and the formula is as follows: (9) in, is the first derivative of the Lorentzian norm function; Adaptive step size (10) Among them, the two step lengths and Calculated according to the following formula: (11) (12) In formulas (11) and (12), as well as , is the first-order derivative of the objective function; Represents the truncation parameter, (13) In formula (13), (14) (15) No. k In the iterations, the regularization parameter is adaptively adjusted (16) In formula (16), The regularization term parameters set for the first iteration are adaptively adjusted by comparing the gap between the regularization term parameters of each iteration and the first iteration.

2. The method for enhancing satellite-borne microwave scatterometer data based on improved adaptive bilateral total variation regularization algorithm and multi-orbit coverage strategy according to claim 1, characterized in that: The method for normalizing the incident angle in step 1 is: Select an angle within the range of the incident angle of the satellite-borne microwave scatterometer As the normalized incident angle, according to the approximate linear dependence between the backscattering coefficient value and the incident angle, as shown in formula (1), (1) In the formula, is the backscattering coefficient value, is the incident angle, A is the incident angle normalized to The backscatter coefficient value, B represents right Dependencies; According to formula (1), the parameters are determined by linear regression of the backscattering coefficient Estimates , and then determine according to formula (2) Estimated value of ; (2) Thus, the backscattering coefficient value after normalization of the incident angle is obtained.

3. The method for enhancing satellite-borne microwave scatterometer data based on improved adaptive bilateral total variation regularization algorithm and multi-orbit coverage strategy according to claim 1, characterized in that: In step 1, the longitude and latitude information in the satellite-borne microwave scatterometer data is used to select the ROI according to the longitude and latitude.

Images