Satellite-borne SAR self-focusing method and system for long-time integral bunching mode, computer and storage medium
By employing a self-focusing method, block processing, and multiple algorithms to accurately estimate and compensate for errors, the problem of defocusing in spaceborne SAR images under long-term integral focusing mode was solved, achieving high-quality fine focusing effects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUZHOU AEROSPACE INFORMATION RES INST
- Filing Date
- 2025-11-28
- Publication Date
- 2026-05-12
AI Technical Summary
In long-term integral focusing mode, spaceborne SAR images are prone to defocusing. Traditional autofocusing techniques cannot effectively handle high-order Doppler information and navigation information errors, resulting in a decline in imaging quality.
By employing a self-focusing method, through steps such as block processing, Fourier transform, minimum entropy distance alignment algorithm, Savitzky-Golay filter, and high-order fitting, the distance migration error and phase error are accurately estimated and compensated, thus achieving fine focusing.
With an ultra-long synthetic aperture time, fine focusing is achieved across all scenes, solving the image defocusing problem, improving imaging quality and reliability, and reducing system complexity and cost.
Smart Images
Figure CN122017831A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to synthetic aperture radar (SAR) signal processing technology, specifically to a spaceborne SAR self-focusing method, system, computer, and storage medium for long-duration integral focusing mode. Background Technology
[0002] Spaceborne synthetic aperture radar (SAR) has the advantage of being unaffected by climate and environment, enabling it to continuously and stably acquire surface information under adverse weather conditions such as clouds, fog, rain, snow, and night. This capability allows ultra-high-resolution spaceborne SAR to provide crucial surface observation data even under unfavorable weather conditions [Lan G. Cumming, Frank H. Wong. Synthetic Aperture Radar Imaging: Algorithms and Implementation [M]. Electronic Industry Press, 2012.]. Ultra-high-resolution spaceborne SAR will greatly improve the accuracy and detail of global monitoring, providing richer data resources for Earth science research. Research on ultra-high-resolution spaceborne SAR is of great significance, as it not only enables all-weather, all-time global remote sensing monitoring, enhancing applications in environmental monitoring, natural resource management, disaster assessment and early warning, and other fields, but also provides crucial data support for urban planning, precision agriculture, and other areas.
[0003] Spotlight mode is an important method for achieving ultra-high resolution SAR. Its working principle involves dynamically adjusting the antenna beam direction, keeping the radar line of sight fixed on a specific area for an extended period. This increases the synthetic aperture accumulation time (SAP) over that area, significantly improving azimuth resolution. Currently, the highest publicly disclosed resolution achieved by spotlight mode in spaceborne SAR varies across different satellite systems. For example, Germany's TerraSAR-X satellite achieves 0.25 meters in spotlight mode, the US Capella series satellites achieve 0.3 meters, and my country's GF3 SAR satellite achieves 0.5 meters. Generally, the azimuth resolution of spotlight mode is positively correlated with the SAP. In low-Earth orbit spaceborne SAR, based on simplified azimuth resolution calculations, achieving the currently known highest resolution of 0.25 meters requires a SAP of 4-6 seconds. To obtain true centimeter-level resolution, even longer SAP times and more refined antenna control techniques are needed, leading to the development of long-time integration spotlight mode. This mode is currently known to be carried by my country's Qilu-1 satellite and is an experimental mode. This mode extends the synthetic aperture time to 15-17 seconds, allowing the radar to illuminate the target area for a longer period, thereby acquiring more azimuth information and achieving a maximum azimuth resolution of 0.025 meters. This prolonged staring further improves the resolution, providing more detailed surface images.
[0004] Figure 1This is a simplified geometric diagram representing the long-duration integral spotting mode, where Ls represents the synthetic aperture length corresponding to the synthetic aperture time Ts, T represents the target point, R0 represents the target's shortest slant range, and θ is the target's accumulation angle. In the long-duration integral spotting mode, SAR can achieve higher resolution due to the extended synthetic aperture time. However, this mode has a lower tolerance for errors; any small error will be amplified during imaging. First, in this mode, the Doppler information required for imaging is not limited to the second-order Doppler modulation frequency but also includes higher-order Doppler information. Traditional frequency-domain imaging algorithms are highly dependent on Doppler information; if higher-order Doppler information is not processed accurately, it will lead to severe defocusing of the image. Second, the long-duration integral spotting mode has extremely high requirements for the accuracy of the satellite's GPS navigation information. This is because accurate navigation information is needed during imaging to compensate for errors caused by platform motion; if the navigation information has errors or insufficient accuracy, the image will also exhibit defocusing. Furthermore, the synthetic aperture time (SAP) of long-duration integral spotting mode is long, typically reaching tens of seconds. Spaceborne SAR has a long operating range, and the accumulated angle for targets in the scene can reach 5-6 degrees. Compared to the 1-2 degrees corresponding to the 3-5 second SAP of typical spaceborne spotting, there is a significant angle change. This large angle change means that the direction of illumination on a target varies considerably at different times. This can cause variations in the target's scattering intensity within the SAP time. Different targets have different angle-sensitive scattering characteristics, making error estimation and compensation more difficult.
[0005] The advantage of autofocus technology lies in its elimination of the need for additional sensors to measure platform motion errors. Instead, it achieves fine image processing by analyzing the SAR received data itself. This method not only improves imaging accuracy and reliability but also reduces system complexity and cost. However, traditional autofocus technology cannot be directly applied to long-duration integrating focus modes.
[0006] In summary, to address the issue of defocusing in the novel long-duration integral focusing mode imaging, it is urgent to conduct targeted research on autofocus technology to ensure the quality of ultra-high resolution SAR images and provide high-quality products for subsequent target-level applications. Summary of the Invention
[0007] The purpose of this invention is to provide a spaceborne SAR self-focusing method, system, computer, and storage medium for long-duration integral focusing mode.
[0008] The technical solution to achieve the purpose of this invention is as follows: a spaceborne SAR self-focusing method for long-duration integral focusing mode, comprising the following steps:
[0009] Step 1: For single-view complex images at the SARL1A level, perform full image segmentation based on the set azimuth and range segmentation sizes;
[0010] Step 2: Calculate the power value for any block of image, calculate its contrast based on the power distribution within the block, sort the contrast of the block images, and record the block with the highest contrast among all blocks as the strong scattering region; or sort the mean power of the block images, and record the block with the highest power value among all blocks as the strong scattering region.
[0011] Step 3: Perform Fourier transform on the strong scattering region slice to obtain azimuth frequency domain data. Using out-of-band signal energy as clutter energy, filter the spectrum with high signal-to-clutter ratio. Use the minimum entropy range alignment algorithm to obtain the range alignment curve of the whole frequency band. Compensate to the azimuth frequency domain data to complete range alignment. At the same time, use polynomial fitting to obtain the range migration correction curve of the whole scene for the range alignment curve of the whole frequency band.
[0012] Step 4: Apply the minimum entropy phase correction algorithm to the strong scattering region slice after distance alignment for phase autofocusing processing to obtain the fine focusing result and phase compensation error of the strong scattering region slice data. Plot the phase error compensation curve of the strong scattering region, apply Savitzky-Golay filter for smoothing and then perform high-order fitting to obtain the phase error compensation curve of the whole scene and the whole frequency band.
[0013] Step 5: Perform a two-dimensional Fourier transform on the data of the entire scene to obtain dual-frequency domain data. Apply the range migration correction curve of the entire scene to perform range envelope alignment. Then, perform an inverse range Fourier transform to the range Doppler domain. Apply the phase error compensation curve of the entire scene and the entire frequency band to perform phase error correction on the entire frequency band. Transform the data back to the image domain through the azimuth Fourier transform, and at the same time achieve azimuth compression to obtain the result after fine focusing.
[0014] Furthermore, for any image block, the statistical power value is calculated, and its contrast is calculated based on the power distribution within the block. The calculation formula is as follows:
[0015]
[0016] in This represents a segmented image, where m represents the azimuth sampling point and n represents the range sampling point. This represents the statistical power value of the image block. This indicates that the mean value is calculated, where M represents the total number of azimuth sampling points in the image block, and N represents the total number of range sampling points in the image block.
[0017] Let be a constant, and let it be... Then the above equation simplifies to
[0018] .
[0019] Furthermore, the contrast of the image blocks is sorted, and the block with the highest contrast is denoted as the strong scattering region; or the mean power of the image blocks is sorted, and the block with the highest power value is denoted as the strong scattering region, where:
[0020] The contrast of the segmented image is sorted, and the segment with the highest contrast among all segments is recorded as the strong scattering region. This method is suitable for scenes with many strong points, including urban areas and ports.
[0021] The mean power of the segmented image is sorted, and the segment with the highest power value is identified as the strong scattering region. This method is applicable to uniform scenes, including mountainous areas and forests.
[0022] Furthermore, a Fourier transform is performed on the slice of the strong scattering region to obtain the azimuth frequency domain data. The specific method is as follows:
[0023] A slice of the strongly scattering region in a single-view complex image is denoted as... Where t represents the azimuth information, Representing range information, the following Fourier transform yields the azimuth frequency domain data:
[0024]
[0025] in Let represent the Fourier transform, n represent the range sampling points, m represent the azimuth sampling points, R represent the slant range information of the target, λ represent the wavelength, c represent the speed of light, and σ represent the scattering intensity. and These represent the range migration and phase error that remain after imaging processing due to various types of errors, respectively.
[0026] Furthermore, the azimuth spectrum of the strong scattering region slice is analyzed. Using out-of-band signal energy as clutter energy, spectra with high signal-to-clutter ratios are selected, and the minimum entropy range alignment algorithm is used to obtain the range alignment curve for the entire frequency band. Range alignment is completed by compensating for the azimuth frequency domain data. The specific method is as follows:
[0027] The full-band range alignment curve obtained through the minimum entropy range alignment algorithm The unit is pixels, which are converted to distance-time according to the relationship between pixels and distance-time, and expressed as:
[0028]
[0029] in This represents the distance-time corresponding to the distance alignment curve; hereinafter, we will use "distance alignment curve time" as a metonym. Indicates the distance sampling rate;
[0030] Time to align with the distance curve By applying a Hampel filter and utilizing a median-based nonlinear filtering method, outliers in the range alignment curve are identified and removed while maintaining the basic shape, resulting in the filtered range alignment curve time. Subsequently Perform high-order fitting;
[0031] Let the fitting coefficient be p, the azimuth bandwidth be Ba, the pulse repetition frequency be PRF, the azimuth sampling points of the slice data be M, and the azimuth sampling points corresponding to the azimuth bandwidth be na. Then the range alignment curve of the processed full-bandwidth data is obtained as follows: :
[0032]
[0033] in This represents a polynomial calculation function with x as the fitting coefficient, y as the independent variable, and z as the number of sampling points. This represents the floor function;
[0034] Align the processed distance obtained above with the curve time. Compensation applied to the azimuth frequency domain data:
[0035]
[0036] at this time This item should be basically corrected to zero.
[0037] Furthermore, for the range alignment curves across the entire frequency band, a polynomial fitting method is used to obtain the range migration correction curves for the entire scene, and range alignment processing is performed. The formula is as follows:
[0038]
[0039] Where Ma represents the number of azimuth sampling points for the entire scene data, and Na represents the number of azimuth sampling points corresponding to the entire scene bandwidth Ba.
[0040] Furthermore, phase error compensation curves for the strong scattering region are plotted, and after smoothing with a Savitzky-Golay filter, high-order fitting is performed to obtain phase error compensation curves for the entire scene and frequency band. The specific method is as follows:
[0041] set up To obtain the phase error compensation curve for this strong scattering region, a high-order fit is performed after smoothing using a Savitzky-Golay filter. Let the fitting coefficient be q. Since the azimuth frequency domain data of the entire scene and the azimuth frequency domain data of a portion of the scene have the same bandwidth, they are essentially equivalent to having the same frequency domain error. The phase error compensation curve for the entire scene and the entire frequency band is then... :
[0042]
[0043] A spaceborne SAR autofocusing system for long-duration integral focusing mode is provided. The system implements the aforementioned spaceborne SAR autofocusing method for long-duration integral focusing mode, achieving autofocusing of the spaceborne SAR in this mode. The system is divided into five modules, each executed separately:
[0044] Module 1 performs block processing on single-view complex images at the SARL1A level, based on the set azimuth block size and range block size.
[0045] Module 2 calculates the power value of any block image, calculates its contrast based on the power distribution within the block, sorts the contrast of the block images, and identifies the block with the highest contrast among all blocks as a strong scattering region; or sorts the mean power of the block images and identifies the block with the highest power value among all blocks as a strong scattering region.
[0046] Module 3 performs Fourier transform on the strong scattering region slice to obtain azimuth frequency domain data. Using out-of-band signal energy as clutter energy, it filters out spectra with high signal-to-clutter ratio and uses the minimum entropy range alignment algorithm to obtain the range alignment curve of the entire frequency band. It then compensates the azimuth frequency domain data to complete the range alignment. At the same time, it uses polynomial fitting to obtain the range migration correction curve for the entire scene from the range alignment curve of the entire frequency band.
[0047] Module 4 applies the minimum entropy phase correction algorithm to the strong scattering region slice after distance alignment for phase autofocusing processing, obtains the fine focusing result and phase compensation error of the strong scattering region slice data, plots the phase error compensation curve of the strong scattering region, applies Savitzky-Golay filter for smoothing and performs high-order fitting to obtain the phase error compensation curve of the whole scene and the whole frequency band.
[0048] Module 5 performs a two-dimensional Fourier transform on the data from the entire scene to obtain dual-frequency domain data. It then applies the range migration correction curve of the entire scene for range envelope alignment, and performs an inverse range Fourier transform to the range Doppler domain. Finally, it applies the phase error compensation curve of the entire scene and frequency band to correct the phase error of the entire frequency band. The data is then transformed back to the image domain through an azimuth Fourier transform, while simultaneously achieving azimuth compression to obtain the finely focused result.
[0049] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned spaceborne SAR autofocusing method for long-duration integral focusing mode, thereby achieving spaceborne SAR autofocusing for long-duration integral focusing mode.
[0050] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the aforementioned spaceborne SAR autofocusing method for long-duration integral focusing mode is implemented, thereby achieving spaceborne SAR autofocusing for long-duration integral focusing mode.
[0051] Compared with existing technologies, the significant advantages of this invention are: 1) It can accurately estimate and compensate for range migration error and phase error in the entire scene under imaging error conditions, resulting in finely focused images. Compared with traditional autofocusing technology (directly applying range alignment algorithm + PGA algorithm), this invention can achieve fine focusing of the entire large scene image under ultra-long synthetic aperture time conditions of more than 15 seconds, and can be compiled into software as a plug-in to be embedded into the imaging processing program of spaceborne SAR long integral spotting mode for automatic execution, without repeated parameter adjustment, which can effectively solve the SAR image defocusing problem caused by insufficient navigation information accuracy, parameter calculation error and other factors. 2) The range alignment based on partial frequency bands and the full-scene migration error estimation method based on Hample filtering and high-order fitting can effectively solve the problem of large estimation error of traditional range alignment algorithm under ultra-long synthetic aperture time conditions. Attached Figure Description
[0052] Figure 1 A simplified geometric diagram of the long-time integral clustering mode.
[0053] Figure 2 The flowchart shows the minimum entropy distance alignment algorithm.
[0054] Figure 3 This is a flowchart of the minimum entropy phase correction algorithm.
[0055] Figure 4 Flowchart of the full-scene fine-focusing algorithm.
[0056] Figure 5 This is a flowchart of a spaceborne SAR autofocusing method for long-duration integral focusing mode.
[0057] Figure 6 This is the original image of a long-term integral clustering mode of a scene from the Qilu-1 satellite.
[0058] Figure 7 The results of traditional self-focusing technology for a long-term integral focusing mode of Qilu-1 satellite are shown.
[0059] Figure 8This invention provides a finely focused image for a scene from the Qilu-1 satellite, using a long-term integral focusing mode.
[0060] Figure 9 Images comparing focusing effects in multiple scenes. Detailed Implementation
[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0062] like Figure 5 As shown, the present invention provides a spaceborne SAR self-focusing method for long-duration integral focusing mode, and the specific implementation steps are as follows:
[0063] Step 1: Full-scene image segmentation
[0064] For SARL1A level single-view complex images, the entire image is first divided into blocks according to the predefined azimuth and range block sizes. The block size can be flexibly changed according to the specific image size and requirements.
[0065] Step 2: Image power block statistics
[0066] Obtain any block of data from step 1, calculate its power value, and calculate its contrast based on the power distribution within the block. The calculation formula is as follows:
[0067]
[0068] in This represents a block of image data, where m represents the azimuth sampling point and n represents the range sampling point. This represents the statistical power value of the image block. This indicates that the mean value is calculated, where M represents the total number of azimuth sampling points in the image block, and N represents the total number of range sampling points in the image block. As can be seen from the formula... Let be a constant, and let it be... Then the above formula can be simplified to
[0069]
[0070] Step 3: Screening for strong scattering regions
[0071] This invention provides two methods for filtering strong scattering regions. The first method involves sorting the contrast of the image blocks obtained in step 2, identifying the block with the highest contrast as the strong scattering region. The second method involves sorting the mean power of the image blocks, identifying the block with the highest power value as the strong scattering region. Each method has its advantages. The contrast-based method is primarily suitable for scenes with many strong points, such as urban areas, ports, and airports, while the power-based method is primarily suitable for uniform scenes, such as mountainous areas, woodlands, deserts, forests, and farmland.
[0072] Screening strong scattering regions is necessary for two reasons. First, because the long integral spotting mode has a large amount of data, the computational workload of directly estimating the range migration error and phase error of the entire scene is too large and the efficiency is very low. Second, because the target quality in strong scattering regions is high, it is suitable for error estimation. It is less affected by weak scattering regions and can estimate the error value more accurately.
[0073] Step 4: Acquisition of azimuth frequency domain data
[0074] The strongly scattering region slice of the single-view complex image obtained in step 3 is denoted as Where t represents the azimuth information, This slice represents range information. Since it's image domain data, a Fourier transform is needed to obtain the azimuth frequency domain data, as follows:
[0075]
[0076] in Let represent the Fourier transform, n represent the range sampling points, m represent the azimuth sampling points, R represent the slant range information of the target, λ represent the wavelength, c represent the speed of light, and σ represent the scattering intensity. and These represent the range migration and phase error that remain after imaging processing due to various types of errors, respectively.
[0077] Step 5: Distance migration correction based on partial spectrum
[0078] As described in the background section, the synthetic aperture time of long-time integral focusing mode is long, and the accumulation angle of the target in the scene can reach 5-6 degrees. The scattering intensity of the target may vary across the entire frequency band, and there may also be partial frequency band gaps due to the illumination angle. Traditional range alignment algorithms are highly dependent on the data signal-to-noise ratio. Problems such as missing frequency bands and large variations in frequency band scattering characteristics can cause range alignment algorithms based on full-band data to fail. To address this issue, this invention proposes a range migration correction method based on a portion of the spectrum. First, the azimuth spectrum obtained in step 4 is analyzed. Taking the out-of-band signal energy as clutter energy, a spectrum with a high signal-to-clutter ratio is obtained. Range alignment processing is performed on this portion of the frequency band, and the minimum entropy range alignment algorithm is used to obtain the range alignment curve for the entire frequency band. The flowchart of the algorithm is as follows: Figure 2 As shown.
[0079] The full-band range alignment curve obtained through the minimum entropy range alignment algorithm The unit is pixels. Converting it to distance-time according to the relationship between pixels and distance-time can be expressed as:
[0080]
[0081] in This represents the distance-time corresponding to the distance alignment curve; hereinafter, we will use "distance alignment curve time" as a metonym. This represents the range sampling rate. The time for the range alignment curve is also considered. It applies a Hampel filter, using a median-based nonlinear filtering method to identify and remove outliers in the range alignment curve while maintaining its basic shape, resulting in the filtered range alignment curve time. The purpose of this step is to remove the range migration estimation error caused by the inconsistency in scattering intensity within the frequency band.
[0082] Then on A high-order fit is performed, the fitting order of which can be adjusted according to the specific data. Let the fitting coefficient be p, the azimuth bandwidth be Ba, the pulse repetition frequency be PRF, the azimuth sampling points of the slice data be M, and the azimuth sampling points corresponding to the azimuth bandwidth be na. Then the range alignment curve of the processed full-bandwidth data can be obtained as follows: :
[0083]
[0084] in This represents a polynomial calculation function with x as the fitting coefficient, y as the independent variable, and z as the number of sampling points. This represents the floor function. Finally, the processed distance obtained above is aligned with the curve time. Compensation applied to the azimuth frequency domain data:
[0085]
[0086] at this time This item should be basically corrected to zero.
[0087] Step 6: Calculation of distance migration correction curve for the entire scene
[0088] Obtained in step 5 The range alignment curve for this strong scattering region is shown. Since the azimuth frequency domain data of the entire scene and the azimuth frequency domain data of a part of the scene have the same bandwidth, they are essentially equivalent to having the same frequency domain error. Therefore, the distance migration correction curve for the entire scene can be obtained using polynomial fitting:
[0089]
[0090] Where Ma represents the number of azimuth sampling points for the entire scene data, and Na represents the number of azimuth sampling points corresponding to the entire scene bandwidth Ba. In fact... It can be approximated as to An interpolation calculation is performed, and the method of this invention uses a polynomial fitting interpolation method.
[0091] Step 7: Phase compensation and focusing in strong scattering regions
[0092] Step 5 applies the minimum entropy phase correction algorithm to the aligned data for phase autofocus processing. The minimum entropy phase correction algorithm is an adaptive phase error estimation algorithm based on overall image information. Essentially, it solves an optimization problem to find the best image focus by compensating for the phase. The focus effect of an image can be measured by its entropy; the better the focus, the lower the image entropy, and the clearer the image. Therefore, the optimization criterion of this algorithm is to minimize the image entropy value, and under this criterion, it searches for the optimal result of the compensated phase. The flowchart of this algorithm is as follows: Figure 3 As shown. This algorithm can obtain fine focusing results and phase compensation errors for slice data of strong scattering regions. .
[0093] Step 8: Calculation of phase error compensation curve for the entire scene
[0094] Obtained in step 7 The phase error compensation curve for this strong scattering region is typically obtained using the minimum entropy phase correction algorithm. The curve usually doesn't exhibit abrupt outliers, only jagged fluctuations. This can be smoothed using a Savitzky-Golay filter before high-order fitting. The fitting order can be adjusted based on the specific data. Let the fitting coefficient be q. Since the azimuth frequency domain data for the entire scene and the azimuth frequency domain data for a portion of the scene have the same bandwidth, they are essentially equivalent to having the same frequency domain error. Therefore, the phase error compensation curve for the entire scene and the entire frequency band can be obtained as follows: :
[0095]
[0096] Step 9: Full-scene fine-tuning
[0097] Perform a two-dimensional Fourier transform on the data from the entire scene to obtain dual-frequency domain data, and apply the data obtained in step 6. Perform range envelope alignment, then perform inverse range-to-Fourier transform to the range-Doppler domain, and apply the results obtained in step 8. Phase error correction is performed across the entire frequency band, and finally the data is transformed back into the image domain through azimuth Fourier transform, while azimuth compression is achieved to obtain the finely focused result.
[0098] Example
[0099] To verify the effectiveness of the present invention, a full-scene fine focusing processing experiment was conducted on the measured data of the long-time integral focusing mode of the Qilu-1 SAR satellite. The results show that, compared with traditional autofocus technology, the present invention can more effectively solve the image defocusing problem under ultra-long synthetic aperture time and obtain better focusing results.
[0100] Table 1 shows the data parameters for this scene; the synthesis time for the aperture is 15.2 seconds. Figure 6 The original image of the entire scene. Figure 7 This represents the focusing result of traditional autofocus techniques (directly applying distance alignment algorithms + PGA algorithms). Figure 8 This is the result of fine focusing using the technology of this invention. Because the full-scene image is large, details cannot be well displayed in the large image. Figure 9 The comparison results of three small scenes are presented. Visually, it can be seen that the technical method of this invention produces the best focused image effect and basically solves the problem of severe defocus in the original image.
[0101] Image entropy can be used to measure the focus quality of an image. The image entropy of a focused image is typically less than the image entropy of a defocused image. The lower the image entropy, the more concentrated the image energy, and the higher the focus quality. The definition of image entropy is as follows:
[0102]
[0103] Where, x ij This represents the grayscale value of the pixel in the i-th row and j-th column of the image. Representing an image The entropy.
[0104] Calculate according to the above definition Figure 9 The image entropy of the three different methods in the three scenarios shows that the image entropy of the present invention is the smallest and the focusing effect is the best.
[0105] Table 1 Data parameters of a certain Qilu-1 satellite
[0106]
[0107] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0108] The embodiments described above are merely illustrative.
[0109] Several embodiments of this application have been described in detail, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.
Claims
1. A spaceborne SAR self-focusing method for long-duration integral focusing mode, characterized in that, The steps are as follows: Step 1: For single-view complex images at the SARL1A level, perform full image segmentation based on the set azimuth and range segmentation sizes; Step 2: Statistically calculate the power value for any block of image, calculate its contrast based on the power distribution within the block, sort the contrast of the block images, and record the block with the highest contrast among all blocks as the strong scattering region. Alternatively, the mean power of the segmented images can be sorted, and the segment with the highest power value among all segments can be denoted as the strong scattering region. Step 3: Perform Fourier transform on the strong scattering region slice to obtain azimuth frequency domain data. Using out-of-band signal energy as clutter energy, filter the spectrum with high signal-to-clutter ratio. Use the minimum entropy range alignment algorithm to obtain the range alignment curve of the whole frequency band. Compensate to the azimuth frequency domain data to complete range alignment. At the same time, use polynomial fitting to obtain the range migration correction curve of the whole scene for the range alignment curve of the whole frequency band. Step 4: Apply the minimum entropy phase correction algorithm to the strong scattering region slice after distance alignment for phase autofocusing processing to obtain the fine focusing result and phase compensation error of the strong scattering region slice data. Plot the phase error compensation curve of the strong scattering region, apply Savitzky-Golay filter for smoothing and then perform high-order fitting to obtain the phase error compensation curve of the whole scene and the whole frequency band. Step 5: Perform a two-dimensional Fourier transform on the data of the entire scene to obtain dual-frequency domain data. Apply the range migration correction curve of the entire scene to perform range envelope alignment. Then, perform an inverse range Fourier transform to the range Doppler domain. Apply the phase error compensation curve of the entire scene and the entire frequency band to perform phase error correction on the entire frequency band. Transform the data back to the image domain through the azimuth Fourier transform, and at the same time achieve azimuth compression to obtain the result after fine focusing.
2. The spaceborne SAR self-focusing method for long-duration integral focusing mode according to claim 1, characterized in that, For any given image block, the statistical power value is used to calculate its contrast based on the power distribution within the block. The calculation formula is as follows: ； in This represents a segmented image, where m represents the azimuth sampling point and n represents the range sampling point. This represents the statistical power value of the image block. This indicates that the mean value is calculated, where M represents the total number of azimuth sampling points in the image block, and N represents the total number of range sampling points in the image block. Let be a constant, and let it be... Then the above equation simplifies to 。 3. The spaceborne SAR self-focusing method for long-duration integral focusing mode according to claim 1, characterized in that, Sort the image blocks by contrast, and denote the block with the highest contrast as the strong scattering region; or sort the image blocks by mean power, and denote the block with the highest power value as the strong scattering region, where: The contrast of the segmented image is sorted, and the segment with the highest contrast among all segments is recorded as the strong scattering region. This method is suitable for scenes with many strong points, including urban areas and ports. The mean power of the segmented image is sorted, and the segment with the highest power value is identified as the strong scattering region. This method is applicable to uniform scenes, including mountainous areas and forests.
4. The spaceborne SAR self-focusing method for long-duration integral focusing mode according to claim 1, characterized in that, Perform a Fourier transform on a slice of the strong scattering region to obtain the azimuth frequency domain data. The specific method is as follows: A slice of the strongly scattering region in a single-view complex image is denoted as... Where t represents the azimuth information, Representing range information, the following Fourier transform yields the azimuth frequency domain data: ； in Let represent the Fourier transform, n represent the range sampling points, m represent the azimuth sampling points, R represent the slant range information of the target, λ represent the wavelength, c represent the speed of light, and σ represent the scattering intensity. and These represent the range migration and phase error that remain after imaging processing due to various types of errors, respectively.
5. The spaceborne SAR self-focusing method for long-duration integral focusing mode according to claim 1, characterized in that, The azimuth spectrum of the strongly scattering region slice is analyzed. Using out-of-band signal energy as clutter energy, spectra with high signal-to-clutter ratios are selected. The minimum entropy range alignment algorithm is then used to obtain the full-band range alignment curve. Range alignment is completed by compensating for the azimuth frequency domain data. The specific method is as follows: The full-band range alignment curve obtained through the minimum entropy range alignment algorithm The unit is pixels. It is converted to distance-time according to the relationship between pixels and distance-time, and expressed as: ； in This represents the distance-time corresponding to the distance alignment curve; hereinafter, we will use "distance alignment curve time" as a metonym. Indicates the distance sampling rate; Time to align with the distance curve By applying a Hampel filter and utilizing a median-based nonlinear filtering method, outliers in the range alignment curve are identified and removed while maintaining the basic shape, resulting in the filtered range alignment curve time. Subsequently Perform higher-order fitting; Let the fitting coefficient be p, the azimuth bandwidth be Ba, the pulse repetition frequency be PRF, the total number of azimuth sampling points in the slice data be M, and the total number of azimuth sampling points corresponding to the azimuth bandwidth be na. Then the time to obtain the range alignment curve of the processed full-bandwidth data is: : ； in This represents a polynomial calculation function with x as the fitting coefficient, y as the independent variable, and z as the number of sampling points. This represents the floor function; Align the processed distance obtained above with the curve time. Compensation applied to the azimuth frequency domain data: ； at this time This item should be basically corrected to zero.
6. The spaceborne SAR self-focusing method for long-duration integral focusing mode according to claim 5, characterized in that, Time-range alignment curves obtained from slices of strong scattering regions The distance migration correction curve for the entire scene is obtained by interpolation using polynomial fitting. The distance alignment process is performed using the following formula: ； Where Ma represents the number of azimuth sampling points for the entire scene data, and Na represents the number of azimuth sampling points corresponding to the entire scene bandwidth Ba.
7. The spaceborne SAR self-focusing method for long-duration integral focusing mode according to claim 6, characterized in that, Phase error compensation curves for the strong scattering region are plotted, smoothed using a Savitzky-Golay filter, and then subjected to high-order fitting to obtain phase error compensation curves for the entire scene and frequency band. The specific method is as follows: set up To obtain the phase error compensation curve for this strong scattering region, a high-order fit is performed after smoothing using a Savitzky-Golay filter. Let the fitting coefficient be q. Since the azimuth frequency domain data of the entire scene and the azimuth frequency domain data of a portion of the scene have the same bandwidth, they are essentially equivalent to having the same frequency domain error. The phase error compensation curve for the entire scene and the entire frequency band is then... : 。 8. A spaceborne SAR autofocusing system for long-duration integral focusing mode, characterized in that, The method for autofocusing spaceborne SAR in long-duration integral focusing mode as described in any one of claims 1-7 is implemented to achieve autofocusing of spaceborne SAR in long-duration integral focusing mode. It is divided into five modules, each executed separately: Module 1 performs block processing on single-view complex images at the SARL1A level, based on the set azimuth block size and range block size. Module 2 calculates the power value for any block of image, calculates its contrast based on the power distribution within the block, sorts the contrast of the block images, and identifies the block with the highest contrast among all blocks as the strong scattering region. Alternatively, the mean power of the segmented images can be sorted, and the segment with the highest power value among all segments can be denoted as the strong scattering region. Module 3 performs Fourier transform on the strong scattering region slice to obtain azimuth frequency domain data. Using out-of-band signal energy as clutter energy, it filters out spectra with high signal-to-clutter ratio and uses the minimum entropy range alignment algorithm to obtain the range alignment curve of the entire frequency band. It then compensates the azimuth frequency domain data to complete the range alignment. At the same time, it uses polynomial fitting to obtain the range migration correction curve for the entire scene from the range alignment curve of the entire frequency band. Module 4 applies the minimum entropy phase correction algorithm to the strong scattering region slice after distance alignment for phase autofocusing processing, obtains the fine focusing result and phase compensation error of the strong scattering region slice data, plots the phase error compensation curve of the strong scattering region, applies Savitzky-Golay filter for smoothing and performs high-order fitting to obtain the phase error compensation curve of the whole scene and the whole frequency band. Module 5 performs a two-dimensional Fourier transform on the data from the entire scene to obtain dual-frequency domain data. It then applies the range migration correction curve of the entire scene for range envelope alignment, and performs an inverse range Fourier transform to the range Doppler domain. Finally, it applies the phase error compensation curve of the entire scene and frequency band to correct the phase error of the entire frequency band. The data is then transformed back to the image domain through an azimuth Fourier transform, while simultaneously achieving azimuth compression to obtain the finely focused result.
9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the spaceborne SAR autofocusing method for long-duration integral focusing mode as described in any one of claims 1-7, thereby realizing spaceborne SAR autofocusing for long-duration integral focusing mode.
10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the spaceborne SAR autofocusing method for long-duration integral focusing mode as described in any one of claims 1-7, thereby realizing spaceborne SAR autofocusing for long-duration integral focusing mode.