A method for geometric correction of wide swath SAR imagery

Abstract

The present application provides a kind of geometric correction method of wide SAR image, first grade SAR image is divided grid in distance direction and Doppler direction, establishes two ellipsoid polar coordinate system, the pixel coordinate of first grade image is converted into two coordinates under two ellipsoid polar coordinate system, the latitude of the grid point is combined, the pixel coordinate of first grade image is converted into the pixel coordinate of second grade image;Establish the fitting relationship of whole image of second grade image pixel coordinate to first grade image pixel coordinate, and every grid point is selected as the center, the fitting relationship of every grid point is established, and then the pixel coordinate of every second grade image is obtained Corresponding first grade image pixel coordinate;For every row of second grade image pixel, the position of pixel in first grade image is obtained using fitting polynomial, and the minimum value is detected by difference, and the position of minimum value is removed according to the position of minimum value Abnormal data.The present application can face global range, especially wide image, and has strong practicability.

Description

Technical Field

[0001] This invention relates to geometric correction technology for synthetic aperture radar (SAR) images, specifically to a geometric correction method and system for wide-swath SAR images. Background Technology

[0002] The coverage of the images reaches thousands of kilometers, and the pixel size of a single image exceeds 100,000*100,000, which brings new problems to the geometric correction of the entire image. For example, the positioning accuracy of wide-swath SAR system-level geometric correction products decreases at the image edges, and SAR images in the North and South Poles are severely distorted. Solving these positioning problems caused by the excessive width is of great significance to the field of wide-swath SAR positioning.

[0003] SAR geometric correction is the process of converting SAR Level 1 images from the image coordinate system to the geographic coordinate system. Currently, the model used for SAR image geometric correction is the single-image SAR positioning model, which can be mainly divided into rigorous geometric models (Curlander JC. Location of Spaceborne Sar Imagery[J]. IEEE Transactions on Geoscience & Remote Sensing, 1982, GE-20(3):359-364) and general geometric models (Zhang Guo, Fei Wenbo, Li Zhen, et al. Experiment and analysis of replacing the rigorous imaging geometric model of spaceborne SAR with RPC[J]. Acta Geodaetica et Cartographica Sinica, 2010, 39(3):7). ​​The geometric model is used to convert the SAR Level 1 map into a Level 2 map with geographic information. For geometric correction, its improvement is related to the antenna phase center position and velocity error, SAR system time error, atmospheric transmission delay error, and imaging processing-introduced error. Geometric positioning accuracy can be improved by correcting various errors (Ding Chibiao, Liu Jiayin, Lei Bin, et al. Preliminary study on system-level geometric positioning accuracy of Gaofen-3 SAR satellite [J]. Journal of Radar, 2017, 6(1):6.); or by correcting the RD model using the error equivalence method to improve geometric positioning accuracy (Qiu Xiaolan, Liu Jiayin, You Hongjian, et al. A SAR geometric correction method based on error equivalence RD model correction.). Various parameters need to be calibrated and corrected, and the average elevation of the area needs to be obtained to achieve accurate positioning, making the corrected secondary map more accurate.

[0004] Currently, although there has been much research on the mapping of the North and South Poles, many problems still exist, including the lack of a unified mapping standard. Generally, it is necessary to distinguish between polar and non-polar regions. For example, Mercator charts, which are commonly used in non-polar regions, have significant distortion near the poles, while the polar spherical projection, which is more suitable for polar regions, is not suitable for other regions. (Wang Haibo, Zhang Hanwu, Zhang Pingping, et al. Polar grid coordinate representation method based on polar spherical projection [J]. Command Control and Simulation, 2016, 38(2):4.) For wide-swath imagery, its coverage is extensive, sometimes encompassing both polar and non-polar regions. This makes it difficult to segment the image for geometric correction. As of June 2020, my country's IcePathfinder optical satellite covered latitudes of 85°S and 85°S, exceeding the observation limit of mainstream international Landsat satellites (82.5°S and 82.5°S). The relevant team independently developed geometric correction and on-orbit radiometric calibration techniques (Zhang, Ying, Zhaohui Chi, Fengming Hui, Teng Li, Xuying Liu, Baogang Zhang, Xiao Cheng, and Zhuoqi Chen. 2021. "Accuracy Evaluation on Geolocation of the Chinese First Polar Microsatellite (Ice Pathfinder) Imagery" Remote Sensing 13, no. 21: 4278.). However, no solutions have been found for geometric correction of wide-swath SAR images in polar regions, highlighting the need to address this issue.

[0005] SAR image geometric correction employs a positioning model to locate each point in the primary SAR image, obtaining the three-dimensional position of each point. The size of the secondary image is determined by drawing a grid on the Earth's surface. The primary image data is then resampled into the secondary image, outputting GeoTIFF. Current geometric correction methods, when applied to wide-swath imagery, suffer from severe distortion in high-latitude regions, particularly in the Arctic and Antarctic, resulting in low positioning accuracy and an inability to adapt to global areas. Summary of the Invention

[0006] The purpose of this invention is to propose a geometric correction method and system for wide-swath SAR images.

[0007] The technical solution for achieving the objective of this invention is: a geometric correction method for wide-swath SAR images, comprising the following steps:

[0008] Step 1, homogenization of the second-level graph grid coordinates based on the dual-ellipsoidal polar coordinate system:

[0009] The SAR Level 1 image is divided into grids in the range and Doppler directions, and two ellipsoidal polar coordinate systems are established. The pixel coordinates of the Level 1 image are converted into two coordinates in the dual ellipsoidal polar coordinate system. Combined with the latitude of the grid points, the pixel coordinates of the Level 1 image are converted into pixel coordinates of the Level 2 image.

[0010] Step 2, gradual fitting of image grid points in dual ellipsoidal polar coordinates:

[0011] Establish a full-image fitting relationship from the pixel coordinates of the second-level image to the pixel coordinates of the first-level image. Select grid points with each grid point as the center, establish the fitting relationship for each grid point, and then obtain the pixel coordinates of the first-level image corresponding to each pixel coordinate of the second-level image.

[0012] Step 3, Anomaly detection of the successive fitting curve:

[0013] For each row of pixels in the secondary image, the pixel position in the primary image is obtained by fitting a polynomial, the minimum value is detected by difference, and abnormal data is removed based on the position where the minimum value appears.

[0014] Further, in step 1, the coordinates of the second-level graph grid are homogenized based on the dual-ellipsoidal polar coordinate system. The specific method is as follows:

[0015] Step 1.1: According to the configuration file of the SAR Level 1 image, divide the Level 1 image into grid points, and use the distance equation, Doppler equation and Earth equation to solve the three-dimensional coordinates of the four corner points and grid points of the Level 1 image in the geocentric geofixed coordinate system.

[0016] For the three-dimensional coordinates (x, y, z) of a point P in the geocentric Earth-fixed coordinate system, they are expressed in two ellipsoidal polar coordinate systems as follows:

[0017]

[0018]

[0019] Where θ∈[-90°,90°], α∈[-90°,90°], β∈(-180°,180°]; θ, α and β are both mathematical variables, and θ represents latitude. Indicates longitude; α and β have no practical meaning; R e ,R p These are the Earth's major and minor radii, respectively.

[0020] From the polar coordinate expressions of the two ellipsoids above, we can derive...

[0021]

[0022] in

[0023] For data where the latitudes of the four corner points are all between 50° North and 50° South, the grid point coordinates of the data are transformed into... For data where the latitude of the four corner points does not satisfy the condition of being between 50° North and 50° South, the grid point coordinates of the data are transformed into the (β, α, h) form.

[0024] Step 1.2: Combine the latitude of the grid points to convert the pixel coordinates of the first-level image into the pixel coordinates of the second-level image;

[0025] Based on the four corner points The coordinates of the plane or β-α plane determine the size of the secondary image:

[0026] Table 1 shows the coordinates of the four corner points in different coordinate systems.

[0027]

[0028] for The maximum value in the middle. for The minimum value is obtained by analogy with θ. max ,θ min ,α max ,α min ,β max ,β min The width and height of the secondary image are determined by the maximum and minimum values ​​as follows:

[0029]

[0030]

[0031] Grid point P i Pixel coordinates (x) in a primary image L1 i ,y L1 i ), corresponding The coordinates in the coordinate system are The corresponding coordinates in the β-α coordinate system are (β i ,α i ), the corresponding pixel coordinates (x) in the secondary image L2 i ,y L2 i ):

[0032]

[0033]

[0034] in yes The pixel spacing in direction, representing the longitude spanned by one pixel, is... θ It refers to the pixel spacing in the θ direction, which represents the interval spanning one latitude. α It is the pixel spacing in the α direction. pix β It is the pixel spacing in the β direction, pix θ =pix β .

[0035] Furthermore, in step 2, the gradual fitting of image grid points in dual ellipsoidal polar coordinates is specifically performed as follows:

[0036] Step 2.1, let grid point P i Pixel coordinates (x) in a primary image L1 i ,y L1 i ), the corresponding pixel coordinates (x) in the secondary image L2 i ,y L2 i The fitting relationship between the pixel coordinates of the second-level image and the pixel coordinates of the first-level image is established as follows:

[0037]

[0038] Where A nm and B nm These are the fitting coefficients for the entire graph, where n and m are non-negative integers, and Q is the fitting order.

[0039] Step 2.2: Using each grid point as the center, select K*K grid points, where K is a positive integer and K>4, and perform fitting to establish the fitting relationship for each grid point as follows:

[0040]

[0041] Where a nm and b nm These are the fitting coefficients, where n and m are non-negative integers, and Q is the fitting order.

[0042] Step 2.3, for each position in the secondary image, the corresponding primary image coordinates (x, y, y) are... L1 i ,y L1 i ), using the fitting coefficient A of the entire graph nm and B nm To determine the coverage area of ​​its grid points, the calculation method is as follows:

[0043]

[0044] Where GridX and GridY are the grid point indices in the x and y directions, respectively, w and h are the width and length of the first-level image, and M is the number of grid points;

[0045] Step 2.4, according to (x L1 i ,y L1 i The fitting coefficients of the grid points to which the grid points belong are used to recalculate their corresponding positions in the first-level graph.

[0046] Further, in step 3, the anomaly detection of the successive fitting curve is performed using the following method:

[0047] Step 3.1: Extract the pixel coordinates (x, y) of the primary image corresponding to each row of secondary image pixels. L1 j ,y L1 j ),j∈[1,L2_width], whereL2_width is the width of the secondary image;

[0048] Step 3.2, for the coordinate sequence (x L1 1,y L1 1),...,(x L1 j ,y L1 j ),...,(x L1 L2_width-1 ,y L1 L2_width-1 Perform difference operations:

[0049]

[0050] The differenced sequence (Δx) is obtained L1 1,Δy L1 1),...,(Δx L1 j ,Δy L1 j ),...,(Δx L1 L2_width-1 ,Δy L1 L2_width-1 )

[0051] Step 3.3, obtain |Δx L1 1|The pixel position j at the minimum value. If j≤1 or j≥L2_width, there will be no problem during resampling. If 1<j<L2_width, abnormal data will occur during resampling, Δx. L1 j If the smallest pixel position j is located to the right of the center point of the secondary image, then the data to the right of j is set to zero; otherwise, the data to the left of j is set to zero.

[0052] A geometric correction system for wide-swath SAR images, based on the aforementioned geometric correction method for wide-swath SAR images, realizes the geometric correction of wide-swath SAR images.

[0053] A computer device includes 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 geometric correction method for the wide-swath SAR image.

[0054] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the geometric correction method for the wide-swath SAR image.

[0055] Compared with existing technologies, the significant advantages of this invention are: 1) The second-level image grid coordinate homogenization method based on the dual-ellipsoidal polar coordinate system breaks through the current limitation that geometric correction is only applicable to non-polar regions, making geometric correction applicable globally, especially for wide-format images, and thus highly practical; 2) The gradual fitting method for image grid points under dual-ellipsoidal polar coordinates ensures seamless connection of regional grid fitting under the condition that the polynomial order remains unchanged, effectively improving the positioning accuracy of each region in the image and correcting positioning deviations in image boundary regions; 3) The gradual fitting curve anomaly detection method solves the problem of out-of-bounds abnormal data caused by the folding of fitting curves due to the excessive number of pixels in a small number of wide-format images. The three-point method proposed in this invention, while balancing the advantages of existing technologies, proposes a processing approach and key steps based on the dual-ellipsoidal coordinate system. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the North and South Pole regions.

[0057] Figure 2 This is a diagram showing the relationship between the observation orbit and the location of the Arctic observation area in the geocentric-geostatic coordinate system.

[0058] Figure 3 This is a schematic diagram of the Arctic observation area within the latitude and longitude plane.

[0059] Figure 4 For observation area data in Schematic diagram of the projection of the plane (left) and the β-α plane (right).

[0060] Figure 5 This is a schematic diagram of the fitted curve.

[0061] Figure 6 This is a schematic diagram showing the corresponding curves of the horizontal coordinate of row 120000 in the second-level graph and the distance coordinate in the first-level graph.

[0062] Figure 7 This is a diagram illustrating abnormal data.

[0063] Figure 8 This is a diagram illustrating the corrected data. Detailed Implementation

[0064] 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.

[0065] This invention provides a geometric correction processing method for wide-swath SAR images, the specific implementation process of which is as follows: Figure 1 As shown, the main steps are:

[0066] 1) A method for homogenizing the coordinates of a second-level graph grid based on a dual-ellipsoidal polar coordinate system

[0067] First, based on the parameters provided in the configuration file of the SAR Level 1 image, grid points are divided within the Level 1 image. Then, by simultaneously solving the distance equation, Doppler equation, and Earth equation, the three-dimensional coordinates (x, y, z) of the image center point, four corner points, and grid points in the geocentric-geostatic coordinate system are obtained. Based on these three-dimensional coordinates, the longitude and latitude are calculated.

[0068] The three-dimensional coordinates (x, y, z) of a point P in a geocentric Earth-fixed coordinate system can be expressed in two ellipsoidal polar coordinate systems as follows:

[0069]

[0070]

[0071] Where θ∈[-90°,90°], α∈[-90°,90°], β∈(-180°,180°],

[0072] From the polar coordinate expressions of the two ellipsoids above, we can derive...

[0073]

[0074] in

[0075] To ensure smooth subsequent operations, the data needs to be evaluated: for data where the latitude of the four corner points is between 50° North and 50° South, the grid point coordinates of this data need to be converted... For data where the latitude of the four corner points does not satisfy the condition of being between 50° North and 50° South, the grid point coordinates of the data are transformed into the form (β, α, h).

[0076] Based on the four corner points The coordinates of the plane (or β-α plane) are used to determine the dimensions of the secondary image as follows:

[0077] The coordinates of the four corner points are represented in different coordinate systems as shown in the table below.

[0078]

[0079] for The maximum value in the middle. for The minimum value is obtained by analogy with θ. max ,θ min ,α max ,α min ,β max ,β min .

[0080] Using these maximum and minimum values, the width and height of the secondary image can be determined by the following formula:

[0081]

[0082]

[0083] Grid point P i Pixel coordinates (x) in a primary image L1 i ,y L1 i ), corresponding The coordinates in the coordinate system are The corresponding coordinates in the β-α coordinate system are (β i ,α i ), the corresponding pixel coordinates (x) in the secondary image L2 i ,y L2 i );

[0084]

[0085]

[0086] in yes The pixel spacing in direction, representing the longitude spanned by one pixel, is... θ It refers to the pixel spacing in the θ direction, which represents the interval spanning one latitude. α It is the pixel spacing in the α direction. pix β It is the pixel spacing in the β direction, pix θ =pix β .

[0087] This method of coordinate determination can perfectly solve the geometric correction problem caused by abrupt changes in meridians between the Arctic and Antarctic regions on the latitude and longitude plane.

[0088] 2) Gradual fitting method for image grid points in dual ellipsoidal polar coordinates

[0089] While the grid division obtained in the dual ellipsoidal polar coordinate system can solve the image distortion caused by uneven grid division in high latitude and polar regions, the number of pixels in a single direction of wide-format images reaches hundreds of thousands. Increasing the number of grid points is no longer sufficient to improve the fitting accuracy. Based on the image pixel coordinates obtained in step 1), we propose a gradual fitting method based on grid points.

[0090] First, the SAR Level 1 image data is uniformly divided into M grid points along both the range and Doppler directions, where grid point P... i Pixel coordinates (x) in a primary image L1 i ,y L1 i ), the corresponding pixel coordinates (x) in the secondary image L2 i ,y L2 i First, calculate the fitting coefficients A of the geometric transformation of the entire image. nm and B nm ,

[0091]

[0092] Where A nm and B nm These are the fitting coefficients for the entire graph, where n and m are non-negative integers, and Q is the fitting order, typically 2 or 3.

[0093] Then, taking each grid point as the center, select K*K grid points, where K is a positive integer and K>4, and perform fitting to obtain the fitting coefficients for that grid point. The grid point fitting polynomial is as follows:

[0094]

[0095] Where a nm and b nm These are the fitting coefficients, where n and m are non-negative integers, and Q is the fitting order, typically 2 or 3.

[0096] For each position coordinate (x) in the secondary graph L1 ,y L1 First, use the fitting coefficient A of the entire graph. nm and B nm Calculate which grid point coverage area it belongs to, the calculation method is as follows:

[0097]

[0098] Where GridX and GridY are the grid point indices in the x and y directions, respectively, w and h are the width and length of the first-level image, and M is the number of grid points.

[0099] Therefore, (x) L1 i ,y L1 i The fitting coefficients of the grid points to which the grid points belong are used to recalculate their corresponding positions in the first-level graph.

[0100] The rigor of a tight grid fit lies in the fact that each grid point has its own set of fitting coefficients, rather than a region having its own set of fitting coefficients. There may be slight differences in the fitting coefficients between adjacent grid points, but these differences are sufficient to ensure the smoothness of the data in the regions between grid points.

[0101] This rigorous grid fitting setting allows us to use fewer fitting coefficients to complete the fitting of wide-span primary image to secondary image coordinates.

[0102] like Figure 2 As shown, for light gray grid points, choosing K=5, that is, a 5*5 grid of points centered on the grid point, allows us to obtain fitting coefficients. Similarly, for dark gray grid points, we can also obtain corresponding fitting coefficients. It can be seen that there are K*(K-1) points that are the same when fitting adjacent grid points. As long as the fitting residual is kept within 1 pixel, the boundary area between dark gray and light gray grid points can be seamlessly connected during fitting.

[0103] 3) Method for detecting anomalies in successive fitting curves

[0104] For wide-swath data, covering an area larger than typical imagery, the image size is very large. Position mapping from level-two to level-one maps uses a fitting method. Polynomial fitting only fits the data for a given region and has no fitting capability for points outside the given data range. Therefore, for geometric correction from level-one to level-two maps, areas with no data along the edges in the level-two map should be represented as zero. However, some fitting coefficients can lead to outliers appearing along the edges of the level-two map. To address potential outliers during processing, a gradual fitting curve verification method is proposed.

[0105] The testing method first extracts the pixel coordinates (x, y) of the primary image corresponding to each row of secondary image pixels. L1 j ,y L1 j ),j∈[1,L2_width],

[0106] Then, for the coordinate sequence (x) L1 1,y L1 1),...,(x L1 j ,y L1 j ),...,(x L1 L2_width-1 ,y L1 L2_width-1 Perform difference operations

[0107]

[0108] The differenced sequence (Δx) is obtained L1 1,Δy L1 1),...,(Δx L1 j ,Δy L1 j ),...,(Δx L1 L2_width-1 ,Δy L1 L2_width-1 )

[0109] Finally, determine |Δx j | represents the pixel position j at the minimum value. If j ≤ 1 or j ≥ L2_width, no problems will occur during resampling. If 1 < j < L2_width, abnormal data will appear during resampling.

[0110] Example

[0111] To verify the effectiveness of the present invention, the following experimental design was conducted.

[0112] I. Condition Design

[0113] 1. Track Design

[0114] The satellite flies from 10°E to 45°N longitude to 5°E to 33°N longitude, with the following orbital parameters:

[0115] Table 1 Simulation Track Parameters

[0116] orbital parameters high track length Satellite speed Lifting rail Side view Parameter value 810km 1500km 7km / s Descending orbit Left side

[0117] 2. Design of the location and scope of the observation area

[0118] The simulation area, ranging from 7.22°E to 32.88°E and from 25.30°N to 44.50°N, is a seemingly rectangular curved surface (1500km * 1525km). The simulated ground elevation is 0m. The relationship between the orbit and the observation area is as follows: Figure 3As shown, the gray lines represent the orbit, and the gray area represents the observation area. A schematic diagram of the observation area on a latitude-longitude plane is given, as follows: Figure 4 On a plane divided by equal intervals of latitude and longitude, the observed image appears to broaden at higher latitudes. This is because as latitude increases on Earth, the actual distance between meridians decreases, resulting in a broadening phenomenon on an equally divided plane.

[0119] 3. Resolution and Projection Plane Design

[0120] The observation resolution is set as shown in Table 2.

[0121] Table 2 Simulation Observation Resolution

[0122]

[0123] At the above resolution settings, the location of the observation area on the Earth projection plane of the secondary map is as follows: Figure 5 As shown.

[0124] 4. Fitting Form and Order Design

[0125] The projection relationship between the first-order and second-order plots is fitted using a bivariate third-order polynomial, with Q = 3. The formula is as listed above.

[0126] II. Simulation Data

[0127] 1. A method for homogenizing coordinates of a second-level graph grid based on a dual-ellipsoidal polar coordinate system

[0128] The method for homogenizing the coordinates of a second-level map grid based on a dual-ellipsoidal polar coordinate system is a necessary step in the wide-swath geometric correction method for global regions. First, we present the practical problems encountered in the geometric correction process for wide-swath data in some regions, and then verify the effectiveness of our proposed method.

[0129] like Figure 1 As shown, in high-latitude regions, the intervals between meridians vary significantly across different latitudes. Especially in the Arctic / Antarctic regions, the latitude and longitude grid resembles a spiderweb, with meridians converging at a single point near the poles. This demonstrates that the intervals between meridians in a single SAR image vary significantly with latitude. For secondary maps of mid- and low-latitude regions, the same longitude resolution is generally used. Clearly, this approach is not suitable for high-latitude regions, especially the Arctic and Antarctic.

[0130] Based on the previous simulation conditions, a satellite flight trajectory was set so that it would fly over the North Pole. The observation area is approximately 1500km * 1525km. The relationship between the observation trajectory and the location of the North Pole observation area in the geocentric Earth-fixed coordinate system is shown in the figure below. Figure 2 As shown.

[0131] Depend on Figure 3As can be seen, the data on the latitude and longitude plane of the Arctic region are severely distorted and cannot be geometrically corrected according to the latitude and longitude plane of this region.

[0132] Based on the method of homogenizing the grid coordinates of the second-level map in the dual ellipsoidal polar coordinate system, the data of the observation area are homogenized by using the coordinates of the β-α surface when the latitude is greater than 50°.

[0133] like Figure 4 As shown, the data in the observation area on the left is... The image shows the projection of the observed area data onto the β-α plane. A comparison reveals that this method has a significant effect on grid coordinate homogenization in high-latitude regions, especially the Arctic and Antarctic regions.

[0134] 2. Gradual Fitting Method for Image Grid Points in Double Ellipsoid Polar Coordinates

[0135] First, the necessity of using a successive mesh fitting method in the wide-span geometric correction problem is explained. Table 3 presents the fitting residual statistics for dividing the entire map into 60*60 and 180*180 mesh points.

[0136] Table 3. Statistics of Residuals from Full-Graph Fitting

[0137]

[0138] It is evident that for wide-format scenes, due to the large image size (150,000 pixels in one direction), increasing the number of grid points does not effectively improve fitting accuracy. Increasing the fitting order can improve accuracy to some extent, but high-order fitting of binary variables is quite sensitive to initial value settings, and certain difficulties arise when solving for the fitting parameters. Therefore, this invention does not increase the order but adopts a gradual fitting method using grid points.

[0139] The seamless connection of the grid point fitting is verified below. Two adjacent grid points in the first-level map are randomly selected, and a grid with K=15 squares is taken with each point as the center. The statistics of the fitting residuals are shown in Table 4.

[0140] Table 4. Statistics of Residuals from Mesh Lattice Fitting

[0141]

[0142]

[0143] Meanwhile, the residuals of the common grid points of the two sets of lattices are subtracted to obtain the fitting error of the two sets of coefficients.

[0144] Table 5. Statistics of the fitting error for the two sets of fitting coefficients.

[0145]

[0146] The two tables above show that the boundary accuracy of the region grid-based fitting method is at the sub-pixel level, and the variance statistics indicate that this accuracy is quite stable. In other words, the region grid-based fitting method can achieve high-precision, seamless fitting.

[0147] For each position in the secondary plot, first use the fitting coefficient of the entire plot to find its corresponding position in the primary plot. Then, determine the nearest grid point based on this position and obtain the fitting coefficient of that point. Finally, use this fitting coefficient to find its precise position in the primary plot again.

[0148] The image has 150,000 pixels in one direction, 60 grid points in one direction, and a pixel interval of 2,500 pixels between grid points. Within a range of 2,500 pixels centered on a grid point, the fitting coefficient of that grid point is used for accurate projection.

[0149] If the fitting coefficients are obtained using a 60*60 grid, the position of any pixel of valid data in the secondary image projected onto the primary image is within a range of 2500 pixels centered on its corresponding grid point.

[0150] Statistical analysis of all valid data in the secondary maps revealed that they all correspond to the grid range of their respective primary map. Furthermore, based on the analysis of fitting the regional grid points, the fitting error between a grid point and its adjacent grid points is within 1 pixel. Therefore, even if a point maps to an adjacent grid range with a very small probability, the fitting is still accurate.

[0151] 3. Validate the anomaly detection method for successive fitting curves.

[0152] Under normal circumstances, the fitted curve is monotonic within the grid layout range, such as Figure 5 As shown. However, in some rare cases, it will be as follows: Figure 6 As shown, the relationship between the data in rows 140,000 and columns 1 to 180,000 in the bottom region of the second-level plot and the distance coordinate X in the first-level plot is illustrated. It can be seen that in the bottom row of data in the second-level plot, at column 121313, the curve has an extreme point and then declines. Since data after column 121313 should not be mapped to valid data, this point clearly maps to valid data in the first-level plot, resulting in abnormal data in the lower right corner of the second-level plot. Figure 7 As shown. Using the method proposed in this invention, the modified result is as follows. Figure 8 As shown.

[0153] 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.

[0154] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, 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 modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A geometric correction method for wide-swath SAR images, characterized in that, Includes the following steps: Step 1, homogenization of the second-level graph grid coordinates based on the dual-ellipsoidal polar coordinate system: The SAR Level 1 image is divided into grids in the range and Doppler directions, and two ellipsoidal polar coordinate systems are established. The pixel coordinates of the Level 1 image are converted into two coordinates in the dual ellipsoidal polar coordinate system. Combined with the latitude of the grid points, the pixel coordinates of the Level 1 image are converted into pixel coordinates of the Level 2 image. Step 2, gradual fitting of image grid points in dual ellipsoidal polar coordinates: Establish a full-image fitting relationship from the pixel coordinates of the second-level image to the pixel coordinates of the first-level image. Select grid points with each grid point as the center, establish the fitting relationship for each grid point, and then obtain the pixel coordinates of the first-level image corresponding to each pixel coordinate of the second-level image. Step 3, Anomaly detection of the successive fitting curve: For each row of secondary image pixels, the pixel position in the primary image obtained by fitting a polynomial is used to detect the minimum value by difference, and abnormal data is removed based on the position where the minimum value appears. in: Step 2, gradual fitting of image grid points in dual ellipsoidal polar coordinates, the specific method is as follows: Step 2.1, Set grid points Pixel coordinates in a primary image The corresponding pixel coordinates in the secondary image The fitting relationship between the pixel coordinates of the second-level image and the pixel coordinates of the first-level image is established as follows: ； in and These are the fitting coefficients for the entire graph. It is a non-negative integer. It is the fitting order; Step 2.2: Using each grid point as the center, select K*K grid points, where K is a positive integer and K>4, and perform fitting to establish the fitting relationship for each grid point as follows: ； in and These are the fitting coefficients. It is a non-negative integer. It is the fitting order; Step 2.3: For each position in the secondary image, the corresponding primary image coordinates... Using the fitting coefficients of the entire graph and To determine the coverage area of ​​its grid points, the calculation method is as follows: ； in These are the grid point indices in the x and y directions, respectively. These are the width and length of the first-level image. It is the number of grid points; Step 2.4, according to The fitting coefficients corresponding to the grid points to which they belong are used to recalculate their corresponding positions in the first-level graph.

2. The geometric correction method for wide-swath SAR images according to claim 1, characterized in that, Step 1: Uniformize the coordinates of the second-level graph grid based on the dual-ellipsoidal polar coordinate system. The specific method is as follows: Step 1.1: According to the configuration file of the SAR Level 1 image, divide the Level 1 image into grid points, and use the distance equation, Doppler equation and Earth equation to solve the three-dimensional coordinates of the four corner points and grid points of the Level 1 image in the geocentric geofixed coordinate system. For the three-dimensional coordinates (x, y, z) of a point P in the geocentric Earth-fixed coordinate system, they are expressed in two ellipsoidal polar coordinate systems as follows: ； ； in , , , ; , , , All are mathematical variables. Indicates latitude, Indicates longitude. , It has no practical significance; These are the Earth's major and minor radii, respectively. From the polar coordinate expressions of the two ellipsoids above, we can derive... ； in ; For data where the latitudes of the four corner points are all between 50° North and 50° South, the grid point coordinates of the data are transformed into... For data where the latitude of the four corner points does not meet the requirement of being between 50° North and 50° South, the grid point coordinates of the data are transformed into... form; Step 1.2: Combine the latitude of the grid points to convert the pixel coordinates of the first-level image into the pixel coordinates of the second-level image; Based on the four corner points noodles or The coordinates of the plane determine the size of the secondary image: Let the four corner points be , , and Its geocentric coordinates are respectively , , and , The coordinates are respectively , , and , The coordinates are respectively , , and ; set up for The maximum value in the middle. for The minimum value is obtained similarly. The width and height of the secondary image are determined by the maximum and minimum values ​​as follows: when ; when ; Grid points Pixel coordinates in a primary image Corresponding The coordinates in the coordinate system are Corresponding The coordinates in the coordinate system are The corresponding pixel coordinates in the secondary image : when ; when ; in yes The pixel spacing in the direction represents the longitude spanned by one pixel. yes The pixel spacing in a direction represents the interval spanned by one latitude. yes Pixel spacing in direction, , yes Pixel spacing in direction, .

3. The geometric correction method for wide-swath SAR images according to claim 1, characterized in that, Step 3, anomaly detection of the successive fitting curve, the specific method is as follows: Step 3.1: Extract the pixel coordinates of the primary image corresponding to each row of secondary image pixels. ,in The width of the secondary image; Step 3.2, for the coordinate sequence Perform difference operations: ； Obtain the difference sequence ; Step 3.3, obtain | |The smallest pixel position j, if or If so, there will be no problem during the resampling process. Abnormal data may occur during the resampling process. If the smallest pixel position j is located to the right of the center point of the secondary image, then the data to the right of j is set to zero; otherwise, the data to the left of j is set to zero.

4. A geometric correction system for wide-swath SAR images, characterized in that, Based on the geometric correction method for wide-swath SAR images according to any one of claims 1-3, geometric correction of wide-swath SAR images is achieved.

5. 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 geometric correction method for wide-swath SAR images according to any one of claims 1-3.

6. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the geometric correction method for wide-swath SAR images according to any one of claims 1-3.

Images