A navigation satellite bistatic InSAR image registration method based on sequential translation

Abstract

This invention discloses a sequential translation-based method for bistatic InSAR image registration of navigation satellites. First, the invention extracts temporal strong points based on simulated resolution units and determines a candidate set of strong points to be registered based on the radiation radius of a proposed reference region. Then, following a sequential order, the primary image is dynamically changed during registration, and coarse registration is performed based on the maximum correlation coefficient criterion, transforming the one-to-many relationship between strong points in the primary and secondary images into a one-to-one relationship. Next, within the expanded registration area of ​​the secondary image, translation processing is used, and fine registration is performed based on the maximum correlation coefficient criterion to determine the position of the points to be registered on the secondary image. This invention overcomes the limitations of traditional image registration methods, is applicable to GNSS-based InSAR systems, and thus improves the temporal coherence of target points within navigation satellite temporal SAR images, laying the foundation for high-precision deformation monitoring applications.

Description

Technical Field

[0001] This invention relates to the field of bistatic synthetic aperture radar technology, and specifically to a navigation satellite bistatic InSAR image registration method based on sequential translation. Background Technology

[0002] Bistatic Synthetic Aperture Radar (GNSS-based BiSAR) is a bistatic SAR system that uses in-orbit navigation satellites as external radiation sources, with the receiver located near the ground. The receiver can be airborne, vehicle-mounted, or even fixed. Due to the short reorbiting period and abundant in-orbit satellite resources of navigation satellites, the near-ground receiver can receive signals from at least eight satellites at any given time. This allows for 3D deformation inversion through information fusion from different angles, making it suitable for monitoring rapidly deforming scenarios. Deformation monitoring applications require long-term observation of targets to continuously track their deformation. This necessitates finding a large number of corresponding points in the temporal SAR image to improve the temporal correlation of target points within the image, i.e., registration processing.

[0003] However, due to the low signal-to-noise ratio and highly asymmetric bistatic configuration of GNSS-based InSAR systems, many traditional registration methods are no longer applicable. First, the low signal-to-noise ratio of navigation satellite signals results in target points in time-series SAR images exhibiting a point spread function pattern, lacking characteristic information. Furthermore, the narrow bandwidth causes the resolution units of each target to span pixels, making it impossible to distinguish specific objects compared to optical images. Therefore, traditional feature-based registration algorithms are no longer suitable. Second, the offset of navigation satellite trajectories exhibits periodicity within a certain time range, meaning that adjacent trajectories show an approximate translational pattern. However, adjacent trajectories from airborne and spaceborne platforms will randomly offset from a certain reference position. This characteristic causes the positional correlation of targets in the acquired SAR images to gradually deteriorate with increasing time intervals; if the interval is too long, target points may even become completely uncorrelated. Therefore, registration methods based on a fixed reference day cannot be applied to applications requiring long-term observation. Thus, an image registration method suitable for GNSS-based InSAR systems is urgently needed. Summary of the Invention

[0004] In view of this, the present invention provides a navigation satellite bistatic InSAR image registration method based on sequential translation, which can improve the temporal coherence of each target point in the navigation satellite temporal SAR image.

[0005] The sequential translation-based navigation satellite bistatic InSAR image registration method of the present invention includes:

[0006] Step 1: Determine the main image and auxiliary image sequentially according to the temporal SAR image sequence; extract the strong points in the main image and auxiliary image based on the simulation resolution unit template;

[0007] In the image I, if the amplitude of the pixel P at the center point of the simulation resolution unit template is the maximum value within the coverage area of ​​the simulation resolution unit template, then the center point is considered to be a strong point in the image I.

[0008] Step 2: For each strong point in the main image, construct a corresponding reference region in the auxiliary image, and extract the strong points in the auxiliary image within the reference region to construct a candidate registration point set; the reference region is: the region where the simulation resolution unit template is similarly enlarged by an integer multiple with the strong point as the center point;

[0009] Step 3: Perform coarse registration for each strong point in the main image: Based on the simulated resolution unit template, extract the effective resolution units of the strong points in the main image, as well as the effective resolution units of each candidate point to be registered in the auxiliary image corresponding to the strong point in the main image; calculate the complex correlation coefficient between the effective resolution units of the strong points in the main image and the effective resolution units of each candidate point to be registered in the corresponding auxiliary image, and the candidate point to be registered with the largest correlation coefficient is the point to be registered for the strong point in the main image.

[0010] Step 4: For each strong point in the main image, perform fine registration: In the expanded registration range area centered on the point to be registered in the auxiliary image, calculate the correlation coefficient between the simulation resolution unit of the strong point in the main image and the simulation resolution unit at different translation positions by translating the simulation resolution unit template. The maximum correlation coefficient corresponds to the fine registration position of the strong point in the main image on the auxiliary image.

[0011] Repeat steps 1 to 4 to complete the coarse and fine registration of all strong points in the time-series SAR image, forming a registration link for each strong point in the image.

[0012] A preferred simulation resolution element template is:

[0013]

[0014] Where A and R are the maximum number of pixels in the north-south and east-west directions of the simulated resolution unit template, respectively, determined by the satellite trajectory and system parameters; f ij A value of 1 indicates that the object is inside the simulated resolution cell outline, and a value of 0 indicates that the object is outside the simulated resolution cell outline. i = 1, 2, ..., a, ... A; j = 1, 2, ..., r, ... R;

[0015] The simulation resolution unit is represented by the following fuzzy function, and the 3dB resolution unit contour is taken. P1 is the target point, and P2 is a point around the target point P1, which is the target point P1 imaged and simulated based on the real trajectory and real system parameters.

[0016]

[0017]

[0018] Where, Φ TA and Φ RA These are the unit vectors of the transmitter and receiver relative to the target point P1, respectively; β is the bibase angle; Θ is the vector along the angle bisector of β; ω E Ξ and Ξ are the equivalent angular velocity and equivalent direction of motion of the transmitter; p is the range pulse compression result; m A It is the result of azimuth pulse compression; λ is the wavelength, c is the speed of light; the superscript T indicates matrix rank transformation.

[0019] Preferably, the reference region is an elliptical domain centered on the strong point, with a radiation radius Ra = {n, e} and a major radius e at an angle θ with the horizontal axis; n and e are the two-dimensional resolution of the navigation satellite bistatic InSAR system, in pixels; and These represent the range resolution and azimuth resolution of the simulated resolution unit template, respectively. It is the unit vector along the x-axis.

[0020] Preferably, in step 3, the formula for calculating the correlation coefficient is:

[0021]

[0022] Among them, M and M′ i Let M and M′ be the resolution unit matrices of the strong point in the main image and the i-th candidate registration point in the auxiliary image, respectively. They are obtained by expanding pixel units outwards from the strong point in the main image and the i-th candidate registration point in the auxiliary image, respectively. i The size is the same as the contour matrix F of the simulation resolution unit template; the superscript H indicates the conjugate transpose, "|||| 2 "" indicates squaring each element in the matrix, and "||" indicates taking the absolute value of each element in the matrix.

[0023] Preferably, in step 4, the registration range is expanded by: taking the point to be registered in the auxiliary image as the center, expanding upward and downward by R pixels and left and right by A pixels respectively, thereby forming an expanded registration range of size (2R+1)×(2A+1); where A and R are the maximum number of pixels in the north-south and east-west directions of the simulated resolution unit template, respectively, which are determined by the satellite trajectory and system parameters.

[0024] Preferably, the formula for calculating the correlation coefficient in step 4 is:

[0025]

[0026] M is the resolution unit matrix of the main image strong points, which is obtained by expanding pixel units outward from the main image strong points as the center. The size of M is the same as the contour matrix F of the simulation resolution unit template. The matrix represents the resolvable cell at the translation position of the auxiliary image; the superscript T indicates matrix transposition, the superscript H indicates taking the conjugate transposition, and "|||| 2 "" indicates squaring each element in the matrix, and "||" indicates taking the absolute value of each element in the matrix.

[0027] The better ones also include:

[0028] Step 5, correct the registration link:

[0029] If the registration link is interrupted, the interrupted image is skipped, and coarse and fine matching are performed directly with the subsequent images to form a registration link;

[0030] If a strong point on the auxiliary image is a registration point for multiple strong points on the main image, then the correlation coefficients between each strong point in the main image and the strong point in the auxiliary image are compared, and the strong point in the auxiliary image is assigned to the strong point in the main image corresponding to the largest correlation coefficient.

[0031] Beneficial effects:

[0032] This invention first extracts temporal strong points based on simulated resolution units and determines a candidate set of strong points to be registered based on the radiation radius of the proposed reference region to adapt to the system characteristics of bistatic InSAR for navigation satellites. Then, following a sequential order, the master image during registration is dynamically changed, and coarse registration is performed based on the maximum correlation coefficient criterion, converting the one-to-many relationship between strong points in the master and auxiliary images into a one-to-one relationship. This reduces the temporal decorrelation effect caused by the periodic offset of the navigation satellite trajectory, thereby improving the temporal coherence among strong points in the navigation satellite temporal SAR image. Next, within the expanded auxiliary image registration area, translation processing is used, and fine registration is performed based on the maximum correlation coefficient criterion to determine the position of the points to be registered on the auxiliary image, thereby improving the accuracy of the registration results. Finally, the registration results are corrected. Skip registration solves the problem of inconsistent link lengths of registration points, overcomes the decorrelation effect caused by long time intervals, and redistributes the registration results of strong points where registration links intersect based on the maximum correlation coefficient criterion, thereby improving the temporal coherence of each target point in the navigation satellite temporal SAR image. This invention breaks through the limitations of traditional image registration methods and is applicable to GNSS-based InSAR systems, thereby improving the temporal coherence of target points in navigation satellite time-series SAR images and laying the foundation for high-precision deformation monitoring applications. Attached Figure Description

[0033] Figure 1 This is a flowchart of the algorithm of the present invention;

[0034] Figure 2 This is a schematic diagram of the reference area in this invention;

[0035] Figure 3 The strength extraction results of the present invention are used in the examples provided.

[0036] Figure 4 The results of the registration point set of the present invention are used in the examples described;

[0037] Figure 5 The results of sequential registration using the present invention in the illustrated embodiments;

[0038] Figure 6 The average correlation coefficient results for the illustrated embodiments are obtained using the methods of the present invention and the conventional methods, respectively.

[0039] Figure 7 The north-south average registration offset results for the given embodiments are obtained using the present invention and the conventional method, respectively.

[0040] Figure 8 The average registration offset in the east-west direction for the examples are obtained by using the present invention and the conventional method, respectively. Detailed Implementation

[0041] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0042] This invention provides a sequential translation-based bistatic InSAR image registration method for navigation satellites. First, based on a time-series image set and simulated resolution units, temporal strong points are extracted. Then, the reference region radius is calculated to generate a set of points to be registered. Next, coarse registration is performed sequentially to obtain corresponding points. Then, fine registration is performed within the expanded sub-image registration range. Both registrations are based on the maximum correlation coefficient criterion. Finally, the registration results are corrected to obtain a complete registration link, thereby improving the temporal coherence of target points within the navigation satellite's time-series SAR image. The algorithm flowchart of this invention is shown below. Figure 1 As shown, the specific steps are as follows:

[0043] Step 1: Calculate the simulation resolution unit and extract the temporal strength points on each image.

[0044] First, point target simulation is performed using the actual trajectory and system parameters from the first image. That is, assuming imaging simulation is performed on target point P1, and P2 is a point surrounding target P1, the resulting resolution cell is:

[0045]

[0046] Where Φ TA and Φ RA These are the unit vectors of the transmitter and receiver relative to the target point P1, respectively; β is the bibase angle; Θ is the vector along the angle bisector of β; and ω... E Ξ and Ξ are the equivalent angular velocity and equivalent direction of motion of the transmitter. p is the range pulse compression result, m A This is the result of azimuth pulse compression, where λ is the wavelength and c is the speed of light. The theoretical 3dB resolution unit can then be expressed as:

[0047]

[0048] The simulated resolution element template outline is written in matrix form:

[0049]

[0050] The matrix F has a size of A×R, where A is the number of north-south pixels in the resolving unit of target A, and R is the number of east-west pixels. Since the resolving unit is a blur function, the 3dB contour is represented by an ellipse, and this matrix consists of 0s and 1s. A value of 1 indicates that the pixel at that position in the extracted contour is a pixel within the resolving unit of that point and will participate in the subsequent correlation coefficient calculation. A value of 0 indicates that the pixel at that position is not a pixel within the resolving unit of that point and will not participate in the correlation coefficient calculation. The value at the center of the matrix is ​​always 1, which is the position of the amplitude peak point of the resolving unit of that point, denoted as the a-th row and r-th column.

[0051] Traversing each pixel in the temporal SAR image, i.e., globally shifting the obtained resolution unit template across each image, if the amplitude at pixel P at the center point of the template is the maximum value within the template's coverage area, then that point is considered a strong point in that image I, i.e.:

[0052]

[0053] Where I R×A This represents the image information of image I within the coverage area of ​​template F, where "*" indicates a dot product operation between matrices. This local maxima method extracts the set of strong points on each image, i.e., P = {P1, P2, ..., P...}. d ,…,P D}

[0054] Step 2: Calculate the radiation radius of the reference area and generate a set of registration points for each strong point.

[0055] Based on the simulation resolution unit template, let its range resolution be... Azimuth resolution is Assuming the two directions are approximately perpendicular, the cosine of the angle θ between the resolving unit and the horizontal direction is: in It is the unit vector along the x-axis.

[0056] Because GNSS-based InSAR systems have low two-dimensional resolution, simulated resolution cells typically contain multiple pixels, rendering traditional pixel-based registration methods inapplicable. Therefore, based on simulated resolution cells, the concept of a reference region radiation radius Ra = {n, e} is introduced. Its unit is pixels, usually taken as an integer multiple of the system's two-dimensional resolution, i.e., the simulated resolution cell, centered on a strong point, such as... Figure 2 As shown by the small, solid, dark gray ellipse, with n as the minor radius and e as the major radius, an elliptical domain is formed that makes an angle θ with the horizontal axis, as follows. Figure 2 The hollow gray ellipse represents the reference region used when generating the set of points to be registered from this strong point. Figure 2 Small solid ellipses in light gray represent strong points in the auxiliary image.

[0057] Let the two-dimensional coordinates of the p-th strong point on day d be... The set of all strong points extracted on day d+1 is m represents the number of strong points extracted on that day, and Then the points in the set of points to be registered for the p-th strong point on day d. The satisfied relationship is:

[0058]

[0059] in:

[0060]

[0061] The size of the set of points to be registered will vary depending on the location of the strong points; all points that satisfy the above inequality will be uniformly formed. Figure 1 The three light gray solid ellipses inside the medium gray hollow ellipse are the points in the set of points to be registered.

[0062] After the above processing, the set of points to be registered corresponding to the strong points each day will be obtained, denoted as Q = {Q1, Q2, ..., Q...} d ,…,Q D},in

[0063] Step 3: Dynamically change the main image during registration according to the sequential order, and perform coarse registration based on the maximum correlation coefficient criterion to obtain corresponding points.

[0064] Registration is performed sequentially, with the image from day d as the primary image and the image from day d+1 as the secondary image. Let the position of the p-th strong point in the primary image be: Its corresponding set of registration points is The position of the i-th point to be registered in the point set is: First: Extract the resolvable unit of this point on the main image, that is: Centered on the image, extend a-1, Aa, r-1, and Rr pixel units upwards, downwards, leftwards, and rightwards respectively, forming the main image strong pixel resolution unit matrix M, denoted as:

[0065]

[0066] Where m ar express The position on the main image corresponds to the pixel value on the main image. Similarly, the resolution unit matrix M′ corresponding to the registration point on the auxiliary image can be obtained. i .

[0067] Then, the simulated resolving unit contour matrix F is compared with the resolving unit matrix M of the main image to be registered and the corresponding matrix resolving unit matrix M′ of the to be registered point. i Dot product operation is performed to extract the effective resolvable units of each strong point. Then, the strong points in the main image are... The extracted effective resolving units and the effective resolving units of each strong point in their respective registration point set are subjected to complex correlation coefficient calculations. The point with the highest correlation coefficient in the auxiliary image is selected as the registration point, thus converting the one-to-many relationship of strong points in the primary and secondary images into a one-to-one relationship. The formula for calculating the correlation coefficient is:

[0068]

[0069] Step 4: Within the registration range centered on the point to be registered in the auxiliary image, fine registration is performed by translating the simulation resolution unit template and based on the maximum correlation coefficient criterion, thereby achieving the position correction of the point to be registered.

[0070] Suppose strong points in the main image Within the set of registration points of the auxiliary image Points achieved coarse registration. First, within the auxiliary image, using... Centered on the data, expand the data upwards and downwards by R pixels and horizontally by A pixels, thus forming a registration region of size (2R+1)×(2A+1), denoted as:

[0071]

[0072] Where s (A+1)×(R+1) express The position on the main image corresponds to the pixel value on the auxiliary image, and each number in matrix S is a complex number. During translation registration, if the simulated resolution unit is translated by u and v pixels in the two-dimensional direction within the registration area, then matrix F is expanded as follows:

[0073]

[0074] Calculate S*F and extract only the data from row u+1 to row u+1+a and column v+1 to column v+1+r, denoted as:

[0075]

[0076] The formula for calculating the correlation coefficient under this translation registration is:

[0077]

[0078] For different translation positions within the expanded registration range, we can obtain a set of all correlation coefficient calculation results, denoted as ρ, which has R×A elements. The precise registration position of this point on the auxiliary image is then:

[0079]

[0080] Based on the above method, all strong points in all images are processed in the same way, thereby achieving coarse and fine registration of all strong points in time-series SAR images.

[0081] Step 5: Correct the registration results. Use skip registration for points with short registration links and redistribute registration links where they intersect.

[0082] After completing the sequential registration steps described above, strong points in each image will form their own registration links. Due to the low signal-to-noise ratio and bistatic configuration of GNSS-based InSAR systems, differences in the focus position and focus shape of resolution cells lead to: 1. Inconsistent link lengths for each point; 2. Intersecting registration links, meaning a strong point in the secondary image may be a registration point for multiple strong points in the primary image. Phenomenon 1 affects the number of deformation observation points, resulting in decreased observation accuracy. To overcome this problem and increase the number of deformation observation points, a skip-type sequential registration can be adopted, skipping at the moment of the first link interruption, for example, for points that have not yet formed registration links. Regenerate the set of points to be registered on day d+2 and perform sequential registration. For phenomenon 2, it is necessary to redistribute the registration results of strong points with intersections in the sets of points to be registered based on the maximum correlation coefficient criterion. That is, compare the correlation coefficients of these points with the registration results and assign the registration results to the points with larger correlation coefficients. This completes the correction of the sequential registration results.

[0083] The following describes the results of the measured data processing. In this embodiment, based on the parameters in Table 1, an artificial lake in Changshu, Jiangsu Province, was used as the experimental scenario. The scenario area was 350,000 square meters. A dedicated receiver for receiving navigation satellite signals was placed there to receive the reflected signals emitted by navigation satellites into the scenario and perform subsequent imaging and registration processing. A total of 22 SAR images were obtained from June 29 to July 25, 2021.

[0084] Table 1

[0085]

[0086] Point target simulation imaging was performed using satellite orbit data from day 1, and the 3dB resolution cell outline and peak point position of the target were extracted. The final two-dimensional size of the ideal resolution cell outline in the same orbit was 13m × 5m, and the peak position was (7, 3). Strong points were extracted from 22-day images using this simulated resolution cell outline. The distribution of the extracted strong points is as follows: Figure 3 As shown, the set of points to be registered corresponding to strong points on the main image is as follows: Figure 4 As shown, the result of sequential registration according to the method proposed in this invention is as follows: Figure 5 As shown.

[0087] In the sequential registration results, strong points with a link length greater than 2 days within images from consecutive days are considered to be the same target point. In this embodiment, a total of 943 independent strong points were extracted from 22 days of images, of which 116 points had a registration link length of more than 5 days and 21 points had a registration link length of more than 10 days, which can provide sufficient information for long-term monitoring of deformation in a 350,000 square meter scene.

[0088] To further illustrate the effectiveness of the proposed method, all the aforementioned strong points were processed using a traditional registration method. The traditional registration method uses the image from the first day as the master image, and registers each subsequent day's image with the image from the first day. Based on the link length of each strong point obtained from sequential registration, the 22-day registration link length obtained by the traditional registration method was truncated. Then, the average correlation coefficient of each strong point for each day under both registration methods was calculated. The results... Figure 6 As shown in the figure, the correlation coefficient of sequential registration is much higher than that of traditional registration, and sequential registration can overcome the decorrelation effect caused by long time intervals.

[0089] The average daily registration offset under the two registration methods is as follows: Figure 7 (North-South direction) and Figure 8 As shown in the figure (north-south direction), the average offset of the proposed method is smaller than that of the traditional registration method in both the north-south and east-west directions. Especially when the registration link length exceeds 13 days, the registration offset of the traditional registration method increases significantly due to the effect of temporal decorrelation. This verifies the effectiveness of the proposed method, which can improve the shortcomings of traditional methods and enhance the temporal coherence of target points within navigation satellite time-series SAR images.

[0090] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A navigation satellite bistatic InSAR image registration method based on sequential translation, characterized in that, include: Step 1: Determine the main image and auxiliary image sequentially according to the temporal SAR image sequence; Based on the simulation resolution unit template, strong points are extracted from the main image and the auxiliary image; The simulation resolution unit template is as follows: Where A and R are the maximum number of pixels in the north-south and east-west directions of the simulated resolution unit template, respectively, determined by the satellite trajectory and system parameters; f ij A value of 1 indicates that the object is inside the simulated resolution cell outline, and a value of 0 indicates that the object is outside the simulated resolution cell outline. i = 1, 2, ..., a, ... A; j = 1, 2, ..., r, ... R; The simulation resolution unit is represented by the following fuzzy function, and the 3dB resolution unit contour is taken. P1 is the target point, and P2 is a point around the target point P1, which is the target point P1 imaged and simulated based on the real trajectory and real system parameters. Where, Φ TA and Φ RA These are the unit vectors of the transmitter and receiver relative to the target point P1, respectively; β is the bibase angle; Θ is the vector along the angle bisector of β; ω E Ξ and Ξ are the equivalent angular velocity and equivalent direction of motion of the transmitter; p is the range pulse compression result; m A This is the result of azimuth pulse compression; λ is the wavelength, c is the speed of light; the superscript T indicates matrix rank transformation; In image I, the simulation resolution unit template is globally translated. If the amplitude of pixel P at the center point of the simulation resolution unit template is the maximum value within the coverage area of ​​the simulation resolution unit template, then the center point is considered to be a strong point in image I. Step 2: For each strong point in the main image, construct a corresponding reference region in the auxiliary image, and extract the strong points in the auxiliary image within the reference region to construct a candidate set of points to be registered; the reference region is an elliptical domain centered on the strong point, with a radiation radius Ra = {n, e} and a major radius e at an angle θ with the horizontal axis; n and e are the two-dimensional resolution of the navigation satellite bistatic InSAR system, in pixels; and These represent the range resolution and azimuth resolution of the simulated resolution unit template, respectively. It is the unit vector along the x-axis; Step 3: Perform coarse registration for each strong point in the main image: Based on the simulated resolution unit template, extract the effective resolution units of the strong points in the main image, as well as the effective resolution units of each candidate point to be registered in the auxiliary image corresponding to the strong point in the main image; calculate the complex correlation coefficient between the effective resolution units of the strong points in the main image and the effective resolution units of each candidate point to be registered in the corresponding auxiliary image, and the candidate point to be registered with the largest correlation coefficient is the point to be registered for the strong point in the main image. Step 4: For each strong point in the main image, perform fine registration: In the expanded registration range area centered on the point to be registered in the auxiliary image, calculate the correlation coefficient between the simulation resolution unit of the strong point in the main image and the simulation resolution unit at different translation positions by translating the simulation resolution unit template. The maximum correlation coefficient corresponds to the fine registration position of the strong point in the main image on the auxiliary image. Repeat steps 1 to 4 to complete the coarse and fine registration of all strong points in the time-series SAR image, forming a registration link for each strong point in the image.

2. The method as described in claim 1, characterized in that, In step 3, the formula for calculating the correlation coefficient is: Among them, M and M i Let M and M' be the resolution unit matrices of the strong point in the main image and the i-th candidate registration point in the auxiliary image, respectively. They are obtained by expanding pixel units outwards from the strong point in the main image and the i-th candidate registration point in the auxiliary image, respectively. i The size of ′ is the same as the contour matrix F of the simulation resolution unit template; the superscript H indicates the conjugate transpose, "|||| 2 "" indicates squaring each element in the matrix, and "||" indicates taking the absolute value of each element in the matrix.

3. The method as described in claim 1, characterized in that, In step 4, the registration range is expanded by extending the area by R pixels vertically and A pixels horizontally, centered on the point to be registered in the auxiliary image, thus forming an expanded registration range of size (2R+1)×(2A+1). Here, A and R are the maximum number of pixels in the north-south and east-west directions of the simulation resolution unit template, respectively, which are determined by the satellite trajectory and system parameters.

4. The method as described in claim 3, characterized in that, The formula for calculating the correlation coefficient in step 4 is as follows: M is the resolution unit matrix of the main image strong points, which is obtained by expanding pixel units outward from the main image strong points as the center. The size of M is the same as the contour matrix F of the simulation resolution unit template. The matrix represents the resolvable unit at the translation position of the auxiliary image; the superscript T indicates matrix transposition, and the superscript H indicates taking the conjugate transposition. 2 "" indicates squaring each element in the matrix, and "||" indicates taking the absolute value of each element in the matrix.

5. The method as described in claim 1, characterized in that, Also includes: Step 5, correct the registration link: If the registration link is interrupted, the interrupted image is skipped, and coarse and fine matching are performed directly with the subsequent images to form a registration link; If a strong point on the auxiliary image is a registration point for multiple strong points on the main image, then the correlation coefficients between each strong point in the main image and the strong point in the auxiliary image are compared, and the strong point in the auxiliary image is assigned to the strong point in the main image corresponding to the largest correlation coefficient.

Images