Multi-pass chromatography three-dimensional imaging method based on actual measurement airborne SAR (Synthetic Aperture Radar)

Through SIFT precise registration and PGA phase correction technology, the problem of the same name point offset and low accuracy in the onboard multi-avald SAR system is solved, and the coherence between channels and three-dimensional reconstruction accuracy is improved. It is suitable for high-resolution SAR images deal with.

CN118918151APending Publication Date: 2024-11-08NORTHWESTERN POLYTECHNICAL UNIV
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410759763.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-13
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Due to the unstable flight trajectory and large differences in observation angles, the existing multi-avald pass through TomoSAR technology based on the multi-avald pass through the airborne system is difficult to be applicable to airborne SAR data.

Method used

The SIFT precise registration method is used to register images of multi-avalent SAR images, and through PGA phase correction technology, combined with phase and power information, the pixel point with the smallest difference in normalized power phase is selected as the point of the same name, improving the coherence between channels. and three-dimensional reconstruction accuracy.

Benefits of technology

The problem of point offset of the same name is initially solved, the coherence between channels is improved, the accuracy of three-dimensional reconstruction is enhanced, and it is suitable for processing high-resolution SAR images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918151A_ABST
    Figure CN118918151A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-pass tomography three-dimensional imaging method based on an actual measurement airborne SAR (Synthetic Aperture Radar), which comprises the following steps: taking each SAR image obtained by multi-pass as a main image, and respectively calculating a total coherence coefficient mean value of other images and the main image; performing image registration on the group of main images and auxiliary images with the highest total coherence coefficient mean value to obtain registered images; performing phase correction on the registered image to obtain a corrected image; calculating the nearest reference slope distance of all points in the elevation grid; all homonymy points on the flight are screened out; the focusing energy of the origin of the coordinates is calculated, and all points on the elevation grids are traversed to obtain an imaging result; according to the method, the problems of incapability of imaging and low precision caused by homonymy point offset and observation angle change of the airborne SAR are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of tomoSAR (TomoSAR) three-dimensional imaging, and in particular relates to a multi-flight tomographic three-dimensional imaging method based on measured airborne SAR. Background Art

[0002] Tomographic Synthetic Aperture Radar (TomoSAR) is an advanced radar imaging technology that combines the principles of synthetic aperture radar (SAR) and tomography. By observing the same target from multiple perspectives and angles, TomoSAR can achieve three-dimensional reconstruction of the target structure. Airborne multi-flight TomoSAR three-dimensional imaging technology is an extended application of TomoSAR technology. It uses multiple flights over the same area to obtain SAR data under different baselines. Through different baselines, SAR observation data from different directions and perspectives can be obtained, which can be used for tomographic three-dimensional reconstruction. Compared with airborne array SAR systems, traditional airborne multi-flight SAR systems usually have a larger field of view and coverage, but the resolution is lower for long-distance observation targets. Multi-flight TomoSAR observations require a time span of several weeks or even longer for multi-channel acquisition, depending on the duration of the flight mission and the number of images required. Therefore, there will be a large difference in observation angles between different flyby SAR images, resulting in a weakening of coherence between channels. In addition, due to the unstable flight trajectory, the same pixel point in different channels of the airborne multi-pass TomoSAR will produce a larger offset compared to the spaceborne multi-pass TomoSAR system, making the deformation difference difficult to ignore.

[0003] Most of the existing TomoSAR imaging technologies based on multi-flyby SAR systems are oriented towards spaceborne data.

[0004] The spaceborne multi-pass TomoSAR imaging process can be summarized as follows:

[0005] (1) Perform multi-channel complex image registration on the acquired multi-channel single-view complex image sequence.

[0006] (2) De-skewing the registered multi-channel image to remove the reference slant range phase.

[0007] (3) Improve the quality of SAR tomography by performing phase correction on multi-channel complex images.

[0008] (4) After preprocessing the multi-channel complex images, the data is subjected to tomographic processing to achieve three-dimensional reconstruction of the target.

[0009] Since there is a big difference between the data obtained by spaceborne SAR and the data obtained by airborne SAR, the TomoSAR technology based on the spaceborne multi-flight system is not well suited for the imaging of airborne multi-flight SAR technology. Generally speaking, the orbit of the spaceborne SAR system is relatively stable, while the flight trajectory of the airborne multi-flight SAR system is not stable. Therefore, in the multi-channel data obtained by the airborne multi-flight SAR system, the pixel points representing the same target in different channels will produce a larger offset, and even reach the deviation of the resolution size. In addition, since the height of the spaceborne SAR system platform is high enough, the error caused by the change of the observation angle can be ignored. Moreover, the resolution of the image data obtained by the spaceborne SAR system is generally lower than that of the airborne SAR system. Summary of the invention

[0010] The purpose of the present invention is to provide a multi-flight tomography three-dimensional imaging method based on measured airborne SAR to solve the problems of airborne SAR's homonymous point offset and the inability to image due to changes in observation angles and low precision.

[0011] The present invention adopts the following technical scheme: a multi-flight tomographic three-dimensional imaging method based on measured airborne SAR, comprising:

[0012] Step 1: Taking each SAR image obtained by multiple flights as the main image, respectively calculating the total coherence coefficient mean of other images and the main image; the other images are images other than the main image among the SAR images obtained by multiple flights;

[0013] Step 2: Perform image registration on a group of primary images and auxiliary images with the highest total coherence coefficient mean value to obtain a registered image;

[0014] Step 3: Perform phase correction on the registered image to obtain a corrected image;

[0015] Step 4: Position the two-dimensional coordinates of the pixel points of the corrected image into the geographic coordinate system, and construct a new three-dimensional coordinate system with any point of the corrected image as the coordinate origin and the azimuth-slant range-elevation directions as the coordinate axes. In this coordinate system, construct an elevation grid and calculate the nearest reference slant distances of all points in the elevation grid.

[0016] Step 5: Calculate the steering vectors of different elevations of each pixel point based on the nearest reference slant distance, and calculate the normalized power phase difference between the coordinate origin of the correction image and the neighboring pixel points centered on this point on other SAR images. Select the pixel point with the smallest normalized power phase difference as the same-name point, and screen out all the same-name points on the flyby;

[0017] Step 6: Calculate the scattered power at the coordinate origin and traverse all points on the elevation grid to obtain the imaging result.

[0018] Furthermore, the calculation formula of the normalized power phase difference in step 5 is:

[0019]

[0020] In the formula, E c (u, v) is the normalized power difference between the pixel and the neighboring pixels centered on the pixel on other flying SAR images, P c (u, v) is the normalized phase difference between the pixel and the neighboring pixels centered on the pixel in other passes; r r is the grey absolute correlation degree and r r ∈[0,1]; c is the number of times flown, (u,v) is the pixel coordinates in other flown neighborhood windows; W is the neighborhood window.

[0021] Further,

[0022]

[0023] In the formula, g m (x, r) is the value of the pixel with coordinates (x, r) on the main image; g c (u,v) is the value of the neighborhood pixel point centered on (x,r) on other SAR images; (u,v) is the coordinate of the neighborhood pixel point; c is the number of times it has flown; and W is the neighborhood window.

[0024] Further,

[0025]

[0026] In the formula, φ m (x, r) is the phase of the pixel with coordinates (x, r) on the main image; φ c (u, v) is the phase of the neighboring pixel centered at (x, r) on other flyby SAR images, c is the number of flybys, (u, v) is the coordinates of the neighboring pixel; R m (s l ) is the coordinate (x, r) on the main image in the elevation direction s l Slope distance at c (s l ) is the cth flight in the height direction s l The slope distance at .

[0027] Furthermore, the calculation formula of the total coherence coefficient mean in step 1 is:

[0028]

[0029] In the formula, w s With h sare the pixel width and height of the SAR image, E(·) is the operation of the expected value of the pixel in the image, g1 and g2 are the main image and the auxiliary image, M is the number of times of flying, and m is the mth group of image pairs. The beneficial effects of the present invention are:

[0030] The present invention adopts SIFT-based precise registration to preliminarily solve the problem of homonymous point offset; the present invention adopts the PGA phase correction method to solve the problem of observation angle difference of airborne multi-flyover SAR data; the present invention accurately aligns homonymous points of the same pixel on the main image, thereby improving the coherence between channels;

[0031] In view of the problem of unstable orbit of airborne multi-flyover SAR system, the present invention adopts SIFT precise registration method to preliminarily solve the problem of offset between same-name points and ensure accurate registration of images; in view of the problem of observation angle difference of airborne multi-flyover SAR data, multi-channel phase correction technology is adopted to improve the coherence between different channels and ensure subsequent tomography processing; the combined phase and power information further improves the offset problem of same-name points, increases coherence, improves the accuracy of three-dimensional reconstruction, and also has certain advantages in processing high-resolution SAR images. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The experimental data obtained by the airborne multi-pass N-SAR system in Example 1 in Weinan, Shaanxi Province, Figure 1 (a) is the optical image of the target building. Figure 1 (b) is the amplitude diagram of airborne data;

[0033] Figure 2 is the mean value of the total coherence coefficient of different sequences of airborne multi-flyover SAR data in Example 1;

[0034] Figure 3 Comparison of amplitude maps before and after registration of the target building area by the two methods in Example 1;

[0035] Figure 4 1 is the BF imaging algorithm in Example 1 and the reconstruction result of Example 1. DETAILED DESCRIPTION

[0036] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] The present invention discloses a multi-flight tomography three-dimensional imaging method based on measured airborne SAR, comprising:

[0038] Step 1: Taking each SAR image obtained by multiple flights as the main image, respectively calculate the total coherence coefficient mean of other images and the main image; the other images are images other than the main image in each SAR image obtained by multiple flights.

[0039]

[0040] Among them, the coherence coefficient of each image pair is expressed as

[0041]

[0042] Where E(·) is the expected value of pixels in the image, g1 and g2 are the main image and auxiliary image respectively, and w s With h s are the pixel width and height of the SAR image respectively. Each SAR image is taken as the main image, and the mean total coherence coefficient of the remaining images and the main image is calculated. The sequence with the highest mean total coherence coefficient is selected as the final experimental sequence. M is the number of passes, and m is the mth group of image pairs.

[0043] Step 2: Perform image registration on a group of primary images and auxiliary images with the highest total coherence coefficient mean value to obtain the registered image.

[0044] The purpose of image registration is to make the pixels at the same position in the two images after registration, that is, the same points, correspond to the same target in the measured scene, that is, the same resolution unit. Before registration, the elements at the same position in the two images correspond to different resolution units in the measured scene, so the phase difference obtained cannot reflect the actual ground undulation. Therefore, image registration is a key step in TomoSAR 3D measurement. Image registration is mainly divided into the following steps.

[0045] ① Cross-correlation method for rough registration

[0046] Before fine registration, it is necessary to perform pixel-level registration on the complex image pairs, that is, coarse registration. For coarse registration, this technology uses the cross-correlation image method. The cross-correlation method is based on the statistical characteristics of the image correlation coefficient between the main and auxiliary images. The offset corresponding to the maximum value is the offset between the complex image pairs. When the image satisfies the ergodicity, the sample average can be used to estimate the mean in the appropriately sized area. When the window slides, the correlation coefficient between the matching window and the sliding window can be expressed as:

[0047]

[0048] Where △i and △j are the coordinate translations of the image in the slant range and azimuth directions respectively. γ(△i,△j) represents the correlation coefficient between the auxiliary image g2(i,j) and the main image g1(i,j) after the translation (△i,△j). When its modulus is the largest, the registration result between the main and auxiliary images is the best. At this time, the offset between the main and auxiliary images is the translation (△i,△j).

[0049] The cross-correlation algorithm can be performed in both the spatial domain and the frequency domain. In the frequency domain, a pair of InSAR images is subjected to two-dimensional cross-correlation, with G1 and G2 representing the main and auxiliary images in the frequency domain after Fourier transform, and (u, v) representing the pixel position on the cross-power spectrum. The cross-power spectrum P(u, v) of the main and auxiliary images is:

[0050]

[0051] in, is the complex conjugate of G2(U,V). Performing an inverse discrete Fourier transform on P(u,v) yields the impulse function:

[0052] F -1 (P(u,v))=F -1 (e j2π(u△i+v△j) )=δ(△i,△j) (4)

[0053] The peak value of correlation appears at the position where the corresponding image pair is exactly aligned. The position of the peak value is also the displacement vector required, that is, the coordinates (△i,△j) corresponding to the maximum value of the impulse function are the offset. The cross-correlation can be quickly calculated through fast Fourier transform to achieve the purpose of rough alignment.

[0054] ②Precise registration based on scale-invariant feature transformation

[0055] Fine registration is sub-pixel registration, so before fine registration, the coarsely registered main and auxiliary images can be interpolated and resampled. Commonly used interpolation functions include sinc function, bilinear interpolation, etc. By searching each pixel after interpolation, a sub-pixel registration result can be obtained, and finally the optimal offset between the main and auxiliary images can be obtained.

[0056] Generally speaking, the amplitude and phase errors and deformations between the image data acquired by spaceborne SAR and airborne multi-array SAR systems are small, so the traditional SAR image registration algorithm can be used. However, for airborne multi-flyover SAR systems, the traditional registration algorithm is not applicable. This is because the close-range and long-range variations of airborne multi-flyover SAR systems are large, and the images will have certain local deformations. The N-SAR measured data used in the experiment also has such problems. Therefore, this technology uses the SIFT (Scale-invariant Feature Transform) algorithm to eliminate the translation, rotation and scale differences between images, thereby improving coherence and increasing the accuracy of three-dimensional reconstruction.

[0057] The SIFT algorithm is a classic algorithm for detecting and describing local features in images. It constructs a scale space by performing multi-scale pixel sampling and Gaussian smoothing filtering on the original image, and then searches and describes key points in the scale space to generate key point descriptors. These key point descriptors are invariant to scale and rotation, so they can be used to match feature points between different images. The SIFT algorithm steps can be divided into the following steps:

[0058] A. Generate DoG (Difference of Gaussian) scale space

[0059] The Gaussian scale space L(i, j, σ) of an image is defined as the convolution of the original image I(i, j) with a 2D Gaussian function G(i, j, σ) of a variable scale, that is:

[0060] L(i,j,σ)=G(i,j,σ)*I(i,j) (5)

[0061]

[0062] Where (i, j) is the pixel coordinate, and σ is the variance of the Gaussian normal distribution. The DoG pyramid of the image is established by subtracting adjacent images in two Gaussian scale spaces:

[0063]

[0064] Where D(i,j,σ) is the DoG pyramid. c is the scale ratio of adjacent images in scale space, a constant.

[0065] B. Find local extreme points that meet the conditions

[0066] In the DoG scale space, extreme points are usually regarded as significant feature points of the image. By traversing each sampling point and comparing it with the surrounding adjacent points, it is determined whether the point is a local extreme point in its domain and scale space. Because precise registration is sub-pixel image registration, in order to achieve higher accuracy requirements, the algorithm further processes the feature points. Surface fitting is performed using a three-dimensional quadratic function to obtain a more accurate position of the extreme points. A threshold is set to eliminate low contrast, and the Hessian matrix is ​​used to remove the edge response interference caused by the Gaussian difference operation to optimize the feature point detection results.

[0067] C. Determine the direction of the feature points.

[0068] After finding the position of the feature point, the main direction of the feature point is determined according to the gradient direction of the pixels in the neighborhood of the feature point. The modulus m(i, j) and direction (i, j) of the gradient of the Gaussian smoothed image at point (i, j) can be expressed by equations (2-21) and (2-22) respectively:

[0069]

[0070] D. Generate keypoint descriptors.

[0071] The local feature points found above are the key points we need. Take a 16×16 pixel area with the key point as the center, divide this area into 4×4 sub-blocks, divide the gradient directions of all pixels in the sub-block into 8 directions, and count the gradient histograms of the 8 directions of each sub-block to get a seed point. Each key point is composed of 4×4 seed points, and each seed point has 8 directions, thus forming a 4×4×8=128 feature vector, which has rotation invariance, scale invariance, etc. In order to ensure that the vector has a certain degree of illumination invariance, it also needs to be normalized.

[0072] E. Feature point matching

[0073] The SIFT matching algorithm determines the similarity based on the Euclidean distance between feature descriptors. The smaller the Euclidean distance, the higher the similarity. The matching criterion used here is the nearest neighbor matching method (k-nearest neighbor, KNN). The feature point sets of the original image and the matching image are defined as A = {X1, X2, ..., X K} and B={Y1,Y2,...,Y K}, the feature distance function of the feature point can be expressed as:

[0074]

[0075] Among them, X in , X jn Represent the nth feature component of the key point respectively. According to the feature distance function, the corresponding value and the matching degree between the feature points are obtained. For each feature point, the KNN algorithm is used to obtain the nearest neighbor value d between the point and the feature point in another picture. n and the next nearest neighbor value d m , and then calculate the ratio of these two values ​​d n / d m When the ratio is less than the set threshold, it indicates that the point has a high degree of match with the nearest neighbor point, and it can be regarded as the best matching point for the point.

[0076] F. Matching pair optimization.

[0077] After obtaining the matching point information, the RANSAC algorithm is then used to optimize the matching to obtain the best matching information of the image feature points. The standard RANSAC algorithm is used to calculate the similarity in the model. The algorithm randomly selects a supporting sample set and then finds the set with the largest number of samples as the basis of the final model.

[0078] After completing the above steps, the auxiliary image is resampled using the obtained transformation matrix and calibrated to the space of the main image to obtain the SIFT-based registration result.

[0079] In terms of image registration, the correlation function between the interferometric complex image pairs can be used as the registration metric and the registration can be performed through fast Fourier transform. However, for high-resolution multi-pass airborne data, the above method is not applicable due to the large changes in close and long distances and the existence of certain local deformations. In terms of phase error correction, a method based on the intermediate output of PS-INSAR can be used to compensate for the phase error, but compared with the PGA method, this method has low computational efficiency and insufficient adaptability.

[0080] Step 3: Perform phase correction on the registered image to obtain the corrected image.

[0081] Phase Gradient Autofocus (PGA) algorithm is usually used in synthetic aperture radar imaging to correct phase errors caused by platform motion or system parameter errors. If the phase error of multi-channel data of an airborne multi-flying SAR system is estimated and compensated based on the PGA algorithm, it is necessary to rely on some simple resolution units: azimuth-range resolution units containing only one strong scattering element. Therefore, before performing phase correction, it is necessary to first determine whether there are special units in the multi-channel SAR data and detect these special units.

[0082] Assume that in a resolution unit (x0, y0) within the target range, there is only one scattering unit with a height of zero, and the scattering unit is only located within one resolution unit. At this time, the backscattering characteristic function γ(s) of the scattering unit can be approximately regarded as an ideal pulse signal:

[0083] γ(s)=g·δ(s) (11)

[0084] Where γ is the backscattering coefficient of the scattering unit:

[0085]

[0086] Where A is the amplitude of γ, is the phase of γ. For the convenience of later analysis, the azimuth-slant range index (x0, y0) is omitted here. Then the spectrum of γ(s) containing the phase error can be expressed as:

[0087]

[0088] Where g(x) is obtained by Fourier transform of γ(s), and the derivative of equation (2-25) is:

[0089]

[0090] The derivative of the phase error can be obtained as:

[0091]

[0092] By integrating the phase error derivative, the phase error is obtained as:

[0093]

[0094] Through the above steps, the phase error value can be obtained, and the phase error compensation can be realized. This is the fundamental starting point of the PGA algorithm. However, the above analysis is based on the assumption that g(x) is a continuous signal and the height of the strong scattering element is zero. In general, the observation data we obtain are discrete sampling values ​​of the continuous spectrum, and the strong scattering unit should not always be at the zero point.

[0095]

[0096] Among them, s0 is the elevation of the scattering unit. The estimated value s of the scattering unit position is obtained by some methods. The g(x m ) still has phase error, assuming The corresponding linear phase term 2πx m s0 is removed from the obtained result:

[0097]

[0098] In a similar way to processing continuous signals, the difference operation is performed on two adjacent sample values ​​in the result after removing the linear phase:

[0099]

[0100] make:

[0101] △F(x m )=F(x m )-F(x m-1 ) (twenty one)

[0102]

[0103] Then (21) can be expressed as:

[0104]

[0105] if is a small value, then Taylor expansion and retaining first-order terms:

[0106]

[0107] Substituting the above formula into (24), we get:

[0108]

[0109] From this we can get the phase error difference as:

[0110]

[0111] Phase Error The estimate of is the sum of the differences in phase errors, that is:

[0112]

[0113] At this point, the original signal complex value g(x m ,y,r) and The effect of phase error can be removed by multiplication:

[0114]

[0115] Step 4: Locate the two-dimensional coordinates of the pixel points of the corrected image in the geographic coordinate system, construct a new three-dimensional coordinate system with any point of the corrected image as the coordinate origin and the azimuth-slant range-elevation directions as coordinate axes, construct an elevation grid in this coordinate system, and calculate the nearest reference slant distances of all points in the elevation grid.

[0116] Specifically:

[0117] Select any point (x, r) in the corrected image as the reference pixel point P, locate it in the geographic coordinate system from the azimuth-slant range direction, and construct a new three-dimensional coordinate system with point P as the coordinate origin and azimuth-slant range-elevation as the coordinate axes. s ,0.83ρ s ] is the spacing to construct the elevation grid, where ρ s is the elevation imaging resolution ρ of the airborne TomoSAR system s .

[0118] To solve the problem of accurate deslant and same-name point mismatch in airborne multi-flight data, the BF (Beamforming) algorithm is used to obtain the accurate nearest reference slant range R. ref The nearest reference slant distance R is obtained from the data collected by the radar POS machine and the scene data. ref First, construct a three-dimensional imaging grid and select a point s on the grid l , calculate the nearest reference slope distance R ref.

[0119] To construct a three-dimensional imaging grid, it is necessary to convert the system posture information and the two-dimensional SAR image into a new coordinate system in which the z-axis is perpendicular to the earth's tangent plane and the x-axis is the velocity direction.

[0120] First, each pixel in the corrected image is positioned in the geographic coordinate system to achieve the same coordinate system as the POS machine’s posture information. For the pixel point P at (x, r) on the corrected image, the pixel point is positioned based on the slant range-Doppler based on the posture information of the reference flight. The azimuth-slant range information of the pixel point and its corresponding position in the Gauss Kruger coordinate system (X P ,Y P ,Z P ) should satisfy the following relationship:

[0121]

[0122] Among them (E t ,N t ,U t ,V Et ,V Nt ,V Ut ) is the three-dimensional position and speed information of the aircraft in the northeast sky coordinate system corresponding to the azimuth time t given by the POS machine, (E P ,N P ,U P ) is the three-dimensional position of point P in the northeast celestial coordinate system, f d is the Doppler frequency, λ is the wavelength, and R is the instantaneous slant range.

[0123] For spaceborne SAR, the positioning solution of the above model requires complex iteration based on distance, Doppler, and earth model. However, the flight altitude of airborne SAR is limited and the observation area is small, so the corresponding Gauss Krüger coordinate system information can be directly solved. For the known average height h of the imaging area m , calculate the coordinates of the reference point P relative to the aircraft position:

[0124]

[0125] in:

[0126]

[0127] Solving (32) yields:

[0128]

[0129] Then the ground coordinates corresponding to the point (x, r) on the original SAR two-dimensional image can be obtained as:

[0130]

[0131] To convert to the geographic coordinate system, you can call the GaussProInverse function of MATLAB to finally get the longitude and latitude positioning results of point P (Latp, Lonp), thus unifying the position coordinate systems of all passing aircraft provided by the POS machine.

[0132] Then, the longitude, latitude, and altitude data of all the POS machines passing by, as well as the component of the aircraft speed in the northeast sky (ENU) coordinate system (V Et ,V Nt ,V Ut ) to perform coordinate conversion. Select the reference point P with known position information as the coordinate origin, and uniformly convert it to a coordinate system with the z-axis direction perpendicular to the earth's tangent plane and the flight speed direction as the x-axis. Data in the geographic coordinate system (Lat t ,Lon t ,Height t ) to the ENU coordinate system (E t ,N t ,U t ) can directly call the geodetic2enu function in matlab. Then, according to the aircraft attitude data, the ENU coordinate system position data is converted to the coordinate system to be determined. Given the aircraft heading angle ψ, the aircraft position and carrier speed after coordinate conversion are:

[0133]

[0134] A new three-dimensional coordinate system is constructed with point P as the origin and azimuth-slope-elevation as the coordinate axes, and △s∈[0.50ρ s ,0.83ρ s ] is the spacing to construct the elevation grid. For a certain scattering point l, reverse positioning is performed in the new coordinate system to estimate the position of the scattering point at zero Doppler under the reference flight, and obtain the aircraft posture information at this time, thereby calculating the reference closest slant distance R of the echo of the scattering point under the reference flight ref . Construct a three-dimensional imaging grid in the azimuth-slant range-elevation coordinate system. Take point P as the coordinate origin and construct a grid with △s∈[0.50ρ s ,0.83ρ s ] is the height grid with spacing, where ρ s is the elevation imaging resolution of the airborne TomoSAR system.

[0135] Step 5: Calculate the steering vectors of different elevations of each pixel point based on the nearest reference slant distance, and calculate the normalized power phase difference between the coordinate origin of the correction image and the neighboring pixel points centered on this point on other SAR images. Select the pixel point with the smallest normalized power phase difference as the same-name point, and screen out all the same-name points on the flyby.

[0136] Combined with R ref The vector formed by all the complex values ​​at (x, r) can finally be calculated to get (x, r, s l ) at the focus result:

[0137]

[0138] Among them, g is a vector composed of all the complex data of the two-dimensional SAR image at each pixel point, and a is the steering vector constructed at the pixel point (x, r):

[0139] g=[g1(x,r),g2(x,r),…,gn(x,r)…,gM(x,r)]T (39)

[0140] a(s l )=[a1(s l ),a2(s l ),…,a m (s l ),…,a M (s l )] T (40)

[0141] in,

[0142] In (x,r,s l ), R m (r,s l )=R ref .

[0143] Although the airborne multi-pass data has undergone multi-channel complex image registration, the inter-channel coherence is still not high enough, which indicates that there are still homonymous points that are not accurately aligned with the same pixel on the main image. Due to the large difference in observation angles and unstable flight tracks, the airborne high-resolution multi-pass data will cause scattering points from different altitudes in the same resolution unit in the reference flyby image to be scattered to the neighboring pixels of the point in other flyby SAR images, resulting in homonymous point mismatch. At the same time, due to the influence of coherent speckle noise, there are significant differences in the power and phase of the signals between different passes in the resolution unit.

[0144] Since the premise of searching for BF imaging is to align all the same-name points on the flyby SAR images to the same pixel, in order to search for the specific resolution unit neighborhood of the reference flyby image in other flyby SAR images and find the pixel point that is most relevant to the power and phase of the target resolution unit, a discrimination criterion combining minimization of power and phase difference is adopted. For a pixel (x, r) on the reference flyby SAR image, the normalized power difference between the pixel and the neighboring pixel points centered on the pixel on other flyby SAR images is calculated as:

[0145] in,

[0146] In the formula, g m (x, r) is the value of the pixel with coordinates (x, r) on the main image; g c (u,v) is the value of the neighboring pixel point centered at (x,r) on other SAR images; (u,v) is the coordinate of the neighboring pixel point; c is the number of times the pixel point has flown; W is the neighborhood window. Theoretically, the closer the power characteristics of the pixel point in the window are to the power characteristics of the central reference point, the better the E c The closer (u,v) is to zero.

[0147] The phase difference between the same-name points in different passes is theoretically caused by the slant distance between the radar observation point and the target point. After compensating for the phase difference, the normalized phase difference between the target pixel on the reference pass image and the field pixel centered on the pixel on other passes is defined as:

[0148]

[0149] When the normalized phase difference is zero, the selected pixel is the pixel with the most similar phase characteristics. However, the measured data is usually affected by factors such as noise, and the positioning results of the same-name points obtained by minimizing the normalized power difference or minimizing the phase difference are usually not completely consistent. In order to comprehensively consider these two normalized differences, the gray absolute correlation degree r is introduced. r describes the similarity of the normalized power and phase difference indicators, and r r ∈[0,1]. r = 1, the two are completely similar, and the optimal same-name point can be found no matter which one is used; r r = 0, the normalized phase difference is used as the reference for searching. So far, the normalized difference of joint phase and power (NDPP) is formed, which is expressed as:

[0150]

[0151] Find the pixel point corresponding to the minimum value, which is the point with the same name:

[0152]

[0153] Step 6: Calculate the focusing energy of the coordinate origin and traverse all points on the elevation grid to obtain the imaging result.

[0154] Based on all the points with the same name selected by (45), combined with Rref, we can substitute

[0155]

[0156] The TomoSAR focusing signal at that location can be calculated, and the above calculations are performed on all points of the elevation grid to finally obtain the TomoSAR imaging results.

[0157] Example 1

[0158] The experimental data was obtained by using the airborne multi-pass N-SAR system in Weinan, Shaanxi Province, with a total of 8 images. Figure 1 (a) is the optical image of the target building. Figure 1 (b) is the amplitude map of the airborne data, from which a single target is cropped. The image size is 400×400. The NSAR system parameters are shown in Table 1.

[0159] Table 1 Airborne multi-flyover NSAR system and scenario parameters

[0160]

[0161] Figure 2 is the mean total coherence coefficient of different sequences of airborne multi-flyover SAR data. It can be seen that the sequence with the third SAR image as the main image has the highest mean total coherence coefficient, so the third scene is selected as the main image.

[0162] Table 2. Comparison of coherence coefficients of the registration results with the main image being the third image

[0163]

[0164] Figure 3 Comparison of amplitude maps of the target building area before and after registration by two methods. Figure 3 (a) is the main image, Figure 3 (b) is the auxiliary image before registration, Figure 3 (c) is the auxiliary image after the traditional cross-correlation method registration. Figure 3 (d) is the auxiliary image after SIFT registration according to the present invention.

[0165] from Figure 3By comparing with Table 2, it can be concluded that after SIFT registration, the coherence coefficients of all image pairs are higher than those of the traditional cross-correlation registration method, and are greatly improved compared to before registration, indicating that the SIFT registration method has a significant effect on the registration of multi-pass airborne SAR data.

[0166] Table 3 Three evaluation indicators of point cloud obtained by BF imaging algorithm and this embodiment

[0167]

[0168] Figure 4 (a) is the reconstruction result of the BF imaging algorithm. Figure 4 (b) is the reconstruction result of this embodiment. Figure 4 As can be seen from Table 3, the point cloud generated by this embodiment achieves better results, indicating that the present invention obtains a point cloud with higher concentration and continuity and less confusion, and can effectively improve the result of three-dimensional reconstruction of building targets.

[0169] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-flight tomographic three-dimensional imaging method based on measured airborne SAR, characterized in that: include: Step 1: Take each SAR image obtained by multiple flights as the main image, and calculate the mean of the total coherence coefficients of other images and the main image respectively; The other images are images other than the main image among the SAR images obtained by multiple flights; Step 2: Perform image registration on a group of primary images and auxiliary images with the highest total coherence coefficient mean value to obtain a registered image; Step 3: Perform phase correction on the registered image to obtain a corrected image; Step 4: Position the two-dimensional coordinates of the pixel points of the corrected image into the geographic coordinate system, and construct a new three-dimensional coordinate system with any point of the corrected image as the coordinate origin and the azimuth-slant range-elevation directions as the coordinate axes. In this coordinate system, construct an elevation grid and calculate the nearest reference slant distances of all points in the elevation grid. Step 5: Calculate the steering vectors of different elevations of each pixel point based on the nearest reference slant distance, and calculate the normalized power phase difference between the coordinate origin of the correction image and the neighboring pixel points centered on this point on other SAR images. Select the pixel point with the smallest normalized power phase difference as the same-name point, and screen out all the same-name points on the flyby; Step 6: Calculate the focusing energy of the coordinate origin and traverse all points on the elevation grid to obtain the imaging result.

2. The method of multi-flight tomography 3D imaging based on measured airborne SAR according to claim 1, characterized in that: The calculation formula for the normalized power phase difference in step 5 is: In the formula, E c (u, v) is the normalized power difference between the pixel and the neighboring pixels centered on the pixel on other flying SAR images, P c (u, v) is the normalized phase difference between the pixel and the neighboring pixels centered on the pixel in other passes; r r is the grey absolute correlation degree and r r ∈[0,1]; c is the number of times flown, (u,v) is the pixel coordinates in other flown neighborhood windows; W is the neighborhood window.

3. The method of multi-flight tomography 3D imaging based on measured airborne SAR according to claim 2, characterized in that: c=1,2,…,M,c≠m,(u,v)∈W In the formula, g m (x, r) is the value of the pixel with coordinates (x, r) on the main image; g c (u,v) is the value of the neighborhood pixel point centered on (x,r) on other SAR images; (u,v) is the coordinate of the neighborhood pixel point; c is the number of times it has flown; and W is the neighborhood window.

4. The method of multi-flight tomography 3D imaging based on measured airborne SAR according to claim 2, characterized in that: c=1,2,…,M,c≠m,(u,v)∈W In the formula, φ m (x, r) is the phase of the pixel with coordinates (x, r) on the main image; φ c (u, v) is the phase of the neighboring pixel centered at (x, r) on other flyby SAR images, c is the number of flybys, (u, v) is the coordinates of the neighboring pixel; R m (s l ) is the coordinate (x, r) on the main image in the elevation direction s l Slope distance at c (s l ) is the cth flight in the height direction s l The slope distance at .

5. The method of multi-flight tomography 3D imaging based on measured airborne SAR according to claim 1, characterized in that: The calculation formula for the mean value of the total coherence coefficient in step 1 is: In the formula, w s With h s are the pixel width and height of the SAR image, E(·) is the operation of the expected value of the pixel in the image, g1 and g2 are the main image and auxiliary image, M is the number of times of flying, and m is the mth group of image pairs.

Citation Information

Cited By

  • Sea surface elevation measurement method and device based on multi-angle SAR (Synthetic Aperture Radar) image in same flight

    CN119879837A

  • Method and device for reconstructing airborne multi-pass InSAR image

    CN120595293A

  • Chromatographic SAR imaging method combining homonymy point focusing and subspace spectrum estimation

    CN120802265A