阅读说明：本技术 一种两次分块策略的广域sar复图像序列快速配准方法 (Wide-area SAR complex image sequence rapid registration method adopting twice blocking strategy ) 是由 杨波 江利明 徐华平 汪汉胜 黄荣刚 周志伟 于 2021-09-24 设计创作，主要内容包括：本发明公开了一种两次分块策略的广域SAR复图像序列快速配准方法,以感兴趣区域中心位置为中心,裁剪长时间SAR图像序列,确定主图像和辅图像；将长时间SAR图像序列重叠的区域裁剪出来作为裁剪SAR图像；对每个辅图像块进行粗配准；进行不重叠划分获得辅孙图像块和主孙图像块,采用控制点的辅孙图像块的行列配准偏移量与主孙图像块的中心元素位置,拟合多项式模型的参数,并行计算出任一辅图像每个辅图像块中每个像元的配准偏移量；对任意辅图像块的边界延拓图像进行并行重采样,输出精配准SAR辅图像块。利用本发明可以实现广域长时间SAR复图像序列高精度图像配准。(The invention discloses a method for quickly registering a wide area SAR complex image sequence by a twice blocking strategy, which comprises the steps of taking the central position of an interested area as the center, cutting the long-time SAR image sequence, and determining a main image and an auxiliary image; cutting out the overlapped region of the long-time SAR image sequence as a cut SAR image; performing coarse registration on each auxiliary image block; performing non-overlapping division to obtain auxiliary grandchild image blocks and main grandchild image blocks, fitting parameters of a polynomial model by adopting the row-column registration offset of the auxiliary grandchild image blocks of the control points and the central element position of the main grandchild image blocks, and calculating the registration offset of each pixel in each auxiliary image block of any auxiliary image in parallel; and carrying out parallel resampling on the boundary continuation image of any auxiliary image block, and outputting a precise registration SAR auxiliary image block. The invention can realize the high-precision image registration of the wide-area long-time SAR complex image sequence.)

1. A wide-area SAR complex image sequence rapid registration method adopting a twice blocking strategy is characterized by comprising the following steps:

the method comprises the following steps: selecting longitude and latitude and elevation of the central position of the region of interest on Google earth, calculating the row and column positions of the central position of the region of interest in each SAR image, cutting long-time SAR image sequences with the same size by taking the central position of the region of interest as the center, selecting an image with the highest coherence coefficient from the long-time SAR image sequences as a main image, and remaining auxiliary images;

step two: calculating the line registration offset of all the auxiliary images relative to the main image based on the maximum correlation coefficient criterion; according to the row-column registration offset of the auxiliary image relative to the main image and the central row-column position of the region of interest of the main image, cutting out the overlapped region of all the long-time SAR image sequences as a cut SAR image;

step three: carrying out image partitioning on the cut SAR image according to a set pixel size and a set boundary overlapping pixel to obtain an image block, wherein the image block is divided into a main image block and auxiliary image blocks, carrying out rough registration on each auxiliary image block in parallel based on a maximum correlation coefficient criterion to obtain the row-column registration offset of any auxiliary image block relative to the main image block, searching all auxiliary image block sets corresponding to the main image block according to the row-column registration offset of the auxiliary image blocks, and carrying out boundary continuation on all the auxiliary image block sets according to the size of an interpolation kernel function in fine registration to obtain a new auxiliary image block set;

step four: performing non-overlapping division on the auxiliary image blocks subjected to coarse registration in the step four to obtain auxiliary grandchild image blocks, and performing non-overlapping division on the main image blocks to obtain main grandchild image blocks; calculating a correlation coefficient between the image blocks of the descendants and the image blocks of the descendants; calculating the row-column registration offset between the image blocks of the descendants and the image blocks of the descendants, and recording the positions of central elements of the image blocks of the descendants; screening the central pixel of the high-coherence descendant image block as a control point, and fitting by adopting the row-column registration offset of the descendant image block of the control point and the central element position of the descendant image block to obtain the parameter of the polynomial model; according to the fitted polynomial model parameters, the registration offset of each pixel in each auxiliary image block of any auxiliary image is calculated in parallel;

step five: and according to the row-column registration offset of any auxiliary image pixel fitted in the fourth step, performing parallel resampling on the boundary continuation image of any auxiliary image block by adopting an interpolation kernel function with a set size, outputting a precise registration SAR auxiliary image block with the size same as that of the auxiliary image block, and performing geometric averaging on the auxiliary image block in the overlapped area.

2. The method for fast registering the wide-area SAR complex image sequence with the double blocking strategy according to claim 1, wherein the coherence coefficient p in the first step is_cCalculating by fast fourier transform:

ρ_c＝norm{FFT^-1[FFT(S^m)FFT(S^s*)]}

wherein norm {. cndot } represents a normalization function, S^mRepresenting the main image, S^sRepresenting the secondary image and the conjugation operator.

3. The method for fast registering the wide-area SAR complex image sequence with the double block strategy according to claim 2, characterized in that in the third step,

number of rows and columns nrow × ncol of image block:

wherein, blocksize and overlap respectively represent the pixel size of the image block and the overlapped pixel size of the image block, rowsize is the number of lines of the clipped SAR image, and closesize is the number of columns of the clipped SAR image.

4. The method for fast registering a twice-blocking-strategy wide-area SAR complex image sequence according to claim 3, wherein in the third step, the image blocking of the clipped SAR image according to the set pixel size and the set boundary overlapping pixels to obtain the image block comprises the following steps:

the line and column positions of the cut SAR image obtained from the step two are

And (3) cutting the cut SAR image again according to the size of nrow × ncol, and partitioning the image block according to the set pixel size and the set boundary overlapped pixels to obtain the image block.

5. Wide area SA of a two-partition strategy according to claim 4The R complex image sequence rapid registration method is characterized in that in the fourth step, a correlation coefficient gamma between a master image block and a slave image block is calculated_(i,j)Based on the following formula:

whileAnd respectively representing pixel values of the primary and secondary image blocks with two-dimensional indexes of (i, j) with row-column numbers of (x, y).

6. The method for fast registration of a twice-blocking-strategy wide-area SAR complex image sequence according to claim 5, wherein in the fourth step, the polynomial model is

Wherein, a₀,a₁,a₂,a₃,a₄,a₅,b₀,b₁,b₂,b₃,b₄,b₅Is a polynomial model parameter and is used as a model parameter,andfor the row-column registration offset, x, between the Sunday and Sunday image blocks_(i,j),y_(i,j)Is the central element position of the image block of the descendant.

Technical Field

The invention relates to a synthetic aperture radar data processing method, in particular to a wide area SAR complex image sequence rapid registration method adopting a twice blocking strategy, and belongs to the technical field of interferometric synthetic aperture radar image data processing.

Background

Synthetic Aperture Radar (SAR) is an active microwave imaging Radar capable of working all day long and all weather, and has a certain penetration capability to the earth surface and has the advantages of high breadth, high space-time resolution and the like. Since the 20 th century and the 50 th era, SAR has been one of international leading-edge technologies for earth observation. The long-time sequence Interferometric Synthetic Aperture Radar (InSAR) and tomosynthesis Aperture Radar (tomoSAR) technologies developed on the basis of SAR can efficiently and highly accurately acquire three-dimensional terrain and elevation transformation information covered continuously by a plurality of SAR complex images with certain view angle differences. Pixel Offset Tracking (POT) techniques predict the rate of movement of an object in a scene by exploiting the SAR amplitude sequence variation observed over time. These techniques are widely used in the fields of military reconnaissance (topographic mapping, target identification, three-dimensional battlefield simulation, weapon precision strike), resource exploration (oil and gas field monitoring, mineral resource exploitation, groundwater extraction), environmental monitoring and disaster assessment (ground subsidence monitoring, building stability, orthotopic seismic wave offset measurement, mountain landslide monitoring and identification), industrial engineering (city planning, railway site selection), global changes (volcanic corrosion and swelling, frozen soil degradation, glacier drift, polar ice layer changes, etc.), astronomy research (planet detection), and the like.

The practical performance of the technology is restricted by the registration precision of the SAR complex image sequence, so that the realization of the fast high-precision wide-area SAR complex image sequence registration is the basic premise and key link for applying the technologies. The SAR complex image sequence registration refers to that an SAR image at a certain reference moment is selected as a main image, and the SAR auxiliary image sequences of the same scene observed for a long time are registered with the main image. On one hand, the satellite-borne SAR image is large in width and many in image sequence registration images, so that the image registration is high in computational complexity. On the other hand, the wide-area wide SAR images observed at different angles have a geometric distortion effect, so that the boundary pixels are difficult to meet the high-precision resampling requirement when the wide-area SAR images are subjected to block registration.

The current image registration focuses on the proposal of a similarity criterion method, but an image sequence registration framework which can meet the requirement of high-precision registration and can be quickly realized and relieve the geometric distortion effect is lacked from the engineering realization. Although many proposed similarity Registration criteria Based on features, Correlation, mutual information amount and even the latest depth Learning and the like can effectively achieve the requirement of fast and high-precision Registration of SAR images, the proposed similarity Registration criteria are limited to a small-range scene region with insignificant geometric distortion effect, such as the Image Registration method Based on the maximum Cross-Correlation coefficient criterion proposed by Pallotta in its article, "advanced SAR Image Registration Using a modified SAR-SIFT Algorithm and delay Registration, and the proposed Image Registration method Based on the Selected SAR Image Matching criterion" proposed by Paul in its article, "advanced SAR-SIFT Algorithm and delay Matching-Matching Local Matching SAR" in the IEEE Journal of estimated SAR in Applied similarity Registration Algorithm and delay Matching, the proposed similarity Registration method Based on the features of the Selected SAR Image Matching Algorithm in the Applied SAR Image Matching Algorithm and Matching, and the proposed similarity Registration method Based on the IEEE Matching Algorithm and Matching Algorithm in its article Matching search and Matching Algorithm, the proposed by Paul in its article Matching SAR Image Registration Algorithm and Matching Algorithm, and Matching Algorithm and Matching Algorithm in its interest, 2019) the image registration method based on deep learning is proposed.

Disclosure of Invention

The invention mainly aims to provide a method for quickly registering a wide area SAR complex image sequence with a twice blocking strategy aiming at the problems of geometric distortion, difficulty in high-precision estimation of sub-pixel offset and resampling of boundary pixels in the precise image registration process, low registration calculation efficiency of a plurality of wide area images and the like in the wide area long-time SAR complex image sequence.

The above object of the present invention is achieved by the following technical solutions:

a wide-area SAR complex image sequence rapid registration method adopting a twice blocking strategy comprises the following steps:

the method comprises the following steps: selecting longitude and latitude and elevation of the central position of the region of interest on Google earth, calculating the row and column positions of the central position of the region of interest in each SAR image, cutting long-time SAR image sequences with the same size by taking the central position of the region of interest as the center, selecting an image with the highest coherence coefficient from the long-time SAR image sequences as a main image, and remaining auxiliary images;

step two: calculating the line registration offset of all the auxiliary images relative to the main image based on the maximum correlation coefficient criterion; according to the row-column registration offset of the auxiliary image relative to the main image and the central row-column position of the region of interest of the main image, cutting out the overlapped region of all the long-time SAR image sequences as a cut SAR image;

step three: carrying out image partitioning on the cut SAR image according to a set pixel size and a set boundary overlapping pixel to obtain an image block, wherein the image block is divided into a main image block and auxiliary image blocks, carrying out rough registration on each auxiliary image block in parallel based on a maximum correlation coefficient criterion to obtain the row-column registration offset of any auxiliary image block relative to the main image block, searching all auxiliary image block sets corresponding to the main image block according to the row-column registration offset of the auxiliary image blocks, and carrying out boundary continuation on all the auxiliary image block sets according to the size of an interpolation kernel function in fine registration to obtain a new auxiliary image block set;

step four: performing non-overlapping division on the auxiliary image blocks subjected to coarse registration in the step four to obtain auxiliary grandchild image blocks, and performing non-overlapping division on the main image blocks to obtain main grandchild image blocks; calculating a correlation coefficient between the image blocks of the descendants and the image blocks of the descendants; calculating the row-column registration offset between the image blocks of the descendants and the image blocks of the descendants, and recording the positions of central elements of the image blocks of the descendants; screening the central pixel of the high-coherence descendant image block as a control point, and fitting by adopting the row-column registration offset of the descendant image block of the control point and the central element position of the descendant image block to obtain the parameter of the polynomial model; according to the fitted polynomial model parameters, the registration offset of each pixel in each auxiliary image block of any auxiliary image is calculated in parallel;

step five: and according to the row-column registration offset of any auxiliary image pixel fitted in the fourth step, performing parallel resampling on the boundary continuation image of any auxiliary image block by adopting an interpolation kernel function with a set size, outputting a precise registration SAR auxiliary image block with the size same as that of the auxiliary image block, and performing geometric averaging on the auxiliary image block in the overlapped area.

The coherence coefficient ρ in the above-mentioned step one_cCalculating by fast fourier transform:

wherein norm {. cndot } represents a normalization function, S^mRepresenting the main image, S^sRepresenting the secondary image and the conjugation operator.

In the third step as described above, the first step,

number of rows and columns nrow × ncol of image block:

wherein, blocksize and overlap respectively represent the pixel size of the image block and the overlapped pixel size of the image block, rowsize is the number of lines of the clipped SAR image, and closesize is the number of columns of the clipped SAR image.

In the third step, the image partitioning of the cropped SAR image according to the set pixel size and the set boundary overlapping pixels to obtain the image block includes the following steps:

the line and column positions of the cut SAR image obtained from the step two are

The clipped SAR image is clipped again according to the size of nrow × ncol, and the clipped SAR image is obtained by blocking according to the set pixel size and the set boundary overlapped pixelsAn image block.

In step four as described above, the correlation coefficient Γ between the master and slave grandchild image blocks is calculated_(i,j)Based on the following formula:

whileAnd respectively representing pixel values of the primary and secondary image blocks with two-dimensional indexes of (i, j) with row-column numbers of (x, y).

In step four, as described above, the polynomial model is

Wherein, a₀,a₁,a₂,a₃,a₄,a₅,b₀,b₁,b₂,b₃,b₄,b₅Is a polynomial model parameter and is used as a model parameter,andfor the row-column registration offset, x, between the Sunday and Sunday image blocks_(i,j),y_(i,j)Is the central element position of the image block of the descendant.

Compared with the prior art, the invention has the following beneficial effects:

(1) and (4) universality. The invention provides a wide-area wide-width remote sensing image high-precision registration-oriented framework, and the framework can be suitable for high-precision registration of image sequences obtained by various sensors on various platforms by changing specific registration criteria and resampling in the framework.

(2) High precision. According to the invention, the first-step geographic positioning principle can control the registration offset of all SAR images within a tens order range, the second-step full-scene SAR image coarse registration strategy can control the registration offset of all SAR images within a units order range, the third-step auxiliary image block coarse registration strategy can control the registration accuracy of all SAR image blocks within a sub-pixel level, and the fourth-step and fifth-step auxiliary sun image block fine registration strategies can control the registration accuracy of all SAR image blocks within a thousandth pixel level.

(3) And (5) practicability. The invention can greatly improve the operation efficiency of the registration of a plurality of SAR images through the parallel registration of a plurality of images and the block parallel computation of the images, and the two-time blocking strategy can greatly relieve the geometric distortion effect caused by different observation angles.

Drawings

FIG. 1 is a schematic flow diagram of the present invention;

FIG. 2 is a Google Earth optical image of the Beijing cistron together with an 8.70Km 7.37Km experimental study area containing the capital International airport cut out using the principle of geocoding for geometric localization;

FIG. 3 is a power diagram of main and auxiliary image blocks after registration finally obtained by using the invention for 4 satellite-borne TerrraSAR-X/TanDEM-X time sequence SAR complex images.

Detailed Description

The present invention will be described in further detail with reference to examples for the purpose of facilitating understanding and practice of the invention by those of ordinary skill in the art, and it is to be understood that the present invention has been described in the illustrative embodiments and is not to be construed as limited thereto.

A method for quickly registering a wide-area SAR complex image sequence by a twice blocking strategy specifically comprises the following steps:

the method comprises the following steps: roughly selecting longitude and latitude and elevation (lon, lat, hei) of the position of the center of interest on Google earth, and transforming the position of the center of interest under the WGS84 coordinate system to the position under the geocentric inertial coordinate systemAccording to an SAR distance equation and a Doppler equation;

wherein, | - | represents the Euclidean distance,<·,·>representing the vector inner product. In combination with the satellite ephemeris data,is the speed of movement of the satellite relative to the ground,determining the time t at which the satellite images the center of interest for the spatial position of the satellite and the spatial position of the satelliteFurther obtaining the distance r, f between the center of the region of interest and the satellite at the imaging moment_dcIs the doppler center frequency.

By using imaging fast and slow time formula

Wherein t is_min、r_min、ρ_r、f_prfRespectively the SAR image first phase element imaging time, the distance between a satellite and a first phase element target, the SAR image range resolution and the pulse repetition frequency,representing a ceiling operator. The integer value is determined as the row-column position (row, col) of the center of the region of interest in the SAR image. Calculating the row-column position (row) of the center of the region of interest in each SAR image₀,col₀) In the row and column position (row) of each SAR image₀,col₀) And as the center, cutting the long-time SAR image sequence with the same size, selecting the image with the highest coherence coefficient from the long-time SAR image sequence as a main image, and remaining the auxiliary image.

Correlation coefficient rho_cCan be calculated by fast fourier transformation:

wherein norm {. cndot } represents a normalization function, S^mRepresenting the main image, S^sRepresenting the secondary image and the conjugation operator.

Step two: calculating the line-row registration offset (delta m, delta n) of the auxiliary image in the long-time SAR image sequence relative to the main image in the long-time SAR image sequence based on the maximum correlation coefficient criterion, and calculating the center line-row position (row) of the interested region of the main image according to the line-row registration offset (delta m, delta n) of the auxiliary image relative to the main image₀,col₀) And cutting out the overlapped region of all the long-time SAR image sequences, and recording the size of the cut SAR image as (rowsize), wherein the rowsize is the number of lines of the cut SAR image, and the closeness is the number of columns of the cut SAR image.

Step three: carrying out image partitioning on the cut SAR image according to the size of 1024 multiplied by 1024 pixels and the boundary overlapping of 128 pixels to obtain image blocks, wherein the image blocks are divided into main image blocks and auxiliary image blocks, and the method specifically comprises the following steps:

calculating the number of rows and columns of the image block nrow × ncol:

wherein block size 1024 and overlap 128 denote the pixel size of the image block and the overlapping pixel size of the image block, respectively.

The line and column positions of the cut SAR image obtained from the step two are

The clipped SAR image is clipped again in the size of nrow × ncol, and is blocked by 128 pixels in the size of 1024 × 1024 and overlapping the boundary to obtain an image block (main image block/auxiliary image block).

Performing coarse registration on each auxiliary image block in parallel by adopting a maximum correlation coefficient criterion in the second step to obtain the row-column registration offset of any auxiliary image block with the number i relative to the main image blockFinding a line-row registration offset from a primary image blockCorresponding set of all secondary image blocksAnd according to the size kernelsize of the interpolation kernel function in the fine registration, all auxiliary image blocks are collectedCarrying out boundary continuation to obtain a new auxiliary image block set

Step four: for step three 1024 x 1024 size auxiliary image block setEach sub-image block in (2) is divided into 4 × 4 auxiliary grandchild image blocks, and then the set of sub-image blocks is subjected to non-overlapping 256 × 256 divisionThe 4 × 4 secondary grandchild image blocks of each secondary image block in (b) are interpolated 256 or 512 times.

For step three 1024 x 1024 size master image block setEach main image block is divided into 4 × 4 grandchild image blocks, and then the main image block set is divided into 256 × 256 non-overlapping partitionsThe 4 × 4 descendant image blocks of each of the master image blocks in (1) are interpolated 256 or 512 times.

And establishing a two-dimensional index (i, j) for the numbers of the two-time blocking in the second step and the four-time blocking in the fourth step, wherein the two-dimensional index (i, j) represents the grandchild image block with the number j in the image block with the number i. The following coherent estimation method independent of the phase is adopted to calculate the correlation coefficient gamma between the master image block and the auxiliary image block in parallel_(i,j)I.e. by

Wherein

WhileAnd respectively representing pixel values of the primary and secondary image blocks with two-dimensional indexes of (i, j) with row-column numbers of (x, y).

Based on the maximum real correlation coefficientThen, calculate the grandchild image blockAuxiliary sun image blockRow-column registration offset betweenAndand records the central element positions (x) of the 4 x 4 descendant image blocks_(i,j),y_(i,j))。

Screening rho_(i,j)>Taking the central pixel of the 0.5 high-coherence image block of the descendant image block as a control point, adopting the row-column registration offset of the descendant image block of the control point and the central element position of the descendant image block, and fitting the registration offset of all pixels of the descendant image block through a polynomial model, namely

Wherein, a₀,a₁,a₂,a₃,a₄,a₅,b₀,b₁,b₂,b₃,b₄,b₅Is a polynomial model parameter. Establishing a registration offset model of the central pixel of a high-coherence auxiliary grandchild image in 4 x 4 auxiliary grandchild image blocks, and determining a correlation coefficient gamma between the main grandchild image block and the auxiliary grandchild image block_(i,j)As the model weight, the parameters of the polynomial model can be obtained by adopting a coherent weighted least square method.

And according to the fitted polynomial model parameters, calculating the registration offset of each pixel in each auxiliary image block of any auxiliary image in parallel.

Step five: according to the row-column registration offset of any auxiliary image pixel fitted in the step four, adopting an interpolation kernel function with the set size of kernelsize to carry out random numbering asi sub-image blockBoundary extension image ofParallel resampling is performed. Output and auxiliary image blockFine-registering the SAR secondary image blocks of the same size and geometrically averaging the secondary image blocks of the overlapping area.

Examples of the embodiments

To illustrate the effectiveness of the present invention, the following 4 TerrraSAR-X/TanDEM-XSAR image data validation experiments were performed. The SAR observation area is the capital international airport of the Beijing cistron, fig. 2 is a Google Earth optical image of the Beijing cistron, a black frame is an experimental research area which is cut out by adopting the geometric positioning of the geocoding principle of the step one of the invention and contains the capital international airport by 8.70Km multiplied by 7.37Km, and the longitude and latitude coordinates of four corner points are respectively (116.5635,40.0323), (116.5453,40.1095), (116.6484,40.0440), (116.6304, 40.1211).

Fig. 3 is a power diagram of a main and auxiliary image block after registration finally obtained by using 4 satellite-borne terrnasar-X/TanDEM-X time sequence SAR complex images, wherein the image (b) is a main image, and the images (a), (c) and (d) are auxiliary images, so that the invention can effectively realize high-precision registration of a wide-area complex scene SAR complex image sequence, and proves that the invention can effectively solve the problems of difficult high-precision registration of a geometric distortion area in the wide-area long-time SAR complex image sequence and difficult accurate resampling of image boundary pixels in the image fine registration process and the like existing in the registration of the wide-area complex image sequence. Table 1 shows the comparison of the time spent by the present invention and the unused parallel computation based on the maximum correlation coefficient criterion, from which it can be seen that the present invention significantly improves the computation efficiency of the wide-area long-time sequence spaceborne SAR complex image registration.

TABLE 1 calculation schedule of the results of the practice of the present invention

The invention	Time (hours)
		Non-parallel computing	3.7
Parallel computing	0.15

The foregoing detailed description is given for the sole purpose of illustration, and is not to be construed as limiting the scope of the invention, as those skilled in the art will recognize that various modifications, additions and substitutions can be made to the detailed description without departing from the spirit of the invention or exceeding the scope of the claims set forth below.