阅读说明：本技术 Sar卫星图像与光学图像自动匹配检索方法 (SAR satellite image and optical image automatic matching retrieval method ) 是由 葛雨辰 熊召龙 赖作镁 于 2021-08-31 设计创作，主要内容包括：本发明公开的一种SAR卫星图像与光学图像自动匹配检索方法,鲁棒性好,匹配精度高,计算速度快,本发明通过下述技术方案实现：首先对待检测的SAR图像采用滑窗的手段进行重叠等间隔采样；采用循环匹配机制分别计算所有光学模板图像向SAR图像序列的匹配结果,以及所有SAR图像序列向光学模板图像的匹配结果,合并两条循环匹配支路的SAR图像块序列结果至全图；并送入筛选模块,筛选各自分支的距离度量矩阵结果符合门限的特征点对,组成有效匹配点对集合；然后利用获取的相关矩阵,计算出SAR特征点坐标和光学特征点坐标,合并两组坐标结果,筛选距离最小的N组匹配点对；采用RANSAC算法将异常值筛除,获得自动匹配检索结果。(The invention discloses an automatic matching retrieval method of SAR satellite images and optical images, which has the advantages of good robustness, high matching precision and high calculation speed, and is realized by the following technical scheme: firstly, overlapping and equally-spaced sampling is carried out on an SAR image to be detected by adopting a sliding window method; respectively calculating matching results of all the optical template images to the SAR image sequences and matching results of all the SAR image sequences to the optical template images by adopting a circular matching mechanism, and combining the results of the SAR image block sequences of the two circular matching branches to a full graph; sending the data to a screening module, screening characteristic point pairs of which the distance measurement matrix results of the respective branches conform to a threshold to form an effective matching point pair set; then, calculating SAR characteristic point coordinates and optical characteristic point coordinates by using the obtained correlation matrix, combining two sets of coordinate results, and screening N sets of matching point pairs with the minimum distance; and (4) screening abnormal values by adopting a RANSAC algorithm to obtain an automatic matching retrieval result.)

1. An SAR satellite image and optical image automatic matching retrieval method is characterized in that: firstly, overlapping and equally-spaced sampling is carried out on an SAR image to be detected by adopting a sliding window method, and an SAR image block sequence with the size consistent with that of an optical template image is collected; respectively calculating matching results of all the optical template images to the SAR image sequences and matching results of all the SAR image sequences to the optical template images by adopting a circular matching mechanism, combining SAR image block sequence results of two circular matching branches to a whole graph according to the distance error of circular matching, and respectively obtaining the matching results of the optical template images to the SAR whole graph; and matching results of the SAR full image to the optical template image; then, the matching results obtained by the two circular matching branches are sent to a screening module, and characteristic point pairs with the distance measurement matrix results of the respective branches meeting a threshold are screened to form an effective matching point pair set; then, calculating the SAR characteristic point coordinates obtained by starting and matching the optical effective matching points and the optical characteristic point coordinates obtained by starting and matching the SAR effective matching points by using the correlation matrix obtained by the two circular matching branches, combining two groups of coordinate results, calculating the distance between the coordinate results and the effective matching point pair set, and screening N groups of matching point pairs with the minimum distance; and finally, screening abnormal values by adopting a random sampling consensus algorithm RANSAC to obtain an automatic matching retrieval result.

2. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: in the method of adopting a sliding window means to carry out overlapped equal-interval sampling on the SAR image to be retrieved, according to the size W of a set window_{Optical system}×h_{Optical system}And step length s (s < min (W)_{Optical system},h_{Optical system}) ) is intercepted to obtain an SAR image block sequence [ I ] with the length of M_sar1,I_sar2......I_sarM]Simultaneously recording the coordinate position of each image block and the optical image I_{Optical system}Matching, and finally obtaining an image block candidate sequence of [ (I)_{Optical system},I_sar1),(I_{Optical system},I_sar2),......(I_{Optical system},I_sarM)]Wherein W is_{Optical system}，h_{Optical system}Respectively an optical image I_{Optical system}The length and width dimensions of (1).

3. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: and dividing the SAR image block to be retrieved and the optical image into a circular matching branch A and a circular matching branch B, wherein the circular matching branch A adopts a circular matching mechanism to calculate the matching result of the optical template image to all SAR image sequences, and the circular matching branch B adopts a circular matching mechanism to calculate the matching result of all SAR image sequences to the optical template image.

4. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: combining matching points in the SAR image coordinate overlapping area in the two branches according to the distance error of the circular matching, combining the SAR image block sequence result to the full map, and respectively obtaining the matching result of the optical template image to the SAR full map: matching point pairs in the optical-SAR direction, a distance measurement matrix and a correlation matrix, and matching results of the SAR overall image to the optical template image are as follows: SAR-matching point pairs of optical directions, distance measurement matrix, correlation matrix.

5. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: circularly matching the optical image point to the SAR image point, circularly matching the SAR image point to the optical image point, and screening points of which the distance measurement matrixes of the branches conform to a threshold by a screening module to form an effective matching point pair set; obtaining the matching output from the effective matching point of the optical image to the SAR image according to the correlation matrix of the circular matching branch A and obtaining the matching output from the effective matching point of the SAR image to the optical image according to the correlation matrix of the circular matching branch B; merging the two branch output results, and screening out N pairs of matching point pairs with the minimum distance difference with the effective matching point pair set in the merged result; and finally, sending the N points with the closest matching distances into an abnormal value elimination module, eliminating the abnormal values in the matching point pairs by adopting a RANSAC consistency algorithm, and eliminating the abnormal values to obtain and output a final matching result.

6. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: the method comprises the steps that an optical image point circularly matches branches to an SAR image point, the SAR image point circularly matches branches to the optical image point, a screening module screens points of which the distance measurement matrixes of the branches meet a threshold to form an effective matching point pair set (X)^all _{Optical system}，X^all _SAR) (ii) a Matching the phases of branch A according to a loopObtaining effective matching point X of optical image by using correlation matrix^all _{Optical system}To SAR image X^result _SARThe correlation matrix of the matching output and the circular matching branch B obtains the effective matching point X of the SAR image^all _SARTo the optical image X^result _{Optical system}The matching output of (1); merging two branch output results (X)^all _{Optical system}，X^result _SAR) And (X)^result _{Optical system},X^all _SAR) And screening out the set (X) of the pairs of points which are effectively matched in the merged result^all _{Optical system}，X^all _SAR) N pairs of matching point pairs (X) with minimum distance difference^N _{Optical system}，X^N _SAR) (ii) a Finally, sending the N points with the closest matching distance into an abnormal value elimination module, and eliminating the matching point pair (X) by adopting a RANSAC consistency algorithm^N _{Optical system}，X^N _SAR) And (4) screening abnormal values of the abnormal data to obtain and output a final matching result.

7. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: firstly, extracting a dense feature description subset P of all pixel points of the whole optical imageⁱⁿ _{Optical system}：Pⁱⁿ _{Optical system}＝{(x,f_x),x∈I_{Optical system}And SAR image block sequence [ I }_sar1,I_sar2......I_sarM]The dense feature description subset P of all pixel points of the whole graphⁱⁿ _SAR＝[Pⁱⁿ _SAR1,Pⁱⁿ _SAR2,...,Pⁱⁿ _SAR3]，

Dense feature description subset P from optical imagesⁱⁿ _{Optical system}Starting, obtaining a feature point description subset which is correspondingly matched with all SAR image block sequence members: p^middle _SARi＝{(x',f_x'),x'∈I_sariIn which I_{Optical system}For optical images, I_sariFor the ith SAR image block, x is the pixel in the image, f_xIs a feature descriptor of pixel x, x' being from I_{Optical system}Middle imageStarting from element x, in_sariCorresponding matched pixel coordinates, f, obtained in_x'Is a descriptor of its characteristics.

8. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: loop matching branch A from P^middle _SARiStarting from the starting point, the same matching algorithm is adopted to return to obtain the corresponding optical image block sequence matching result P^out _{Optical system}＝{(x”,f_x”),x”∈I_{Optical system}P is calculated using the distance function dⁱⁿ _{Optical system}And P^out _{Optical system}All optical image pixel coordinate distance measurement matrix D in between_SARi：D_SARi＝{d(x-x”)|x',x∈I_{Optical system}，x”∈I_{Optical system}，x'∈I_sariA merging module according to [ D ]_SAR1,D_SAR2,...D_SARM]Value and original image coordinate corresponding to each pixel point, all SAR block matching results are merged into a big image according to the minimum confidence level value, and complete matching results (X) of the SAR original image are merged_{Optical system}，X_SAR) And the combined distance matrix D^compose _SARThe final matching result of matching the same original coordinate point in the overlapping area of the optical template image to the SAR image is the coordinate of the SAR characteristic point corresponding to the minimum distance difference in the set, and the final matching result is according to (X)_{Optical system}，X_SAR) Obtaining a correlation matrix Cor of an optical image to a SAR image_optical-SARWherein x' is selected from the group consisting of_sariStarting at pixel x' in I_{Optical system}Corresponding matched pixel coordinates, f, obtained in_x”Is a descriptor of its characteristics.

9. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: the circulation matching branch B extracts the dense feature description subset P of all pixel points of the whole image of the optical template image obtained in the circulation matching branch Aⁱⁿ _{Optical system}＝{(x,f_x),x∈I_{Optical system}}, and SAR image block sequence [ I_sar1,I_sar2......I_sarM]The dense feature description subset P of all pixel points of the whole graphⁱⁿ _SAR：Pⁱⁿ _SAR＝[Pⁱⁿ _SAR1,Pⁱⁿ _SAR2,...,Pⁱⁿ _SAR3]And P isⁱⁿ _SARi＝{(x,f_x),x∈I_sari}; dense feature descriptor subset P from SAR image block sequencesⁱⁿ _SAR＝[Pⁱⁿ _SAR1,Pⁱⁿ _SAR2,...,Pⁱⁿ _SAR3]Starting from the starting point, obtaining each SAR image block feature subset P by adopting a feature matching method the same as the cyclic matching branch Aⁱⁿ _SARiCorresponding set of optical image block feature points P^middle _{Optical i}＝{(x',f_x'),x'∈I_{Optical system}In which I_{Optical system}For optical images, I_sariIs the ith SAR image block, wherein x is the pixel in the image, f_xIs a feature descriptor of pixel x, x' being from I_sariStarting at pixel x, at_{Optical system}Corresponding matched pixel coordinates, f, obtained in_x'Is a descriptor of its characteristics.

10. The SAR satellite image and optical image automatic matching retrieval method of claim 1, characterized in that: loop matching branch B from P^middle _{Optical i}Starting from the starting point, the same matching algorithm is adopted to return to obtain the matching result P of the corresponding SAR image block sequence^out _SARi＝{(x”,f_x”),x”∈I_sariP is calculated using the distance function dⁱⁿ _SARiAnd P^out _SARiAll optical image pixel coordinate distance measurement matrix D in between_SARi：D_SARi＝{d(x-x”)|x',x∈I_sari，x”∈I_sari，x'∈I_{Optical system}A merging module according to [ D ]_SAR1,D_SAR2,...D_SARM]Value and original image coordinate corresponding to each pixel point, all SAR block matching results are merged into a big image according to the minimum confidence level value, and complete matching results (X) of the SAR original image are merged_SAR，X_{Optical system}) And the combined distance matrixD^compose _SARWhen the SAR image overlapping region is matched with the optical template image, the final matching result of the original coordinate point matching is the SAR characteristic point coordinate corresponding to the minimum distance difference value in the set, and finally the method is carried out according to the (X)_SAR，X_{Optical system}) Obtaining a correlation matrix Cor of the SAR image to the optical image_SAR-optics(ii) a The screening module screens points of which the circulating matching results of the branches conform to a threshold to form a matching point pair, starting from a starting source image point x, calculating a matching point x 'of a target source image, reversely matching the matching point x' to a starting source image feature point x ', and finally obtaining the distance between the x and the x'; the screening module sets a threshold Th, screens characteristic points with the backtracking distance smaller than the threshold, and respectively selects the characteristic points from (X)_{Optical system},X_SAR) And (X)_SAR,X_{Optical system}) Obtaining better matching characteristic point pairs (X)¹ _{Optical system}，X¹ _SAR) And (X)² _SAR,X² _{Optical system}) (ii) a The merging module merges the high-quality matching characteristic point pairs obtained by the two circular matching branches in the screening module to form an effective matching point pair set: will (X)² _SAR,X² _{Optical system}) The point pair sequence is exchanged to (X)² _{Optical system}，X² _SAR) And with (X)¹ _{Optical system}，X¹ _SAR) The combined results are named (X)^all _{Optical system}，X^all _SAR) (ii) a The screening module obtains a correlation matrix Cor from the optical image to the SAR image according to the circular matching branch A_optical-SARInputting X^all _{Optical system}A feature point set is obtained to obtain an output feature point set X^result _SAR：

X^result _SAR＝X^all _{Optical system}·Cor_optical-SAR(ii) a Correlation matrix Cor of SAR image obtained from cyclic matching branch B to optical image_SAR-opticsInputting X^all _SARA feature point set is obtained to obtain an output feature point set X^result _{Optical system}：X^result _{Optical system}＝X^all _SAR·Cor_SAR-optics(ii) a The screening module merges the loop matching branch output results (X)^all _{Optical system}，X^result _SAR) And (X)^result _{Optical system},X^all _SAR) And calculate the sum (X)^all _{Optical system}，X^all _SAR) N pairs of matching point pairs (X) with minimum distance difference^N _{Optical system}，X^N _SAR): and eliminating the abnormality by adopting a RANSAC consistency algorithm to obtain a final matching result.

Technical Field

The invention relates to the technical field of image processing, in particular to a method for matching an optical image with an SAR satellite image.

Background

Synthetic Aperture Radar (SAR) image and optical image registration are the prerequisites of application such as image fusion, target change detection, natural disaster assessment, and the like, and belong to the multi-modal image registration category. The optical image usually adopts central projection area imaging or push broom type scanning to obtain data, the imaging result accords with the visual characteristics of human eyes, the imaging resolution is high, the identification is easy, but the optical sensor is easily influenced by the weather and illumination conditions, fuzzy image noise is easily generated, and the working efficiency is limited to a certain extent. Synthetic Aperture Radar (SAR) technology is an active imaging system, and SAR not only enables high resolution imaging, but also has all-weather working capability and is capable of providing high resolution image data. The SAR image belongs to coherent imaging of slant range projection, is generated by a Synthetic Aperture Radar (SAR) system, is an active earth observation system, can carry out observation to the earth all the time and all the weather, has certain earth surface penetration capacity, and has unique phase information which cannot be obtained by other sensors. However, compared with an optical image, the resolution and the signal-to-noise ratio of the SAR image are low, and speckle noise and geometric distortion caused by staying, perspective and radar shadow exist in the imaging result. Therefore, in the heterogeneous matching process of the optical and SAR satellite images, the problem of how to deal with the significant geometric and radiation differences of the two data sources is faced. At present, in the image matching process, due to the limitation of the prior art conditions, the satellite SAR image and the real-time optical image are generally adopted for matching, and because the optical image often contains a significant geographic position error, the optical image cannot directly depend on the geographic coding to provide accurate matching with the satellite SAR image, and needs to be revised by depending on a heterogeneous image matching technology. However, due to the fact that the imaging mechanisms of the optical image and the SAR image are different, the optical image and the SAR image often have a large gray scale difference, and the conventional matching method developed on the single-source image cannot achieve a satisfactory result.

In the actual production process, a common use scenario is to set an optical image of a local view as a target template, and perform retrieval matching in a large-range SAR image. Commonly used large graph search matches two types of mainstream methods. One is to extract the characteristics of edges, textures and the like based on the common image information of the SAR and the optical image, filter in a correlation thermodynamic diagram peak value response mode and directly obtain the global optimal position of the small optical template image in the large SAR image. However, the correlation thermodynamic diagram matching-based method has a large limitation of application scenarios, is only suitable for image pairs with translation differences, and for optical images with missing geocoding information and poor accuracy, the method does not have a use premise because the method cannot perform prior pre-scaling and rotation matching with the SAR image. Another common method is a sparse feature point-based matching method, in which feature points and descriptors of optical and SAR images are extracted first, and then matching between the feature points is performed. Compared with a related heat map, the matching method of the feature points has the capability of solving the image pair comprising the problems of rotation and scaling, but the problem of how to deal with the remarkable geometric and radiation difference of two data sources is faced in the heterogeneous matching process of the optical image and the SAR image. Geometric distortion caused by stay, perspective, radar shadow and the like exists in the SAR image, and the optical image is influenced by illumination effects such as cloud layers, object shadow and the like, so characteristic points extracted by the two images are often greatly different, for example, some characteristic points which are obvious in the optical image are not obvious in the SAR image, and on the contrary, some characteristic points which are obvious in the optical SAR image are not obvious in the optical image, and at the moment, directly adopting sparse matching on the images can cause a plurality of mismatching pairs.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides an automatic matching and retrieving method for SAR satellite images and optical images, which has the advantages of good robustness, high matching precision and high calculation speed and has a cyclic verification mechanism, so as to solve or at least partially solve the technical problem that the matching effect of the optical images and the SAR satellite images in the scene of strong illumination change and strong radiation difference is poor in the prior art.

The invention provides an automatic matching retrieval method of an SAR satellite image and an optical image, which is characterized by comprising the following steps: firstly, overlapping and equally-spaced sampling is carried out on an SAR image to be detected by adopting a sliding window method, and an SAR image block sequence with the size consistent with that of an optical template image is collected; respectively calculating matching results of all the optical template images to the SAR image sequences and matching results of all the SAR image sequences to the optical template images by adopting a circular matching mechanism, combining SAR image block sequence results of two circular matching branches to a whole graph according to the distance error of circular matching, and respectively obtaining the matching results of the optical template images to the SAR whole graph; and matching results of the SAR full image to the optical template image; then, the matching results obtained by the two circular matching branches are sent to a screening module, and characteristic point pairs with the distance measurement matrix results of the respective branches meeting a threshold are screened to form an effective matching point pair set; then, calculating the SAR characteristic point coordinates obtained by starting and matching the optical effective matching points and the optical characteristic point coordinates obtained by starting and matching the SAR effective matching points by using the correlation matrix obtained by the two circular matching branches, combining two groups of coordinate results, calculating the distance between the coordinate results and the effective matching point pair set, and screening N groups of matching point pairs with the minimum distance; and finally, screening abnormal values by adopting a random sampling consensus algorithm RANSAC to obtain an automatic matching retrieval result.

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

aiming at the defects in the prior art, the SAR image to be retrieved is subjected to overlapping equal-interval sampling by adopting a sliding window method, so that different significance effects of global feature points at the same position in different sub-image fields can be ensured, the sub-images can obtain richer and more diverse local features, more accurate results are given in the global feature merging stage, the calculation speed is increased, and the registration precision is improved while the registration speed is ensured.

The method adopts a twice circulating matching mechanism to respectively calculate the matching result of the SAR image to the optical template image and the matching result of the optical template image to the SAR image, can adapt to the scene of strong illumination change and strong radiation difference of the SAR and the optical image to extract common characteristics, ensures the commonality and the traceability of the obtained characteristic points, and realizes the robust heterogeneous image matching.

The method eliminates abnormal values through the RANSAC algorithm, does not need excessive parameter tuning, does not need pre-registration information, and can also check whether the model is linear and whether errors are randomly distributed. Finally, the technical problem that the scene matching effect of the optical image and the SAR satellite image in the strong illumination change and the strong radiation difference is poor in the method in the prior art can be solved or at least partially solved.

Drawings

FIG. 1 is a flow chart of the optical image and SAR image matching algorithm of the present invention;

FIG. 2 is a schematic diagram of an embodiment of the loop matching mechanism of the present invention;

in order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Detailed Description

See fig. 1. According to the method, firstly, overlapping equal-interval sampling is carried out on an SAR image to be detected by adopting a sliding window method, and an SAR image block sequence with the size consistent with that of an optical template image is collected; respectively calculating matching results of all the optical template images to the SAR image sequences and matching results of all the SAR image sequences to the optical template images by adopting a circular matching mechanism, combining SAR image block sequence results of two circular matching branches to a whole graph according to the distance error of circular matching, and respectively obtaining the matching results of the optical template images to the SAR whole graph; and matching results of the SAR full image to the optical template image; then, the matching results obtained by the two circular matching branches are sent to a screening module, and characteristic point pairs with the distance measurement matrix results of the respective branches meeting a threshold are screened to form an effective matching point pair set; then, calculating the SAR characteristic point coordinates obtained by starting and matching the optical effective matching points and the optical characteristic point coordinates obtained by starting and matching the SAR effective matching points by using the correlation matrix obtained by the two circular matching branches, combining two groups of coordinate results, calculating the distance between the coordinate results and the effective matching point pair set, and screening N groups of matching point pairs with the minimum distance; and finally, screening abnormal values by adopting a random sampling consensus algorithm RANSAC to obtain an automatic matching retrieval result.

The method comprises the following specific steps:

1) is treated by means of sliding windowIn the process of carrying out overlapped equal-interval sampling on the searched SAR images, according to the size W of a set window_{Optical system}×h_{Optical system}And step length s (s < min (W)_{Optical system},h_{Optical system}) ) is intercepted to obtain an SAR image block sequence [ I ] with the length of M_sar1,I_sar2......I_sarM]Simultaneously recording the coordinate position of each image block and the optical image I_{Optical system}Matching, and finally obtaining an image block candidate sequence of [ (I)_{Optical system},I_sar1),(I_{Optical system},I_sar2),......(I_{Optical system},I_sarM)]Wherein W is_{Optical system}，h_{Optical system}Respectively an optical image I_{Optical system}The length and width dimensions of (1).

2) And dividing the SAR image block to be retrieved and the optical image into a circular matching branch A and a circular matching branch B, wherein the circular matching branch A adopts a circular matching mechanism to calculate the matching result of the optical template image to all SAR image sequences, and the circular matching branch B adopts a circular matching mechanism to calculate the matching result of all SAR image sequences to the optical template image.

3) Combining matching points in the SAR image coordinate overlapping area in the two branches according to the distance error of the circular matching, combining the SAR image block sequence result to the full map, and respectively obtaining the matching result of the optical template image to the SAR full map: matching point pairs in the optical-SAR direction, a distance measurement matrix and a correlation matrix, and matching results of the SAR overall image to the optical template image are as follows: SAR-matching point pairs of optical directions, distance measurement matrix, correlation matrix.

4) The method comprises the steps that an optical image point circularly matches branches to an SAR image point, the SAR image point circularly matches branches to the optical image point, a screening module screens points of which the distance measurement matrixes of the branches meet a threshold to form an effective matching point pair set (X)^all _{Optical system}，X^all _SAR) (ii) a Obtaining the effective matching point X of the optical image according to the correlation matrix of the circular matching branch A^all _{Optical system}To SAR image X^result _SARThe correlation matrix of the matching output and the circular matching branch B obtains the effective matching point X of the SAR image^all _SARTo the optical image X^result _{Optical system}The matching output of (1); merging two branch output results (X)^all _{Optical system}，X^result _SAR) And (X)^result _{Optical system},X^all _SAR) And screening out the set (X) of the pairs of points which are effectively matched in the merged result^all _{Optical system}，X^all _SAR) N pairs of matching point pairs (X) with minimum distance difference^N _{Optical system}，X^N _SAR)。

5) Finally, sending the N points with the closest matching distance into an abnormal value elimination module, and eliminating the matching point pair (X) by adopting a RANSAC consistency algorithm^N _{Optical system}，X^N _SAR) And (4) screening abnormal values of the abnormal data to obtain and output a final matching result.

See fig. 2. Firstly, extracting a dense feature description subset P of all pixel points of the whole optical imageⁱⁿ _{Optical system}：Pⁱⁿ _{Optical system}＝{(x,f_x),x∈I_{Optical system}And SAR image block sequence [ I }_sar1,I_sar2......I_sarM]The dense feature description subset P of all pixel points of the whole graphⁱⁿ _SAR＝[Pⁱⁿ _SAR1,Pⁱⁿ _SAR2,...,Pⁱⁿ _SAR3]Then extracting the dense feature description subset P of all pixel points of the whole optical imageⁱⁿ _{Optical system}：Pⁱⁿ _{Optical system}＝{(x,f_x),x∈I_{Optical system}And SAR image block sequence [ I }_sar1,I_sar2......I_sarM]The dense feature description subset P of all pixel points of the whole graphⁱⁿ _SAR＝[Pⁱⁿ _SAR1,Pⁱⁿ _SAR2,...,Pⁱⁿ _SAR3]Second, dense feature description subset P from the optical imageⁱⁿ _{Optical system}And starting to obtain a feature point description subset which is correspondingly matched with all SAR image block sequence members. The embodiment uses the ith image block I_sariFor example, the following steps are carried out: its corresponding matching feature point description subset P^middle _SARi＝{(x',f_x'),x'∈I_sariIn which I_{Optical system}For optical images, I_sariFor the ith SAR image block, x is the pixel in the image, f_xIs a feature descriptor of pixel x, x' being from I_{Optical system}Starting at pixel x, at_sariCorresponding matched pixel coordinates, f, obtained in_x'Is a descriptor of its characteristics.

Then, the loop matches branch A from P^middle _SARiStarting from the starting point, the same matching algorithm is adopted to return to obtain the corresponding optical image block sequence matching result P^out _{Optical system}＝{(x”,f_x”),x”∈I_{Optical system}P is calculated using the distance function dⁱⁿ _{Optical system}And P^out _{Optical system}All optical image pixel coordinate distance measurement matrix D in between_SARi：D_SARi＝{d(x-x”)|x',x∈I_{Optical system}，x”∈I_{Optical system}，x'∈I_sari}. Wherein x' is selected from the group consisting of_sariStarting at pixel x' in I_{Optical system}Corresponding matched pixel coordinates, f, obtained in_x”Is a descriptor of its characteristics. The distance function d used above includes, but is not limited to, the L1 distance, the L2 distance, the cosine distance, etc.

The adopted feature descriptor extraction method comprises but is not limited to traditional methods sift, surf, deep learning network features resnet, vgg, hardnet, lisrd and the like; the adopted feature descriptor extraction method comprises but is not limited to traditional methods sift, surf, deep learning network features resnet, vgg, hardnet, lisrd and the like; the matching method adopted in the above method includes but is not limited to traditional matching method Flann, violent matching, deep learning cnn network, superglue, transformer and the like.

By merging modules according to [ D ]_SAR1,D_SAR2,...D_SARM]Value and original image coordinate corresponding to each pixel point, all SAR block matching results are merged into a big image according to the minimum confidence level value, and complete matching results (X) of the SAR original image are merged_{Optical system}，X_SAR) And the combined distance matrix D^compose _SARAnd the final matching result of matching the same original coordinate point of the optical template image to the SAR image overlapping region is the minimum distance difference in the setThe value of the corresponding SAR characteristic point coordinate is finally determined according to (X)_{Optical system}，X_SAR) Obtaining a correlation matrix Cor of an optical image to a SAR image_optical-SAR。

The circulation matching branch B extracts the dense feature description subset P of all pixel points of the whole image of the optical template image obtained in the circulation matching branch Aⁱⁿ _{Optical system}＝{(x,f_x),x∈I_{Optical system}}, and SAR image block sequence [ I_sar1,I_sar2......I_sarM]The dense feature description subset P of all pixel points of the whole graphⁱⁿ _SAR：Pⁱⁿ _SAR＝[Pⁱⁿ _SAR1,Pⁱⁿ _SAR2,...,Pⁱⁿ _SAR3]And P isⁱⁿ _SARi＝{(x,f_x),x∈I_sari}; dense feature descriptor subset P from SAR image block sequencesⁱⁿ _SAR＝[Pⁱⁿ _SAR1,Pⁱⁿ _SAR2,...,Pⁱⁿ _SAR3]Starting from the starting point, obtaining each SAR image block feature subset P by adopting a feature matching method the same as the cyclic matching branch Aⁱⁿ _SARiThe embodiment uses the ith image block I as the corresponding optical image block feature point set_sariFor example, P corresponding thereto^middle _{Optical i}＝{(x',f_x'),x'∈I_{Optical system}In which I_{Optical system}For optical images, I_sariFor the ith SAR image block, x is the pixel in the image, f_xIs a feature descriptor of pixel x, x' being from I_sariStarting at pixel x, at_{Optical system}Corresponding matched pixel coordinates, f, obtained in_x' is a descriptor of its characteristics. Then, the loop matches branch B from P^middle _{Optical i}Starting from the starting point, the same matching algorithm is adopted to return to obtain the matching result P of the corresponding SAR image block sequence^out _SARi＝{(x”,f_x”),x”∈I_sariP is calculated using the distance function dⁱⁿ _SARiAnd P^out _SARiAll optical image pixel coordinate distance measurement matrix D in between_SARi：D_SARi＝{d(x-x”)|x',x∈I_sari，x”∈I_sari，x'∈I_{Optical system}}

The merging module is according to [ D_SAR1,D_SAR2,...D_SARM]Value and original image coordinate corresponding to each pixel point, all SAR block matching results are merged into a big image according to the minimum confidence level value, and complete matching results (X) of the SAR original image are merged_SAR，X_{Optical system}) And the combined distance matrix D^compose _SARWhen the SAR image overlapping region is matched with the optical template image, the final matching result of the original coordinate point matching is the SAR characteristic point coordinate corresponding to the minimum distance difference value in the set, and finally the method is carried out according to the (X)_SAR，X_{Optical system}) Obtaining a correlation matrix Cor of the SAR image to the optical image_SAR-optics

The screening module screens points of which the circulating matching results of the branches conform to a threshold to form a matching point pair, starting from a starting source image point x, calculating a matching point x 'of a target source image, reversely matching the matching point x' to a starting source image feature point x ', and finally obtaining the distance between the x and the x'; the screening module sets a threshold Th, screens characteristic points with the backtracking distance smaller than the threshold, and respectively selects the characteristic points from (X)_{Optical system},X_SAR) And (X)_SAR,X_{Optical system}) Obtaining better matching characteristic point pairs (X)¹ _{Optical system}，X¹ _SAR) And (X)² _SAR,X² _{Optical system})

The merging module merges the high-quality matching characteristic point pairs obtained by the two circular matching branches in the screening module to form an effective matching point pair set: will (X)² _SAR,X² _{Optical system}) The point pair sequence is exchanged to (X)² _{Optical system}，X² _SAR) And with (X)¹ _{Optical system}，X¹ _SAR) The combined results are named (X)^all _{Optical system}，X^all _SAR)。

The screening module obtains a correlation matrix Cor from the optical image to the SAR image according to a circulation matching mechanism adopting a circulation matching branch A_optical-SARInputting X^all _{Optical system}A feature point set is obtained to obtain an output feature point set X^result _SAR：X^result _SAR＝X^all _{Optical system}·Cor_optical-SAR(ii) a Correlation matrix Cor from the obtained SAR image to the optical image_SAR-opticsInputting X^all _SARA feature point set is obtained to obtain an output feature point set X^result _{Optical system}：X^result _{Optical system}＝X^all _SAR·Cor_SAR-optics。

The screening module merges the loop matching branch output results (X)^all _{Optical system}，X^result _SAR) And (X)^result _{Optical system},X^all _SAR) And calculate the sum (X)^all _{Optical system}，X^all _SAR) N pairs of matching point pairs (X) with minimum distance difference^N _{Optical system}，X^N _SAR): as the matching points of the SAR image and the optical image have statistical consistency, the RANSAC consistency algorithm is adopted to eliminate the abnormity and obtain the final matching result.

The foregoing is directed to the preferred embodiment of the present invention and it is noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.