SAR scene matching method based on cross-scale fusion enhancement and point clustering correction

Abstract

The invention provides an SAR scene matching method based on cross-scale fusion enhancement and point clustering correction, and the method comprises the steps: carrying out the multi-scale downsampling processing of an obtained to-be-matched SAR image and a reference SAR image, and correspondingly obtaining a real-time image pyramid and a reference image pyramid; correspondingly obtaining a real-time image weighted fusion enhanced image and a reference image weighted fusion enhanced image by using the real-time image pyramid and the reference image pyramid; carrying out two-dimensional cross-correlation processing on the two obtained weighted fusion enhanced images to obtain a cross-correlation matrix, and generating a rough matching position by adopting the cross-correlation matrix; sequentially carrying out angle feature point detection and clustering processing on the real-time image weighted fusion enhanced image to obtain geometric center coordinates of a plurality of target clusters; and performing fine matching processing by using the geometric center coordinates of the plurality of target clusters and the rough matching positions to obtain a final matching center, and further obtaining a final registration result. In this way, the calculation amount in the SAR scene matching process is reduced, and the matching real-time performance is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of scene matching, and in particular to a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction. Background Art

[0002] Scene matching technology plays a crucial role in navigation and positioning technology. By matching real-time scene images with pre-stored reference images (or maps), this technology accurately determines target locations and has been widely used in numerous fields, including precision guidance, drone and robot navigation. However, in practical applications, particularly in complex environments, achieving high-precision and robust real-time matching has become a key issue hindering the further development of scene matching technology. This issue is particularly prominent in the specific context of SAR (synthetic aperture radar) scene matching, and an effective solution is urgently needed.

[0003] To address the numerous challenges in scene matching, researchers have conducted extensive research and proposed a variety of matching algorithms. Among them, the traditional cross-correlation matching algorithm is a template matching method based on image grayscale information. Its core concept is to calculate a similarity measure between each subregion in the template image and the target image to identify the best matching location. This algorithm typically includes key steps such as image preprocessing, sliding window traversal and similarity calculation, and matching location determination. During the image preprocessing stage, the reference and target images are smoothed, such as using Gaussian filtering, to reduce noise interference on grayscale values. During the sliding window traversal and similarity calculation, the target image is slid pixel by pixel within the reference image. At each sliding position, the grayscale correlation between the template and the subregion is calculated, resulting in a correlation peak map between the two images. Finally, based on the calculated correlation peak map, the maximum peak is found and used as the location of the final matching point. Furthermore, point-based matching algorithms are also an important research direction in scene matching. These algorithms detect significant key points in the image, such as corners, edges, and spots, and generate unique descriptors for each keypoint, thereby achieving cross-image feature matching. Currently, mainstream point feature matching algorithms include SIFT (Scale-Invariant Feature Transform), SURF (Speeded Up Robust Features), and ORB (Oriented Fast and Rotated BRIEF). These algorithms generally include key steps such as keypoint detection, feature descriptor generation, feature matching, and post-matching processing. During the keypoint detection stage, the image is filtered to varying degrees to extract features of varying scales and orientations, identifying unique and stable points within the image that are insensitive to changes in scale, rotation, and illumination. When generating feature descriptors, the local region centered on the feature point is rotated to its dominant orientation. For example, SIFT determines the dominant orientation by calculating the gradient directional histogram, ensuring rotational invariance of the descriptor. The magnitude and direction of the gradient are then calculated within the feature point's neighborhood, which is then divided into subregions. The gradient directional histogram of each subregion is then calculated to generate a multidimensional vector. Ultimately, a discriminative vector (descriptor) is generated for each feature point, making it robust to changes in illumination, rotation, and scale. In the feature matching phase, after obtaining a set of feature points, similarity calculations are performed on these matching points to find one-to-one matching pairs in the feature point sets of the two images. In the post-matching processing phase, appropriate screening methods are selected based on the transformation model between the images to eliminate incorrect matching pairs and retain reliable matches that meet geometric consistency.

[0004] Although the above-mentioned existing technologies have achieved certain results in the field of scene matching, many problems still exist. Traditional cross-correlation algorithms require pixel-by-pixel sliding and calculation of similarity metrics, resulting in huge computational complexity and poor real-time performance. At the same time, the algorithm is more sensitive to nonlinear distortions such as rotation and scaling, and the matching accuracy will be significantly reduced under noise interference. Although point feature-based algorithms have a certain degree of robustness in complex scenes, they will be seriously interfered with by coherent speckle noise in SAR images. Coherent speckle noise can affect feature point detection and descriptor generation, leading to the occurrence of mismatches. In addition, in low-texture areas such as sea surfaces and snow, feature points are sparse, which significantly increases the matching failure rate. Moreover, this type of algorithm relies on a stable distribution of feature points and has poor adaptability to local occlusion, making it difficult to meet matching requirements in complex environments. Summary of the Invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction, comprising:

[0008] Obtaining a SAR image to be matched and a reference SAR image;

[0009] Perform multi-scale downsampling processing on the SAR image to be matched and the reference SAR image respectively, and obtain the real-time image pyramid and the reference image pyramid respectively;

[0010] Perform filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and perform weighted fusion processing based on the filtering results to obtain a weighted fusion enhancement image of the real-time image and a weighted fusion enhancement image of the reference image;

[0011] Perform two-dimensional cross-correlation processing on the weighted fusion enhanced image of the real-time image and the weighted fusion enhanced image of the reference image to obtain a cross-correlation matrix, and use the cross-correlation matrix to generate a rough matching position;

[0012] Perform corner feature point detection and clustering processing on the real-time weighted fusion enhanced image to obtain the geometric center coordinates of multiple target clusters;

[0013] Based on the geometric center coordinates and rough matching positions of multiple target clusters, the matching SAR image and the reference SAR image are precisely matched to obtain the final matching center.

[0014] The final matching center is used to perform registration processing on the SAR image to be matched and the reference SAR image to generate the final registration result.

[0015] Optionally, multi-scale downsampling processing is performed on the SAR image to be matched and the reference SAR image respectively, and a real-time image pyramid and a reference image pyramid are obtained accordingly, including:

[0016] Perform multiple wavelet transform processes on the SAR image to be matched and the reference SAR image respectively, and output the results of each wavelet transform process, thereby obtaining the wavelet transform results of the SAR image to be matched at multiple scales and the wavelet transform results of the reference SAR image at multiple scales;

[0017] The wavelet transform results of the SAR image to be matched at multiple scales are used to form a real-time image pyramid;

[0018] The wavelet transform results of the reference SAR image at multiple scales are used to construct an upper reference image pyramid.

[0019] Optionally, filtering processing is performed on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and weighted fusion processing is performed based on the filtering results to obtain a weighted fusion enhanced image of the real-time image and a weighted fusion enhanced image of the reference image, including:

[0020] Obtain a two-dimensional Gaussian gamma mixture filter at multiple directional scales;

[0021] A two-dimensional Gaussian gamma hybrid filter at multiple directional scales is used to filter each layer in the real-time image pyramid and the reference image pyramid, respectively, to obtain a first filtering result and a second filtering result; the first filtering result includes the filtering results of each layer in the real-time image pyramid at multiple directional scales; the second filtering result includes the filtering results of each layer in the reference image pyramid at multiple directional scales;

[0022] Performing edge enhancement convolution processing on the first filtering result and the second filtering result respectively, to obtain a first convolution result and a second convolution result respectively;

[0023] The first convolution result and the second convolution result are summed according to the corresponding layers, and a plurality of first summation results and a plurality of second summation results are obtained accordingly;

[0024] After adjusting the multiple first summation results to a first standard size, performing weighted fusion processing to obtain a real-time image weighted fusion enhanced image;

[0025] After adjusting the multiple second summation results to a second standard size, performing weighted fusion processing to obtain a weighted fusion enhanced image of the reference image;

[0026] The first standard size is a size corresponding to the first layer of the plurality of first summation results;

[0027] The second standard size is a size corresponding to a first layer of the plurality of second summation results.

[0028] Alternatively, the two-dimensional Gaussian gamma mixture filter at multiple scales is expressed as:

[0029]

[0030] Where H represents a two-dimensional Gaussian-gamma hybrid filter, α represents the first Gaussian function parameter, β represents the second Gaussian function parameter, σ represents the third Gaussian function parameter, the first Gaussian function parameter, the second Gaussian function parameter, and the third Gaussian function parameter are used to control the shape, decay rate, and standard deviation of the Gaussian function, respectively. Γ(·) represents the gamma function, Γ(α) represents the value of the gamma function after substituting the first Gaussian function parameter α, exp(·) represents the exponential function, and m φ represents the first direction, r φ Indicates the second direction,

[0031] m φ =mcosθ h -rsinθ h ;

[0032] r φ =msinθ h +rcosθ h ;

[0033] θ h represents the rotation angle in the hth direction, m represents the horizontal coordinate of the spatial domain, r represents the vertical coordinate of the spatial domain, h represents the hth direction, m and r are integers between [-5, 5], h = 1, 2, 3, 4.

[0034] Optionally, the real-time graph weighted fusion enhanced graph is represented as:

[0035]

[0036] The reference image weighted fusion enhancement graph is represented as:

[0037]

[0038] Among them, G a represents the real-time graph weighted fusion enhanced graph, G b Represents the reference image weighted fusion enhancement image, i represents the i-th layer of the pyramid, n represents the total number of layers of the pyramid, It represents the first summation result after the pyramid of the i-th layer is adjusted to the first standard size. It represents the second summation result after the pyramid of the i-th level is adjusted to the second standard size.

[0039] Optionally, performing two-dimensional cross-correlation processing on the weighted fusion enhanced image of the real-time image and the weighted fusion enhanced image of the reference image to obtain a cross-correlation matrix, and using the cross-correlation matrix to generate a coarse matching position, including:

[0040] Perform two-dimensional cross-correlation processing on the weighted fusion enhanced image of the real-time image and the weighted fusion enhanced image of the reference image to obtain a cross-correlation matrix;

[0041] Eliminate the edge positions in the cross-correlation matrix to obtain a cross-correlation elimination matrix;

[0042] Obtain the position coordinates corresponding to the maximum cross-correlation value in the cross-correlation elimination matrix to obtain the maximum cross-correlation position coordinates;

[0043] The coarse matching position is calculated using the maximum cross-correlation position coordinates;

[0044] Among them, the cross-correlation matrix is ​​expressed as:

[0045]

[0046] Among them, k represents the abscissa of the cross-correlation matrix, l represents the ordinate of the cross-correlation matrix, C(k,l) represents the value of the cross-correlation matrix under the (k,l) coordinate, X, Y represent the abscissa size and ordinate size of the weighted fusion enhanced image of the reference image, P, Q represent the abscissa size and ordinate size of the weighted fusion enhanced image of the real-time image, x, y represent the abscissa value and ordinate value of the weighted fusion enhanced image of the reference image, G b (x, y) represents the pixel value of the reference image weighted fusion enhanced image at the (x, y) coordinate, G a (x+k-1, y+l-1) represents the pixel value of the weighted fusion enhanced image of the real-time image at the coordinate (x+k-1, y+l-1), k and l are both integers, k∈[1,X+P-1], l∈[1,Y+Q-1];

[0047] The coarse matching position is expressed as:

[0048] x R =2*(Xk p +1)+P;

[0049] y R =2*(Yl p +1)+Q;

[0050] x R Indicates the horizontal coordinate of the rough matching position, y R Indicates the ordinate of the rough matching position, k pIndicates the horizontal coordinate of the position corresponding to the maximum cross-correlation value in the cross-correlation elimination matrix, l p Indicates the vertical coordinate of the position where the cross-correlation value in the cross-correlation elimination matrix is ​​the largest.

[0051] Optionally, corner feature point detection and clustering processing are performed on the real-time image weighted fusion enhanced image in sequence to obtain the geometric center coordinates of multiple target clusters, including:

[0052] The FAST algorithm is used to detect corner feature points in the real-time weighted fusion enhanced image;

[0053] Perform DBSCAN clustering on the corner feature points to obtain the clustering results;

[0054] Sort the clustering results according to the number of cluster points contained in the clustering results, and take the first N clustering results to form the target cluster;

[0055] The coordinate points in each target cluster are averaged to obtain the geometric center coordinates of multiple target clusters.

[0056] Optionally, based on the geometric center coordinates and rough matching positions of the multiple target clusters, a fine matching process is performed on the SAR image to be matched and the reference SAR image to obtain a final matching center, including:

[0057] The SAR image to be matched is cropped into multiple SAR sub-images to be matched using the geometric center coordinates of multiple target clusters;

[0058] Based on the coarse matching position and the multiple SAR sub-images to be matched, the reference SAR image is correspondingly cropped to obtain multiple fine matching reference areas corresponding to the multiple SAR sub-images to be matched;

[0059] Performing two-dimensional cross-correlation processing on multiple SAR sub-images to be matched and multiple precise matching reference areas to obtain a precise matching position of each SAR sub-image to be matched, and calculating a confidence value based on the precise matching position;

[0060] The fine matching position and confidence value are used to perform offset calculation, and the final matching center is obtained based on the result of the offset calculation and the coarse matching position calculation.

[0061] Optionally, the result of the offset calculation is expressed as:

[0062]

[0063] in, Represents the result of the offset calculation, Indicates the offset horizontal coordinate, Indicates the offset ordinate, x j Indicates the horizontal coordinate corresponding to the precise matching position of the jth SAR sub-image to be matched, yj Indicates the vertical coordinate corresponding to the precise matching position of the jth SAR sub-image to be matched, α j represents the confidence level corresponding to the exact matching position of the jth SAR sub-image to be matched, and N represents the total number of SAR sub-images to be matched.

[0064] In a second aspect, the present invention provides a SAR scene matching device based on cross-scale fusion enhancement and point clustering correction, the SAR scene matching device based on cross-scale fusion enhancement and point clustering correction includes: an acquisition unit, a downsampling unit, a filtering unit, a matching processing unit, a clustering unit, and a registration unit;

[0065] The acquisition unit is used to: acquire the SAR image to be matched and the reference SAR image;

[0066] The downsampling unit is used to perform multi-scale downsampling processing on the SAR image to be matched and the reference SAR image, respectively, to obtain a real-time image pyramid and a reference image pyramid;

[0067] The filtering unit is used to perform filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and perform weighted fusion processing based on the filtering results to obtain a weighted fusion enhancement image of the real-time image and a weighted fusion enhancement image of the reference image;

[0068] The matching processing unit is used to: perform two-dimensional cross-correlation processing on the real-time image weighted fusion enhanced image and the reference image weighted fusion enhanced image to obtain a cross-correlation matrix, and use the cross-correlation matrix to generate a rough matching position;

[0069] The clustering unit is used to perform corner feature point detection and clustering processing on the real-time weighted fusion enhanced image in sequence to obtain the geometric center coordinates of multiple target clusters;

[0070] The matching processing unit is further used to: perform fine matching processing on the SAR image to be matched and the reference SAR image based on the geometric center coordinates and rough matching positions of the multiple target clusters to obtain a final matching center;

[0071] The registration unit is used to perform registration processing on the to-be-matched SAR image and the reference SAR image using the final matching center to generate a final registration result.

[0072] The present invention provides a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction, comprising the following steps: acquiring a to-be-matched SAR image and a reference SAR image; performing multi-scale downsampling processing on the to-be-matched SAR image and the reference SAR image, respectively, to obtain a real-time image pyramid and a reference image pyramid; performing filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and performing weighted fusion processing based on the filtering processing results, to obtain a real-time image weighted fusion enhancement image and a reference image weighted fusion enhancement image; performing two-dimensional cross-correlation processing on the real-time image weighted fusion enhancement image and the reference image weighted fusion enhancement image to obtain a cross-correlation matrix, and using the cross-correlation matrix to generate a coarse matching position; performing corner feature point detection and clustering processing on the real-time image weighted fusion enhancement image in sequence, to obtain geometric center coordinates of multiple target clusters; performing fine matching processing on the to-be-matched SAR image and the reference SAR image based on the geometric center coordinates of the multiple target clusters and the coarse matching positions, to obtain a final matching center; and performing registration processing on the to-be-matched SAR image and the reference SAR image using the final matching center to generate a final registration result. In the present invention, firstly, the processing efficiency and noise resistance performance are significantly improved through multi-scale downsampling and weighted fusion processing. Specifically, pyramid construction greatly reduces the image size, reduces the pixel base for cross-correlation calculation, solves the high computational complexity problem caused by pixel-by-pixel sliding of traditional algorithms, improves the real-time performance of matching processing, and multi-directional filtering integration effectively suppresses coherent speckle noise and reduces matching errors; secondly, feature point clustering and hierarchical matching overcome the defects of the point feature method, clusters into target clusters after corner point detection, replaces single point features with geometric centers, reduces dependence on the stability of individual features, and clustering is based on regional statistical characteristics, which is highly robust to noise and local occlusion, thereby improving adaptability in complex environments; finally, the "coarse-fine" two-stage matching strategy comprehensively optimizes performance, that is, coarse matching is used to quickly locate the approximate area, and fine matching is used to fine-tune the calibration based on the cluster center and coarse position, ultimately achieving efficient and accurate alignment and improving noise resistance. In summary, the present invention alleviates the problems of large computational complexity, severe noise interference, and poor adaptability to local occlusion in existing SAR scene matching technologies through a hierarchical matching strategy of multi-scale processing, filter fusion, and clustering guidance, thereby improving the real-time performance of matching.

[0073] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A schematic flow chart of a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction provided by an embodiment of the present invention;

[0075] Figure 2 A schematic diagram of a weighted fusion enhancement graph of a real-time image and a weighted fusion enhancement graph of a reference image is exemplarily shown;

[0076] Figure 3 A schematic diagram of a cross-correlation matrix is ​​shown exemplarily;

[0077] Figure 4 A schematic diagram of multiple SAR sub-images to be matched obtained by cropping according to the geometric centers of multiple target clusters is exemplarily shown;

[0078] Figure 5 A schematic diagram of the final registration result is exemplarily shown;

[0079] Figure 6 A schematic structural diagram of a SAR scene matching device based on cross-scale fusion enhancement and point clustering correction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0080] The present invention proposes a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction, aiming to: reduce computational complexity and improve real-time performance through multi-scale downsampling and directional filtering; enhance robustness to nonlinear distortion by utilizing multi-directional edge features; and suppress noise interference and improve the matching success rate in low-texture areas through a slice weighted correction mechanism combined with confidence screening.

[0081] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0082] In order to alleviate the problems of large computational complexity, severe noise interference, and poor adaptability to local occlusion in existing SAR scene matching technologies, while improving the real-time performance of matching, an embodiment of the present invention provides a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction. Figure 1 A flow chart of a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction provided by an embodiment of the present invention is shown as follows: Figure 1 Shown, including:

[0083] S101: Acquire a SAR image to be matched and a reference SAR image.

[0084] It should be noted that, in this embodiment, the SAR image to be matched may be a SAR scene image acquired in real time, and the reference SAR image may be a pre-stored benchmark SAR image (or map).

[0085] S102 : Perform multi-scale downsampling processing on the SAR image to be matched and the reference SAR image respectively, and obtain a real-time image pyramid and a reference image pyramid accordingly.

[0086] Optionally, S102 may specifically include:

[0087] Perform multiple wavelet transform processes on the SAR image to be matched and the reference SAR image respectively, and output the results of each wavelet transform process, thereby obtaining the wavelet transform results of the SAR image to be matched at multiple scales and the wavelet transform results of the reference SAR image at multiple scales;

[0088] The wavelet transform results of the SAR image to be matched at multiple scales are used to form a real-time image pyramid;

[0089] The wavelet transform results of the reference SAR image at multiple scales are used to construct an upper reference image pyramid.

[0090] In the embodiment of the present invention, when the SAR image to be matched is a and the reference SAR image is b, multiple wavelet transform processes are performed on the SAR image to be matched a and the reference SAR image b, respectively, to obtain the wavelet transform results a1, a2, ... a1 of the SAR image to be matched at multiple scales. n , the wavelet transform results of the reference SAR image at multiple scales b1, b2, ... b n , where a n represents the result of the nth wavelet transform corresponding to the SAR image to be matched, b n It is the result of the nth wavelet transform corresponding to the reference SAR image. n is obtained by performing a wavelet transform on the result of the n-1th wavelet transform, that is, a2 is obtained by performing a wavelet transform on a1. Preferably, in this embodiment, n generally takes values ​​of 1, 2, or 3. Since the number of wavelet transforms is equivalent to the number of pyramid levels, in this embodiment of the present invention, the real-time image pyramid and the reference image pyramid generally have three levels.

[0091] S103: Perform filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and perform weighted fusion processing based on the filtering processing results to obtain a weighted fusion enhancement image of the real-time image and a weighted fusion enhancement image of the reference image.

[0092] Optionally, S103 may specifically include:

[0093] Obtain a two-dimensional Gaussian gamma mixture filter at multiple directional scales;

[0094] A two-dimensional Gaussian gamma hybrid filter at multiple directional scales is used to filter each layer in the real-time image pyramid and the reference image pyramid, respectively, to obtain a first filtering result and a second filtering result; the first filtering result includes the filtering results of each layer in the real-time image pyramid at multiple directional scales; the second filtering result includes the filtering results of each layer in the reference image pyramid at multiple directional scales;

[0095] Performing edge enhancement convolution processing on the first filtering result and the second filtering result respectively, to obtain a first convolution result and a second convolution result respectively;

[0096] The first convolution result and the second convolution result are summed according to the corresponding layers, and a plurality of first summation results and a plurality of second summation results are obtained accordingly;

[0097] After adjusting the multiple first summation results to a first standard size, performing weighted fusion processing to obtain a real-time image weighted fusion enhanced image;

[0098] After adjusting the multiple second summation results to a second standard size, performing weighted fusion processing to obtain a weighted fusion enhanced image of the reference image;

[0099] The first standard size is a size corresponding to the first layer of the plurality of first summation results;

[0100] The second standard size is a size corresponding to a first layer of the plurality of second summation results.

[0101] It should be noted that, in this embodiment, the plurality of first summation results may be adjusted to the first standard size using bilinear interpolation or other interpolation methods, and the second summation result may be adjusted to the second standard size using the aforementioned interpolation method.

[0102] Alternatively, the two-dimensional Gaussian gamma mixture filter at multiple scales is expressed as:

[0103]

[0104] Where H represents a two-dimensional Gaussian-gamma hybrid filter, α represents the first Gaussian function parameter, β represents the second Gaussian function parameter, σ represents the third Gaussian function parameter, the first Gaussian function parameter, the second Gaussian function parameter, and the third Gaussian function parameter are used to control the shape, decay rate, and standard deviation of the Gaussian function, respectively. Γ(·) represents the gamma function, Γ(α) represents the value of the gamma function after substituting the first Gaussian function parameter α, exp(·) represents the exponential function, and m φ represents the first direction, r φ Indicates the second direction,

[0105] m φ =mcosθ h -rsinθ h ;

[0106] r φ =msinθ h +rcosθ h ;

[0107] θ h represents the rotation angle in the hth direction, m represents the horizontal coordinate of the spatial domain, r represents the vertical coordinate of the spatial domain, h represents the hth direction, m and r are integers between [-5, 5], h = 1, 2, 3, 4.

[0108] Based on this, we can get θ h The values ​​can be 45°, 90°, 135° and 180°, and the two-dimensional Gaussian gamma hybrid filter corresponding to each angle is represented as H 45 、H 90 、H 135 and H 180 Therefore, the matrix H corresponding to the 4-layer 11×11 two-dimensional Gaussian gamma mixture filter is finally obtained.

[0109] Optionally, the real-time graph weighted fusion enhanced graph is represented as:

[0110]

[0111] The reference image weighted fusion enhancement graph is represented as:

[0112]

[0113] Among them, G a represents the real-time graph weighted fusion enhanced graph, G b Represents the reference image weighted fusion enhancement image, i represents the i-th layer of the pyramid, n represents the total number of layers of the pyramid, It represents the first summation result after the pyramid of the i-th layer is adjusted to the first standard size. It represents the second summation result after the pyramid of the i-th level is adjusted to the second standard size.

[0114] In this embodiment, taking a1 as an example, a two-dimensional Gaussian gamma hybrid filter in four directions is used to perform filtering, and the enhancement result of a1 in the corresponding direction is obtained: G 45 , G 90 , G 135 and G 180 .

[0115]

[0116] Among them, G 45 It represents the enhancement result of a1 when the rotation angle is 45°, G 90 It represents the enhancement result of a1 when the rotation angle is 90°, G 135 It represents the enhancement result of a1 when the rotation angle is 135°, G 180 It represents the enhancement result of a1 when the rotation angle is 180°, and * represents convolution processing.

[0117] In this embodiment, G45 , G 90 , G 135 and G 180 The four direction enhancement results are added together to obtain the first filtering result corresponding to a1 Then construct a centrosymmetric difference kernel h0 of length 11, h0 = [-1, 0, 0, 0, 0, 2, 0, 0, 0, 0, -1], and perform the multiplication of the two kernels along the row and column directions respectively. After the convolution operation, the sum is added to get the first summation result corresponding to a1. Specifically expressed as:

[0118]

[0119] Among them, (G ver ,G hor ) indicates The first convolution result after edge enhancement convolution processing, G ver ,G hor They represent the processing results of column-based enhanced convolution and row-based enhanced convolution, respectively, and * represents convolution processing.

[0120] Reference The process of a2,…a n ,b1,b2,…b n , and also perform enhancement processing, edge enhancement convolution and corresponding layer addition processing in four directions in turn, and finally obtain multiple first summation results and multiple second summation results

[0121] Will All adjusted to The same size, and then perform weighted fusion processing to obtain the real-time graph weighted fusion enhanced graph G a ;Will All adjusted to The same size, and then perform weighted fusion processing to obtain the real-time graph weighted fusion enhanced graph G b .

[0122] in addition, Figure 2 The following is a schematic diagram showing a weighted fusion enhancement graph of a real-time graph and a weighted fusion enhancement graph of a reference graph. Figure 2 Figure (a) shows an exemplary schematic diagram of a SAR image to be matched. Figure 2 FIG. (b) shows a schematic diagram of a reference SAR image. Figure 2 Figure (c) shows an exemplary diagram of a real-time graph weighted fusion enhanced graph. Figure 2 Figure (d) shows an exemplary schematic diagram of the reference image weighted fusion enhancement image. Figure 2 It can be seen that the targets in both the real-time image weighted fusion enhanced image and the reference image weighted fusion enhanced image are clearer than those in the original image, and the edge features are more obvious.

[0123] S104 , performing two-dimensional cross-correlation processing on the weighted fusion enhanced image of the real-time image and the weighted fusion enhanced image of the reference image to obtain a cross-correlation matrix, and using the cross-correlation matrix to generate a coarse matching position.

[0124] Optionally, S104 may specifically include:

[0125] Perform two-dimensional cross-correlation processing on the weighted fusion enhanced image of the real-time image and the weighted fusion enhanced image of the reference image to obtain a cross-correlation matrix;

[0126] Eliminate the edge positions in the cross-correlation matrix to obtain a cross-correlation elimination matrix;

[0127] Obtain the position coordinates corresponding to the maximum cross-correlation value in the cross-correlation elimination matrix to obtain the maximum cross-correlation position coordinates;

[0128] The coarse matching position is calculated using the maximum cross-correlation position coordinates;

[0129] Among them, the cross-correlation matrix is ​​expressed as:

[0130]

[0131] Among them, k represents the abscissa of the cross-correlation matrix, l represents the ordinate of the cross-correlation matrix, C(k,l) represents the value of the cross-correlation matrix under the (k,l) coordinate, X, Y represent the abscissa size and ordinate size of the weighted fusion enhanced image of the reference image, P, Q represent the abscissa size and ordinate size of the weighted fusion enhanced image of the real-time image, x, y represent the abscissa value and ordinate value of the weighted fusion enhanced image of the reference image, G b (x, y) represents the pixel value of the reference image weighted fusion enhanced image at the (x, y) coordinate, G a (x+k-1, y+l-1) represents the pixel value of the weighted fusion enhanced image of the real-time image at the coordinate (x+k-1, y+l-1), k and l are both integers, k∈[1,X+P-1], l∈[1,Y+Q-1];

[0132] The coarse matching position is expressed as:

[0133] x R =2*(Xk p +1)+P;

[0134] y R =2*(Yl p +1)+Q;

[0135] x R Indicates the horizontal coordinate of the rough matching position, y R Indicates the ordinate of the rough matching position, k p Indicates the horizontal coordinate of the position corresponding to the maximum cross-correlation value in the cross-correlation elimination matrix, l p Indicates the vertical coordinate of the position where the cross-correlation value in the cross-correlation elimination matrix is ​​the largest.

[0136] In addition, the specific operation of removing edge positions in the cross-correlation matrix to obtain the cross-correlation removal matrix may be: clipping the cross-correlation matrix to the same size as the reference SAR image.

[0137] Figure 3 The schematic diagram of the cross-correlation matrix is ​​shown as an example. Figure 3 As shown, it can be seen that the cross-correlation matrix is ​​peak-shaped, and the position of the peak corresponds to the position of the SAR image to be matched in the reference SAR image.

[0138] S105 , performing corner feature point detection and clustering processing on the real-time image weighted fusion enhanced image in sequence to obtain geometric center coordinates of multiple target clusters.

[0139] Optionally, S105 may specifically include:

[0140] The FAST algorithm is used to detect corner feature points in the real-time weighted fusion enhanced image;

[0141] Perform DBSCAN clustering on the corner feature points to obtain the clustering results;

[0142] Sort the clustering results according to the number of cluster points contained in the clustering results, and take the first N clustering results to form the target cluster;

[0143] The coordinate points in each target cluster are averaged to obtain the geometric center coordinates of multiple target clusters.

[0144] In this embodiment, N may be a positive integer greater than or equal to 2. Preferably, N may be 5.

[0145] S106 , based on the geometric center coordinates and rough matching positions of the multiple target clusters, perform fine matching processing on the SAR image to be matched and the reference SAR image to obtain a final matching center.

[0146] Optionally, S106 may specifically include:

[0147] The SAR image to be matched is cropped into multiple SAR sub-images to be matched using the geometric center coordinates of multiple target clusters;

[0148] Based on the coarse matching position and the multiple SAR sub-images to be matched, the reference SAR image is correspondingly cropped to obtain multiple fine matching reference areas corresponding to the multiple SAR sub-images to be matched;

[0149] Performing two-dimensional cross-correlation processing on multiple SAR sub-images to be matched and multiple precise matching reference areas to obtain a precise matching position of each SAR sub-image to be matched, and calculating a confidence value based on the precise matching position;

[0150] The fine matching position and confidence value are used to perform offset calculation, and the final matching center is obtained based on the result of the offset calculation and the coarse matching position calculation.

[0151] Optionally, the result of the offset calculation is expressed as:

[0152]

[0153] in, Represents the result of the offset calculation, Indicates the offset horizontal coordinate, Indicates the offset ordinate, x j Indicates the horizontal coordinate corresponding to the precise matching position of the jth SAR sub-image to be matched, y j Indicates the vertical coordinate corresponding to the precise matching position of the jth SAR sub-image to be matched, α j represents the confidence level corresponding to the exact matching position of the jth SAR sub-image to be matched, and N represents the total number of SAR sub-images to be matched. It should be noted that the total number of SAR sub-images to be matched corresponds to the number N of clustering results in the target cluster.

[0154] In this embodiment, the segmentation process of multiple fine matching reference areas includes: using the geometric center coordinates of the multiple target clusters obtained in the above steps, cropping the SAR image to be matched into N small areas, and calculating the cropping center (x cf ,y cf ), where x cf Indicates the horizontal coordinate of the cropping center of the corresponding area in the reference SAR image, y cf Indicates the ordinate of the cropping center of the corresponding area in the reference SAR image:

[0155] x cf =x R +x c -P0,

[0156] y cf =y R +y c -Q0;

[0157] Among them, P0 and Q0 are the horizontal and vertical coordinate sizes of the SAR image to be matched, respectively. c ,y c ) represents the center coordinates of multiple SAR sub-images to be matched, that is, the geometric center coordinates of the target cluster, x c ,y c They represent the horizontal coordinate of the geometric center of the target cluster and the vertical coordinate of the geometric center of the target cluster, respectively. R Indicates the horizontal coordinate of the rough matching position, y R Indicates the vertical coordinate of the rough matching position. According to the cropping center (x cf ,y cf ), the reference SAR image is cropped to form a precisely matched reference area. According to the above steps, a total of N corresponding SAR sub-images to be matched and precisely matched reference areas are obtained.

[0158] Furthermore, a two-dimensional cross-correlation process is performed on the multiple SAR sub-images to be matched and the multiple reference areas to be precisely matched (the specific two-dimensional cross-correlation process can be referred to the above embodiment, which will not be described in detail in this embodiment). The maximum peak position of the cross-correlation matrix after matching is calculated to obtain the pixel offset Then, record the value of the maximum peak, remove the values ​​within 10*10 around the maximum peak, and then find the second largest value of the cross-correlation matrix after removal, and get the ratio of the maximum peak value to the second largest value, the correlation peak ratio R j .

[0159] According to the correlation peak ratio, the confidence level α is determined j , the confidence calculation formula is as follows:

[0160]

[0161] Among them, α j Indicates the confidence level corresponding to the precise matching position of the jth SAR sub-image to be matched, R j Indicates the correlation peak ratio corresponding to the jth SAR sub-image to be matched.

[0162] Finally, the final matching center is expressed as:

[0163]

[0164] Among them, (x F ,y F ) represents the coordinate value of the final matching center, x F ,y F Represent the horizontal and vertical coordinates of the final matching center respectively.

[0165] Figure 4A schematic diagram of multiple SAR sub-images to be matched obtained by cropping according to the geometric centers of multiple target clusters is exemplarily shown. Figure 4 Figure (a) shows a schematic diagram of the structure of multiple target clusters. Figure 4 Figure (b) exemplarily shows a structural diagram of multiple SAR sub-images to be matched.

[0166] S107 , performing registration processing on the SAR image to be matched and the reference SAR image using the final matching center to generate a final registration result.

[0167] It should be noted that, in this embodiment, the final registration result may be a mosaic of matching results or a chessboard of matching results.

[0168] Figure 5 A schematic diagram showing the final registration result is shown as an example. Figure 5 FIG. (a) shows an exemplary structural diagram of a SAR reference image. Figure 5 Figure (b) shows an exemplary structural diagram of the SAR image to be matched. Figure 5 Figure (c) shows an exemplary structural diagram of the matching result mosaic. Figure 5 Figure (d) shows an exemplary structural diagram of the matching result chessboard graph.

[0169] An embodiment of the present invention provides a SAR scene matching method based on cross-scale fusion enhancement and point clustering correction, comprising: obtaining a to-be-matched SAR image and a reference SAR image; performing multi-scale downsampling processing on the to-be-matched SAR image and the reference SAR image, respectively, to obtain a real-time image pyramid and a reference image pyramid; performing filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and performing weighted fusion processing based on the filtering processing results, to obtain a weighted fusion enhancement image of the real-time image and a weighted fusion enhancement image of the reference image; performing two-dimensional cross-correlation processing on the weighted fusion enhancement image of the real-time image and the weighted fusion enhancement image of the reference image to obtain a cross-correlation matrix, and using the cross-correlation matrix to generate a coarse matching position; performing corner feature point detection and clustering processing on the weighted fusion enhancement image of the real-time image in sequence to obtain geometric center coordinates of multiple target clusters; performing fine matching processing on the to-be-matched SAR image and the reference SAR image based on the geometric center coordinates of the multiple target clusters and the coarse matching positions to obtain a final matching center; and performing registration processing on the to-be-matched SAR image and the reference SAR image using the final matching center to generate a final registration result. In the embodiments of the present invention, firstly, multi-scale downsampling and weighted fusion processing are used to significantly improve processing efficiency and noise resistance. Specifically, pyramid construction significantly reduces the image size, reduces the pixel base for cross-correlation calculation, solves the high computational complexity problem caused by pixel-by-pixel sliding in traditional algorithms, improves the real-time performance of matching processing, and multi-directional filtering integration effectively suppresses coherent speckle noise and reduces matching errors. Secondly, feature point clustering and hierarchical matching overcome the defects of the point feature method. After corner point detection, the points are clustered into target clusters, and the geometric center is used to replace the single point feature, which reduces the dependence on the stability of individual features. Clustering is based on regional statistical characteristics and is highly robust to noise and local occlusion, thereby improving adaptability in complex environments. Finally, the "coarse-fine" two-stage matching strategy comprehensively optimizes performance, that is, coarse matching is used to quickly locate the approximate area, and fine matching is used to fine-tune the calibration based on the cluster center and coarse position, ultimately achieving efficient and accurate registration and improving noise resistance. In summary, the embodiments of the present invention alleviate the problems of existing SAR scene matching technologies such as high computational complexity, severe noise interference, and poor adaptability to local occlusion through a hierarchical matching strategy that combines multi-scale processing, filter fusion, and clustering guidance, thereby improving the real-time performance of matching.

[0170] In summary, the beneficial effects of the SAR scene matching method based on cross-scale fusion enhancement and point clustering correction provided by the embodiment of the present invention are:

[0171] 1. An image pyramid is constructed through multi-scale wavelet transforms, combined with a multi-directional Gaussian-gamma hybrid filter to extract cross-scale edge features. This significantly reduces computational complexity while enhancing robustness to nonlinear distortions such as rotation and scaling. During the fine matching phase, clustering-based adaptive region selection and a multi-slice weighted correction mechanism are employed to dynamically focus on high-confidence regions, avoiding matching failures in low-texture areas (such as sea surfaces and snow).

[0172] 2. Construct edge responses in four directions (45°, 90°, 135°, and 180°) at multiple scales, and highlight the main edge structure through Gaussian filtering smoothing and weighted fusion (the larger the scale, the higher the weight).

[0173] 3. Calculate the regional confidence based on the ratio of primary to secondary correlation peaks, and assign lower weights to the offsets of low-confidence areas (such as noise interference areas) to ensure the accuracy and robustness of the final matching results.

[0174] 4. Use the DBSCAN density clustering algorithm to dynamically screen areas with dense feature points, eliminate noise points, select the top N clusters by sorting the number of points within the cluster, and define the precise matching sub-regions based on the geometric center to avoid the blindness of fixed area selection.

[0175] In order to verify the effectiveness of the SAR scene matching method based on cross-scale fusion enhancement and point clustering correction provided by the embodiment of the present invention, a simulation experiment was also conducted as follows:

[0176] Real-time SAR images (SAR images to be matched) from different regions were selected, including deserts, hills, cities, rivers, and farmland. The real-time SAR images were sized at 512*512, 256*256, and 128*128, corresponding to reference SAR images of 1024*1024, 512*512, and 256*256 sizes, for a total of 20 pairs of images. The matching time and accuracy of different algorithms were tested using MATLAB R2023b software on the same Windows platform, using MATLAB's built-in timing. The true values ​​of the matching pixel coordinates were the true values ​​marked during the SAR simulator simulation. The time consumed by each algorithm is shown in Table 1 below:

[0177] Table 1 Comparison of computational time between various algorithms and the method of the present invention

[0178]

[0179] For each algorithm, the pixel coordinates of the upper left corner of each real-time SAR image in the reference SAR image were calculated and compared with the true coordinates to compare the pixel error of each image. Finally, the average error of the 20 image pairs was calculated. The average error values ​​of each algorithm are shown in the following table:

[0180] Table 2 Comparison of matching accuracy between various algorithms and the method of the present invention

[0181]

[0182] From the comparison of Table 1 and Table 2 above, it can be seen that the matching time and matching accuracy of the method of the present invention are better than those of the traditional algorithm, and the robustness in various scenarios is stronger, which shows that the method of the present invention is suitable for SAR precise matching tasks.

[0183] The method provided in the embodiment of the present invention can be applied to electronic devices. Specifically, the electronic devices can be desktop computers, portable computers, smart mobile terminals, servers, etc., which are not limited in the embodiment of the present invention.

[0184] Based on the same inventive concept, an embodiment of the present invention further provides a SAR scene matching device based on cross-scale fusion enhancement and point clustering correction. Figure 6 The present invention provides a schematic structural diagram of a SAR scene matching device based on cross-scale fusion enhancement and point clustering correction. Figure 6 As shown, it includes: an acquisition unit 601, a downsampling unit 602, a filtering unit 603, a matching processing unit 604, a clustering unit 605 and a registration unit 606;

[0185] The acquisition unit 601 is used to: acquire the SAR image to be matched and the reference SAR image;

[0186] The downsampling unit 602 is used to perform multi-scale downsampling processing on the SAR image to be matched and the reference SAR image, respectively, to obtain a real-time image pyramid and a reference image pyramid;

[0187] The filtering unit 603 is used to perform filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and perform weighted fusion processing based on the filtering results to obtain a weighted fusion enhanced image of the real-time image and a weighted fusion enhanced image of the reference image.

[0188] The matching processing unit 604 is used to perform two-dimensional cross-correlation processing on the weighted fusion enhanced image of the real-time image and the weighted fusion enhanced image of the reference image to obtain a cross-correlation matrix, and generate a coarse matching position using the cross-correlation matrix;

[0189] The clustering unit 605 is used to perform corner feature point detection and clustering processing on the real-time image weighted fusion enhanced image in sequence to obtain the geometric center coordinates of multiple target clusters;

[0190] The matching processing unit 604 is further configured to: perform fine matching processing on the SAR image to be matched and the reference SAR image based on the geometric center coordinates and rough matching positions of the multiple target clusters to obtain a final matching center;

[0191] The registration unit 606 is used to perform registration processing on the to-be-matched SAR image and the reference SAR image using the final matching center to generate a final registration result.

[0192] It should be noted that the terms "first," "second," and the like are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention.

[0193] In the description of this specification, the reference terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" mean that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0194] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the above-mentioned disclosed embodiments by viewing the drawings and the disclosed content. In the description of the present invention, the word "comprising" does not exclude other components or steps, "one" or "an" does not exclude multiple situations, and the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0195] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention cannot be considered to be limited to these descriptions. For those skilled in the art of the present invention, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the scope of protection of the present invention.

Claims

1. A SAR scene matching method based on cross-scale fusion enhancement and point clustering correction, characterized in that: include: Obtaining a SAR image to be matched and a reference SAR image; Performing multi-scale downsampling processing on the to-be-matched SAR image and the reference SAR image respectively, to obtain a real-time image pyramid and a reference image pyramid respectively; Performing filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and performing weighted fusion processing based on the results of the filtering processing to obtain a weighted fusion enhanced image of the real-time image and a weighted fusion enhanced image of the reference image; Performing two-dimensional cross-correlation processing on the real-time image weighted fusion enhanced image and the reference image weighted fusion enhanced image to obtain a cross-correlation matrix, and using the cross-correlation matrix to generate a coarse matching position; Performing corner feature point detection and clustering processing on the real-time image weighted fusion enhanced image in sequence to obtain geometric center coordinates of multiple target clusters; Based on the geometric center coordinates of the multiple target clusters and the rough matching positions, a fine matching process is performed on the SAR image to be matched and the reference SAR image to obtain a final matching center; The SAR image to be matched and the reference SAR image are registered using the final matching center to generate a final registration result.

2. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 1 is characterized in that: The multi-scale downsampling processing is performed on the to-be-matched SAR image and the reference SAR image respectively to obtain a real-time image pyramid and a reference image pyramid, including: Performing multiple wavelet transform processes on the to-be-matched SAR image and the reference SAR image, respectively, and outputting the results of each wavelet transform process, thereby correspondingly obtaining wavelet transform results of the to-be-matched SAR image at multiple scales and wavelet transform results of the reference SAR image at multiple scales; The real-time image pyramid is formed by wavelet transform results of the SAR image to be matched at multiple scales; The wavelet transform results of the reference SAR image at multiple scales are used to form the reference image pyramid.

3. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 1, characterized in that: The filtering process is performed on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and weighted fusion processing is performed based on the results of the filtering process to obtain a real-time image weighted fusion enhanced image and a reference image weighted fusion enhanced image, including: Obtain a two-dimensional Gaussian gamma mixture filter at multiple directional scales; Using the two-dimensional Gaussian gamma hybrid filter at the multi-directional scales, filtering each layer in the real-time image pyramid and the reference image pyramid, respectively, to obtain a first filtering result and a second filtering result; the first filtering result includes the filtering results of each layer in the real-time image pyramid at the multiple directional scales; the second filtering result includes the filtering results of each layer in the reference image pyramid at the multiple directional scales; Performing edge enhancement convolution processing on the first filtering result and the second filtering result respectively, to obtain a first convolution result and a second convolution result respectively; Summing the first convolution result and the second convolution result according to corresponding layers, to obtain a plurality of first summation results and a plurality of second summation results; After adjusting the multiple first summation results to a first standard size, performing weighted fusion processing to obtain the real-time image weighted fusion enhanced image; After adjusting the plurality of second summation results to a second standard size, performing weighted fusion processing to obtain the reference image weighted fusion enhanced image; The first standard size is a size corresponding to the first layer of the plurality of first summation results; The second standard size is a size corresponding to a first layer of the plurality of second summation results.

4. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 3 is characterized in that: The two-dimensional Gaussian gamma hybrid filter under multi-directional scales is expressed as: Wherein, H represents a two-dimensional Gaussian gamma hybrid filter, α represents a first Gaussian function parameter, β represents a second Gaussian function parameter, σ represents a third Gaussian function parameter, the first Gaussian function parameter, the second Gaussian function parameter and the third Gaussian function parameter are used to control the shape, attenuation rate and standard deviation of the Gaussian function, respectively, Γ(·) represents a gamma function, Γ(α) represents the value of the gamma function after substituting the first Gaussian function parameter α, exp(·) represents an exponential function, and m φ represents the first direction, r φ Indicates the second direction, m φ =mcosθ h -rsinθ h ; r φ =msinθ h +rcosθ h ; θ h represents the rotation angle in the hth direction, m represents the horizontal coordinate of the spatial domain, r represents the vertical coordinate of the spatial domain, h represents the hth direction, m and r are integers between [-5, 5], h = 1, 2, 3, 4.

5. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 3, characterized in that: The real-time graph weighted fusion enhanced graph is represented as: The reference image weighted fusion enhancement image is expressed as: Among them, G a Denotes the real-time graph weighted fusion enhanced graph, G b Represents the reference image weighted fusion enhancement image, i represents the i-th layer of the pyramid, n represents the total number of layers of the pyramid, It represents the first summation result after the pyramid of the i-th layer is adjusted to the first standard size. It represents the second summation result after the i-th pyramid level is adjusted to the second standard size.

6. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 1, characterized in that: The step of performing two-dimensional cross-correlation processing on the weighted fusion enhanced image of the real-time image and the weighted fusion enhanced image of the reference image to obtain a cross-correlation matrix, and using the cross-correlation matrix to generate a rough matching position includes: Performing two-dimensional cross-correlation processing on the real-time image weighted fusion enhanced image and the reference image weighted fusion enhanced image to obtain the cross-correlation matrix; Eliminating edge positions in the cross-correlation matrix to obtain a cross-correlation elimination matrix; Obtaining the position coordinates corresponding to the maximum cross-correlation value in the cross-correlation elimination matrix to obtain the maximum cross-correlation position coordinates; Calculating the rough matching position using the maximum cross-correlation position coordinates; Wherein, the cross-correlation matrix is ​​expressed as: Wherein, k represents the abscissa of the cross-correlation matrix, l represents the ordinate of the cross-correlation matrix, C(k,l) represents the value of the cross-correlation matrix under the (k,l) coordinate, X, Y represent the abscissa size and ordinate size of the weighted fusion enhancement image of the reference image, P, Q represent the abscissa size and ordinate size of the weighted fusion enhancement image of the real-time image, x, y represent the abscissa value and ordinate value of the weighted fusion enhancement image of the reference image, G b (x, y) represents the pixel value of the reference image weighted fusion enhanced image at the (x, y) coordinate, G a (x+k-1, y+l-1) represents the pixel value of the weighted fusion enhanced image of the real-time image at the coordinate (x+k-1, y+l-1), k and l are both integers, k∈[1,X+P-1], l∈[1,Y+Q-1]; The coarse matching position is expressed as: x R =2*(X-k p +1)+P; and R =2*(Yl p +1)+Q; x R Indicates the horizontal coordinate of the rough matching position, y R Indicates the ordinate of the rough matching position, k p Indicates the horizontal coordinate of the position corresponding to the maximum cross-correlation value in the cross-correlation elimination matrix, l p Indicates the vertical coordinate of the position where the cross-correlation value in the cross-correlation elimination matrix is ​​the largest.

7. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 1, characterized in that: The step of sequentially performing corner feature point detection and clustering processing on the real-time image weighted fusion enhanced image to obtain geometric center coordinates of multiple target clusters includes: Using the FAST algorithm to detect corner feature points in the real-time image weighted fusion enhanced image; Performing DBSCAN clustering processing on the corner feature points to obtain a clustering result; Sort the clustering results according to the number of points in the clusters contained in the clustering results, and take the first N clustering results to form the target cluster; The coordinate points in each target cluster are averaged to obtain the geometric center coordinates of the multiple target clusters.

8. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 1, characterized in that: The step of performing fine matching processing on the to-be-matched SAR image and the reference SAR image based on the geometric center coordinates of the multiple target clusters and the rough matching positions to obtain a final matching center includes: Using the geometric center coordinates of the multiple target clusters, the to-be-matched SAR image is cropped into a plurality of to-be-matched SAR sub-images; Based on the coarse matching position and the multiple SAR sub-images to be matched, the reference SAR image is subjected to corresponding cropping processing to obtain multiple fine matching reference areas corresponding to the multiple SAR sub-images to be matched; performing two-dimensional cross-correlation processing on the plurality of SAR sub-images to be matched and the plurality of reference areas for precise matching to obtain a precise matching position of each SAR sub-image to be matched, and calculating a confidence value based on the precise matching position; An offset calculation is performed using the fine matching position and the confidence value, and the final matching center is obtained based on a result of the offset calculation and the coarse matching position.

9. The SAR scene matching method based on cross-scale fusion enhancement and point clustering correction according to claim 8, characterized in that: The result of the offset calculation is expressed as: in, Represents the result of the offset calculation, Indicates the offset horizontal coordinate, Indicates the offset ordinate, x j Indicates the horizontal coordinate corresponding to the precise matching position of the jth SAR sub-image to be matched, y j Indicates the vertical coordinate corresponding to the precise matching position of the jth SAR sub-image to be matched, α j represents the confidence level corresponding to the exact matching position of the jth SAR sub-image to be matched, and N represents the total number of SAR sub-images to be matched.

10. A SAR scene matching device based on cross-scale fusion enhancement and point clustering correction, characterized in that: The SAR scene matching device based on cross-scale fusion enhancement and point clustering correction includes: an acquisition unit, a downsampling unit, a filtering unit, a matching processing unit, a clustering unit, and a registration unit; The acquisition unit is used to: acquire the SAR image to be matched and the reference SAR image; The downsampling unit is used to perform multi-scale downsampling processing on the SAR image to be matched and the reference SAR image, respectively, to obtain a real-time image pyramid and a reference image pyramid; The filtering unit is used to perform filtering processing on each layer of the real-time image pyramid and the reference image pyramid at multiple directional scales, and perform weighted fusion processing based on the results of the filtering processing to obtain a weighted fusion enhancement image of the real-time image and a weighted fusion enhancement image of the reference image; The matching processing unit is used to: perform two-dimensional cross-correlation processing on the real-time image weighted fusion enhanced image and the reference image weighted fusion enhanced image to obtain a cross-correlation matrix, and use the cross-correlation matrix to generate a rough matching position; The clustering unit is used to: perform corner feature point detection and clustering processing on the real-time image weighted fusion enhanced image in sequence to obtain geometric center coordinates of multiple target clusters; The matching processing unit is further configured to: perform fine matching processing on the to-be-matched SAR image and the reference SAR image based on the geometric center coordinates of the multiple target clusters and the rough matching position to obtain a final matching center; The registration unit is used to perform registration processing on the to-be-matched SAR image and the reference SAR image using the final matching center to generate a final registration result.

Images