Multi-modal satellite remote sensing image matching method and system based on phase information

Abstract

The application discloses a kind of multi-modal satellite remote sensing image matching method and system based on phase information, first the two multi-modal images to be matched are respectively subjected to median filtering processing, to weaken the influence of noise as far as possible, then respectively on two images using phase consistency minimum moment and non-maximum suppression extraction image feature points, then using the phase consistency amplitude and angle of extension respectively constructs feature descriptor, and the Euclidean distance between feature descriptors is measured to screen matching points, finally using RANSAC algorithm to the matching point obtained is eliminated false match, obtains the final correct matching homonymic point.The application is matched based on phase information, in the matching process, using the phase consistency information of image to extract feature points and construct feature descriptor, can effectively improve the resistance of matching method to the nonlinear radiation difference between multi-modal images.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing image processing, and in particular relates to a multimodal satellite remote sensing image matching method and system based on phase information. Background Art

[0002] As satellite-borne sensors have gradually expanded from single optical sensors to multiple sensors, including microwave, laser, and infrared, my country's ability to acquire multimodal satellite remote sensing data has greatly improved. Multimodal remote sensing images are typically acquired by different sensors, each providing data at different wavelengths, resolutions, and information about different ground features. By fusing multimodal remote sensing images, the limitations of single-sensor images, which often lack sufficient ground feature information, can be overcome, improving the ability to acquire ground information. Image fusion is premised on matching points with the same name across multimodal images.

[0003] However, there are significant nonlinear radiometric differences between multimodal remote sensing images, meaning that their variations in grayscale values ​​and contrast are nonlinear. Because traditional gradient-based matching methods perform poorly when dealing with nonlinear radiometric differences, it is difficult to obtain stable and reliable homonymous points between multimodal images. Therefore, it is necessary to develop a new method for homonymous point matching in multimodal remote sensing images. This method needs to be able to effectively handle nonlinear radiometric differences, improve the accuracy and stability of homonymous point matching, and thus provide a reliable foundation for subsequent image fusion processing. Such a method will help improve the comprehensive utilization efficiency of multimodal remote sensing images and provide more accurate and comprehensive information for fields such as Earth observation and resource management. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a multimodal satellite remote sensing image matching method and system based on phase information. During the matching process, the phase consistency information of the image is used to extract feature points and construct feature descriptors, which can effectively improve the matching method's resistance to nonlinear radiation differences between multimodal images.

[0005] In order to achieve the above object, the present invention provides a multimodal satellite remote sensing image matching method based on phase information, comprising the following steps:

[0006] Step 1: Perform median filtering on the two images to be matched respectively;

[0007] Step 2: Extract feature points from the two images using phase-consistent minimum moment and non-maximum suppression.

[0008] Step 3, construct feature point descriptors using the expanded phase consistency amplitude and angle;

[0009] Step 4: Calculate the Euclidean distance between the descriptors of the two images in sequence, and select the points with the same name according to the set ratio threshold and matching threshold;

[0010] Step 5: Use the RANSAC algorithm to remove incorrectly matched points from the matched points with the same name.

[0011] Moreover, in step 1, one image is used as the original image and the other image is used as the reference image. The grayscale value of each pixel in the original image and the reference image is set to the median grayscale value of all pixels in the k1×k1 neighborhood window of the pixel to minimize the influence of noise.

[0012] Furthermore, in step 2, the phase consistency minimum moment maps of the original image and the reference image after median filtering are respectively calculated. The calculation formula of the phase consistency minimum moment is:

[0013]

[0014] Where, Indicates the minimum phase consistency moment of any pixel point on the current scale image; It represents the sum of the squares of the x-direction components of phase consistency in different directions; It represents the sum of the products of the x-direction component and the y-direction component of the phase consistency in different directions; It represents the sum of the squares of the y-direction components of phase consistency in different directions; The specific calculation formula is:

[0015]

[0016]

[0017]

[0018] Where θ represents a discrete direction set set artificially, and PC2(θ) represents the phase consistency map of the current scale image at direction θ. The calculation formula is as follows:

[0019]

[0020] Where W(·) is the weight factor, which is used to assign larger weights to multiple frequencies with greater consistency; A n is the amplitude of the nth scale; ξ is the minimum value used to prevent division by 0; For noise compensation; The symbol indicates that when the value is less than 0, it is set to zero, and when it is greater than 0, it is not processed; ΔΦ n (θ) represents the phase deviation function with scale n and direction θ, which is calculated as follows:

[0021]

[0022] Where, Φ n (θ) represents the phase component with scale n and direction θ; It represents the average value of the phase components in different directions with a scale of n.

[0023] The phase consistency minimum moment is calculated for each pixel corresponding to the original image and the reference image to obtain the phase consistency minimum moment maps of the original image and the reference image. Non-maximum suppression is performed on the two phase consistency minimum moment maps respectively. The maximum value position obtained after non-maximum suppression is the feature point of the two images.

[0024] Moreover, in step 3, according to the feature point positions calculated in step 2, a phase consistency window of a certain size on the corresponding image is extracted respectively, and the phase consistency information in the window is used to construct a descriptor. By modifying the scale and angle in the image phase consistency calculation formula (5), the phase consistency maps of images of different scales in different directions can be calculated, and the phase consistency maps of different scales in each direction are accumulated to obtain phase consistency maps in different directions. Then, the phase consistency maps in different directions are mapped to the x and y directions respectively and accumulated to obtain phase consistency amplitude maps in the x and y directions. The phase consistency amplitude and angle of each pixel can be calculated based on the phase consistency amplitude maps in the x and y directions. The calculation formulas for the phase consistency amplitude m and phase consistency angle θ at pixel (i, j) are as follows:

[0025]

[0026]

[0027] Where m x 、m y Represent the phase consistency amplitude of pixel (i, j) in the x and y directions respectively.

[0028] The descriptor corresponding to the feature point is constructed based on the calculated phase consistency amplitude map and angle map of the window. Specifically, the amplitude map and angle map windows are divided into k2×k2 units respectively, and the phase consistency angle direction in each unit is assigned to N directions. The phase consistency amplitude calculated to each direction is Gaussian-weighted and accumulated in the corresponding direction according to the distance to form an N-direction histogram. The x-direction of the histogram is 1 to N directions, and the y-direction is the corresponding accumulated value. Finally, the histogram in the k2×k2 units is represented by the corresponding y-value, that is, each histogram can be converted into a vector of length N, and the k2×k2 units can generate a descriptor of k2×k2×N dimensions.

[0029] Moreover, in step 4, the Euclidean distance between descriptors is calculated using the L2 norm to determine the similarity of the descriptors. The Euclidean distance calculation formula between feature descriptors is as follows:

[0030]

[0031] Where D(P,Q) represents the Euclidean distance between points P and Q, (x1,y1) are the coordinates of point P on the original image, and (x2,y2) are the coordinates of point Q on the reference image.

[0032] Calculate the Euclidean distance between each feature point on the original image and all feature points on the reference image, and sort the calculated Euclidean distances to obtain the minimum distance and the second minimum distance. If the ratio of the minimum distance to the second minimum distance is less than or equal to the set ratio threshold δ, it means that the distance of the matching point pair is relatively stable and it is considered to be a correctly matched point of the same name. Otherwise, it means that the distance of the matching point pair is unreliable and it is removed. Then, the matching point pairs are further screened according to the matching threshold γ. If the Euclidean distance between the matching point pairs is less than γ, it is considered to be a valid match and it is retained. Otherwise, the point is considered to be an incorrectly matched point and is removed.

[0033] Moreover, in step 5, after obtaining the matching point sets X and Y according to steps 1-4, a random sampling consistency strategy based on an affine model is used to select the optimal point set as the correctly matched point set. The specific operations are as follows:

[0034] Randomly select three or more pairs of points with the same name in the point sets X and Y, and match the point pairs (x i ,y i ) and (u i ,v i ) is expressed as homogeneous coordinates and expressed in matrix form as follows:

[0035]

[0036] Where X and Y represent the matching point sets in the source image and target image, respectively.

[0037] According to the formula H=Y T X(X T X) -1 Calculate the least squares solution to obtain the transformation matrix H. Use the calculated affine transformation matrix H to map all matching points X to obtain the predicted matching point position X′. Calculate the Euclidean distance L between the predicted matching point position X′ and the actual matching point X. If L is less than the set threshold η, it is considered an inlier; otherwise, it is considered an outlier. Count the total number of inliers in the point set X. Repeat the above steps and select the data set with the largest number of inliers as the optimal data set. The point pairs contained in this data set are the correctly matched points of the same name in the present invention.

[0038] The present invention also provides a multimodal satellite remote sensing image matching system based on phase information, which is used to implement the multimodal satellite remote sensing image matching method based on phase information as described above.

[0039] Furthermore, the invention comprises a processor and a memory, wherein the memory is used to store program instructions, and the processor is used to call the program instructions in the memory to execute the multi-modal satellite remote sensing image matching method based on phase information as described above.

[0040] Alternatively, it includes a readable storage medium having a computer program stored thereon, and when the computer program is executed, it implements the multimodal satellite remote sensing image matching method based on phase information as described above.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1) Matching is based on phase information. During the matching process, both feature point extraction and feature descriptor construction utilize the phase consistency information of the image, which can effectively improve the matching method's resistance to nonlinear radiation differences between multimodal images;

[0043] 2) Using phase consistency minimum moment and non-maximum suppression to extract image feature points, more feature points with more obvious features and higher repeatability between images can be extracted;

[0044] 3) Using phase consistency strength and angle to construct descriptors can improve the descriptor's ability to describe image structure information and increase the correct matching rate of homonymous points. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Flowchart of a method according to an embodiment of the present invention.

[0046] Figure 2 This is a connection diagram showing the results of matching 6 image pairs using the SIFT algorithm.

[0047] Figure 3 This is a connection diagram showing the results of matching 6 image pairs using the SAR-SIFT algorithm.

[0048] Figure 4 This is a connection diagram of the matching results of 6 groups of image pairs using the method proposed in this invention.

[0049] Figure 5 This is the checkerboard registration diagram of the results of matching 6 groups of image pairs using the method proposed in this invention. DETAILED DESCRIPTION

[0050] The application provides a multi-modal satellite remote sensing image matching method and system based on phase information.

[0051] Embodiment one

[0052] As shown in the figure, the application provides a multi-modal satellite remote sensing image matching method based on phase information, which includes the following steps: Figure 1

[0053] Step 1, median filtering processing is performed on two images to be matched respectively.

[0054] One image is taken as an original image, and the other image is taken as a reference image. Median filtering processing is performed on the original image and the reference image respectively to weaken the influence of noise as much as possible. In this embodiment, the gray value of each pixel point is set as the median value of the gray values of all pixel points in the 3*3 neighborhood window of the pixel.

[0055] Step 2, feature points are extracted on two images based on phase consistency minimum moments.

[0056] The phase consistency minimum moment diagram of the original image and the reference image after median filtering processing is calculated respectively. The calculation formula of the phase consistency minimum moment is as follows:

[0057]

[0058] In the formula, PCmin represents the phase consistency minimum moment of any pixel point on the current scale image; represents the sum of squares of x direction components of different direction phase consistencies; represents the sum of products of x direction components and y direction components of different direction phase consistencies; represents the sum of squares of y direction components of different direction phase consistencies; Specific calculation formula is as follows:

[0059]

[0060]

[0061]

[0062] In the formula, θ represents a set of discrete directions set artificially, and in this embodiment, θ is set as 30°, 60°, 90°, 120°, 150° and 180°; PC2(θ) represents the phase consistency diagram of the current scale image in direction θ, and the calculation formula is as follows:

[0063] ​​

[0064] Where W(·) is the weight factor, which is used to assign larger weights to multiple frequencies with greater consistency; A n is the amplitude of the nth scale; ξ is the minimum value used to prevent division by 0; For noise compensation; The symbol indicates that when the value is less than 0, it is set to zero, and when it is greater than 0, it is not processed; ΔΦ n (θ) represents the phase deviation function with scale n and direction θ, which is calculated as follows:

[0065]

[0066] Where, Φ n (θ) represents the phase component with scale n and direction θ; It represents the average value of the phase components in different directions with a scale of n.

[0067] The phase consistency minimum moment is calculated for each pixel corresponding to the original image and the reference image to obtain the phase consistency minimum moment maps of the original image and the reference image. Non-maximum suppression is performed on the two phase consistency minimum moment maps respectively. The maximum value position obtained after non-maximum suppression is the feature point of the two images.

[0068] Step 3: construct descriptors for the feature points extracted in step 2.

[0069] According to the feature point positions calculated in step 2, phase consistency windows of a certain size are extracted from the corresponding images, and the phase consistency information in the window is used to construct a descriptor. By modifying the scale and angle in the image phase consistency calculation formula (Formula (5)), the phase consistency maps of images of different scales in different directions can be calculated. The phase consistency maps of different scales in each direction are accumulated to obtain phase consistency maps in different directions. The phase consistency maps in different directions are then mapped to the x and y directions and accumulated to obtain phase consistency amplitude maps in the x and y directions. The phase consistency amplitude and angle of each pixel can be calculated based on the phase consistency amplitude maps in the x and y directions. The calculation formulas for the phase consistency amplitude m and phase consistency angle θ at pixel (i, j) are as follows:

[0070]

[0071]

[0072] Where m x 、m y Represent the phase consistency amplitude of pixel (i, j) in the x and y directions respectively.

[0073] The descriptor corresponding to the feature point is constructed based on the calculated phase consistency amplitude map and angle map of the window. In this embodiment, the amplitude map and angle map windows are divided into 6×6 units respectively, and the phase consistency angle direction in each unit is assigned to 8 directions (i.e., one direction every 22.5 degrees), and the phase consistency amplitude calculated to each direction is Gaussian-weighted and accumulated in the corresponding direction according to the distance to form an 8-direction histogram. The x-direction of the histogram is 1 to 8 directions, and the y-direction is the corresponding accumulated value. Finally, the histogram in the 6×6 units is represented by the corresponding y value, that is, each histogram can be converted into a vector of length 8, and the 6×6 units can generate a descriptor of 6×6×8 dimensions.

[0074] Step 4: Calculate the Euclidean distance between the descriptors of the two images in sequence, and select the points with the same name based on the ratio threshold and matching threshold.

[0075] After obtaining the feature descriptors corresponding to the feature points, the image feature point matching will be converted to feature point descriptor matching, that is, the similarity of the descriptors corresponding to a pair of feature points in the original image and the reference image will be used to determine whether they are points of the same name. The present invention uses the L2 norm to calculate the Euclidean distance between descriptors to determine the similarity of descriptors. The Euclidean distance calculation formula between feature descriptors is as follows:

[0076]

[0077] Where D(P,Q) represents the Euclidean distance between points P and Q, (x1,y1) are the coordinates of point P on the original image, and (x2,y2) are the coordinates of point Q on the reference image.

[0078] The shorter the Euclidean distance, the higher the descriptor similarity is considered, and the greater the possibility that the feature point pair is a homonymous point. Since the Euclidean distance is calculated for each feature point on the original image and all feature points on the reference image, the calculated Euclidean distances are sorted to obtain the minimum distance and the second minimum distance. A ratio threshold δ is set to preliminarily screen the matching point pairs. In this embodiment, the ratio threshold δ is set to 1. If the ratio of the minimum distance to the second minimum distance is less than or equal to δ, it means that the distance of this matching point pair is relatively stable and is considered to be a correctly matched homonymous point. On the contrary, if the distance ratio between the two is greater than δ, it means that the distance of this matching point pair is unreliable, which may be noise or an erroneous match and needs to be eliminated. The matching point pairs are further screened according to the matching threshold γ. In this embodiment, the matching threshold γ is set to 100. For each matching point pair preliminarily screened, if the Euclidean distance between them is less than γ, it is considered a valid match and is retained. Otherwise, the point is considered to be an erroneous match and needs to be eliminated.

[0079] Step 5: Use the RANSAC algorithm to remove incorrectly matched points from the matched points with the same name.

[0080] After obtaining the matching point sets X and Y according to steps 1-4, a random sampling consistency strategy based on an affine model is used to select the optimal point set as the correctly matched point set. The specific method is as follows:

[0081] Randomly select three or more pairs of points with the same name in the point sets X and Y, and solve the affine transformation matrix H based on the least squares method. The solution of the transformation matrix is ​​as follows: i ,y i ) and (u i ,v i ) is expressed as homogeneous coordinates and expressed in matrix form as follows:

[0082]

[0083] Where X and Y represent the matching point sets in the source image and target image, respectively.

[0084] According to the formula H=Y T X(X T X) -1 Calculate the least squares solution to obtain the transformation matrix H. Use the calculated affine transformation matrix H to map all matching points X to obtain the predicted matching point position X'. Calculate the Euclidean distance L between the predicted matching point position X' and the actual matching point X. If L is less than the set threshold η (η is 3 in this embodiment), it is considered an inlier, otherwise it is considered an outlier, and the total number of inliers contained in the point set X is counted. Repeat the above operation and take the set of data with the largest number of inliers as the best data set. The point pairs contained in this data set are the points of the same name correctly matched by the present invention. In order to improve the robustness of the algorithm, this embodiment sets the number of loops to 10,000 times.

[0085] Six image pairs were selected and matched using the SIFT algorithm, SAR-SIFT algorithm and the algorithm of the present invention to verify the reliability of the matching algorithm proposed in the present invention. The experimental results are displayed by connecting lines with the same-name points. Figure 2 、 Figure 3 、 Figure 4It can be seen that the SIFT algorithm did not match the correct homonymous points in any of the six image pairs. SAR-SIFT matched the correct homonymous points in the first three image pairs, but the number of correctly matched points was small. It was also unable to match the correct homonymous points in the last three optical and SAR remote sensing image pairs. Compared with the SIFT and SAR-SIFT algorithms, the present invention matched a large number of stable and reliable homonymous points in all six image pairs. This is mainly because the grayscale information between optical and SAR images has significant nonlinear differences, which greatly reduces the robustness of the SIFT algorithm and SAR-SIFT algorithm based on gradient information to construct descriptors. The present invention uses the phase consistency information of the image to construct descriptors, which can effectively resist the nonlinear changes in grayscale information between images.

[0086] In order to verify the matching effect of the present invention, the 6 sets of image pairs that are matched with the same-name point pairs are used for chessboard registration, and the quality of the matching algorithm is indirectly evaluated by observing the edge connection of the chessboard grid points. Figure 5 As shown, from Figure 5 It can be seen that there is no particularly obvious misalignment at the edges of the chessboards in the matching results of the six sets of images. When the six sets of images are magnified and observed, no particularly large deviation is found at the edges of the six sets of images, which shows that the matching of the same-name points by the present invention is very reliable.

[0087] Example 2

[0088] Based on the same inventive concept, the present invention also provides a multimodal satellite remote sensing image matching system based on phase information, including a processor and a memory, the memory is used to store program instructions, and the processor is used to call the program instructions in the memory to execute the above-mentioned multimodal satellite remote sensing image matching method based on phase information.

[0089] Example 3

[0090] Based on the same inventive concept, the present invention also provides a multimodal satellite remote sensing image matching system based on phase information, including a readable storage medium, on which a computer program is stored. When the computer program is executed, a multimodal satellite remote sensing image matching method based on phase information as described above is implemented.

[0091] In specific implementation, the method proposed in the technical solution of the present invention can be automatically run by those skilled in the art using computer software technology. System devices that implement the method, such as computer-readable storage media that store the corresponding computer program of the technical solution of the present invention and computer equipment that runs the corresponding computer program, should also be within the scope of protection of the present invention.

[0092] The specific embodiments described herein are merely illustrative of the spirit of the application. Various modifications or changes in the specific embodiments described herein can occur to those skilled in the art to which the application pertains without departing from the spirit of the application, and it is understood that such modifications or changes are to be considered as within the scope of the application as defined by the appended claims.

Claims

1. A multimodal satellite remote sensing image matching method based on phase information, characterized in that: The following steps are involved: Step 1: Perform median filtering on the two images to be matched respectively; Step 2: Extract feature points from the two images using phase-consistent minimum moment and non-maximum suppression. Step 3, construct feature point descriptors using the expanded phase consistency amplitude and angle; According to the position of the feature points calculated in step 2, a phase consistency window of a certain size is extracted from the corresponding image, and the phase consistency information in the window is used to construct a descriptor; by modifying the scale and angle in the image phase consistency calculation formula, the phase consistency maps of images of different scales in different directions can be calculated, and the phase consistency maps of different scales in each direction are accumulated to obtain phase consistency maps in different directions. Then, the phase consistency maps in different directions are mapped to the x and y directions respectively and accumulated to obtain the phase consistency amplitude maps in the x and y directions. The phase consistency amplitude and angle of each pixel can be calculated based on the phase consistency amplitude maps in the x and y directions; the calculation formulas for the phase consistency amplitude m and phase consistency angle θ at pixel (i, j) are as follows: Where m x 、m y Represents the phase consistency amplitude of pixel (i, j) in the x and y directions respectively; Step 4: Calculate the Euclidean distance between the descriptors of the two images in sequence, and select the points with the same name according to the set ratio threshold and matching threshold; The L2 norm is used to calculate the Euclidean distance between descriptors to determine the similarity of descriptors. The Euclidean distance calculation formula between feature descriptors is as follows: Where D(P,Q) represents the Euclidean distance between points P and Q, (x1,y1) is the coordinate of point P on the original image, and (x2,y2) is the coordinate of point Q on the reference image. Calculate the Euclidean distance between each feature point on the original image and all feature points on the reference image, and sort the calculated Euclidean distances to obtain the minimum distance and the second minimum distance. If the ratio of the minimum distance to the second minimum distance is less than or equal to the set ratio threshold δ, it means that the distance of the matching point pair is relatively stable and it is considered to be a correctly matched homonymous point. Otherwise, it means that the distance of the matching point pair is unreliable and it is removed. Then, the matching point pairs are further screened according to the matching threshold γ. If the Euclidean distance between the matching point pairs is less than γ, it is considered to be a valid match and retained. Otherwise, the point is considered to be a wrong matching point and removed. Step 5: Use the RANSAC algorithm to remove incorrectly matched points from the matched points with the same name.

2. The multimodal satellite remote sensing image matching method based on phase information according to claim 1, characterized in that: In step 1, one image is used as the original image and the other image is used as the reference image. Median filtering is performed on the original image and the reference image respectively, that is, the grayscale value of each pixel is set to the median of the grayscale values ​​of all pixels in the k1×k1 neighborhood window of the pixel.

3. The multimodal satellite remote sensing image matching method based on phase information according to claim 1, characterized in that: In step 2, the phase consistency minimum moment maps of the original image and the reference image after median filtering are calculated respectively. The calculation formula of the phase consistency minimum moment is: Where, Represents the minimum phase consistency moment of any pixel point on the current scale image; It represents the sum of the squares of the x-direction components of phase consistency in different directions; It represents the sum of the products of the x-direction component and the y-direction component of the phase consistency in different directions; It represents the sum of the squares of the y-direction components of phase consistency in different directions; The specific calculation formula is: Where θ represents a discrete direction set set artificially, and PC2(θ) represents the phase consistency map of the current scale image at direction θ. The calculation formula is as follows: Where W(·) is the weight factor, which is used to assign larger weights to multiple frequencies with greater consistency; A n is the amplitude of the nth scale; ξ is the minimum value used to prevent division by 0; For noise compensation; The symbol indicates that when the value is less than 0, it is set to zero, and when it is greater than 0, it is not processed; ΔΦ n (θ) represents the phase deviation function with scale n and direction θ, which is calculated as follows: Where, Φ n (θ) represents the phase component with scale n and direction θ; It represents the average value of the phase components in different directions with a scale of n; The phase consistency minimum moment is calculated for each pixel corresponding to the original image and the reference image to obtain the phase consistency minimum moment maps of the original image and the reference image. Non-maximum suppression is performed on the two phase consistency minimum moment maps respectively. The maximum value position obtained after non-maximum suppression is the feature point of the two images.

4. The multimodal satellite remote sensing image matching method based on phase information according to claim 1, characterized in that: The descriptor corresponding to the feature point in step 3 is constructed based on the calculated phase consistency amplitude map and angle map of the window. Specifically, the amplitude map and angle map windows are divided into k2×k2 units respectively, and the phase consistency angle direction in each unit is assigned to N directions. The phase consistency amplitude calculated to each direction is Gaussian-weighted and accumulated in the corresponding direction according to the distance to form an N-direction histogram. The x-direction of the histogram is 1 to N directions, and the y-direction is the corresponding accumulated value. Finally, the histogram in the k2×k2 units is represented by the corresponding y value, that is, each histogram can be converted into a vector of length N, and the k2×k2 units can generate a descriptor of k2×k2×N dimensions.

5. The multimodal satellite remote sensing image matching method based on phase information according to claim 1, characterized in that: In step 5, after obtaining the matching point sets X and Y according to steps 1-4, a random sampling consistency strategy based on an affine model is used to select the optimal point set as the correctly matched point set. The specific operations are as follows: Randomly select three or more pairs of points with the same name in the point sets X and Y, and match the point pairs (x i ,y i ) and (u i ,v i ) is expressed as homogeneous coordinates and expressed in matrix form as follows: Where X and Y represent the matching point sets in the source image and target image respectively; According to the formula H=Y T X(X T X) -1 Calculate the least squares solution to obtain the transformation matrix H. Use the calculated affine transformation matrix H to map all matching points X to obtain the predicted matching point position X′. Calculate the Euclidean distance L between the predicted matching point position X′ and the actual matching point X. If L is less than the set threshold η, it is considered an inlier; otherwise, it is considered an outlier. Count the total number of inliers contained in the point set X. Repeat the above operation and take the set of data with the largest number of inliers as the optimal data set. The point pairs contained in this data set are the correctly matched points of the same name by the present invention.

6. A multimodal satellite remote sensing image matching system based on phase information, characterized in that: The invention comprises a processor and a memory, wherein the memory is used to store program instructions, and the processor is used to call the program instructions in the memory to execute a multimodal satellite remote sensing image matching method based on phase information as described in any one of claims 1 to 5.

7. A multimodal satellite remote sensing image matching system based on phase information, characterized in that: It includes a readable storage medium, on which a computer program is stored. When the computer program is executed, a multimodal satellite remote sensing image matching method based on phase information as described in any one of claims 1 to 5 is implemented.

Images