A Remote Sensing Image Registration Method, Device, and Storage Medium

Abstract

The present invention relates to a remote sensing image registration method, device and storage medium. The method includes the following steps: Step S1, perform feature detection and description on two given remote sensing images I1 and I2 to be registered, and construct an initial matching set S based on the similarity of feature descriptors; Step S2, based on the consistency constraint of corresponding neighborhood elements and the relative position consistency constraint of neighborhood elements for each pair of matches, establish an objective function for non-rigid matching, eliminate incorrect matches through a threshold, and retain the optimal inlier set corresponding to the correct matches after iterative optimization #imgabs0#. Among them, in the first iteration, the loss function only considers the consistency constraint of neighborhood elements; Step S3, perform transformation estimation on the retained optimal inlier set #imgabs1#, and output the registered remote sensing image. Compared with the prior art, the present invention has the advantage of strong robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of image registration, and more particularly to a remote sensing image registration method, device and storage medium. Background Art

[0002] Image registration aims to geometrically calibrate two or more images in the spatial domain. The images to be registered may be obtained by different sensors at different times and from different perspectives. Image registration can integrate multi-source information to achieve the purpose of fusion. After years of development, image registration, as one of the cornerstones of image stitching, three-dimensional reconstruction, target detection and tracking, change detection, positioning and navigation, image retrieval, etc., has been widely applied in many fields such as medicine, aerospace, and national defense.

[0003] Image registration has received extensive attention and research from scholars in the past few decades. The main research can be roughly divided into the following two categories:

[0004] 1) Gray-based registration methods

[0005] Construct a similarity matching criterion using the gray information of the image. The method is simple and easy to implement. Especially when the images to be registered lack significant structural information, this type of method shows good registration results.

[0006] However, because it strongly depends on the gray values of the images, its stability is not high, and it is easily affected by factors such as uneven illumination and noise, which affect its registration performance.

[0007] 2) Feature-based registration methods

[0008] First, perform feature detection and description on the reference image and the image to be registered, and then construct an initial match based on the similarity of the feature descriptors. Then, combine spatial geometric constraints and consistency constraints, etc. to eliminate false matches. Finally, calculate the parameters of the geometric transformation based on the correct matches to achieve image registration.

[0009] Since this type of method uses robust features, it can process complex images. In addition, the number of feature points is much less than that of pixel points, and the computational complexity is relatively low, so it can be applied to real-time applications. Compared with ordinary visible light images, remote sensing images usually contain more noise due to different imaging mechanisms. For example, Synthetic Aperture Radar (SAR) images are distributed with speckle noise due to the principle of coherent imaging. In addition, remote sensing images collected at different times are prone to uneven illumination. Therefore, feature-based registration methods play a dominant role in the field of remote sensing image registration.

[0010] In the feature-based registration method, a two-step strategy is usually adopted. In the first step, an algorithm capable of detecting and extracting significant and discriminative feature points is first used to detect and describe the feature points, and an initial match is constructed through a certain similarity measure. Due to the existence of noise and the limitations of feature descriptors, the initial match inevitably contains incorrect matches. In the second step, incorrect matches are removed by imposing geometric constraints or consistency constraints on the initial match set, and finally, the geometric transformation parameters are calculated based on the remaining correct matches. Typical methods of this strategy include methods that rely on parametric models, represented by the RANSAC method (M.A. Fischler and R.C. Bolles, “Random sample consensus: A paradigm for model fitting with application to image analysis and automated cartography,” Commun. ACM, vol. 24, no. 6, pp. 381–395, Jun. 1981), and methods based on non-parametric models, represented by the LPM method (J. Ma, J. Zhao, J. Jiang, H. Zhou, and X. Guo, “Locality Preserving Matching,” International Journal of Computer Vision, vol. 127, no. 5, pp. 512–531, Sep. 2018).

[0011] Although the above methods have achieved good results in many fields, when the image pairs to be registered encounter complex non-rigid deformations, the performance of the methods that rely on parametric models drops significantly. In addition, the characteristics of remote sensing images, such as noise, low inlier ratio, repetitive structures, etc., also pose significant challenges to the methods based on non-parametric models.

[0012] Therefore, there is an urgent need for a method with strong robustness that is applicable to remote sensing image registration. Summary of the Invention

[0013] The purpose of the present invention is to provide a remote sensing image registration method, device, and storage medium that are applicable to remote sensing image registration and have high robustness, in order to overcome the defects of the above-mentioned existing technologies.

[0014] The purpose of the present invention can be achieved through the following technical solutions:

[0015] According to a first aspect of the present invention, a remote sensing image registration method is provided, and the method includes the following steps:

[0016] Step S1: Perform feature detection and description on two given remotely sensed images I 1 and I 2 and construct an initial matching set S based on the similarity of feature descriptors;

[0017] Step S2: Based on the neighborhood element consistency constraint and the relative position consistency constraint of neighborhood elements, establish an objective function for non-rigid matching, eliminate incorrect matches through a threshold, and retain the optimal inlier set corresponding to the correct matches after iterative optimization Among them, in the first iteration, the loss function only considers the neighborhood element consistency constraint;

[0018] Step S3: Perform transformation estimation on the retained optimal inlier set and output the registered remotely sensed image.

[0019] Preferably, the specific content of step S1 is as follows:

[0020] Use manual feature descriptors or deep learning-based feature descriptors to perform feature detection and description on two given remotely sensed images I 1 and I 2 respectively, and then construct an initial matching set according to the similarity measure where x i and y i are the spatial coordinates of two corresponding feature points respectively, and N is the number of initial matches.

[0021] Preferably, step S2 includes the following sub-steps:

[0022] Step S2.1: Respectively construct the k-nearest neighbors of the feature point x i and its corresponding point y i in the Euclidean space and and use the Jaccard distance to perform difference measurement on and to represent the neighborhood element consistency constraint corresponding to each pair of matches;

[0023] Step S2.2: Respectively construct the ordered sequences σ(x i ) and σ(y i ) composed of the respective k-nearest neighbors of the feature point x i and its corresponding point y i ), as well as the overlapping ordered sequences i about the feature point x i and its corresponding point y and and based on the Levenshtein distance, perform difference measurement on the overlapping ordered sequences and Perform differential measurement, which is characterized by the relative position consistency constraint of neighborhood elements;

[0024] Step S2.3: Establish an objective function for non-rigid matching, eliminate incorrect matches through a threshold, and retain the optimal inlier set corresponding to the correct matches after iterative optimization Among them, the loss function in the first iteration only considers the neighborhood element consistency constraint, and the loss function in subsequent iterations comprehensively considers the neighborhood element consistency constraint and the relative position consistency constraint of neighborhood elements.

[0025] Preferably, the specific steps of step S2.1 are as follows:

[0026] 1) Construct the k-nearest neighbors of the feature point x i and its corresponding point y i in the Euclidean space, and the expressions are: and The expression is:

[0027]

[0028]

[0029] Among them, |·| represents the number of elements in the set;

[0030] 2) Use the Jaccard distance to perform differential measurement on and The expression is:

[0031]

[0032] Among them, is denoted as n i , satisfying 0 ≤ n i ≤ k; is denoted as 2k - n i ;

[0033] 3) Correct the differential measurement The expression is:

[0034]

[0035] Among them, a is a constant, satisfying 0 < a < 1, used to control the attenuation degree.

[0036] Preferably, the specific steps of step S2.2 are as follows:

[0037] 1) Define the ordered sequences σ(x i ) and σ(y i formed by the k-nearest neighbors of the feature point x i ) and its corresponding point y i), and regarding the feature point x i and its corresponding point y i of the overlapping ordered sequence and

[0038]

[0039]

[0040] 2) Calculate the difference metric between and based on the Levenshtein distance:

[0041]

[0042] wherein, is defined as:

[0043]

[0044] wherein, is the remaining ordered sequence after deducting the first element, represents the nth element starting from 0;

[0045] 3) Normalize the difference metric between and Preferably, the step S2.3 is specifically:

[0046] Construct the objective function to be optimized

[0047]

[0048]

[0049] wherein, is the set of potential inliers, S is the initial matching set; C is the loss function, and the expression is:

[0050]

[0051] In the formula, the loss function C includes two terms. The first term is the penalty term for matching that violates the consistency constraint, and the second term is the regularization term for suppressing outliers; λ > 0 is used to control the balance between these two terms;

[0052] Adopt an N×1 binary vector p = [p 1 , p 2 , …, p n to indicate the correctness of each match, and transform the loss function C into:

[0053]

[0054] Among them, p i ∈ {0, 1}, and p i = 0 indicates that the match (x i , y i ) is an incorrect match, and p i = 1 indicates that the match (x i , y i ) is a correct match;

[0055] The correctness judgment p i expression for each match is:

[0056]

[0057] Optimal inlier set

[0058]

[0059] The loss function of the first iteration only considers using the neighborhood element consistency constraint and satisfies:

[0060]

[0061] In the formula, is the difference measure of and using the Jaccard distance;

[0062] The loss function of subsequent iterations comprehensively considers the neighborhood element consistency constraint and the neighborhood element relative position consistency constraint and satisfies:

[0063]

[0064] In the formula, is the difference measure of the overlapping ordered sequences and based on the Levenshtein distance.

[0065] Preferably, the subsequent iteration process satisfies: constructing the neighborhood of each feature point in the next iteration using the match set retained after removing incorrect matches in the previous iteration.

[0066] Preferably, step S3 is specifically:

[0067] Using thin plate spline TPS to estimate the transformation of the optimal inlier set to map the two remote sensing images to the same coordinate system, realizing image registration and information fusion.

[0068] According to a second aspect of the present invention, there is provided an electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, any one of the methods described above is implemented.

[0069] According to a third aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, any one of the methods described above is implemented.

[0070] Compared with the prior art, the present invention has the following advantages:

[0071] 1) The local non-rigid distortion in the remote sensing image of the present invention is modeled by using the local neighborhood elements of the feature points and their relative position consistency. The model does not depend on any specific model and can be applied to a variety of potential transformations in the remote sensing image;

[0072] 2) The present invention uses an exponential function to control the Jaccard distance and The difference measure The attenuation degree of It has a long-tail distribution and can prevent excessive punishment of outliers;

[0073] 3) Since the initial matching set may contain many false matches, the present invention continuously eliminates false matches through iterative optimization, which makes it more robust;

[0074] 4) During the first iteration of the present invention, the loss function only considers the consistency constraint of the neighborhood elements of the feature point, which speeds up the matching speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0077] Example

[0078] like Figure 1 As shown, this embodiment provides a remote sensing image registration method for realizing single-source or multi-source image registration to achieve information fusion, including the following steps:

[0079] Step S1: for two given remote sensing images to be registered,1 and I 2 Perform feature detection and description, and construct an initial matching set S based on the similarity of feature descriptors. Specifically:

[0080] First, use traditional handcrafted feature descriptors such as SIFT (Scale-invariant feature transform) or deep learning-based feature descriptors such as LIFT (Learned invariant feature transform) to perform feature detection and description on two images respectively, and then construct an initial matching set according to the similarity metric where x i and y i are the spatial coordinates of two corresponding feature points respectively, and N is the number of initial matches.

[0081] Step S2: For non-rigid deformations such as geomorphic changes that are prone to occur in remote sensing images, establish an objective function for non-rigid matching based on the consistency constraint of neighborhood elements and the consistency constraint of the relative positions of neighborhood elements. Eliminate incorrect matches through thresholds, and retain the optimal inlier set corresponding to the correct matches after iterative optimization Among them, in the first iteration, the loss function only considers the consistency constraint of neighborhood elements; specifically, it includes the following sub-steps:

[0082] Step S2.1: Construct the k-nearest neighbors of the feature point x i and its corresponding point y i in the Euclidean space and and use the Jaccard distance to and perform difference measurement, which is characterized as the consistency constraint of neighborhood elements corresponding to each pair of matches; specifically, it includes the following content:

[0083] 1) Construct the k-nearest neighbors of the feature point x i and its corresponding point y i in the Euclidean space and The expression is:

[0084]

[0085]

[0086] where |·| is the number of elements in the set; ideally, if (x i , y i ) is a correct match, then their k-nearest neighbors and will be exactly the same. However, in real life, due to the existence of objective factors such as noise, outliers, and image degradation, and will be very difficult to be the same. However, for correct matches, there are usually correct matches clustered in their neighborhoods, while there will be few matches in the neighborhoods of incorrect matches that are consistent with them.

[0087] 2) Use the Jaccard distance for and to measure the difference, and the expression is:

[0088]

[0089] Obviously, if and are exactly the same, then makes Conversely, if and are completely different, then makes

[0090] For the convenience of description, denote as n i (0 ≤ n i ≤ k), so is 2k - n i

[0091]

[0092] 3) It is easy to find that with respect to n i the first-order derivative and the second-order derivative are both less than 0. Therefore, is actually a decreasing concave function, which is contrary to the actual situation. Usually, the value should decrease at a decreasing rate as the value of n i increases, that is, should be a decreasing convex function. To solve this problem, correct the difference measurement , and the expression is:

[0093]

[0094] where a is a constant, satisfying 0 < a < 1, used to control the degree of attenuation. After correction, has a long-tailed distribution and can prevent excessive punishment for outliers.

[0095] Step S2.2. The difference metric in Step S2.1 aims to utilize the consistency of the neighborhood elements of the feature points while ignoring the relationship between the relative positions of the neighborhood elements. When the value of k for neighborhood construction is selected too large, it is easy for incorrect matches to be misidentified as correct matches because they satisfy the neighborhood element consistency. For example, in an extreme case, when n i = N, whether it is a correct match or an incorrect match, their corresponding neighborhood elements are the same, and it is difficult to distinguish them. To solve the above problem, a method for the recognizability of the relative positions of neighborhood elements is proposed, specifically as follows:

[0096] 1) Define the ordered sequences σ(x i ) and σ(y i ) composed of the respective k-nearest neighbors of the feature point x i and its corresponding point y i , as well as the overlapping ordered sequences i and i of the feature point x and its corresponding point y

[0097]

[0098]

[0099] It can be found that and have the same elements but different relative orders.

[0100] 2) Based on the Levenshtein distance, calculate the difference metric between and :

[0101]

[0102] Among them, is defined as:

[0103]

[0104] Among them, is the ordered sequence remaining after deducting the first element from , and represents the nth element starting from 0;

[0105] In information theory, the Levenshtein distance between two ordered sequences is the minimum number of edits required to transform one sequence into the other, where each edit (deletion, insertion, and substitution) can only operate on one element. Therefore, when two ordered sequences have more elements that maintain the same relative order, their Levenshtein distance will be smaller. Conversely, if there are significant differences in the relative order of the elements of two ordered sequences, their Levenshtein distance will be larger.

[0106] 3) Multiply the Levenshtein distance by to map the value to [0, 1), for normalization and the dissimilarity measure of

[0107] Step S2.3: Establish the objective function for non-rigid matching, eliminate incorrect matches through a threshold, and retain the optimal inlier set corresponding to the correct matches after iterative optimization Among them, the loss function in the first iteration only considers the consistency constraint of neighborhood elements, and the loss function in subsequent iterations comprehensively considers the consistency constraint of neighborhood elements and the consistency constraint of the relative positions of neighborhood elements; specifically:

[0108] 1) Construct the objective function to be optimized

[0109]

[0110] Among them, is the potential inlier set, S is the initial matching set; C is the loss function, and the expression is:

[0111]

[0112] In the formula, the loss function C includes two terms. The first term is the penalty term for matches that violate the consistency constraint, and the second term is the regularization term for suppressing outliers; λ > 0, which is used to control the balance between these two terms;

[0113] 2) Use an N×1 binary vector p = [p 1 , p 2 , …, p n to indicate the correctness of each match, and transform the loss function C into:

[0114]

[0115] Among them, p i ∈{0, 1}, p i = 0 indicates that the match (x i , y i ) is an incorrect match, p i=1 indicates matching (x i ,y i ) is a correct match;

[0116] The correctness judgment p of each match i The expression is:

[0117]

[0118] 3) Optimal interior point set

[0119]

[0120] 4) The specific iterative optimization process is as follows:

[0121] Given that the initial matching set may contain a large number of false matches, an iterative optimization strategy is adopted;

[0122] In order to speed up the matching process, the loss function of the first iteration only considers the consistency constraint of the neighborhood elements, satisfying:

[0123]

[0124] In the formula, To use Jaccard distance and The difference measure of

[0125] At this point, the algorithm has been able to remove a large number of false matches, and the set of matches that remain is called Subsequent iterations use the matching set retained after removing the wrong matches in the previous iteration to construct the neighborhood of each feature point in the next iteration. The loss function comprehensively considers the consistency constraints of neighborhood elements and the consistency constraints of the relative positions of neighborhood elements, satisfying:

[0126]

[0127] It is worth noting that the construction of the neighborhood of each feature point is based on the initial complete matching set S. Compared with S, it has a higher proportion of inliers. Therefore, in the next iteration, the construction of the neighborhood of each feature point is based on After about 3 iterations, the algorithm converges, and the remaining matches form the optimal set of interior points.

[0128] In the formula, Based on the Levenshtein distance, the overlapping ordered sequences and The difference measurement performed.

[0129] Step S3: When the optimal inlier set is obtained perform transformation estimation between the two images based on the optimal inlier set To adapt to the local non-rigid deformation existing in remote sensing images, the existing Thin Plate Spline (TPS) is used for transformation estimation, so that the two remote sensing images can be mapped to the same coordinate system, achieving the purpose of image registration and information fusion.

[0130] In summary, a remote sensing image registration method based on the consistency of feature point neighborhoods and their relative positions given in this embodiment uses a robust feature detection and feature description operator to detect and describe features of two given overlapping remote sensing images, constructs an initial matching set according to the similarity of the feature description operator; uses the consistency of correct matches in the local neighborhood to construct an objective function to be optimized, filters out incorrect matches through a threshold, and retains correct matches; finally, based on the retained correct matches, establishes a transformation relationship between the two remote sensing images to achieve remote sensing image registration. The present invention models using the local neighborhood elements of feature points and their relative position consistency for the local non-rigid distortion existing in remote sensing images. This model does not depend on any specific model, so it can be applied to various potential transformations in remote sensing images. Even if there are a large number of incorrect matches in the initial matching set, it can still maintain good robustness.

[0131] The electronic device of the present invention includes a central processing unit (CPU), which can execute various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or computer program instructions loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0132] Multiple components in the device are connected to the I / O interface, including: an input unit, such as a keyboard, a mouse, etc.; an output unit, such as various types of displays, speakers, etc.; a storage unit, such as a disk, an optical disc, etc.; and a communication unit, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit allows the device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0133] The processing unit executes the various methods and processes described above, such as methods S1 to S3. For example, in some embodiments, methods S1 to S3 may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed onto the device via the ROM and / or the communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S3 described above may be executed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S3 by any other suitable means (e.g., by means of firmware).

[0134] The functions described above herein can be performed at least in part by one or more hardware logic components. By way of example and not limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0135] The program code for implementing the methods of the present invention may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, a special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on the remote machine or server.

[0136] In the context of the present invention, a machine-readable medium may be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0137] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A remote sensing image registration method, characterized in that, the method comprises the following steps: Step S1: Perform feature detection and description on two given remotely sensed images I 1 and I 2 and construct an initial matching set S based on the similarity of feature descriptors; Step S2: Based on the neighborhood element consistency constraint and the neighborhood element relative position consistency constraint, establish an objective function for non-rigid matching, eliminate incorrect matches through a threshold, and retain the optimal inlier set corresponding to the correct matches after iterative optimization. Among them, in the first iteration, the loss function only considers the neighborhood element consistency constraint. Specifically, it includes the following sub-steps: Step S2.

1. Construct the feature points x i and their corresponding points y i in the k-nearest neighbors in Euclidean space and and use the Jaccard distance for and to perform differential measurement, which is characterized by the consistency constraint of each pair of matching corresponding neighborhood elements; Step S2.

2. Respectively construct feature points x i and their corresponding points y i into ordered sequences σ(x i ) and σ(y i ) composed of their respective k-nearest neighbors, as well as the overlapping ordered sequences i of feature points x i and their corresponding points y and And perform difference measurement on the overlapping ordered sequences and based on the Levenshtein distance, which is characterized as the relative position consistency constraint of neighborhood elements; Step S2.3: Establish an objective function for non-rigid matching, eliminate incorrect matches through a threshold, and retain the optimal inlier set corresponding to the correct matches after iterative optimization. Among them, the loss function in the first iteration only considers the consistency constraint of neighborhood elements, and the loss function in subsequent iterations comprehensively considers the consistency constraint of neighborhood elements and the consistency constraint of the relative positions of neighborhood elements. Step S3. Perform transformation estimation on the retained optimal inlier set and output the registered remote sensing image.

2. The remote sensing image registration method according to claim 1, characterized in that, the specific step S1 is: Use manual feature descriptors or deep learning-based feature descriptors to perform feature detection and description on two given remote sensing images I 1 and I 2 respectively, and then construct an initial matching set according to the similarity measure where x i and y i are the spatial coordinates of two corresponding feature points respectively, and N is the number of initial matches.

3. The remote sensing image registration method according to claim 1, characterized in that, the specific step S2.1 is: 1) Construct the feature points x i and their corresponding points y i in the k-nearest neighbors in Euclidean space and The expression is: wherein, |·| is the number of elements for calculating the set; 2) Use the Jaccard distance for and to perform differential measurement. The expression is: Among them, denoted as n i , satisfying 0 ≤ n i ≤ k; denoted as 2k - n i ; 3) Modify the difference metric The expression is as follows: where a is a constant satisfying 0 < a < 1 and is used to control the degree of attenuation.

4. The remote sensing image registration method according to claim 1, characterized in that, the specific step S2.2 is: 1) Define the feature point x i and its corresponding point y i The ordered sequences σ(x i ) and σ(y i ) composed of their respective k-nearest neighbors, and the overlapping ordered sequences i about the feature point x i and its corresponding point y and 2) Calculate the dissimilarity measure between and based on the Levenshtein distance: Among them, is defined as: Among them, is the ordered sequence remaining after deducting the first element, represents the nth element starting from 0; 3) Normalization and difference metric 5. The remote sensing image registration method according to claim 1, characterized in that, the specific step S2.3 is: Construct the objective function to be optimized Among them, is the set of potential interior points, S is the initial matching set; C is the loss function, and its expression is: In the formula, the loss function C includes two terms. The first term is the penalty term for violating the consistency constraint matching, and the second term is the regularization term for suppressing outliers; λ>0, which is used to control the balance between these two terms; Adopt an N×1 binary vector p = [p 1 , p 2 , …, p n to indicate the correctness of each match, and transform the loss function C into: Among them, p i ∈ {0, 1}, p i = 0 indicates that the match (x i , y i ) is an incorrect match, and p i = 1 indicates that the match (x i , y i ) is a correct match; Correctness judgment p for each match i The expression is: Optimal inner point set The loss function for the first iteration only considers using the neighborhood element consistency constraint and satisfies: In the formula, is the difference measurement of and using the Jaccard distance; The loss function for subsequent iterations comprehensively considers the neighborhood element consistency constraint and the neighborhood element relative position consistency constraint and satisfies: In the formula, is the difference metric performed on the overlapping ordered sequences and based on the Levenshtein distance.

6. The remote sensing image registration method according to claim 5, characterized in that, the subsequent iteration process satisfies: constructing the neighborhood of each feature point for the next iteration by using the matching set remaining after removing the wrong matches in the previous iteration.

7. The remote sensing image registration method according to claim 1, characterized in that, the specific step S3 is: Use thin plate spline TPS to estimate the transformation of the optimal inlier set Perform transformation estimation, map the two remote sensing images to the same coordinate system, and achieve image registration and information fusion.

8. An electronic device, comprising a memory and a processor, and a computer program is stored on the memory, characterized in that, when the processor executes the program, it implements the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, on which a computer program is stored, characterized in that, when the program is executed by the processor, it implements the method according to any one of claims 1 to 7.

Images