Multi-source remote sensing image registration method and device

Abstract

The invention relates to the technical field of image processing, in particular to a multi-source remote sensing image registration method and device. Comprising the following steps: respectively carrying out feature point detection on a received multi-source remote sensing image pair to be registered to obtain a first point feature and a second point feature; the to-be-registered multi-source remote sensing image pair comprises a reference image and a to-be-registered image; performing multi-scale PIIFD feature description on the first point feature and the second point feature to obtain a first point feature description parameter and a second point feature description parameter; for the first point feature and the second point feature, feature point matching and wrong matching point removal are carried out by utilizing description parameters of the first point feature and the second point feature, effective matching point pairs are screened out, and a first matching feature point set and a second matching feature point set are obtained; and performing transformation matrix estimation and image transformation based on the first matching feature point set and the second matching feature point set to obtain a to-be-registered image aligned with the reference image, and finally obtaining a registered image pair.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a multi-source remote sensing image registration method and device. Background Art

[0002] Image registration refers to the process of aligning two or more image data in the same coordinate system using an effective transformation strategy. It is widely used in image fusion, image change detection, image stitching and image mosaicking.

[0003] In recent years, the emergence of large-scale structured data, the improvement of hardware computing power, and the emergence of new regularization methods have led to rapid development of deep learning technology. In particular, combining deep learning technology to solve the problem of remote sensing image registration has also attracted widespread attention from scholars. However, the remote sensing image registration method based on deep learning relies too much on a large number of existing training samples, resulting in poor generalization ability, and it is difficult to meet the needs of multi-source remote sensing image registration in complex and changing scenes. Therefore, the present invention focuses on the study of non-deep learning remote sensing image registration methods, which can be mainly divided into three categories:

[0004] (1) Grayscale-based image registration methods. This type of method uses the grayscale information of the entire image as the basis to establish a similarity measure between the image to be registered and the reference image. It is simple to implement, but is sensitive to grayscale changes and noise, has a large amount of calculation and low efficiency, and is not suitable for image matching with nonlinear deformation.

[0005] (2) Image registration methods based on transform domains. This type of method is processed in transform domains such as Fourier transform and wavelet transform. It has a certain degree of robustness against translation, rotation and scaling, but it also has limitations. The method is only applicable to images with linear positive correlation in grayscale attributes, and the images must strictly satisfy the defined transformation relationship.

[0006] (3) Feature-based image registration methods. This type of method is insensitive to grayscale changes, noise, etc., and has low computational complexity and high efficiency, and has been widely used. This type of method is mainly implemented through feature detection, feature description, feature matching and image transformation. The key is how to detect the local invariant features of the image, such as point features, line features, surface (i.e., region) features, etc. Compared with line features and region features, point features are widely used in image registration tasks because of their excellent properties of not changing with changes in illumination, translation, scaling, and projection. The main remote sensing image registration methods based on point features include the SIFT (Scale-invariant feature transform) algorithm, the ORB (Oriented fast and Rotate Brief) algorithm, and their improved methods.

[0007] With the development of remote sensing imaging technology in recent years, aerospace and aviation remote sensing have entered the era of massive data. Visible light, infrared, SAR, laser and other multi-type sensor systems can obtain multi-source remote sensing image data of multiple sensors, multiple phases, multiple resolutions and multiple observation angles in the same area. Multi-source remote sensing images are rich in types and have different features, which brings new challenges to image registration technology. There is an urgent need to study general and robust multi-source remote sensing image registration methods.

[0008] In view of the problem of multi-source remote sensing image registration, Li et al. proposed a method based on radiation-variation insensitive feature transform (RIFT) to solve the problem of pixel intensity difference and nonlinear radiation distortion, and has good performance for heterogeneous image registration. This method uses phase consistency (PC) to construct multiple maximum index mappings (MIM) with different rotation degrees for feature description. It has strong robustness to nonlinear radiation distortion and certain rotation invariance. However, due to the loss of a large amount of detail information when optimizing MIM, it has poor adaptability to rotation distortion. Gao et al. successively proposed a multi-source remote sensing image registration method based on multi-scale partial intensity invariant feature descriptor (MS-PIIFD) and a multi-scale local main direction histogram feature (MS-HLMO). The MS-HLMO method has strong robustness to nonlinear radiation distortion and good registration performance for heterogeneous images, but its robustness to rotation distortion needs to be improved. On the contrary, the MS-PIIFD method is highly robust to rotational distortion, but less robust to nonlinear radiation distortion, and the accuracy of heterogeneous image registration needs to be improved.

[0009] In summary, the current multi-source remote sensing image registration method based on point features has poor robustness and cannot handle the problem of accurate multi-source remote sensing image registration of complex scenes with numerous changes, so as to meet the needs of dynamic monitoring of complex regional scenes under multi-source observation conditions. Summary of the invention

[0010] The purpose of the present invention is to overcome the defects of the prior art and propose a multi-source remote sensing image registration method and device.

[0011] In order to achieve the above object, the present invention proposes a multi-source remote sensing image registration method, comprising:

[0012] Step 1) performing feature point detection on the received multi-source remote sensing image pair to be registered, respectively, to obtain a first point feature and a second point feature; the multi-source remote sensing image pair to be registered includes a reference image and an image to be registered;

[0013] Step 2) Perform multi-scale PIIFD feature description on the first point feature and the second point feature respectively to obtain the first point feature description parameter and the second point feature description parameter;

[0014] Step 3) for the first point feature and the second point feature, using their description parameters, perform feature point matching and remove incorrect matching points, screen out valid matching point pairs, and obtain a first matching feature point set and a second matching feature point set;

[0015] Step 4) Based on the first matching feature point set and the second matching feature point set, transformation matrix estimation and image transformation are performed to obtain an image to be registered that is aligned with the reference image, and finally a registered image pair is obtained.

[0016] Preferably, the step 1) comprises:

[0017] Step 1-1) Using the FAST corner detection algorithm, perform feature point detection on the reference image I1 and the image to be registered I2, respectively, to obtain the first original image point feature and the second original image point feature;

[0018] Step 1-2) constructing phase consistency maps for the reference image I1 and the image to be registered I2 respectively, and performing feature point detection on the phase consistency maps to obtain first PC map point features and second PC map point features;

[0019] Step 1-3) The first original image point feature and the second original image point feature are respectively fused with the corresponding PC image point features by removing duplicate point features to obtain the first point feature and the second point feature.

[0020] Preferably, the steps 1-2) include:

[0021] For the reference image I1 and the image to be registered I2, phase consistency maps are constructed respectively by using a phase consistency method to obtain a first phase consistency map and a second phase consistency map;

[0022] The FAST corner point detection algorithm is used to perform feature point detection on the first phase consistency image and the second phase consistency image to obtain the first PC image point features and the second PC image point features.

[0023] Preferably, the step 2) comprises:

[0024] For the input multi-source remote sensing image pairs I1 and I2 to be registered, Gaussian pyramids are created respectively to obtain a first Gaussian pyramid and a second Gaussian pyramid.

[0025] For the first point feature and the second point feature, a partial intensity invariant descriptor method is used on the first Gaussian pyramid and the second Gaussian pyramid respectively to perform multi-scale feature point description to obtain the first point feature description parameter and the second point feature description parameter.

[0026] Preferably, the step 3) comprises:

[0027] For the first point feature and the second point feature, using their description parameters, adopting the BBF matching method, taking cosine similarity as the metric, perform feature point matching to obtain the initial matching point pair;

[0028] By using the consistency of the main direction and spatial distribution of the feature points, the wrong matching point pairs are removed, and the valid matching point pairs are screened out to obtain the first matching feature point set and the second matching feature point set.

[0029] Preferably, the step 4) comprises:

[0030] Based on the first matching feature point set and the second matching feature point set constructed by the screened valid matching point pairs, a geometric model is selected as a transformation model between the image pairs, and the parameters of the transformation matrix corresponding to the transformation model are estimated;

[0031] Based on the transformation matrix obtained by the above estimation, the image to be registered is transformed to obtain the image to be registered that is aligned with the reference image, and finally a registered image pair is obtained.

[0032] Preferably, the geometric model includes: similarity transformation, affine transformation and perspective transformation; wherein the similarity transformation is a combination of basic transformations of translation, rotation and scaling.

[0033] Preferably, the transformation model satisfies the following formula:

[0034]

[0035] Where H is the transformation matrix, (x2, y2) T is the coordinate of the midpoint of the image to be registered, (x1, y1) T is (x2,y2) T Coordinates of the homonymous points on the reference image;

[0036] The transformation matrix H is solved using the least squares method;

[0037] The transformation matrix H of similarity transformation Similarity Satisfy the following formula:

[0038]

[0039] Where S represents the scaling factor, θ represents the rotation angle, (t x ,ty ) represents the translation amount.

[0040] The transformation matrix H of the affine transformation Affine Satisfy the following formula:

[0041]

[0042] Transformation matrix H for perspective transformation Projective Satisfies the following formula,

[0043]

[0044] Among them, a 11 , a 12 , a 13 , a 21 , a 22 , a 23 , a 31 , a 32 are the elements at the corresponding positions of the matrix respectively.

[0045] On the other hand, the present invention provides a computer device / equipment / system, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0046] In yet another aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the above method when executed by a processor.

[0047] Compared with the prior art, the advantages of the present invention are:

[0048] 1. By constructing a phase congruency (PC) map, feature point detection and fusion are performed on the original image and the PC map, which retains the feature points of the original image and fuses the feature points of the PC map, improving the robustness of the method to noise, grayscale changes, especially nonlinear radiation distortion, and is suitable for multi-phase heterogeneous source image registration;

[0049] 2. The multi-scale partially intensity invariant descriptor (MS-PIIFD) is introduced to describe the image feature points, which improves the robustness of the method to scaling, rotation, and especially rotational distortion. It is suitable for multi-resolution and multi-observation angle image registration acquired by different sensors (including homogeneous and heterogeneous). BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of the process of the multi-source remote sensing image registration method of the present invention;

[0051] Figure 2 This is the schematic diagram of the FAST corner detection algorithm;

[0052] Figure 3 It is the experimental verification of raw data;

[0053] Figure 4 This is the verification result of the first group of experiments;

[0054] Figure 5 This is the verification result of the second group of experiments;

[0055] Figure 6 This is the verification result of the third group of experiments. DETAILED DESCRIPTION

[0056] The invention discloses a multi-source remote sensing image registration method, comprising the following steps:

[0057] Step 1) performing feature point detection on the received multi-source remote sensing image pair to be registered, respectively, to obtain a first point feature and a second point feature; the multi-source remote sensing image pair to be registered includes a reference image and an image to be registered;

[0058] Step 2) Perform multi-scale PIIFD feature description on the first point feature and the second point feature respectively to obtain the first point feature description parameter and the second point feature description parameter;

[0059] Step 3) for the first point feature and the second point feature, using their description parameters, perform feature point matching and remove incorrect matching points, screen out valid matching point pairs, and obtain a first matching feature point set and a second matching feature point set;

[0060] Step 4) Based on the first matching feature point set and the second matching feature point set, transformation matrix estimation and image transformation are performed to obtain an image to be registered that is aligned with the reference image, and finally a registered image pair is obtained.

[0061] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0062] Example 1

[0063] like Figure 1As shown, an embodiment of the present invention proposes a multi-source remote sensing image registration method. Specifically, it includes: for the input multi-source remote sensing image pair to be registered, including the reference image I1 and the image to be registered I2, feature point detection is performed respectively to obtain the first point feature and the second point feature; for the first point feature and the second point feature, multi-scale PIIFD feature description is performed respectively to obtain the first point feature description parameter and the second point feature description parameter; for the first point feature and the second point feature, using their description parameters, feature point matching and wrong matching point removal are performed to screen out valid matching point pairs to obtain the first matching feature point set and the second matching feature point set; based on the first matching feature point set and the second matching feature point set, transformation matrix estimation and image transformation are performed to obtain the image to be registered aligned with the reference image, and finally the registered image pair is obtained.

[0064] The first step is feature point detection.

[0065] For the input multi-source remote sensing image pair to be registered, including the reference image I1 and the image to be registered I2, feature point detection is performed respectively to obtain the first original image point feature and the second original image point feature; the phase consistency (PC) map is constructed respectively, and feature point detection is performed on the phase consistency map to obtain the first PC map point feature and the second PC map point feature; the first original image point feature and the second original image point feature are fused with their PC map point features respectively to obtain the first point feature and the second point feature. The method is:

[0066] 1.1. For the reference image I1 and the image to be registered I2, the FAST corner detection algorithm is used to perform feature point detection respectively to obtain the first original image point feature and the second original image point feature. The principle of the FAST corner detection algorithm is as follows Figure 2 shown.

[0067] 1.2. For the reference image I1 and the image to be registered I2, respectively, a phase consistency map is constructed, and feature point detection is performed on the phase consistency map to obtain a first PC map point feature and a second PC map point feature;

[0068] 1.2.1. For the reference image I1 and the image to be registered I2, the phase consistency method is used to construct phase consistency maps, respectively, to obtain the first phase consistency map and the second phase consistency map. Using phase consistency to construct the phase consistency map (edge ​​map) can effectively reduce the impact of light intensity and contrast changes between images, and can effectively retain key information such as image texture, edges and corners.

[0069] 1.2.2. For the first phase consistency diagram and the second phase consistency diagram, the FAST corner point detection algorithm is used to perform feature point detection respectively to obtain the first PC diagram point features and the second PC diagram point features.

[0070] 1.3. The first original image point feature and the second original image point feature are respectively fused with their PC image point features by removing duplicate point features to obtain the first point feature and the second point feature.

[0071] The second step is feature point description.

[0072] For the input multi-source remote sensing image pairs I1 and I2 to be registered, Gaussian pyramids are created respectively to obtain the first Gaussian pyramid and the second Gaussian pyramid; and the first point feature and the second point feature are described by multi-scale PIIFD on the first Gaussian pyramid and the second Gaussian pyramid respectively to obtain the first point feature description parameters and the second point feature description parameters. The method is:

[0073] 2.1. For the input multi-source remote sensing image pairs I1 and I2 to be registered, Gaussian pyramids are created respectively to obtain a first Gaussian pyramid and a second Gaussian pyramid.

[0074] 2.2. For the first point feature and the second point feature, PIIFD is used on the first Gaussian pyramid and the second Gaussian pyramid respectively to perform multi-scale feature point description to obtain the first point feature description parameters and the second point feature description parameters.

[0075] The partial intensity invariant descriptor PIIFD is an improved algorithm of SIFT. First, a main direction (0~π) is assigned to each feature point by calculating the local second-order gradient to ensure its rotation invariance. Then, the gradient in the neighborhood of each feature point is calculated with the main direction as the reference plane. The size of the neighborhood is fixed to 40×40 pixels. Finally, a descriptor vector with a length of 128 is generated for feature point description.

[0076] The third step is feature point matching.

[0077] For the first point feature and the second point feature, BBF (Best Bin First) feature point matching is performed using their description parameters to obtain initial matching point pairs; the main direction and spatial distribution consistency of the feature points are used to remove incorrect matching point pairs, and valid matching point pairs are screened out to obtain the first matching feature point set and the second matching feature point set.

[0078] 3.1. For the first point feature and the second point feature, use their description parameters, adopt the BBF matching method and use cosine similarity as the metric to perform feature point matching and obtain the initial matching point pair.

[0079] 3.2. Using the consistency of the main direction and spatial distribution of feature points, remove the wrong matching point pairs, filter out the valid matching point pairs, and obtain the first matching feature point set and the second matching feature point set. The method is:

[0080] There may be a large number of wrong matching point pairs in the initial matching point pairs, which makes the commonly used wrong matching point removal algorithms, such as the RANSAC algorithm, fail. The present invention uses the consistency of the main direction and spatial distribution of feature points to eliminate wrong matching point pairs, screen out valid matching point pairs, and construct the first matching feature point set and the second matching feature point set.

[0081] The fourth step is transformation matrix estimation and image transformation.

[0082] Based on the first matching feature point set and the second matching feature point set, a transformation matrix is ​​estimated; and based on the transformation matrix, the image to be registered is transformed to obtain the image to be registered aligned with the reference image, and finally a registered image pair is obtained.

[0083] 4.1. Based on the first matching feature point set and the second matching feature point set constructed by the screened valid matching point pairs, a geometric model is selected as the transformation model between the image pairs, and the parameters of the transformation model matrix are estimated. The method is:

[0084] The commonly used geometric transformation models in image registration are similarity transformation, affine transformation, and perspective transformation. The image geometric transformation model is defined as follows:

[0085]

[0086] Where H is the transformation matrix, (x2, y2) T is the coordinate of the midpoint of the image to be registered, (x1, y1) T is (x2,y2) T Coordinates of the points of the same name on the reference image.

[0087] Similarity transformation is a combination of translation, rotation and scaling basic transformations, and its transformation matrix is ​​defined as follows:

[0088]

[0089] Where S represents the scaling factor, θ represents the rotation angle, (t x ,t y ) represents the translation amount.

[0090] The transformation matrix of affine transformation is defined as follows:

[0091]

[0092] The transformation matrix of perspective transformation is defined as follows:

[0093]

[0094] Among them, a 11 , a 12 , a13 , a 21 , a 22 , a 23 , a 31 , a 32 are the elements at the corresponding positions of the matrix respectively.

[0095] Perspective transformation has a more general form, and similarity transformation and affine transformation can be regarded as special cases of perspective transformation. Perspective transformation has 8 degrees of freedom. In theory, only 4 sets of matching point pairs in the overlapping area are needed to solve the perspective transformation matrix of the two images. In practical applications, in order to deal with different spatial distortions (i.e., distortions) that may exist between image pairs, the present invention uses more matching point pairs and adopts the least squares method to solve the transformation matrix H, which has higher robustness. This embodiment adopts affine transformation as the transformation model between image pairs.

[0096] 4.2. Based on the transformation matrix H estimated above, the image to be registered is transformed to obtain the image to be registered that is aligned with the reference image, and finally a registered image pair is obtained.

[0097] The present invention provides a computer device / equipment / system, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0098] The present invention provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the above method are implemented.

[0099] This embodiment selects three sets of challenging scene data, and conducts performance verification experiments on the multi-source remote sensing image registration method provided by the present invention; and compares the performance with three current cutting-edge methods, including RIFT, MS-HLMO, and MS-PIIFD. The first and second sets of data are multi-view, multi-resolution, multi-source homogeneous remote sensing data captured by different visible light sensors; the third set of data is multi-source heterogeneous remote sensing data acquired by visible light sensors and SAR sensors. Figure 3 Three groups of experimental data are given. From left to right in the figure are the 1st to 3rd groups of data, and from top to bottom are the reference image and the image to be registered.

[0100] The quantitative evaluation results of the three groups of experiments are shown in Table 1. At present, the most effective quantitative evaluation indicator for the evaluation of the registration effect of multi-source remote sensing images, especially heterogeneous remote sensing images, is the number of correct matching points. In addition, this verification also focuses on the visualization of the registered result images, which can better reflect people's intuitive visual experience. The visualization results of the first to third groups of experiments are shown in Table 1. Figure 4 , Figure 5 and Figure 6As shown. From left to right in the figure are: RIFT, MS-HLMO, MS-PIIFD and the registration results of the method proposed in the present invention (displayed in a checkerboard manner), and from top to bottom are the overall result map and the enlarged map of the local area. It is worth noting that for the comparison methods that can draw conclusions only through the overall result map, we will no longer display the enlarged map of the local area. It is worth noting that Table 1 gives the number of correct matching points output by the algorithm. The actual number of matching points is generally less than the output of the algorithm. For example, the number of correct matching points output by the RIFT algorithm in the second group of experiments is 6, and the actual number of matching points is 0, which is consistent with Figure 5 The matching results of the RIFT algorithm shown in are consistent.

[0101] Table 1 Quantitative evaluation results - number of correct matching points

[0102]

[0103] Figure 4 , Figure 5 The results of the first and second groups of experiments shown in Table 1 indicate that the proposed method is highly robust to scaling and rotational distortions and is suitable for multi-resolution and multi-observation angle image registration acquired by different sensors (including homogeneous and heterogeneous sensors). Figure 6 The results of the third group of experiments shown in Table 1 indicate that the proposed method has good robustness to nonlinear radiation distortion and noise, and is suitable for multi-phase heterogeneous source image registration.

[0104] The experimental results above show that the multi-source remote sensing image registration method proposed in the present invention is superior to other existing cutting-edge algorithms in terms of effectiveness and versatility, and has good invariance to nonlinear radiation distortion, noise, scaling, rotational distortion, etc. It has high matching accuracy and good robustness for multi-source remote sensing images with multiple phases, multiple resolutions, and multiple observation angles acquired by multiple types of sensors in complex scenes.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention and should be included in the scope of the claims of the present invention.

Claims

1. A multi-source remote sensing image registration method, comprising: Step 1) performing feature point detection on the received multi-source remote sensing image pairs to be registered, and obtaining first point features and second point features; The multi-source remote sensing image pair to be registered comprises a reference image and an image to be registered; Step 2) Perform multi-scale PIIFD feature description on the first point feature and the second point feature respectively to obtain the first point feature description parameter and the second point feature description parameter; Step 3) for the first point feature and the second point feature, using their description parameters, perform feature point matching and remove incorrect matching points, screen out valid matching point pairs, and obtain a first matching feature point set and a second matching feature point set; Step 4) Based on the first matching feature point set and the second matching feature point set, transformation matrix estimation and image transformation are performed to obtain an image to be registered that is aligned with the reference image, and finally a registered image pair is obtained.

2. The multi-source remote sensing image registration method according to claim 1, characterized in that: The step 1) comprises: Step 1-1) Using the FAST corner detection algorithm, perform feature point detection on the reference image I1 and the image to be registered I2, respectively, to obtain the first original image point feature and the second original image point feature; Step 1-2) constructing phase consistency maps for the reference image I1 and the image to be registered I2 respectively, and performing feature point detection on the phase consistency maps to obtain first PC map point features and second PC map point features; Step 1-3) The first original image point feature and the second original image point feature are respectively fused with the corresponding PC image point features by removing duplicate point features to obtain the first point feature and the second point feature.

3. The multi-source remote sensing image registration method according to claim 2, characterized in that: The steps 1-2) include: For the reference image I1 and the image to be registered I2, phase consistency maps are constructed respectively by using a phase consistency method to obtain a first phase consistency map and a second phase consistency map; The FAST corner point detection algorithm is used to perform feature point detection on the first phase consistency image and the second phase consistency image to obtain the first PC image point features and the second PC image point features.

4. The multi-source remote sensing image registration method according to claim 1, characterized in that: The step 2) comprises: For the input multi-source remote sensing image pairs I1 and I2 to be registered, Gaussian pyramids are created respectively to obtain the first Gaussian pyramid and the second Gaussian pyramid; For the first point feature and the second point feature, a partial intensity invariant descriptor method is used on the first Gaussian pyramid and the second Gaussian pyramid respectively to perform multi-scale feature point description to obtain the first point feature description parameter and the second point feature description parameter.

5. The multi-source remote sensing image registration method according to claim 1, characterized in that: The step 3) comprises: For the first point feature and the second point feature, using their description parameters, adopting the BBF matching method, taking cosine similarity as the metric, perform feature point matching to obtain the initial matching point pair; By using the consistency of the main direction and spatial distribution of the feature points, the wrong matching point pairs are removed, and the valid matching point pairs are screened out to obtain the first matching feature point set and the second matching feature point set.

6. The multi-source remote sensing image registration method according to claim 1, characterized in that: The step 4) comprises: Based on the first matching feature point set and the second matching feature point set constructed by the screened valid matching point pairs, a geometric model is selected as a transformation model between the image pairs, and the parameters of the transformation matrix corresponding to the transformation model are estimated; Based on the transformation matrix obtained by the above estimation, the image to be registered is transformed to obtain the image to be registered that is aligned with the reference image, and finally a registered image pair is obtained.

7. The multi-source remote sensing image registration method according to claim 6, characterized in that: The geometric model includes: similarity transformation, affine transformation and perspective transformation; wherein the similarity transformation is a combination of basic transformations of translation, rotation and scaling.

8. The multi-source remote sensing image registration method according to claim 7, characterized in that: The transformation model satisfies the following formula: Where H is the transformation matrix, (x2, y2) T is the coordinate of the midpoint of the image to be registered, (x1, y1) T is (x2,y2) T Coordinates of the homonymous points on the reference image; The transformation matrix H is solved using the least squares method; The transformation matrix H of similarity transformation Similarity Satisfy the following formula: Where S represents the scaling factor, θ represents the rotation angle, (t x ,t y ) represents the translation amount; The transformation matrix H of the affine transformation Affine Satisfy the following formula: Transformation matrix H for perspective transformation Projective Satisfies the following formula, Among them, a 11 , a 12 , a 13 , a 21 , a 22 , a 23 , a 31 , a 32 are the elements at the corresponding positions of the matrix respectively.

9. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 1.

10. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Images