Global and local integrated variational radiometric registration method and system for optical remote sensing images

Abstract

The present invention proposes a global and local integrated variational radiation registration method and system for optical remote sensing images, including: for an input image set consisting of multiple images, based on the geographic information of each image, generating the minimum bounding rectangle of the image set and performing grid division to obtain valid image blocks and overlapping image block pairs with overlapping relationships and their radiation information; based on the radiation difference relationship of the overlapping image block pairs after registration, the radiation difference relationship of the image blocks before and after registration, and based on variational theory and norm regularization theory, constructing a variational radiation registration model that takes into account global-local integration and performing iterative solution to obtain the registration coefficient of each image block; performing weighted linear interpolation on the image block registration coefficient, solving the radiation registration coefficient of each pixel point, and performing linear registration on each pixel point. The present invention corrects and eliminates the global and local radiation differences of the image set, and can effectively retain the image gradient and spatial texture information.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image processing, and relates to a global and local integrated variational radiation registration method and system for optical remote sensing images. Background Art

[0002] Mosaic images covering a wide range of interest areas generated based on multiple remote sensing images are gradually playing an increasingly important role in multiple application fields such as geographic mapping, resource and environmental monitoring, and disaster monitoring. However, due to the influence of factors such as the solar incidence angle, atmosphere, lighting conditions, and acquisition time during the imaging process of satellite remote sensing images, there are often global and local radiation differences between multiple images, which are manifested as distinct brightness, discontinuity, and unnatural transitions in the images, making the generation of seamless remote sensing data a huge challenge. Therefore, in order to avoid obvious overall radiation inconsistency, uneven local radiation transitions, and visible seams in the final mosaic image, radiation registration must be performed before multiple images are stitched together to eliminate the overall and local radiation differences between images.

[0003] The radiation registration methods for multiple remote sensing images can be divided into three categories according to the radiation difference modeling: radiation registration methods based on global models, local models and combined models. The radiation registration method based on the global model expresses the radiation mapping relationship between images as a global linear or nonlinear model, solves the radiation registration coefficient of each image through the path transfer method or the global optimization method, and registers each image. The radiation registration method based on the global model can effectively register the radiation differences of images with simple scenes, low resolution, and small imaging differences, but it is often difficult to overcome the local radiation differences caused by complex scenes, low overlap area ratio, and large imaging differences, such as residual local ground object radiation discontinuity and edge artifacts. Therefore, the radiation registration method based on the local model came into being. Compared with the radiation registration method based on the global model, this type of method generally adopts the idea of ​​regionalization or classification, establishes a local model based on the local area / category information of the image, and obtains the registration parameters of each area / category to achieve local radiation difference registration. The radiation registration method based on the local model can often obtain very good local radiation difference registration results, but the core of this method to eliminate local radiation differences is to effectively segment or classify the image. If the segmentation / classification results do not match the local features of the image, new radiation differences may be introduced, resulting in obvious radiation differences and transition discontinuity between different regions / categories. The radiation registration methods based on the global model and the local model have their own advantages and can complement each other. Therefore, some scholars have proposed radiation registration methods based on combined models. This type of method can generally be divided into two categories. One is the cascaded combined radiation registration model, that is, the registration is performed step by step, generally first global registration, and then local registration; the other is the integrated combined radiation registration model, which simultaneously registers global and local radiation differences in the same model. The cascaded combined model registration method can achieve better overall radiation consistency and natural local radiation transition, but it is still difficult to completely eliminate local radiation differences. For the integrated combined registration model, the result of this method can obtain a better local radiation smooth transition, but it is insufficient for overall radiation consistency registration, and there is still room for improvement in gradient preservation, and the computational efficiency is very low. Compared with the radiometric registration methods based on global and local models, the radiometric registration methods based on combined models can often achieve registration results with overall radiometric consistency and local transition continuity, which is an important development direction for radiometric difference registration in the future. However, the existing methods still cannot achieve satisfactory results in terms of overall radiometric consistency, local radiometric transition, gradient preservation, and processing efficiency.

[0004] In summary, the existing radiation registration methods based on global models, local models and combined models are still difficult to efficiently perform global and local radiation difference registration of multiple images while effectively maintaining image texture details and spatial gradients when facing cross-sensor multi-temporal medium- and high-resolution remote sensing images due to their large sensor differences, large imaging conditions differences, and rich texture details. In this way, it is difficult to obtain radiation registration results with better overall radiation consistency and smooth local radiation transition. Summary of the invention

[0005] The object of the present invention is to provide an integrated variational radiometric registration method for optical remote sensing images taking into account global-local differences in view of the above-mentioned deficiencies in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides a global and local integrated variational radiometric registration method for optical remote sensing images, comprising the following steps:

[0008] Step 1: For an input image set consisting of multiple images, a minimum bounding rectangle of the image set is generated based on the geographic information of each image, and the minimum bounding rectangle is gridded to obtain valid image blocks and overlapping image block pairs with overlapping relationships, and the radiation information and attribute information of all valid image blocks and overlapping image block pairs are obtained;

[0009] Step 2: Based on the radiation information and attribute information, the radiation difference relationship of the overlapped image blocks after registration and the radiation difference relationship of the image blocks before and after registration are obtained, and based on the variational theory and norm regularization theory, a variational radiation registration model that takes into account global-local integration is constructed;

[0010] Step 3: Iteratively solve the global-local integrated variational radiation registration model to obtain the registration coefficient of each valid image block;

[0011] Step 4: Perform weighted linear interpolation on the registration coefficients of the valid image blocks, solve the radiation registration coefficient of each pixel, and perform linear registration on each pixel based on the radiation registration coefficient of each pixel.

[0012] Furthermore, the step 1 includes the following sub-steps:

[0013] Step 1.1: Calculate the minimum bounding rectangle of the image set based on the geographic coordinates and coverage of each image, grid the minimum bounding rectangle, and generate the position and geometric information of all grids based on the geographic coordinates and grid size;

[0014] Step 1.2 traverses each grid and determines the image blocks it contains. If there are pixels with non-zero values ​​in the image block, it is considered to be a valid image block and is cumulatively numbered, otherwise it is not numbered; for valid image blocks falling in the same grid, overlapping image block pairs are formed; after traversing all grids, the radiation information of all valid image blocks and overlapping image block pairs is generated, and then the pixel mean and standard deviation of each valid image block are calculated.

[0015] Furthermore, the attribute information of the valid image blocks and overlapping image block pairs includes the relative coordinates and size of each valid image block in its original image, and the numbers of the valid image blocks adjacent to each valid image block.

[0016] Furthermore, the step 2 includes the following sub-steps:

[0017] Step 2.1: Characterize the radiation difference relationship of the overlapping image blocks with overlapping relationship after registration;

[0018] Step 2.2: Characterize the radiation difference relationship between the image blocks before and after registration;

[0019] Step 2.3: Based on variational theory and norm regularization theory, a variational radiation registration model that takes into account global-local integration is constructed.

[0020] Furthermore, the radiation difference relationship of the overlapping image blocks after registration in step 2.1 is for:

[0021]

[0022] in:

[0023]

[0024]

[0025]

[0026] is the registration coefficient matrix of overlapping image blocks, The dimension is , The dimension is , The dimension is , and Respectively For overlapping image blocks, i.e. valid image blocks and The difference between the mean and standard deviation of the effective image blocks after radiometric registration, is the number of valid image blocks, is the number of overlapping image blocks, Respectively The average value of the original pixels of valid image blocks, Respectively The standard deviation of the original pixels of the valid image blocks, Respectively The multiplicative registration coefficients and additive registration coefficients of valid image blocks are: Respectively The multiplicative registration coefficients and additive registration coefficients of valid image blocks.

[0027] Furthermore, the radiation difference relationship between the image blocks before and after registration in step 2.2 is for:

[0028]

[0029] in:

[0030]

[0031]

[0032]

[0033] is the registration coefficient matrix of the image blocks before and after registration, The dimension is , The dimension is , The dimension is , The dimension is Column vector of They are respectively before and after registration The mean and standard deviation of the radiometric differences of pixels in the valid image blocks.

[0034] Furthermore, in step 2.3, based on variational theory and norm regularization theory, adaptive Norm and Norm constraint, the constructed variational radiation registration model taking into account global-local integration is:

[0035]

[0036] in, is the regularization parameter, express Norm Second power, express Norm Second power, A variational radiation registration model that takes global-local integration into consideration is constructed.

[0037] Furthermore, in step 3, the iterative reweighted least squares method IRLS is solved. In each round of iteration, the problem to be solved is transformed into a weighted least squares optimization problem, and The valid image blocks correspond to Group radiometric registration coefficients.

[0038] Furthermore, in step 4, nine groups of registration coefficients of the valid image block where each pixel is located and its eight neighborhood valid image blocks are used for weighted linear interpolation, and the inverse of the distance from the pixel point to the center point of the nine valid image blocks is used as the weight of the linear interpolation to solve the radiation registration coefficient of each pixel point:

[0039]

[0040] Pixels The multiplicative registration coefficient and additive registration coefficient of ; Represents pixel Valid image block and the registration coefficients of its eight adjacent blocks, Represents pixel The reciprocal of the distance to the center point of the nine valid image blocks is used as the weight of the coefficient and is expressed as:

[0041]

[0042] Get each pixel The multiplicative and additive registration coefficients of Then, linear registration is performed on each pixel.

[0043] In another aspect, the present invention provides a global and local integrated variational radiometric registration system for optical remote sensing images, comprising:

[0044] Module 1: It is used to generate the minimum bounding rectangle of the image set based on the geographic information of each image for an input image set consisting of multiple images, grid the minimum bounding rectangle to obtain valid image blocks and overlapping image block pairs with overlapping relationships, and obtain the radiation information of all valid image blocks and overlapping image block pairs;

[0045] Module 2: It is used to construct a variational radiation registration model that takes into account global-local integration based on the radiation difference relationship of the overlapping image blocks after registration, the radiation difference relationship of the image blocks before and after registration, and the variational theory and norm regularization theory;

[0046] Module 3: It is used to iteratively solve the global-local integrated variational radiation registration model using iterative reweighted least squares method to obtain the registration coefficient of each valid image block;

[0047] Module 4: It is used to perform weighted linear interpolation on the registration coefficients of the valid image blocks, solve the radiation registration coefficient of each pixel, and perform linear registration on each pixel based on the radiation registration coefficient of each pixel.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) By adopting the form of fast image segmentation, constraint models are established based on the radiation difference relationship of overlapping image blocks after registration and the radiation difference relationship of image blocks before and after registration. Compared with the traditional model, it can take into account both global and local radiation differences.

[0050] (2) An adaptive norm constraint is imposed on the two radiation difference terms to make them more consistent with the radiation difference distribution of the remote sensing image, thereby constructing an integrated variational radiation registration model. Compared with the traditional model, it can better register and eliminate the global and local radiation differences of the image, and can effectively retain the image gradient and spatial texture information.

[0051] (3) The iterative solution of the optimization problem based on the iterative reweighted least squares (IRLS) method can converge quickly, improving the shortcomings of the traditional method that is computationally complex and time-consuming. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 is a flow chart of an embodiment of the present invention;

[0054] Figure 2 Schematic diagram of an original remote sensing image with serious radiation difference according to an embodiment of the present invention; wherein (a) is an overall overview of multiple original remote sensing images, (b) is a partial enlarged image of the yellow box in (a), and (c) is a partial enlarged image of the green box in (a);

[0055] Figure 3 This is a schematic diagram of image set segmentation and numbering according to an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of weighted linear interpolation parameter fitting according to an embodiment of the present invention;

[0057] Figure 5 Schematic diagram of the radiometric registration result of an image set according to an embodiment of the present invention, wherein (e) is an overview of the overall registration result of multiple images, (f) is a partial enlarged image of the yellow box in (e), and (g) is a partial enlarged image of the green box in (e). DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings 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 are within the scope of protection of the present invention.

[0059] Example 1

[0060] Remote sensing images acquired from different satellite sensor platforms and at different times often have strong radiation differences, which will greatly reduce the quality and availability of the mosaic results of multiple images. In order to avoid the overall radiation inconsistency, uneven local radiation transition and visible seams in the final mosaic image, it is necessary to perform radiation registration of multiple images before mosaicking. The radiation registration method of multiple remote sensing images can simultaneously eliminate the global and local radiation differences of multiple remote sensing images and better preserve the image gradients, improve the overall radiation consistency and local radiation transition smoothness of the image, and provide high-quality and reliable data resources for subsequent remote sensing applications.

[0061] Please see Figure 1 The present invention provides a global and local integrated variational radiometric registration method for optical remote sensing images, comprising the following steps:

[0062] Step 1: For an input image set consisting of multiple images, a minimum bounding rectangle of the image set is generated based on the geographic information of each image, the minimum bounding rectangle is gridded to obtain valid image blocks and overlapping image block pairs with overlapping relationships, and the radiation information of all valid image blocks and overlapping image block pairs is obtained;

[0063] The implementation method of image rectangular block and radiation information statistics in step 1 is:

[0064] Step 1.1 generates the minimum bounding rectangle and geographic grid information of an image set composed of multiple images as follows: the minimum bounding rectangle of the image set can be calculated based on the geographic coordinates and coverage of each image. Based on the geographic coordinates and grid size, the location and geometric information of all grids can be generated. The grid size is generally a geographic range of 6 kilometers by 6 kilometers, which can be converted into a grid size in pixel units according to the spatial resolution of the image. For the case image set, the upper left corner coordinates of its bounding rectangle are calculated to be (143985, 4743015), and the lower right corner coordinates are (756615, 4029285).

[0065] Figure 2 Taking the original remote sensing images of a certain area with serious radiation differences as an example (hereinafter referred to as the case image set), it can be seen from the figure that the image set has serious inconsistent radiation tones as a whole, and there are obvious radiation differences and visible seams locally.

[0066] Step 1.2 Obtain valid image blocks and overlapping image block pairs, and calculate the radiation mean and standard deviation of each image block: traverse each grid, determine the image blocks it contains, and cumulatively number all image blocks starting from 1. If there are pixels with non-zero values ​​in the image block, it is considered to be a valid image block and numbered, otherwise it is not numbered. For valid image blocks that fall on the same grid, valid overlapping image block pairs are formed. After traversing all grids, the attribute information of all valid image blocks and overlapping image block pairs can be generated, including the relative coordinates and size of each image block in its original image, and the image block number adjacent to each image block. Then calculate the pixel mean and standard deviation of each image block. Image set According to step 1, it can be divided into image blocks, among which the image blocks with overlapping relationship are Yes. Image blocks and numbers are as follows Figure 3 shown.

[0067] For the case image set, rectangular blocks of 6 km by 6 km are used, resulting in 3518 valid image blocks and 1550 pairs of overlapping image blocks.

[0068] Step 2: Based on the radiation difference relationship model of the overlapped image blocks after registration, the radiation difference relationship model of the image blocks before and after registration, and based on the variational theory and norm regularization theory, a variational radiation registration model that takes into account global-local integration is constructed;

[0069] Step 2.1 The implementation method of modeling the radiation difference relationship of the image blocks with overlapping relationship after registration is as follows: For image radiation registration, there is a linear relationship between the pixels of the images before and after registration, which can be expressed as:

[0070]

[0071] In formula (1) For this The radiometric registration result of pixels is: For the The original pixel value of pixels, For the The multiplicative registration coefficient of pixels is For the Additive registration coefficient of pixels.

[0072] Assume that the radiometric registration coefficient of any image block is and , the mean value of the image block after registration is , the standard deviation is , we can get:

[0073]

[0074] From formula (2), we can see that the mean and standard deviation of the registered image can be expressed by the mean, standard deviation and registration coefficient of the image before registration.

[0075] The radiometric difference relationship of the image blocks with overlapping relationship after radiometric registration is constructed as follows: image blocks and The image blocks have overlapping relationship, which is recorded as For overlapping image blocks. The original pixel means of the image blocks are , and the standard deviations are , and their radiation registration coefficients are and , the mean values ​​after registration are , and the standard deviations are , then from formula (2), we can get the mean and standard deviation of the pixels before and after registration have the following relationship:

[0076]

[0077] Therefore, the For overlapping image blocks, i.e. image blocks and The difference in the results after radiometric registration of the image blocks can be expressed as:

[0078]

[0079] For each pair of overlapping image blocks, equation (4) can be listed, and we can get:

[0080]

[0081] Formula (5) can be simplified as:

[0082]

[0083] In formula (6):

[0084]

[0085]

[0086]

[0087] The coefficient matrix in formula (6) The dimension is , for the case image set, its dimension is , in the matrix In For overlapping image blocks, Row, in In the row, The column values ​​are , No. The column values ​​are ; in the In the row, The column value is , No. The column value is , and the rest of the elements are 0. The dimension is For the case image set, its dimension is , represents the difference between the mean and standard deviation of the pixels of the overlapping image blocks after radiometric registration, where For overlapping image blocks (i.e. image blocks and image blocks), corresponding to the The values ​​of the rows are . The dimension is For the case image set, its dimension is , represents the registration coefficient of each image block to be required, where The image block corresponds to The values ​​of the rows are .

[0088] Step 2.2 The implementation method of modeling the radiation difference relationship between the image blocks before and after registration is: From formula (3), we know that The mean and standard deviation of the pixels of the image block after the image block is registered can be expressed by the mean and standard of the pixels before the registration. Therefore, we can get The radiometric difference of the pixels in an image block is:

[0089]

[0090] For each image block, equation (10) can be listed, so we can get:

[0091]

[0092] Formula (11) can be simplified as:

[0093]

[0094] In formula (12):

[0095]

[0096]

[0097]

[0098]

[0099] The coefficient matrix in formula (12) is The dimension is , in the matrix For the case image set, its dimension is , No. The image block corresponds to Row, in In the row, The column values ​​are ; in the In the row, The column value is , and the rest of the elements are 0. The column vector is exactly the same as that in equation (8). The dimension is A column vector of The values ​​of the rows are . The dimension is For the case image set, its dimension is , represents the difference between the mean and standard deviation of the pixels of the image block before and after registration, where image blocks, corresponding to the The values ​​of the rows are .

[0100] Moreover, the implementation method of step 2.3 adaptive norm constraint and variational model construction is as follows: Based on variational theory and norm regularization theory, the radiation difference relationship of the image blocks with overlapping relationship after registration established in steps 2.1 and 2.2, that is, equation (6) and the radiation difference relationship model of the image blocks before and after registration, that is, equation (12), are combined, and adaptive constraints are applied to the two models respectively. Norm and Norm constraint, the energy function of the integrated variational radiation registration model can be expressed as:

[0101]

[0102] In formula (17), is the regularization parameter, express Norm Second power, express Norm Power.

[0103] Step 3: Use iterative reweighted least squares method to iteratively solve the global-local integrated variational radiation registration model to obtain the registration coefficient of each valid image block;

[0104] Moreover, the implementation method of model optimization and solution in step 3 is: Formula (17) is a formula that contains Norm and The unconstrained optimization problem of the norm is difficult to convert into a linear problem for solution, so it can be solved using the iterative reweighted least squares method (IRLS). Therefore, in each round of iteration, the problem to be solved is converted into a weighted least squares optimization problem.

[0105] Step 4: Perform weighted linear interpolation on the registration coefficients of the valid image blocks to solve the radiation registration coefficient of each pixel, and perform linear registration on each pixel based on the radiation registration coefficient of each pixel.

[0106] Step 4 performs weighted linear interpolation on the image block registration coefficients to solve the radiation registration coefficient of each pixel as follows: By optimizing and solving the radiation registration variational model, we can obtain The image blocks correspond to For the case image set, 7036 sets of registration coefficients can be obtained. If the registration coefficient of each image block is directly used to register all pixels in the block, block effects may occur, resulting in a significant reduction in data availability. Figure 4 As shown, use The nine groups of registration coefficients of the image block and its eight neighboring image blocks are weighted linear interpolated to calculate a group of registration coefficients for each pixel. In weighted linear interpolation, the inverse of the distance is used as the interpolation weight. The farther the pixel to be solved is from the center of a certain image block, the less influence the block has on it, and vice versa. Then the pixel The registration coefficient calculation method is shown in formula (18):

[0107]

[0108] In formula (18), Represents pixel The final radiometric registration coefficients. The dimension is A row vector representing a pixel point The The registration coefficients of an image block and its eight adjacent image blocks (a total of 9 image blocks). The dimension is A column vector representing pixel points The reciprocal of the distance to the center of the 9 image blocks is used as the weight of the registration coefficient. The registration coefficient After that, the radiometric registration can be performed on each pixel using formula (1). The radiometric registration coefficients and weights of a certain image block and its eight adjacent image blocks in the case image set are obtained through step 3, as shown in Table 1:

[0109] Table 1 Case image collection Radiometric registration coefficients and weights of an image block and its eight adjacent image blocks

[0110]

[0111] The final solution , after radiometric registration . Figure 5 The final radiation registration result of the case image set is shown. It can be seen that after the radiation registration of the present invention, the global and local radiation differences in the case image set are effectively eliminated, and the resulting images have good overall radiation consistency and natural and smooth local radiation transition.

[0112] Based on the existing multiple remote sensing image radiation registration methods, the present invention adopts the form of fast image segmentation, and establishes constraint models based on the radiation difference relationship of the image blocks with overlapping relationships after registration and the radiation difference relationship of the image blocks before and after registration. Compared with the traditional model, it can take into account the global and local radiation differences at the same time. Adaptive norm constraints are applied to the two radiation difference terms to make them more consistent with the radiation difference distribution of the remote sensing image, thereby constructing an integrated variational radiation registration model. Compared with the traditional model, it can better register and eliminate the global and local radiation differences of the image, and can effectively retain the image gradient and spatial texture information. The iterative solution of the optimization problem based on the iterative reweighted least squares method (IRLS) can converge quickly, improving the shortcomings of the traditional method of complex and time-consuming calculation, with high efficiency and easy to put into practical use.

[0113] Example 2

[0114] This embodiment provides a global and local integrated variational radiometric registration system for optical remote sensing images, including:

[0115] Module 1: For an input image set consisting of multiple images, the minimum bounding rectangle of the image set is generated based on the geographic information of each image, the minimum bounding rectangle is gridded to obtain valid image blocks and overlapping image block pairs with overlapping relationships, and the radiation information of all valid image blocks and overlapping image block pairs is obtained;

[0116] Module 2: It is used to construct a variational radiation registration model that takes into account global-local integration based on the radiation difference relationship of the overlapping image blocks after registration, the radiation difference relationship of the image blocks before and after registration, and the variational theory and norm regularization theory;

[0117] Module 3: It is used to iteratively solve the global-local integrated variational radiation registration model using iterative reweighted least squares method to obtain the registration coefficient of each valid image block;

[0118] Module 4: It is used to perform weighted linear interpolation on the registration coefficients of the valid image blocks, solve the radiation registration coefficient of each pixel, and perform linear registration on each pixel based on the radiation registration coefficient of each pixel.

[0119] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0120] It should be understood that the above description of the preferred embodiment is relatively detailed, and it cannot be considered as limiting the scope of protection of the patent of the present invention. It is not necessary and impossible to list all the embodiments here. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection of the present invention shall be based on the attached claims.

Claims

1. A global and local integrated variational radiometric registration method for optical remote sensing images, characterized in that: The following steps are involved: Step 1: For an input image set consisting of multiple images, a minimum bounding rectangle of the image set is generated based on the geographic information of each image, and the minimum bounding rectangle is gridded to obtain valid image blocks and overlapping image block pairs with overlapping relationships, and the radiation information and attribute information of all valid image blocks and overlapping image block pairs are obtained; The radiation information of the effective image block includes the pixel value of the effective image block, the mean value of the original pixels and the standard deviation of the original pixels; The radiation information of the overlapping image block pair includes the pixel value of the overlapping image block pair, the mean value of the overlapping image block pair after radiation registration, and the standard deviation after radiation registration; The attribute information of the valid image blocks and overlapping image block pairs includes the relative coordinates and size of each valid image block in its original image, and the number of the valid image blocks adjacent to each valid image block; Step 2: Based on the radiation information and attribute information, the radiation difference relationship of the overlapped image block pairs after registration and the radiation difference relationship of the image blocks before and after registration are obtained, and based on the variational theory and norm regularization theory, a variational radiation registration model that takes into account global-local integration is constructed; Step 2 includes the following sub-steps: Step 2.1: Characterize the radiation difference relationship of the overlapping image blocks with overlapping relationship after registration; Radiometric difference relationship between overlapping image blocks after registration for: in: is the registration coefficient matrix of overlapping image blocks, The dimension is , The dimension is , The dimension is , and Respectively For overlapping image blocks, that is, valid image blocks and The difference between the mean and standard deviation of the effective image blocks after radiometric registration, is the number of valid image blocks, is the number of overlapping image blocks, Respectively The average value of the original pixels of valid image blocks, Respectively The standard deviation of the original pixels of the valid image blocks, Respectively The multiplicative registration coefficients and additive registration coefficients of valid image blocks are: Respectively The multiplicative registration coefficients and additive registration coefficients of valid image blocks; Step 2.2: Characterize the radiation difference relationship between the image blocks before and after registration; The relationship between the radiometric differences of image blocks before and after registration for: in: is the registration coefficient matrix of the image blocks before and after registration, The dimension is , The dimension is , The dimension is , The dimension is Column vector of They are respectively The mean and standard deviation radiation differences of pixels in valid image blocks; Step 2.3: Based on variational theory and norm regularization theory, a variational radiation registration model that takes into account global-local integration is constructed; Apply adaptive Norm and Norm constraint, the constructed variational radiation registration model taking into account global-local integration is: in, is the regularization parameter, express Norm Second power, express Norm Second power, To construct a variational radiometric registration model that takes into account global-local integration; Step 3: Iteratively solve the global-local integrated variational radiation registration model to obtain the registration coefficient of each valid image block; Step 4: Perform weighted linear interpolation on the registration coefficients of the valid image blocks, solve the radiation registration coefficient of each pixel, and perform linear registration on each pixel based on the radiation registration coefficient of each pixel.

2. The optical remote sensing image global and local integrated variational radiometric registration method according to claim 1, characterized in that: The step 1 includes the following sub-steps: Step 1.1: Calculate the minimum bounding rectangle of the image set based on the geographic coordinates and coverage of each image, grid the minimum bounding rectangle, and generate the position and geometric information of all grids based on the geographic coordinates and grid size; Step 1.2 traverses each grid and determines the image blocks it contains. If there are pixels with non-zero values ​​in the image block, it is considered to be a valid image block and is cumulatively numbered, otherwise it is not numbered; for valid image blocks falling in the same grid, overlapping image block pairs are formed; after traversing all grids, the radiation information of all valid image blocks and overlapping image block pairs is generated, and then the pixel mean and standard deviation of each valid image block are calculated.

3. The optical remote sensing image global and local integrated variational radiometric registration method according to claim 1, characterized in that: In step 3, the iterative reweighted least squares method IRLS is solved. In each round of iteration, the problem to be solved is transformed into a weighted least squares optimization problem, and the solution is obtained. The valid image blocks correspond to Group radiometric registration coefficients.

4. The optical remote sensing image global and local integrated variational radiometric registration method according to claim 3, characterized in that: In step 4, nine groups of registration coefficients of the valid image block where each pixel is located and its eight neighborhood valid image blocks are used for weighted linear interpolation. The inverse of the distance from the pixel point to the center point of the nine valid image blocks is used as the weight of the linear interpolation to solve the radiation registration coefficient of each pixel point: Pixels The multiplicative registration coefficient and additive registration coefficient of ; Represents pixel Valid image block and the registration coefficients of its eight adjacent blocks, Represents pixel The reciprocal of the distance to the center point of the nine valid image blocks is used as the weight of the coefficient and is expressed as: Get each pixel The multiplicative and additive registration coefficients of Then, linear registration is performed on each pixel.

5. A global and local integrated variational radiometric registration system for optical remote sensing images, characterized in that: include: Module 1: It is used to generate the minimum bounding rectangle of the image set based on the geographic information of each image for an input image set consisting of multiple images, grid the minimum bounding rectangle to obtain valid image blocks and overlapping image block pairs with overlapping relationships, and obtain the radiation information of all valid image blocks and overlapping image block pairs; Module 2: It is used to construct a variational radiation registration model that takes into account global-local integration based on the radiation difference relationship of the overlapping image blocks after registration, the radiation difference relationship of the image blocks before and after registration, and the variational theory and norm regularization theory; Module 3: It is used to iteratively solve the global-local integrated variational radiation registration model using iterative reweighted least squares method to obtain the registration coefficient of each valid image block; Module 4: It is used to perform weighted linear interpolation on the registration coefficients of the valid image blocks, solve the radiation registration coefficient of each pixel point, and perform linear registration on each pixel point based on the radiation registration coefficient of each pixel point; The optical remote sensing image global and local integrated variational radiation registration system is used to execute the steps in the optical remote sensing image global and local integrated variational radiation registration method described in any one of claims 1-4.

Images