An adaptive space-spectral registration and fusion method for large scene satellite remote sensing images

Abstract

An adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery is disclosed. This invention belongs to the field of remote sensing image processing and geographic information technology, specifically relating to an adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery. The purpose of this invention is to solve the problems of low operating efficiency and insufficient adaptability to complex scenes in existing spatial-spectral feature fusion methods, leading to spectral curve shifts after fusion. The process is as follows: 1. Construct a pseudo-panchromatic reference representation based on hyperspectral and panchromatic images; 2. Perform registration using feature point matching at multiple scales; 3. Construct simulated panchromatic components; 4. Obtain the fused large-scene image.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing and geographic information technology, specifically relating to an adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing images. Background Technology

[0002] Hyperspectral remote sensing imagery can provide high-dimensional, continuous spectral information across the visible, near-infrared, and even wider spectral ranges, offering advantages such as finely distinguishing ground material properties and improving interpretation accuracy. However, hyperspectral sensors are typically limited by energy and imaging technology, resulting in relatively low spatial resolution. In contrast, panchromatic imagery offers higher spatial resolution and clearer geometric details, but its spectral dimensions are limited. To simultaneously achieve high spatial and hyperspectral resolution, engineering practice generally requires registering hyperspectral and panchromatic imagery and performing panchromatic sharpening / fusion to generate a fused image that combines spatial detail and spectral information.

[0003] Existing panchromatic sharpening and multi-source fusion methods mainly include component substitution-based methods such as IHS, PCA, Brovey, Gram-Schmidt and their adaptive forms (e.g., GSA), MRA-type methods based on multi-resolution analysis, and recent matrix factorization / low-rank representation and tensor representation methods (e.g., nonnegative matrix factorization, sparse representation, low-rank constraints, coupled matrix / tensor factorization) and deep learning methods (e.g., convolutional networks, generative adversarial networks, Transformers, etc.). Among these, component substitution and MRA methods often rely on simulating panchromatic component construction and detail injection, which can easily lead to spectral distortion, brightness shift, and edge halos when the structure is inconsistent, and are also susceptible to registration errors. Sensitive to imagery, prone to ghosting and misalignment artifacts; matrix / tensor decomposition methods typically require iterative optimization, resulting in high computational complexity and parameter sensitivity, leading to long processing times and high deployment costs in large-scale imagery; while deep learning methods may achieve good visual results on specific data, they generally suffer from limitations such as training data and cost requirements, unstable generalization across sensors / regions, and strong dependence on computing power and operating environment; therefore, there is an urgent need for an integrated registration and fusion technology solution for large-scale remote sensing imagery that, without requiring training and possessing strong generalization capabilities, offers higher operating efficiency than matrix / tensor decomposition methods and outperforms conventional component substitution methods in terms of visual effects and spectral fidelity.

[0004] Furthermore, hyperspectral images and panchromatic images are heterogeneous imaging sources, and they differ in spectral response, contrast distribution, and texture representation. Traditional feature point matching methods based on gray intensity are prone to problems such as unstable matching, increased mismatches, and non-robust affine model estimation in cross-modal scenarios. Under the conditions of wide-swath, high-resolution images (such as tens of thousands of pixels), registration and fusion also face engineering challenges such as large computational load, high memory consumption, and difficulty in balancing processing efficiency.

[0005] Therefore, there is an urgent need for an integrated registration and fusion method that can improve the robustness of cross-modal registration by utilizing structural domain information for large-scale remote sensing data and achieve adaptive spatial-spectral information synergistic enhancement in the fusion stage, so as to improve the spatial detail and spectral fidelity of fused images. Summary of the Invention

[0006] The purpose of this invention is to address the problems of low efficiency and insufficient adaptability to complex scenes in existing spatial-spectral feature fusion methods, which lead to spectral curve shift after fusion. Therefore, this invention proposes an adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing images.

[0007] The specific process of an adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery is as follows:

[0008] Step 1: Construct a pseudo-panchromatic reference representation based on hyperspectral and panchromatic images;

[0009] Step 2: Process the panchromatic image obtained in Step 1. The image is obtained by sequential downsampling to obtain the panchromatic image after the first downsampling. The panchromatic image after the second downsampling The panchromatic image after the third downsampling ;

[0010] The pseudo-panchromatic reference characterization obtained in step 1 The pseudo-panchromatic reference representation is obtained by performing downsampling sequentially after the first downsampling. The pseudo-panchromatic reference representation after the second downsampling The pseudo-panchromatic reference characterization after the third downsampling ;

[0011] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0012] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0013] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0014] Based on scale Affine model of time Compared with the hyperspectral image obtained in step 1 Obtain the registered upsampled hyperspectral image ;

[0015] Step 3: Upsampled hyperspectral image based on the registration from Step 2 Obtain a pseudo-panchromatic image; based on the registered upsampled hyperspectral image from step 2. Obtain intermediate variables Intermediate variables obtained based on pseudo-panchromatic images Based on intermediate variables and The adaptive weights are solved using sampling least squares; the optimal adaptive weights are then determined. Perform nonnegation and normalization to obtain the nonnegation and normalized weights. Based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component;

[0016] Step 4: Register the upsampled hyperspectral image obtained in Step 2. Divided into piece, ;

[0017] The panchromatic image obtained in step 1 Divided into corresponding piece, ;

[0018] The simulated panchromatic components obtained in step 3 Divided into corresponding piece, ;

[0019] Calculate panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained;

[0020] The results of the fusion of each segment are stitched together to obtain a large-scene fused image.

[0021] The beneficial effects of this invention are as follows:

[0022] The purpose of this invention is to overcome the problems of low efficiency and insufficient adaptability to complex scenes in existing spatial-spectral feature fusion methods for large-scene images. It proposes a multi-scale structural domain alignment and adaptive spatial-spectral feature fusion registration and fusion method and plugin for large-scene remote sensing images. This method can efficiently fuse hyperspectral and panchromatic remote sensing images to generate high spectral and high spatial resolution images. The fusion parameters are estimated through a multivariate linear model to improve the spectral fidelity and robustness before and after fusion.

[0023] This method addresses the problems of low versatility, low operating efficiency, and heavy reliance on absolute geometric registration priors in spatial-spectral remote sensing image fusion. It proposes a registration method based on feature point extraction and a fusion method that optimizes the adaptive weight solution of the GSA algorithm by using sampling to solve the least squares estimator. While ensuring the timeliness of the fusion process, it improves the spectral and spatial fidelity before and after fusion compared with the traditional component replacement method.

[0024] To verify the performance of the algorithm proposed in this invention, experiments were conducted on the fusion of multispectral and panchromatic image data from the Gaofen-2 satellite. The experimental results demonstrate the effectiveness of the multi-scale structural domain alignment and adaptive spatial-spectral feature fusion registration and fusion method for large-scene remote sensing images proposed in this invention. Attached Figure Description

[0025] Figure 1 This is a schematic diagram illustrating the implementation process of the present invention;

[0026] Figure 2The images are the original multispectral image, the original panchromatic image, and the fused image of the present invention (512×512 for detail). (a) Original multispectral image of the whole image, (b) Original panchromatic image, (c) Fused image of the present invention (512×512 for detail).

[0027] Figure 3 The images are the original multispectral image, the original panchromatic image, and the fused image of the present invention (512×512 for detail). (a) Original multispectral image of the whole image, (b) Original panchromatic image, (c) Fused image of the present invention (512×512 for detail).

[0028] Figure 4 The images are the original multispectral image, the original panchromatic image, and the image fused by the present invention (512×512 for detail). (a) Original multispectral image of the whole image, (b) Original panchromatic image, (c) Image fused by the present invention (512×512 for detail). Detailed Implementation

[0029] Specific Implementation Method 1: The specific process of this implementation method for adaptive spatial-spectral registration and fusion of large-scene satellite remote sensing images is as follows:

[0030] Step 1: Construct a pseudo-panchromatic reference representation based on hyperspectral and panchromatic images;

[0031] Step 2: Process the panchromatic image obtained in Step 1. The image is obtained by sequential downsampling to obtain the panchromatic image after the first downsampling. The panchromatic image after the second downsampling The panchromatic image after the third downsampling ;

[0032] The pseudo-panchromatic reference characterization obtained in step 1 The pseudo-panchromatic reference representation is obtained by performing downsampling sequentially after the first downsampling. The pseudo-panchromatic reference representation after the second downsampling The pseudo-panchromatic reference characterization after the third downsampling ;

[0033] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0034] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0035] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0036] Based on scale Affine model of time Compared with the hyperspectral image obtained in step 1 Obtain the registered upsampled hyperspectral image ;

[0037] Step 3: Upsampled hyperspectral image based on the registration from Step 2 Obtain a pseudo-panchromatic image; based on the registered upsampled hyperspectral image from step 2. Obtain intermediate variables Intermediate variables obtained based on pseudo-panchromatic images Based on intermediate variables and The adaptive weights are solved using sampling least squares; the optimal adaptive weights are then determined. Perform nonnegation and normalization to obtain the nonnegation and normalized weights. Based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component;

[0038] Step 4: Register the upsampled hyperspectral image obtained in Step 2. Divided into piece, ;

[0039] The panchromatic image obtained in step 1 Divided into corresponding piece, ;

[0040] The simulated panchromatic components obtained in step 3 Divided into corresponding piece, ;

[0041] Calculate panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained;

[0042] The results of the fusion of each segment are stitched together to obtain a large-scene fused image.

[0043] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that: in step 1, a pseudo-panchromatic reference representation is constructed based on hyperspectral images and panchromatic images;

[0044] The specific process is as follows:

[0045] Step 11: Acquire hyperspectral and panchromatic images The hyperspectral image is upsampled to the size of the panchromatic image to obtain the hyperspectral image after upsampling to the size of the panchromatic image. With panchromatic image ;

[0046] in, , Represents the set of real numbers. Indicates hyperspectral image of high, Indicates hyperspectral image width, Indicates hyperspectral image The total number of bands;

[0047]

[0048] in, Indicates hyperspectral image The first band of hyperspectral image; Indicates hyperspectral image The second band of hyperspectral image; Indicates hyperspectral image The Middle Hyperspectral images in each band; Indicates hyperspectral image The Middle Hyperspectral images in each band; ;

[0049] Among them, panchromatic images , Represents the set of real numbers. Represents panchromatic image of high, Represents panchromatic image The width; , ;

[0050] Step 12: Vectorize the hyperspectral image of each band (from 3D to 2D) and arrange them into a dimension-reduced matrix according to the spectral dimensions. ; indicates as:

[0051]

[0052] Vectorization is the process of vectorizing the image of each band (channel) of a hyperspectral image. This involves splitting the image of a single band into columns, stacking all columns sequentially into a column vector, and then concatenating the column vectors of each band in ascending order of band size into a two-dimensional matrix.

[0053] Suppose there is matrix ; ;

[0054] The process of expanding by column (vec operation) is as follows:

[0055] 1. Retrieve column 1: 2. Retrieve the second column: 3. Extract the 3rd column: ;

[0056] 4. Stack them in order: ;

[0057] in,

[0058] Indicates taking The vector, ;

[0059] Indicates taking The vector, ;

[0060] Indicates taking The vector, ;

[0061] Indicates taking The vector, ;

[0062] Step 13: Convert the panchromatic image Vectorization (from two dimensions to one dimension) yields the dimensionality-reduced matrix. ; indicates as:

[0063]

[0064] in, Indicates taking a panchromatic image ;

[0065] Step 14: Based on the dimensionality-reduced matrix and the reduced-dimensional matrix The optimal composite weights are obtained through regression fitting; expressed as:

[0066]

[0067] in, Indicates the optimal composite weight; Indicates the composite weights; Represents the square of the 2-norm; Indicates parameters;

[0068] Step 15: Nonnegate and normalize the optimal composite weights obtained in Step 14 to obtain the nonnegated and normalized optimal composite weights; expressed as:

[0069]

[0070] in,

[0071] Represents the matrix after dimensionality reduction The Middle Vector of hyperspectral image in each band The corresponding optimal synthesis weights;

[0072] Represents the matrix after dimensionality reduction The Middle Vector of hyperspectral image in each band The corresponding optimal synthesis weights;

[0073] Represents the matrix after dimensionality reduction The Middle Vector of hyperspectral image in each band The corresponding optimal composite weights after nonnegation and normalization;

[0074] Step 16: Based on the dimensionality-reduced matrix The Middle Vector of hyperspectral image in each band The corresponding optimal composite weights after nonnegation and normalization and the hyperspectral image from step 11 The Middle Hyperspectral images in each band This yields the pseudo-panchromatic reference characterization (Pseudo-PAN), represented as:

[0075] .

[0076] Pseudo-panchromatic reference characterization It is used for feature point matching, surface matching, and optional ECC subpixel optimization in subsequent registration to improve the stability and robustness of cross-modal registration.

[0077] The other steps and parameters are the same as in Specific Implementation Method 1.

[0078] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that: in step 2, the panchromatic image obtained in step 1... The image is obtained by sequential downsampling to obtain the panchromatic image after the first downsampling. The panchromatic image after the second downsampling The panchromatic image after the third downsampling ;

[0079] The pseudo-panchromatic reference characterization obtained in step 1 The pseudo-panchromatic reference representation is obtained by performing downsampling sequentially after the first downsampling. The pseudo-panchromatic reference representation after the second downsampling The pseudo-panchromatic reference characterization after the third downsampling ;

[0080] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0081] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0082] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ;

[0083] Based on scale Affine model of time Compared with the hyperspectral image obtained in step 1 Obtain the registered upsampled hyperspectral image ;

[0084] The specific process is as follows:

[0085] Step 21: Set the scale , , Indicates the maximum scale value. ;

[0086] Step 22

[0087] The panchromatic image acquired in step 11 Downsampling is performed to obtain the panchromatic image after the first downsampling. ;

[0088] For the panchromatic image after the first downsampling Downsampling is performed to obtain the panchromatic image after the second downsampling. ;

[0089] For the panchromatic image after the second downsampling Downsampling was performed to obtain the panchromatic image after the third downsampling. ;

[0090] Step 23

[0091] The pseudo-panchromatic reference characterization obtained in step 16 Downsampling is performed to obtain the pseudo-panchromatic reference representation after the first downsampling. ;

[0092] Pseudo-panchromatic reference characterization after the first downsampling Downsampling is performed to obtain the pseudo-panchromatic reference representation after the second downsampling. ;

[0093] Pseudo-panchromatic reference characterization after the second downsampling Downsampling is performed to obtain the pseudo-panchromatic reference representation after the third downsampling. ;

[0094] Step 24: Use SIFT to... Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time ;

[0095]

[0096] in,

[0097] Indicates to Extract feature points, and then select the first feature point from the feature matching results. Indicates to Extract feature points, and then select the second feature point from the feature matching results. Indicates to Extract feature points, and perform feature matching on the extracted feature points. One feature point; Indicates to Extract feature points, and perform feature matching on the extracted feature points. One feature point;

[0098] Indicates to Extract feature points, and then select the first feature point from the feature matching results. Indicates to Extract feature points, and then select the second feature point from the feature matching results. Indicates to Extract feature points, and perform feature matching on the extracted feature points. One feature point; Showing Extract feature points, and perform feature matching on the extracted feature points. One feature point;

[0099] Step 25: Set the distance threshold;

[0100] Step 26: Calculate the set of matching point pairs Each pair Euclidean distance, select the set of matching point pairs All Euclidean distances less than the distance threshold The matching point pairs constitute the scale Set of interior points of time ,scale Set of interior points of time Includes right ,scale Set of interior points of time ;

[0101] Based on interior point set Solution scale Global radiometric transformation matrix at time ;

[0102] ,

[0103] in, Representing scale The global radiometric transformation matrix at that time; Represents the global radiometric transformation matrix; Represents the square of the 2-norm;

[0104] Step 27, Scale-based Global radiometric transformation matrix at time Characterization of pseudo-panchromatic reference Perform an affine transformation to obtain the pseudo-panchromatic reference representation after the affine transformation. ;

[0105] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time ;

[0106]

[0107] in,

[0108] Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point;

[0109] Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point;

[0110] Step 28: Calculate the set of matching point pairs Each pair Euclidean distance, select the set of matching point pairs All Euclidean distances less than the distance threshold The matching point pairs constitute the scale Set of interior points of time ;scale Set of interior points of time Includes right ,scale Set of interior points of time ;

[0111] in, Describe the set of interior points The first pair in the middle; Describe the set of interior points The second pair in the middle; Describe the set of interior points The Middle right; Describe the set of interior points The Middle right;

[0112] Based on interior point set Solution scale Global radiometric transformation matrix increment model ;

[0113] ,

[0114] in, Representing scale The global radiative transformation matrix increment model at that time; Represents the square of the 2-norm;

[0115] Step 29, Scale-based Global radiometric transformation matrix increment model Characterization of pseudo-panchromatic reference Perform an affine transformation to obtain the pseudo-panchromatic reference representation after the affine transformation. ;

[0116] Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time ;

[0117]

[0118] in,

[0119] Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point;

[0120] Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point;

[0121] Step 210: Calculate the set of matching point pairs Each pair Euclidean distance, select the set of matching point pairs All Euclidean distances less than the distance threshold The matching point pairs constitute the scale Set of interior points of time ;scale Set of interior points of time Includes right ,scale Set of interior points of time ;

[0122] in, Describe the set of interior points The first pair in the middle; Describe the set of interior points The second pair in the middle; Describe the set of interior points The Middle right; Describe the set of interior points The Middle right;

[0123] Based on interior point set Solution scale Affine model of time ;

[0124] ,

[0125] in, Representing scale Affine model at time; Represents the square of the 2-norm;

[0126] Step 211: Scale obtained from step 210 Affine model of time Compared with the hyperspectral image obtained in step 11 Obtain the registered upsampled hyperspectral image ; indicates as:

[0127] .

[0128] Other steps and parameters are the same as in specific implementation method one or two.

[0129] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that: in step 26... The solution process is as follows;

[0130] The specific process is as follows:

[0131] For the set of interior points A certain person in China , There are two coordinate components , There are two coordinate components ;

[0132] in, express The x and y coordinates; express The x and y coordinates;

[0133] The coordinate system is as follows: the origin is at the top left corner of the image; the x-axis is to the right (column direction), and the y-axis is downward (row direction); the unit is "pixels";

[0134] Let the global radiometric transformation matrix be... There are 6 unknowns ;

[0135] Bundle Convert to vector , Represent real numbers;

[0136] Will x and y coordinates and x and y coordinates Write it as matrix multiplication (two lines): ;

[0137] Set of interior points Included right Stack them row by row to obtain intermediate variables. and ; indicates as:

[0138] ;

[0139]

[0140] Based on vectors intermediate variables , Build ;

[0141] Using the least squares method Solve the problem to obtain the optimal vector. , ;

[0142] in, express The generalized inverse matrix; Dot product;

[0143] Optimal Vector Given a 6x1 column vector, rearrange the first 3 elements into one row, the last 3 elements into one row, and then concatenate them vertically to obtain a... The matrix is ​​the scale. Global radiometric transformation matrix at time .

[0144] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0145] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that, in step 28... The solution process is as follows:

[0146] For the set of interior points A certain person in China , There are two coordinate components , There are two coordinate components ;

[0147] in, express The x and y coordinates; express The x and y coordinates;

[0148] The coordinate system is as follows: the origin is at the top left corner of the image; the x-axis is to the right (column direction), and the y-axis is downward (row direction); the unit is "pixels";

[0149] Suppose a global radiometric transformation matrix incremental model There are 6 unknowns ;

[0150] Bundle Convert to vector , Represent real numbers;

[0151] Will x and y coordinates and x and y coordinates Write it as matrix multiplication (two lines): ;

[0152] Set of interior points Included right Stack them row by row to obtain intermediate variables. and ; indicates as:

[0153] ;

[0154]

[0155] Based on vectors intermediate variables , Build ;

[0156] Using the least squares method Solve the problem to obtain the optimal vector. , ;

[0157] in, express The generalized inverse matrix; Dot product;

[0158] Optimal Vector Given a 6x1 column vector, rearrange the first 3 elements into one row, the last 3 elements into one row, and then concatenate them vertically to obtain a... The matrix is ​​the scale. Global radiometric transformation matrix increment model .

[0159] The other steps and parameters are the same as those in one of the specific implementation methods one to four.

[0160] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that, in step 210... The solution process is as follows:

[0161] For the set of interior points A certain person in China , There are two coordinate components , There are two coordinate components ;

[0162] in, express The x and y coordinates; express The x and y coordinates;

[0163] The coordinate system is as follows: the origin is at the top left corner of the image; the x-axis is to the right (column direction), and the y-axis is downward (row direction); the unit is "pixels";

[0164] Set up an affine model There are 6 unknowns ;

[0165] Bundle Convert to vector , Represent real numbers;

[0166] Will x and y coordinates and x and y coordinates Write it as matrix multiplication (two lines): ;

[0167] Set of interior points Included right Stack them row by row to obtain intermediate variables. and ; indicates as:

[0168] ;

[0169]

[0170] Based on vectors intermediate variables , Build ;

[0171] Using the least squares method Solve the problem to obtain the optimal vector. , ;

[0172] in, express The generalized inverse matrix; Dot product;

[0173] Optimal Vector Given a 6x1 column vector, rearrange the first 3 elements into one row, the last 3 elements into one row, and then concatenate them vertically to obtain a... The matrix is ​​the scale. Affine model of time .

[0174] The other steps and parameters are the same as those in one of the specific implementation methods one to five.

[0175] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that, in step 3, the upsampled hyperspectral image after registration in step 2 is used. Obtain a pseudo-panchromatic image; based on the registered upsampled hyperspectral image from step 2. Obtain intermediate variables Intermediate variables obtained based on pseudo-panchromatic images Based on intermediate variables and The adaptive weights are solved using sampling least squares; the optimal adaptive weights are then determined. Perform nonnegation and normalization to obtain the nonnegation and normalized weights. Based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component;

[0176] The specific process is as follows:

[0177] Step 31: Based on the registered upsampled hyperspectral image from Step 2 Obtain a pseudo-panchromatic image;

[0178] The specific process is as follows:

[0179]

[0180] in,

[0181] This represents the registered upsampled hyperspectral image obtained in step 2. Uniform sampling The set of pixels consisting of the total number of pixels;

[0182] Represents a set of pixels The Middle One pixel;

[0183] This represents the registered upsampled hyperspectral image obtained in step 2. The Middle The weight of each channel;

[0184] This represents the registered upsampled hyperspectral image obtained in step 2. The Middle The first channel The sample hyperspectral image corresponding to each pixel;

[0185] Indicates the first A pseudo-panchromatic image of one pixel;

[0186] Step 32: Upsampled hyperspectral image based on registration from Step 2 Obtain intermediate variables Based on pseudo-panchromatic images Obtain intermediate variables ;

[0187] The specific process is as follows:

[0188]

[0189] in,

[0190] That is to All elements in the first channel are concatenated into a column vector;

[0191] This represents the registered upsampled hyperspectral image obtained in step 2. The first channel of all pixels in the uniformly sampled pixel set. , This represents the registered upsampled hyperspectral image obtained in step 2. The first channel of the first pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The first channel of the second pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first channel of each pixel This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first channel of each pixel;

[0192] This represents the registered upsampled hyperspectral image obtained in step 2. The second channel of all pixels in the uniformly sampled pixel set. , This represents the registered upsampled hyperspectral image obtained in step 2. The second channel of the first pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The second channel of the second pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The second channel of each pixel. This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The second channel of each pixel;

[0193] This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, , This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The second pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel;

[0194] This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, , This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The second pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel;

[0195] Vectorization;

[0196]

[0197] in,

[0198] Indicates intermediate variables. This represents the pseudo-panchromatic image corresponding to the first pixel in the pseudo-panchromatic image. This represents the pseudo-panchromatic image corresponding to the second pixel in the pseudo-panchromatic image. Indicating the first in a pseudo-panchromatic image The pseudo-panchromatic image corresponding to each pixel. Indicating the first in a pseudo-panchromatic image The pseudo-panchromatic image corresponding to each pixel;

[0199] Step 33: Based on intermediate variables and Sampling least squares to solve for adaptive weights;

[0200] The specific process is as follows:

[0201]

[0202] in, Indicates adaptive weights; Indicates coefficient; This represents the optimal adaptive weight;

[0203] Step 34: Optimal adaptive weights Perform nonnegation and normalization to obtain the nonnegation and normalized weights. ; indicates as:

[0204]

[0205] in, Indicates the first Adaptive weights for each channel; Indicates the first Adaptive weights for each channel; Indicates the first Nonnegation and normalized weights of each channel;

[0206] Step 35, based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component.

[0207] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0208] Specific Implementation Method Eight: This implementation method differs from one of Specific Implementation Methods One to Seven in that step 35 is based on the first... Nonnegation and normalized weights of each channel Construct a simulated panchromatic component; represented as:

[0209]

[0210] in,

[0211] Indicates the registered hyperspectral image The One channel;

[0212] This represents the simulated panchromatic component.

[0213] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0214] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that, in step 4, the registered upsampled hyperspectral image obtained in step 2 is used... Divided into piece, ;

[0215] The panchromatic image obtained in step 1 Divided into corresponding piece, ;

[0216] The simulated panchromatic components obtained in step 3 Divided into corresponding piece, ;

[0217] Calculate panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained;

[0218] The fusion results of each segment are stitched together to obtain a large-scene fused image;

[0219] The specific process is as follows:

[0220] Step 41: Register the upsampled hyperspectral image obtained in Step 2. Divided into piece, ;

[0221] The panchromatic image obtained in step 1 Divided into corresponding Block (registered upsampled hyperspectral image) of The position of each block and the panchromatic image of (The positions of all blocks are the same) ;

[0222] The simulated panchromatic components obtained in step 3 Divided into corresponding piece, ;

[0223] in, This indicates the registered upsampled hyperspectral image. The first piece, This indicates the registered upsampled hyperspectral image. The second piece in the middle, This indicates the registered upsampled hyperspectral image. The Middle piece, This indicates the registered upsampled hyperspectral image. The Middle piece;

[0224] Represents panchromatic image The first piece, Represents panchromatic image The second piece in the middle, Represents panchromatic image The Middle piece, Represents panchromatic image The Middle piece;

[0225] Step 42: Calculate the panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained;

[0226] Step 43: Stitch together the fusion results of each segment to obtain a large-scene fused image.

[0227] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0228] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Methods One to Nine in that the panchromatic image is calculated in step 42. The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained;

[0229] The specific process is as follows:

[0230] Step 421: Calculate the panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. ; indicates as: ;

[0231] Step 422, Based on spatial detail items Calculation details :

[0232] in, Indicates Gaussian filtering. The parameters represent the Gaussian filter. The standard deviation is expressed as Gaussian filtering of 1 pixel;

[0233] Step 423, based on details and The fusion results of each block are obtained; represented as:

[0234]

[0235] in, This represents the registered upsampled hyperspectral image obtained in step 2. The Middle The first block Hyperspectral imagery of multiple channels;

[0236] This indicates the first step obtained in step 3. Adaptive weights for each channel.

[0237] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0238] The beneficial effects of the present invention are verified using the following embodiments:

[0239] Example 1:

[0240] The data used in the experiment were multispectral and panchromatic image data from the Gaofen-2 satellite. The resolution of the hyperspectral image was approximately 7600×6900, and the resolution of the panchromatic image was approximately 29600×27200. Figure 1 This is a schematic diagram of the implementation process of the present invention. To more intuitively demonstrate the performance of the proposed method in the panchromatic sharpening task, the present invention divides the original satellite hyperspectral and panchromatic images into small blocks of 128×128 and 512×512, and performs fusion on representative blocks. In the experiment, the Baseline (GS method) and the method of the present invention were used respectively to perform fusion in the same area, and the various fusion indicators were compared. Figure 2 The images are the original multispectral image, the original panchromatic image, and the fused image of the present invention (512×512 for detail). (a) Original multispectral image of the whole image, (b) Original panchromatic image, (c) Fused image of the present invention (512×512 for detail). Figure 3 The images are the original multispectral image, the original panchromatic image, and the fused image of the present invention (512×512 for detail). (a) Original multispectral image of the whole image, (b) Original panchromatic image, (c) Fused image of the present invention (512×512 for detail). Figure 4 The images are the original multispectral image, the original panchromatic image, and the image fused by the present invention (512×512 for detail). (a) Original multispectral image of the whole image, (b) Original panchromatic image, (c) Image fused by the present invention (512×512 for detail).

[0241] Table 1 compares the quality assessment results of the method of this invention with those of the baseline method. The comparison shows that the method of this invention can achieve high-precision fusion under complex backgrounds while ensuring timeliness.

[0242] Table 1 Quantitative Evaluation Indicators

[0243] method EGARS(↓) CC(↑) QNR(↑) SSIM(↑) Baseline method 25.4414 0.8896 0.9619 0.9889 Method of the present invention 3.1519 0.9831 0.9662 0.9953

[0244] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. An adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery, characterized in that: The specific process of the method is as follows: Step 1: Construct a pseudo-panchromatic reference representation based on hyperspectral and panchromatic images; Step 2: Process the panchromatic image obtained in Step 1. The image is obtained by sequential downsampling to obtain the panchromatic image after the first downsampling. The panchromatic image after the second downsampling The panchromatic image after the third downsampling ; The pseudo-panchromatic reference characterization obtained in step 1 The pseudo-panchromatic reference representation is obtained by performing downsampling sequentially after the first downsampling. The pseudo-panchromatic reference representation after the second downsampling The pseudo-panchromatic reference characterization after the third downsampling ; Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ; Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ; Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ; Based on scale Affine model of time Compared with the hyperspectral image obtained in step 1 Obtain the registered upsampled hyperspectral image ; Step 3: Upsampled hyperspectral image based on the registration from Step 2 Obtain a pseudo-panchromatic image; based on the registered upsampled hyperspectral image from step 2. Obtain intermediate variables Intermediate variables obtained based on pseudo-panchromatic images ; Based on intermediate variables and Sampling least squares to solve for adaptive weights; For optimal adaptive weights Perform nonnegation and normalization to obtain the nonnegation and normalized weights. Based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component; Step 4: Register the upsampled hyperspectral image obtained in Step 2. Divided into piece, ; The panchromatic image obtained in step 1 Divided into corresponding piece, ; The simulated panchromatic components obtained in step 3 Divided into corresponding piece, ; Calculate panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained; The results of the fusion of each segment are stitched together to obtain a large-scene fused image.

2. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 1, characterized in that: In step 1, a pseudo-panchromatic reference representation is constructed based on hyperspectral and panchromatic images; The specific process is as follows: Step 11: Acquire hyperspectral and panchromatic images The hyperspectral image is upsampled to the size of the panchromatic image to obtain the hyperspectral image after upsampling to the size of the panchromatic image. With panchromatic image ; in, , Represents the set of real numbers. Indicates hyperspectral image of high, Indicates hyperspectral image width, Indicates hyperspectral image The total number of bands; in, Indicates hyperspectral image The first band of hyperspectral image; Indicates hyperspectral image The second band of hyperspectral image; Indicates hyperspectral image The Middle Hyperspectral images in each band; Indicates hyperspectral image The Middle Hyperspectral images in each band; ; Among them, panchromatic images , Represents the set of real numbers. Represents panchromatic image of high, Represents panchromatic image The width; , ; Step 12: Vectorize the hyperspectral image of each band and arrange it into a dimension-reduced matrix according to the spectral dimension. ; indicates as: in, Indicates taking The vector, ; Indicates taking The vector, ; Indicates taking The vector, ; Indicates taking The vector, ; Step 13: Convert the panchromatic image Vectorization yields the dimensionality-reduced matrix. ; indicates as: in, Indicates taking a panchromatic image ; Step 14: Based on the dimensionality-reduced matrix and the reduced-dimensional matrix The optimal composite weights are obtained through regression fitting; expressed as: in, Indicates the optimal composite weight; Indicates the composite weights; Represents the square of the 2-norm; Indicates parameters; Step 15: Nonnegate and normalize the optimal composite weights obtained in Step 14 to obtain the nonnegated and normalized optimal composite weights; expressed as: in, Represents the matrix after dimensionality reduction The Middle Vector of hyperspectral image in each band The corresponding optimal synthesis weights; Represents the matrix after dimensionality reduction The Middle Vector of hyperspectral image in each band The corresponding optimal synthesis weights; Represents the matrix after dimensionality reduction The Middle Vector of hyperspectral image in each band The corresponding optimal composite weights after nonnegation and normalization; Step 16: Based on the dimensionality-reduced matrix The Middle Vector of hyperspectral image in each band The corresponding optimal composite weights after nonnegation and normalization and the hyperspectral image from step 11. The Middle Hyperspectral images in each band This yields a pseudo-panchromatic reference representation, expressed as: 。 3. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 2, characterized in that: In step 2, the panchromatic image obtained in step 1 is processed. The image is obtained by sequential downsampling to obtain the panchromatic image after the first downsampling. The panchromatic image after the second downsampling The panchromatic image after the third downsampling ; The pseudo-panchromatic reference characterization obtained in step 1 The pseudo-panchromatic reference representation is obtained by performing downsampling sequentially after the first downsampling. The pseudo-panchromatic reference representation after the second downsampling The pseudo-panchromatic reference characterization after the third downsampling ; Using the SIFT method Feature points are extracted, and eigenvalue matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ; Using the SIFT method Feature points are extracted, and eigenvalue matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ; Using the SIFT method Feature points are extracted, and eigenvalue matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time Based on the set of matching point pairs Composition Scale Set of interior points of time Based on interior point set Solution scale Affine model of time ; Based on scale Affine model of time Compared with the hyperspectral image obtained in step 1 Obtain the registered upsampled hyperspectral image ; The specific process is as follows: Step 21: Set the scale , , Indicates the maximum scale value. ; Step 22 The panchromatic image acquired in step 11 Downsampling is performed to obtain the panchromatic image after the first downsampling. ; panchromatic image after the first downsampling Downsampling is performed to obtain the panchromatic image after the second downsampling. ; For the panchromatic image after the second downsampling Downsampling was performed to obtain the panchromatic image after the third downsampling. ; Step 23 The pseudo-panchromatic reference characterization obtained in step 16 Downsampling is performed to obtain the pseudo-panchromatic reference representation after the first downsampling. ; Pseudo-panchromatic reference characterization after the first downsampling Downsampling is performed to obtain the pseudo-panchromatic reference representation after the second downsampling. ; Pseudo-panchromatic reference characterization after the second downsampling Downsampling is performed to obtain the pseudo-panchromatic reference representation after the third downsampling. ; Step 24: Use SIFT to... Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time ; in, Indicates to Extract feature points, and then select the first feature point from the feature matching results. Indicates to Extract feature points, and then select the second feature point from the feature matching results. Indicates to Extract feature points, and perform feature matching on the extracted feature points. One feature point; Indicates to Extract feature points, and perform feature matching on the extracted feature points. One feature point; Indicates to Extract feature points, and then select the first feature point from the feature matching results. Indicates to Extract feature points, and then select the second feature point from the feature matching results. Indicates to Extract feature points, and perform feature matching on the extracted feature points. One feature point; Showing Extract feature points, and perform feature matching on the extracted feature points. One feature point; Step 25: Set the distance threshold; Step 26: Calculate the set of matching point pairs Each pair Euclidean distance, select the set of matching point pairs All Euclidean distances less than the distance threshold The matching point pairs constitute the scale Set of interior points of time ,scale Set of interior points of time Includes right ,scale Set of interior points of time ; Based on interior point set Solution scale Global radiometric transformation matrix at time ; ， in, Representing scale The global radiometric transformation matrix at time; Represents the global radiometric transformation matrix; Represents the square of the 2-norm; Step 27, Scale-based Global radiometric transformation matrix at time Characterization of pseudo-panchromatic reference Perform an affine transformation to obtain the pseudo-panchromatic reference representation after the affine transformation. ; Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time ; in, Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Step 28: Calculate the set of matching point pairs Each pair Euclidean distance, select the set of matching point pairs All Euclidean distances less than the distance threshold The matching point pairs constitute the scale Set of interior points of time ;scale Set of interior points of time Includes right ,scale Set of interior points of time ; in, Represents the set of interior points The first pair in the middle; Represents the set of interior points The second pair in the middle; Represents the set of interior points The Middle right; Represents the set of interior points The Middle right; Based on interior point set Solution scale Global radiometric transformation matrix increment model ; ， in, Representing scale The global radiative transformation matrix increment model at that time; Represents the square of the 2-norm; Step 29, Scale-based Global radiometric transformation matrix increment model Characterization of pseudo-panchromatic reference Perform an affine transformation to obtain the pseudo-panchromatic reference representation after the affine transformation. ; Using the SIFT method Feature points are extracted, and feature matching is performed on the extracted feature points using the FLANN method to obtain the scale. The set of matching points at time ; in, Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Extract feature points, and use the FLANN method to find the first feature point in the feature matching result; Indicates to Extract feature points, and use the FLANN method to find the second feature point in the feature matching result; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Indicates to Feature points are extracted, and the FLANN method is used to perform feature matching on the extracted feature points. One feature point; Step 210: Calculate the set of matching point pairs Each pair Euclidean distance, select the set of matching point pairs All Euclidean distances less than the distance threshold The matching point pairs constitute the scale Set of interior points of time ;scale Set of interior points of time Includes right ,scale Set of interior points of time ; in, Represents the set of interior points The first pair in the middle; Represents the set of interior points The second pair in the middle; Represents the set of interior points The Middle right; Represents the set of interior points The Middle right; Based on interior point set Solution scale Affine model of time ; ， in, Representing scale Affine model at time; Represents the square of the 2-norm; Step 211: Scale obtained from step 210 Affine model of time Compared with the hyperspectral image obtained in step 11 Obtain the registered upsampled hyperspectral image ; indicates as: 。 4. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 3, characterized in that: In step 26 The solution process is as follows; The specific process is as follows: For the set of interior points A certain person in China , There are two coordinate components , There are two coordinate components ; in, express The x and y coordinates; express The x and y coordinates; Let the global radiometric transformation matrix be... There are 6 unknowns ; Bundle Convert to vector , Represent real numbers; Will x and y coordinates and x and y coordinates Rewritten as matrix multiplication: ; Set of interior points Included right Stack them row by row to obtain intermediate variables. and ; indicates as: ； Based on vectors intermediate variables , Build ; Using the least squares method Solve the problem to obtain the optimal vector. , ; in, express The generalized inverse matrix; Dot product; Optimal Vector Given a 6x1 column vector, transform the first 3 elements into one row, the last 3 elements into one row, and then concatenate them vertically to obtain a... The matrix, i.e., the scale Global radiometric transformation matrix at time .

5. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 4, characterized in that: In step 28 The solution process is as follows: For the set of interior points A certain person in China , There are two coordinate components , There are two coordinate components ; in, express The x and y coordinates; express The x and y coordinates; Suppose a global radiometric transformation matrix incremental model There are 6 unknowns ; Bundle Convert to vector , Represent real numbers; Will x and y coordinates and x and y coordinates Rewritten as matrix multiplication: ; Set of interior points Included right Stack them row by row to obtain intermediate variables. and ; indicates as: ； Based on vectors intermediate variables , Build ; Using the least squares method Solve the problem to obtain the optimal vector. , ; in, express The generalized inverse matrix; Dot product; Optimal Vector Given a 6x1 column vector, rearrange the first 3 elements into one row, the last 3 elements into one row, and then concatenate them vertically to obtain a... The matrix is ​​the scale. Global radiometric transformation matrix increment model .

6. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 5, characterized in that: In step 210 The solution process is as follows: For the set of interior points A certain person in China , There are two coordinate components , There are two coordinate components ; in, express The x and y coordinates; express The x and y coordinates; Set up an affine model There are 6 unknowns ; Bundle Convert to vector , Represent real numbers; Will x and y coordinates and x and y coordinates Rewritten as matrix multiplication: ; Set of interior points Included right Stack them row by row to obtain intermediate variables. and ; indicates as: ； Based on vectors intermediate variables , Build ; Using the least squares method Solve the problem to obtain the optimal vector. , ; in, express The generalized inverse matrix; Dot product; Optimal Vector Given a 6x1 column vector, rearrange the first 3 elements into one row, the last 3 elements into one row, and then concatenate them vertically to obtain a... The matrix is ​​the scale. Affine model of time .

7. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 6, characterized in that: In step 3, the upsampled hyperspectral image is based on the registration obtained in step 2. Obtain a pseudo-panchromatic image; based on the registered upsampled hyperspectral image from step 2. Obtain intermediate variables Intermediate variables obtained based on pseudo-panchromatic images ; Based on intermediate variables and Sampling least squares to solve for adaptive weights; For optimal adaptive weights Perform nonnegation and normalization to obtain the nonnegation and normalized weights. Based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component; The specific process is as follows: Step 31: Based on the registered upsampled hyperspectral image from Step 2 Obtain a pseudo-panchromatic image; The specific process is as follows: in, This represents the registered upsampled hyperspectral image obtained in step 2. Uniform sampling The set of pixels consisting of the total number of pixels; Represents a set of pixels The Middle One pixel; This represents the registered upsampled hyperspectral image obtained in step 2. The Middle The weight of each channel; This represents the registered upsampled hyperspectral image obtained in step 2. The Middle The first channel The sample hyperspectral image corresponding to each pixel; Indicates the first A pseudo-panchromatic image of one pixel; Step 32: Upsampled hyperspectral image based on registration from Step 2 Obtain intermediate variables Intermediate variables obtained based on pseudo-panchromatic images ; The specific process is as follows: in, This represents the registered upsampled hyperspectral image obtained in step 2. The first channel of all pixels in the uniformly sampled pixel set. , This represents the registered upsampled hyperspectral image obtained in step 2. The first channel of the first pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The first channel of the second pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first channel of each pixel This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first channel of each pixel; This represents the registered upsampled hyperspectral image obtained in step 2. The second channel of all pixels in the uniformly sampled pixel set. , This represents the registered upsampled hyperspectral image obtained in step 2. The second channel of the first pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The second channel of the second pixel in the uniformly sampled pixel set. This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The second channel of each pixel. This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The second channel of each pixel; This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, , This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The second pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel; This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, , This represents the registered upsampled hyperspectral image obtained in step 2. The first pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The second pixel in the uniformly sampled pixel set One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel, This represents the registered upsampled hyperspectral image obtained in step 2. The first uniformly sampled pixel set in the middle The first pixel One channel; Vectorization; in, Indicates intermediate variables. This represents the pseudo-panchromatic image corresponding to the first pixel in the pseudo-panchromatic image. This represents the pseudo-panchromatic image corresponding to the second pixel in the pseudo-panchromatic image. Indicating the first in a pseudo-panchromatic image The pseudo-panchromatic image corresponding to each pixel. Indicating the first in a pseudo-panchromatic image The pseudo-panchromatic image corresponding to each pixel; Step 33: Based on intermediate variables and Sampling least squares to solve for adaptive weights; The specific process is as follows: in, Indicates adaptive weights; Indicates coefficient; This represents the optimal adaptive weight; Step 34: Optimal adaptive weights Perform nonnegation and normalization to obtain the nonnegation and normalized weights. ; indicates as: in, Indicates the first Adaptive weights for each channel; Indicates the first Adaptive weights for each channel; Indicates the first Nonnegation and normalized weights of each channel; Step 35, based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component.

8. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 7, characterized in that: In step 35, based on the first Nonnegation and normalized weights of each channel Construct a simulated panchromatic component; represented as: in, Indicates the registered hyperspectral image The One channel; This represents the simulated panchromatic component.

9. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 8, characterized in that: In step 4, the registered upsampled hyperspectral image obtained in step 2 is used. Divided into piece, ; The panchromatic image obtained in step 1 Divided into corresponding piece, ; The simulated panchromatic components obtained in step 3 Divided into corresponding piece, ; Calculate panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained; The fusion results of each segment are stitched together to obtain a large-scene fused image; The specific process is as follows: Step 41: Register the upsampled hyperspectral image obtained in Step 2. Divided into piece, ; The panchromatic image obtained in step 1 Divided into corresponding piece, ; The simulated panchromatic components obtained in step 3 Divided into corresponding piece, ; in, This indicates the registered upsampled hyperspectral image. The first piece, This indicates the registered upsampled hyperspectral image. The second piece in the middle, This indicates the registered upsampled hyperspectral image. The Middle piece, This indicates the registered upsampled hyperspectral image. The Middle piece; Represents panchromatic image The first piece, Represents panchromatic image The second piece in the middle, Represents panchromatic image The Middle piece, Represents panchromatic image The Middle piece; Represents the simulated panchromatic component The first piece, Represents the simulated panchromatic component The second piece in the middle, Represents the simulated panchromatic component The Middle piece, Represents the simulated panchromatic component The Middle piece; Step 42: Calculate the panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained; Step 43: Stitch together the fusion results of each segment to obtain a large-scene fused image.

10. The adaptive spatial-spectral registration and fusion method for large-scene satellite remote sensing imagery according to claim 9, characterized in that: In step 42, the panchromatic image is calculated. The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. Based on spatial detail items Calculation details Based on details and The fusion results of each block are obtained; The specific process is as follows: Step 421: Calculate the panchromatic image The Middle piece and simulated panchromatic components The Middle piece The difference yields the spatial detail item. ; indicates as: ; Step 422, Based on spatial detail items Calculation details : in, Indicates Gaussian filtering. The parameters represent the Gaussian filter. The standard deviation is expressed as Gaussian filtering of 1 pixel; Step 423, based on details and The fusion results of each block are obtained; represented as: in, This represents the registered upsampled hyperspectral image obtained in step 2. The Middle The first block Hyperspectral imagery of multiple channels; This indicates the first step obtained in step 3. Adaptive weights for each channel.

Images