Area response difference considered relative radiometric correction method for planar array satellite image

Abstract

The application discloses a kind of area response difference's face array satellite image relative radiation correction method considering difference. The method is first based on NSCT transformation to remove the coarse features of image, then remove the fine features of image by RCV, and then obtain the image mainly containing noise, estimate relative radiation correction coefficient by constructing prior model considering area response difference, and under the constraint of corresponding cloud mask, other images obtained by the same sensor are carried out relative radiation correction, so as to eliminate the imaging quality degradation of face array satellite image caused by inconsistent response of probe element, and improve the radiation quality of face array satellite image. By removing coarse features and fine features, the application can effectively reduce the number of face array satellite images required for estimating relative radiation correction coefficient.

Description

Technical Field

[0001] This invention belongs to the field of relative radiometric correction technology for remote sensing images, and specifically relates to a method for relative radiometric correction of area array satellite images that takes into account regional response differences. Background Technology

[0002] During the imaging process, the non-uniform radiometric distortion caused by inconsistencies in detector response in space, instability in time, and circuit noise not only degrades the visual quality of area array satellite images but also affects image analysis and subsequent processing. Researching relative radiometric correction for area array satellites to eliminate the image quality degradation caused by non-uniform radiometric distortion is of great significance for improving the radiometric quality of area array satellite images.

[0003] For relative radiometric correction of area array satellite imagery, most current correction methods are based on laboratory radiometric calibration, using different radiance values ​​generated by the integrating sphere to establish the response relationship between the grayscale values ​​of the area array satellite imagery and the irradiance at the integrating sphere exit. Common methods include single-point correction, two-point correction, and multi-point correction. Since single-point correction algorithms only compensate for the offset of each pixel, the correction effect deteriorates when the radiance of the target deviates from the calibration point. The effect of two-point correction is limited by the dynamic range; when there is a large inconsistency in the response between pixels, the correction will be biased. Multi-point correction algorithms are essentially an extension of two-point correction, using multi-point correction to approximate the nonlinear response characteristic curve, achieving higher accuracy, but are more complex and difficult to implement. To address the above problems, some have proposed relative radiometric correction methods using multiple images. This method removes image features by superimposing a large number of images, while retaining and enhancing system noise as a common feature of the images. However, this method requires a large number of input images, and the coarse features, fine features, and random noise of the images gradually disappear during the superposition of a large number of images. The methods described above all require the input images to be cloudless or have very little cloud cover. This is because cloud areas have more continuous grayscale values, less noticeable texture changes, and a higher original signal-to-noise ratio. The significant difference in image response between cloudless and cloudless areas can affect the final relative radiometric correction. In area array satellite imagery, non-uniform radiometric distortion exists as point noise across multiple images. All images acquired by the same sensor should have the same point noise; this point noise is system noise. Therefore, the area array relative radiometric correction problem can be transformed into removing system noise from multiple images. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for relative radiometric correction of area array satellite imagery that takes into account regional response differences. This method effectively reduces the number of area array satellite images required to estimate the relative radiometric correction coefficient by removing both coarse and fine features. Furthermore, by considering the impact of regional response differences, it overcomes the limitation that the area array satellite images used for estimating the relative radiometric correction coefficient and for performing relative radiometric correction must be cloud-free.

[0005] To achieve the above objectives, the technical solution provided by this invention is a relative radiometric correction method for area array satellite imagery that takes into account regional response differences, comprising the following steps:

[0006] Step 1: Remove coarse features from the image using Non-Subsampled Profilometry (NSCT).

[0007] Step 1.1: Use NSCT transform to decompose the original image into one low-frequency sub-band and multiple high-frequency sub-bands;

[0008] Step 1.2: Update the low-frequency subband coefficients using bilateral filtering to obtain coarse features in the low-frequency subband. At the same time, update the low-frequency subband coefficients through layered thresholding. By judging whether the high-frequency subband coefficients correspond to the edges and contours of the image, the edge information and image contour information are treated as "noise" and filtered out, while retaining the coarse features in the high-frequency subband.

[0009] Step 1.3: Perform NSCT inverse transform based on the updated low-frequency subband coefficients and high-frequency subband coefficients to reconstruct the coarse features of the original image and remove the coarse features of the image.

[0010] Step 2: Remove fine image features based on the regional covariance matrix;

[0011] Step 2.1: Perform visual feature encoding on the image after removing coarse features;

[0012] Step 2.2: Use the regional covariance matrix combined with visual feature encoding to measure the similarity between the center pixels of the image blocks;

[0013] Step 2.3: Extract and remove fine features to obtain an image containing only system noise and random noise;

[0014] Step 3: Construct and use a prior model that takes into account regional response differences to obtain relative radiometric correction coefficients, and perform relative radiometric correction processing on other images acquired by the same sensor;

[0015] Step 3.1: Construct a priori model that takes into account regional response differences;

[0016] Step 3.2: Optimize the prior model that takes into account regional response differences and eliminate random noise;

[0017] Step 3.3: Perform relative radiometric correction on other images acquired by the same sensor based on the estimated relative radiometric correction coefficients.

[0018] Furthermore, after decomposing the original image using NSCT transform in step 1.1, a low-frequency sub-band and multiple high-frequency sub-bands are obtained. Coarse features are mainly present in the low-frequency sub-band, while image edges, contour information, and noise are contained in the high-frequency sub-band. An image can be considered to consist of two parts: feature information and noise. Feature information is divided into coarse features and fine features, while noise is divided into random noise and systematic noise. Considering that systematic noise is multiplicative noise, the following formula represents the composition of an image:

[0019] Y x,y =C x,y ×F x,y ×K x,y +N x,y (1)

[0020] In the formula, Y x,y C represents the grayscale value of the original image in (x,y). x,y F is the gray value of the coarse feature image at (x,y). x,y N represents the gray value of the fine feature image at (x,y). x,y It is random noise, K x,y This represents the gray value of the system noise image at (x,y). The system noise is applied to the noise-free image using a multiplicative rule.

[0021] Furthermore, the image with coarse features removed in step 1.3 can be represented as follows:

[0022]

[0023] In the formula, T x,y Y represents the gray value at (x,y) of the image after removing coarse features. x,y C represents the grayscale value of the original image in (x,y). x,y F is the gray value of the coarse feature image at (x,y). x,y N represents the gray value of the fine feature image at (x,y). x,y It is random noise, K x,y This represents the grayscale value of the system noise image at (x,y).

[0024] Furthermore, in step 2.1, the image after removing coarse features is segmented into segments of the same size that overlap with each other. The image blocks are then processed, and visual features are encoded for each image block using the following encoding formula:

[0025]

[0026] In the formula, The dimension centered at position (x,y) is Visual features of image blocks V X,Y Each row in the image represents a pixel within the image block, and each column corresponds to seven visual features of the pixel. T X,Y This represents the vectorized intensity value of the image patch at (X,Y). T is calculated using the discrete filter [-1 0 1]. X,Y The first derivative in the horizontal and vertical directions Let T be calculated using the discrete filter [-1 2 -1]. X,Y The second derivatives in the horizontal and vertical directions, where X and Y represent the coordinates of pixels within the image block, respectively.

[0027] Furthermore, the similarity between the center pixels of the image blocks in step 2.2 is calculated as follows:

[0028]

[0029] In the formula, d x,y,i,j This represents the similarity between two pixels at positions (x,y) and (i,j). The dimension centered at position (x,y) is The average visual feature of all pixels within an image block. The dimension centered at position (i,j) is The average visual features of all pixels within the image patch, where R represents the covariance descriptor extracted from the region covariance matrix, calculated as follows:

[0030]

[0031]

[0032]

[0033] In the formula, V X,Y V I,J The dimensions centered at positions (x, y) and (i, j) are respectively... Visual features of image blocks, μ X,Y μ I,J Representing image block V X,Y V I,J The average of the visual features of all pixels within the range.

[0034] Furthermore, the specific calculation method for the fine feature image in step 2.3 is as follows:

[0035]

[0036] In the formula, F x,y T represents the gray value of the fine feature image at (x,y). i,j This represents the gray value at (i,j) of the image after removing coarse features. Pixel (i,j) is within a neighborhood centered at pixel (x,y) with a radius of r and a weight w. x,y,i,j The distance between two pixels (x, y) and (i, j) is defined as follows:

[0037]

[0038] In the formula, d x,y,i,j σ represents the similarity between two pixels at positions (x,y) and (i,j), and σ represents the smoothing parameter.

[0039] Noisy images lacking both coarse and fine features can be calculated using the following formula:

[0040]

[0041] In the formula, Φ x,y T represents an image containing both systematic and random noise. x,y Y represents the gray value at (x,y) of the image after removing coarse features. x,y C represents the grayscale value of the original image in (x,y). x,y F is the gray value of the coarse feature image at (x,y). x,y N represents the gray value of the fine feature image at (x,y). x,y It is random noise, K x,y This represents the grayscale value of the system noise image at (x,y).

[0042] Furthermore, in step 3.1, the impact of outliers generated in the cloud region on the estimation of the relative radiative correction coefficient is considered, and a priori model that takes into account regional response differences is constructed as follows:

[0043]

[0044] In the formula, And ||·||1 represent the square of the L2 norm and the L1 norm of the vector, respectively. This represents the relative radiometric correction factor, whose value is the reciprocal of the system noise, and N represents the image size. It is a diagonal matrix, where M represents the number of images used to calculate the relative radiometric correction coefficients, and λ represents the regularization parameter. This indicates that preliminary corrections are being made to the original image without removing coarse and fine features.

[0045] in, and The definition is as follows:

[0046]

[0047]

[0048] In the formula, This represents the first-order gradient of the initially corrected image. express At position x, the pixel value express The pixel value at position x+1, where x represents the pixel position.

[0049] coefficient and The definition is as follows:

[0050]

[0051] In the formula, It is an identity matrix, and N represents the size of the image.

[0052]

[0053] In the formula, Representing noisy images The reciprocal of the symbol, where H and W represent the height and width of the image, respectively, and N represents the image size. M vectors obtained from the same sensor Combination.

[0054]

[0055] In the formula, Composed of M diagonal matrices Combining elements, where N represents the image size and Y represents the image size. k The definition is as follows:

[0056] Y k =diag(y k (17)

[0057] In the formula, This represents the original image acquired by the same sensor, where N represents the size of the image.

[0058] Furthermore, in step 3.2, auxiliary variables v, z, and s are introduced, and the following definitions are made: The prior model that takes into account regional response differences can then be transformed into the following form:

[0059]

[0060] Minimizing equation (18) can be further transformed into the following iterative steps:

[0061]

[0062]

[0063]

[0064]

[0065]

[0066] In the formula, t represents the number of iterations, a and b represent augmented Lagrange multipliers, and α and β represent augmented Lagrange coefficients.

[0067] Furthermore, in step 3.3, after obtaining the relative radiometric correction coefficient κ, the original image y is corrected using the following formula to obtain a preliminary relative radiometric corrected image. Then, the grayscale values ​​of the cloud region are replaced with their uncorrected original pixel values ​​to obtain the final relative radiometric correction result.

[0068]

[0069] In the formula, the symbol · represents the dot product, and κ represents the relative radiation correction coefficient.

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

[0071] 1) This invention removes coarse and fine features before performing relative radiometric correction, which can effectively reduce the number of array satellite images required to estimate the relative radiometric correction coefficient;

[0072] 2) It takes into account the impact of regional response differences, breaking the limitation that the area array satellite imagery used to estimate the relative radiometric correction coefficient and to perform relative radiometric correction must be cloud-free. Attached Figure Description

[0073] Figure 1 This is a technical flowchart of an embodiment of the present invention. Detailed Implementation

[0074] This invention provides a method for relative radiometric correction of area array satellite images that takes into account regional response differences. The technical solution of this invention will be further described below with reference to the accompanying drawings and embodiments.

[0075] like Figure 1 As shown, the present invention provides a method for relative radiometric correction of area array satellite imagery that takes into account regional response differences, comprising the following steps:

[0076] Step 1: Remove coarse features from the image using Non-Subsampled Contourlet Transform (NSCT).

[0077] Step 1.1: Use NSCT transform to decompose the original image into a low-frequency sub-band and multiple high-frequency sub-bands.

[0078] NSCT is an excellent tool for multi-scale decomposition of 2D images. It uses a non-subsampled pyramid decomposition filter and a non-subsampled directional filter bank to decompose 2D images, thereby obtaining sub-band images at different scales and directions. After decomposing the original image through NSCT transformation, a low-frequency sub-band and multiple high-frequency sub-bands are obtained. Coarse features are mainly present in the low-frequency sub-band, while image edges, contour information, and noise are contained in the high-frequency sub-bands.

[0079] An image can be considered to consist of two parts: feature information and noise. Feature information can be divided into coarse features and fine features. Coarse features are mainly determined by imaging conditions and ground cover types, and belong to low-frequency information. Fine features mainly come from edge details of ground covers and some gray-level abrupt changes, and belong to high-frequency information. Noise is divided into random noise and systematic noise. Considering that systematic noise is multiplicative noise, the following formula can be used to represent the composition of an image:

[0080] Y x,y =C x,y ×F x,y ×K x,y +N x,y (1)

[0081] In the formula, Y x,y C represents the grayscale value of the original image in (x,y). x,y F is the gray value of the coarse feature image at (x,y). x,y N represents the gray value of the fine feature image at (x,y). x,y It is random noise, K x,y This represents the gray value of the system noise image at (x,y). The system noise is applied to the noise-free image using a multiplicative rule.

[0082] Step 1.2: Update the low-frequency subband coefficients using bilateral filtering to obtain coarse features in the low-frequency subband. At the same time, update the low-frequency subband coefficients through layered thresholding. By judging whether the high-frequency subband coefficients correspond to the edges and contours of the image, high-frequency information such as edge information and image contour information are treated as "noise" and filtered out, while retaining the coarse features in the high-frequency subband.

[0083] Step 1.3: Perform NSCT inverse transform based on the updated low-frequency subband coefficients and high-frequency subband coefficients to reconstruct the coarse features of the original image and remove the coarse features of the image.

[0084] An image with coarse features removed can be represented as:

[0085]

[0086] In the formula, T x,y Y represents the gray value at (x,y) of the image after removing coarse features. x,y C represents the grayscale value of the original image in (x,y). x,y F is the gray value of the coarse feature image at (x,y). x,y N represents the gray value of the fine feature image at (x,y). x,y It is random noise, K x,y This represents the grayscale value of the system noise image at (x,y).

[0087] Step 2: Remove fine features from the image based on the regional covariance matrix.

[0088] After removing coarse features, most of the low-frequency information in the image (features determined by imaging conditions and the ground features themselves) has disappeared, while noise and fine features (including edges and contours) are preserved and enhanced. Therefore, by removing the fine features of an image, an image mainly composed of noise can be obtained.

[0089] Step 2.1: Perform visual feature encoding on the image after removing coarse features.

[0090] The image, after removing coarse features, is segmented into segments of equal size with some overlap. The image blocks are then processed, and visual features are encoded for each image block using the following encoding formula:

[0091]

[0092] In the formula, The dimension centered at position (x,y) is Visual features of image blocks V X,Y Each row in the image represents a pixel within the image block, and each column corresponds to seven visual features of the pixel. T X,Y This represents the vectorized intensity value of the image patch at (X,Y). T is calculated using the discrete filter [-1 0 1]. X,Y The first derivative in the horizontal and vertical directions Let T be calculated using the discrete filter [-1 2 -1]. X,YThe second derivatives in the horizontal and vertical directions, where X and Y represent the coordinates of pixels within the image block, respectively.

[0093] Step 2.2: Use the regional covariance matrix combined with visual feature encoding to measure the similarity between the center pixels of the image blocks.

[0094] The visual feature encoding obtained in step 2.1, as an image block descriptor, can implicitly capture local structure. Furthermore, this descriptor is unaffected by noise and is robust to noise. The similarity between the center pixels of the image block is calculated as follows:

[0095]

[0096] In the formula, d x,y,i,j This represents the similarity between two pixels at positions (x,y) and (i,j). The dimension centered at position (x,y) is The average visual feature of all pixels within an image block. The dimension centered at position (i,j) is The average visual features of all pixels within the image patch, where R represents the covariance descriptor extracted from the region covariance matrix, calculated as follows:

[0097]

[0098]

[0099]

[0100] In the formula, V X,Y V I,J The dimensions centered at positions (x, y) and (i, j) are respectively... Visual features of image blocks, μ X,Y μ I,J Representing image block V X,Y V I,J The average of the visual features of all pixels within the range.

[0101] Step 2.3: Extract and remove fine features to obtain an image containing only systematic noise and random noise.

[0102] The specific calculation method is as follows:

[0103]

[0104] In the formula, F x,y T represents the gray value of the fine feature image at (x,y). i,jThis represents the gray value at (i,j) of the image after removing coarse features. Pixel (i,j) is within a neighborhood centered at pixel (x,y) with a radius of r and a weight w. x,y,i,j The distance between two pixels (x, y) and (i, j) is defined as follows:

[0105]

[0106] In the formula, d x,y,i,j σ represents the similarity between two pixels at positions (x,y) and (i,j), and σ represents the smoothing parameter.

[0107] Noisy images lacking both coarse and fine features can be calculated using the following formula:

[0108]

[0109] In the formula, Φ x,y T represents an image containing both systematic and random noise. x,y Y represents the gray value at (x,y) of the image after removing coarse features. x,y C represents the grayscale value of the original image in (x,y). x,y F is the gray value of the coarse feature image at (x,y). x,y N represents the gray value of the fine feature image at (x,y). x,y It is random noise, K x,y This represents the grayscale value of the system noise image at (x,y).

[0110] Step 3: Construct and use a prior model that takes into account regional response differences to obtain relative radiometric correction coefficients, and perform relative radiometric correction processing on other images acquired by the same sensor.

[0111] Step 3.1: Construct a priori model that takes into account regional response differences.

[0112] Considering the impact of outliers generated in cloud regions on the estimation of relative radiative correction coefficients, a priori model that takes into account regional response differences is constructed as follows:

[0113]

[0114] In the formula, And ||·||1 represent the square of the L2 norm and the L1 norm of the vector, respectively. This represents the relative radiometric correction factor, which is also the reciprocal of the system noise. N represents the image size. It is a diagonal matrix, where M represents the number of images used to calculate the relative radiometric correction coefficients, and λ represents the regularization parameter. This indicates that preliminary corrections are being made to the original image without removing coarse and fine features.

[0115] in, and The definition is as follows:

[0116]

[0117]

[0118] In the formula, This represents the first-order gradient of the initially corrected image. express At position x, the pixel value express The pixel value at position x+1, where x represents the pixel position.

[0119] coefficient and The definition is as follows:

[0120]

[0121] In the formula, It is an identity matrix, and N represents the size of the image.

[0122]

[0123] In the formula, Representing noisy images The reciprocal of the symbol, where H and W represent the height and width of the image, respectively, and N represents the image size. M vectors obtained from the same sensor Combination.

[0124]

[0125] In the formula, Composed of M diagonal matrices Combining elements, where N represents the image size and Y represents the image size. k The definition is as follows:

[0126] Y k =diag(y k (17)

[0127] In the formula, This represents the original image acquired by the same sensor, where N represents the size of the image.

[0128] Step 3.2: Optimize the prior model that takes into account regional response differences and eliminate random noise.

[0129] The ADMM algorithm provides a framework for solving optimization problems with linear equality constraints. It allows the original optimization problem to be decomposed into multiple relatively simple sub-optimization problems for iterative solving. This embodiment uses the ADMM algorithm to accelerate the optimization of a priori models that consider regional response differences, enabling multiple images to obtain optimal radiometric correction results and eliminating random noise during the optimization process. This is achieved by introducing auxiliary variables v, z, and s, and defining... The prior model that takes into account regional response differences can then be transformed into the following form:

[0130]

[0131] Minimizing equation (18) can be further transformed into the following iterative steps:

[0132]

[0133]

[0134]

[0135]

[0136]

[0137] In the formula, t represents the number of iterations, a and b represent augmented Lagrange multipliers, and α and β represent augmented Lagrange coefficients.

[0138] Step 3.3: Perform relative radiometric correction on other images acquired by the same sensor based on the estimated relative radiometric correction coefficients.

[0139] System noise is mainly caused by the non-uniform response of the CCD detector. As satellite equipment ages, the response of the CCD detector will also change. Therefore, it is best to update the correction coefficients regularly and then use the updated correction coefficients to perform relative radiometric correction on images acquired by the same sensor with similar imaging times.

[0140] After obtaining the relative radiometric correction coefficient κ, the original image y can be corrected using the following formula to obtain a preliminary relative radiometric corrected image. Then, the grayscale values ​​of the cloud region are replaced with their uncorrected original pixel values ​​to obtain the final relative radiometric correction result.

[0141]

[0142] In the formula, the symbol · represents the dot product, and κ represents the relative radiation correction coefficient.

[0143] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to replace them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for relative radiometric correction of area array satellite imagery that takes into account regional response differences, characterized in that, Includes the following steps: Step 1: Remove coarse features from the image using Non-Subsampled Profilometry (NSCT). Step 1.1: Use NSCT transform to decompose the original image into one low-frequency sub-band and multiple high-frequency sub-bands; After decomposing the original image using NSCT transform, a low-frequency sub-band and multiple high-frequency sub-bands are obtained. Coarse features mainly exist in the low-frequency sub-band, while image edges, contour information, and noise are contained in the high-frequency sub-band. The image can be considered to consist of two parts: feature information and noise. The feature information is divided into coarse features and fine features, and the noise is divided into random noise and systematic noise. Considering that systematic noise is a multiplicative noise, the following formula is used to represent the composition of an image: （1） In the formula, Indicates the original image in grayscale value, It is a coarse feature image in grayscale value at that location Indicating fine feature images in grayscale value at that location It is random noise. Indicates system noise image in The grayscale value at the location is determined by the multiplicative rule applied to the noise-free image. Step 1.2: Update the low-frequency subband coefficients using bilateral filtering to obtain coarse features in the low-frequency subband. At the same time, update the low-frequency subband coefficients through layered thresholding. By judging whether the high-frequency subband coefficients correspond to the edges and contours of the image, the edge information and image contour information are treated as "noise" and filtered out, while retaining the coarse features in the high-frequency subband. Step 1.3: Perform NSCT inverse transform based on the updated low-frequency subband coefficients and high-frequency subband coefficients to reconstruct the coarse features of the original image and remove the coarse features of the image. Step 2: Remove fine image features based on the regional covariance matrix; Step 2.1: Perform visual feature encoding on the image after removing coarse features; Step 2.2: Use the regional covariance matrix combined with visual feature encoding to measure the similarity between the center pixels of the image blocks; Step 2.3: Extract and remove fine features to obtain an image containing only system noise and random noise; Step 3: Construct and use a prior model that takes into account regional response differences to obtain relative radiometric correction coefficients, and perform relative radiometric correction processing on other images acquired by the same sensor; Step 3.1: Construct a priori model that takes into account regional response differences; Considering the impact of outliers generated in cloud regions on the estimation of relative radiative correction coefficients, a priori model that takes into account regional response differences is constructed as follows: （11） In the formula, and Let L2 norm and L1 norm of the vector be the square of the L2 norm and respectively. This represents the relative radiation correction factor, whose value is the reciprocal of the system noise. Indicates the size of the image. It is a diagonal matrix. This indicates the number of images used to calculate the relative radiometric correction factor. Represents the regularization parameter. This indicates a preliminary correction of the original image without removing coarse and fine features, with coefficients... , The definition is as follows: In the formula, It is the identity matrix. Indicates the size of the image; In the formula, Representing noisy images The reciprocal, and These represent the height and width of the image, respectively. Indicates the size of the image. Obtained from the same sensor vectors Combination; Step 3.2: Optimize the prior model that takes into account regional response differences and eliminate random noise; Step 3.3: Perform relative radiometric correction on other images acquired by the same sensor based on the estimated relative radiometric correction coefficients.

2. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 1, characterized in that: The image after removing coarse features in step 1.3 can be represented as follows; （2） In the formula, This indicates the image with coarse features removed. grayscale value at that location Indicates the original image in grayscale value, It is a coarse feature image in grayscale value at that location Indicating fine feature images in grayscale value at that location It is random noise. Indicates system noise image in The grayscale value at that location.

3. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 1, characterized in that: In step 2.1, the image after removing coarse features is segmented into segments of the same size that overlap with each other. The image blocks are then processed, and visual features are encoded for each image block using the following encoding formula: （3） In the formula, Indicated by position The size of the center is Visual features of image blocks , , Each row in the image represents a pixel within the image block, and each column corresponds to seven visual features of the pixel. This represents the vectorized intensity value of the image patch at (X,Y). and These are respectively through discrete filters Calculated The first derivative in the horizontal and vertical directions and These respectively represent the use of discrete filters Calculated The second derivatives in the horizontal and vertical directions, and These represent the coordinates of pixels within the image block.

4. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 3, characterized in that: The similarity between the center pixels of the image blocks in step 2.2 is calculated as follows: （4） In the formula, Indicates position and The similarity between two pixels, Indicated by position The size of the center is The average visual feature of all pixels within an image block. Indicated by position The size of the center is The average visual feature of all pixels within an image block. This represents the covariance descriptor extracted from the regional covariance matrix, and its specific calculation method is as follows: （5） （6） （7） In the formula, , They are based on position , The size of the center is Visual features of image blocks , Representing image blocks , The average of the visual features of all pixels within the range.

5. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 2, characterized in that: The specific calculation method for the fine feature image in step 2.3 is as follows: （8） In the formula, Indicating fine feature images in grayscale value at that location This indicates the image with coarse features removed. grayscale value at that location, pixel In pixels Centered on, with radius r Within the neighborhood of, weight Represents two pixels and The distance between them is defined as follows: （9） In the formula, Indicates position and The similarity between two pixels, This represents the smoothing parameter.

6. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 5, characterized in that: The noisy image lacking both coarse and fine features in step 2.3 can be calculated using the following formula: （10） In the formula, This indicates an image containing both systematic and random noise. This indicates the image with coarse features removed. grayscale value at that location Indicates the original image in grayscale value, It is a coarse feature image in grayscale value at that location Indicating fine feature images in grayscale value at that location It is random noise. Indicates system noise image in The grayscale value at that location.

7. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 1, characterized in that: In step 3.1 and The definition is as follows: （12） （13） In the formula, This represents the first-order gradient of the initially corrected image. express In position pixel values, express In position pixel values, x Indicates pixel position; In the formula, Depend on diagonal matrices Combination Indicates the size of the image. The definition is as follows: In the formula, This refers to the original images acquired by the same sensor. Indicates the size of the image.

8. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 7, characterized in that: In step 3.2, auxiliary variables are introduced. , z , and define , Then, the prior model that takes into account regional response differences can be transformed into the following form: （18） Minimizing equation (18) can be further transformed into the following iterative steps: （19） （20） （21） （22） （23） In the formula, Indicates the number of iterations. and This indicates augmented Lagrange multipliers. and This represents the augmented Lagrange coefficient.

9. The relative radiometric correction method for area array satellite imagery considering regional response differences as described in claim 8, characterized in that: In step 3.3, the relative radiation correction coefficient is obtained. Then, the original image is corrected using the following formula. To obtain preliminary relative radiometrically corrected images Then, the gray values ​​of the cloud region are replaced with their uncorrected original pixel values ​​to obtain the final relative radiometric correction result; （24） In the formula, the symbol Dot product, This represents the relative radiation correction factor.

Images