Area array image system noise removal method based on self-constraint and non-convex low-rank approximation

Abstract

The invention provides an area array image system noise removal method based on self-constraint and non-convex low-rank approximation, and the method comprises the steps: constructing a self-constraint non-convex low-rank approximation cooperation model: directly applying data fidelity constraint to a system noise item, and introducing a noise weight matrix into a target function to quantify the noise intensity spatial distribution difference; depicting non-local self-similarity of image blocks in combination with a non-convex low-rank approximation model, and embedding image regularization to constrain local structure smoothness of the image; a cloud mask matrix is introduced into a data fidelity term, and interference of a cloud coverage area on system noise estimation is shielded; performing iterative optimization on the self-constrained non-convex low-rank approximate collaborative model by adopting an alternating direction multiplier method, and solving a system noise estimated value; and performing noise removal on an input image based on the system noise estimation value, and keeping an original observation value in a cloud region to maintain image quality. According to the scheme provided by the invention, the noise modeling precision, the complex scene adaptability and the operation efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of noise removal of remote sensing image systems, and relates to a technical solution for noise removal of area array image systems based on self-constraint and non-convex low-rank approximation. Background Art

[0002] Remote sensing images are inevitably subject to noise during acquisition, transmission, and storage, severely impacting their visual quality and the accuracy of subsequent analysis and processing. To preserve the effective information in the images, noise removal is a critical step in remote sensing image processing. Image noise can generally be categorized as random noise and systematic noise. The former exhibits strong inconsistencies between images, while the latter exhibits stable distribution and intensity characteristics across images. Systematic noise primarily stems from the response non-uniformity of CCD or CMOS detectors and their associated dark current. In terms of imaging methods, stripe-like noise is common in linear push-broom imagery, while in area array satellite imagery, systematic noise manifests as point noise.

[0003] Currently, research on noise removal in area array imaging systems is relatively limited. Existing methods often rely on laboratory radiometric calibration, using an integrating sphere to provide multiple irradiance levels to establish a mapping between grayscale values and irradiance. Common correction methods include: a one-point method that corrects only for pixel bias, a two-point method constrained by dynamic range, and a multi-point method that performs fitting at multiple irradiance levels. While these methods are effective under experimental conditions, they still face challenges in accuracy and scalability when processing large-scale, high-resolution satellite imagery.

[0004] In contrast, a large number of algorithms have been developed in the field of image denoising for random noise, mainly including model-based methods and deep learning-based methods. Model-based methods rely on effective priors to alleviate the pathological nature of the denoising process. Among them, nonlocal self-similarity (NSS) priors are widely used for image restoration under Gaussian white noise backgrounds. However, system noise often exhibits spatial inhomogeneity and mixed characteristics, which exceed the modeling capabilities of such methods. Deep learning methods learn the complex mapping between noise and clean images from large-scale image pairs through end-to-end training, and have the ability to model complex noise. However, in system noise estimation, a single frame image cannot capture its cross-image consistency, so it is often necessary to rely on multi-frame image superposition to enhance the system noise characteristics.

[0005] Although the multi-frame stacking strategy helps to suppress random noise and highlight systematic noise, it fails to fully utilize the structural consistency of systematic noise in the image sequence, resulting in limited estimation accuracy. In addition, this method is highly dependent on the number of images, which increases the computational cost. It is worth noting that there are significant differences in the performance of systematic noise in cloud areas and cloud-free areas. High-reflectivity cloud areas often mask the characteristics of systematic noise and interfere with systematic noise estimation. To reduce cloud area interference, existing methods usually select images with less cloud cover as training samples. However, in practical applications, especially for wide-bandwidth satellites, obtaining a sufficient number of cloud-free images still faces great challenges. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the present invention provides a noise removal method for an area array imaging system based on self-constraint and non-convex low-rank approximation.

[0007] The technical solution of the present invention is a noise removal method for an area array imaging system based on self-constraint and non-convex low-rank approximation, which performs the following process:

[0008] A self-constrained non-convex low-rank approximate collaborative model is constructed. This involves directly imposing data fidelity constraints on the system noise term and introducing a noise weight matrix into the objective function to quantify the spatial distribution differences in noise intensity. The non-convex low-rank approximate model is combined with the non-local self-similarity of image patches and embedded with graph regularization to constrain the local structural smoothness of the image. Furthermore, a cloud mask matrix is introduced into the data fidelity term to block the interference of cloud coverage areas on the system noise estimation.

[0009] The self-constrained non-convex low-rank approximate collaborative model is iteratively optimized using an alternating direction multiplier method to solve the system noise estimation value;

[0010] The input image is denoised based on the system noise estimate, and the original observations are retained in the cloud region to maintain the image quality.

[0011] Moreover, the noise weight matrix is a diagonal matrix, and its diagonal elements are defined as the inverse of the noise standard deviation of each pixel point.

[0012] Moreover, the construction and implementation method of the non-convex low-rank approximation model includes dividing the input image into partially overlapping reference image blocks, searching for similar image blocks and constructing an image group matrix, and applying a low-rank constraint to each image group matrix to constrain structural similarity.

[0013] Furthermore, when constructing the image group matrix, similar image block search is performed only on the reference image blocks in the non-cloud area, thereby constructing a cloud-free image group.

[0014] Moreover, the cloud mask matrix is a diagonal matrix, and the diagonal elements thereof are defined as values corresponding to pixels in cloud areas being 0, and values corresponding to pixels in non-cloud areas being 1.

[0015] Moreover, when the alternating direction multiplier method is used to iteratively optimize the self-constrained non-convex low-rank approximate collaborative model, the objective function is decomposed into a system noise subproblem, a low-rank approximate image group subproblem and an auxiliary variable subproblem, which are solved separately.

[0016] Moreover, the noise removal stage adopts a sensor-specific correction strategy to apply the offline estimated system noise to all images captured by the same sensor, and realizes noise removal, where the cloud region directly retains the observation value.

[0017] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the noise removal method for the area array imaging system based on self-constraint and non-convex low-rank approximation as described above is implemented.

[0018] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for removing noise from a planar array imaging system based on self-constraint and non-convex low-rank approximation.

[0019] On the other hand, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation.

[0020] Compared with the prior art, the present invention has the following characteristics:

[0021] This invention is suitable for noise removal in array satellite imaging systems. Compared with existing methods, the present invention 1) directly constructs a self-constrained model with system noise as the optimization variable, bypassing the indirect process of "noise image generation and aggregation" in existing methods; 2) quantifies the spatial distribution of noise intensity caused by the non-uniformity of detector response through a noise weight matrix, avoiding the limitations of the global uniform noise assumption; 3) combines non-convex low-rank approximation and graph regularization to respectively characterize the non-local self-similarity and local structural smoothness of the image, enhancing the robustness of noise-signal separation; 4) introduces a cloud mask matrix to block interference from cloud-covered areas, achieving system noise estimation without relying on cloud-free data. Compared with existing methods, the present invention achieves significant improvements in noise modeling accuracy, adaptability to complex scenarios, and algorithm efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Flowchart of an embodiment of the present invention.

[0023] Figure 2Schematic diagram of visual comparison results between the image system noise removal method according to an embodiment of the present invention and the prior art. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without creative work are within the scope of protection of the present invention.

[0025] The present invention provides a method for removing noise from an array imaging system based on self-constraint and non-convex low-rank approximation. The method first constructs a data fidelity constraint with system noise as the core, abandons the existing indirect estimation framework based on "noise image generation-aggregation", and significantly improves the efficiency and accuracy of system noise estimation. Secondly, a regularization mechanism of cloud area mask coupling is designed, and the difference in the statistical characteristics of noise in cloud areas and cloud-free areas is used to separate high reflectivity interference in the objective function, eliminating the need for cloud-free training data. In addition, a non-convex low-rank approximation model is introduced to characterize the non-local self-similarity of image blocks, and a spatially adaptive weight matrix is embedded to quantify the spatial distribution difference of noise intensity. This method achieves high-precision system noise removal with low computational overhead, providing scalable technical support for satellite image quality enhancement.

[0026] See also Figure 1 In an embodiment of the present invention, a method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation is proposed, comprising the following steps:

[0027] Step 1: Build the SC-NCLR model:

[0028] The English name for self-constrained and non-convex low-rank approximation is self-constrained and non-convex low-rank approximation, or SC-NCLR for short. Therefore, the proposed self-constrained non-convex low-rank approximation collaborative model is called the SC-NCLR model.

[0029] The SC-NCLR model directly imposes data fidelity constraints on system noise, eliminating the intermediate steps of noise image generation and aggregation in existing methods, reducing computational complexity and mitigating bias accumulation. Furthermore, a cloud region mask is embedded in the data fidelity and regularization terms to distinguish the noise statistical characteristics of cloud-covered and cloud-free areas, eliminating the need for cloud-free training data. Furthermore, a non-convex low-rank approximation is used to characterize the nonlocal self-similarity of images, and a weight matrix is introduced to quantify differences in the spatial distribution of noise, thereby improving the robustness of system noise estimation. This model addresses the existing problems of strong dependence on cloud-free data, low computational efficiency, and weak ability to model spatial noise heterogeneity.

[0030] This step includes the following sub-steps:

[0031] Step 1.1: Apply data fidelity constraints to system noise

[0032] This approach employs a self-constraint mechanism to directly impose data fidelity constraints on system noise. This differs from traditional approaches that primarily model image noise residuals. This strategy refocuses on the system noise itself, improving the accuracy and efficiency of system noise estimation.

[0033] The embodiment of the present invention considers the spatial heterogeneity of system noise intensity and defines the noise weight matrix W i (i=1,...,M) is used to measure the noise standard deviation of each pixel and construct the following minimization problem:

[0034]

[0035] in, represents the square of the l2 norm, β is the regularization parameter, Φ(s) is the regularization term, s is the system noise, is the estimated value of the system noise. i (i=1,...,M) is the diagonal matrix corresponding to the i-th training image, and M is the number of training images used to remove system noise. i The determination of the diagonal elements in the image depends on the noise standard deviation of each pixel, which is defined as follows:

[0036]

[0037] in is the standard deviation of the noise for each pixel n, where N is the number of pixels in the i-th training image.

[0038] Step 1.2: Model image priors using non-convex low-rank approximation

[0039] In order to effectively characterize the non-local structural correlation of the image, the present invention models the image prior through a non-convex low-rank approximation model. The specific process is as follows:

[0040] First, input the image in is a real number domain, N is the number of pixels of the i-th training image, which is divided into partially overlapping reference image blocks j is the reference image block number, K is the number of reference image blocks in each image, and p is the number of pixels in the reference image block, with a value range of [36,81];

[0041] Then, for each reference image block In its local neighborhood, the improved K nearest neighbor algorithm is used to search for m most similar image blocks. Among them, l is the number of similar image blocks, and these similar image blocks are stacked to form an image group matrix Right now The matrix Each column Represents the reference block One of the similar image blocks, m is the number of similar image blocks, and the preferred recommended value range is [60,120].

[0042] Finally, a low-rank approximation model is constructed for each image group matrix to constrain its structural similarity. Its mathematical expression is:

[0043]

[0044] in is the image group matrix Low-rank approximation of , that is, low-rank approximate image group; is a symmetric positive semidefinite matrix. represents the Moore-Penrose pseudo-inverse, tr(·) is the trace of the matrix, represents the set of positive semidefinite matrices.

[0045] At the same time, in order to characterize the piecewise smoothness of the image, a graph regularization mechanism is introduced. The graph regularization is expressed by the following formula:

[0046]

[0047] in represents the graph Laplacian matrix, for The transpose of

[0048] Adjacency weight matrix The elements of are defined as follows:

[0049]

[0050] in express The xth column of express The yth column of Represents the adjacency weight matrix The xth row and yth column element of , e represents a natural constant, σ 2 is a hyperparameter, adjust and Sensitivity to similarity.

[0051] Angle Matrix The definition is as follows:

[0052]

[0053] The joint optimization model of non-convex low-rank approximation and graph regularization is as follows:

[0054]

[0055] Among them, α and λ represent regularization parameters, which are used to balance the non-convex low-rank approximation terms. and graph regularization term Contributions to minimization problems.

[0056] Step 1.3 introduces the cloud mask matrix to separate the noise statistical characteristics of cloud area and cloud-free area in the objective function

[0057] To solve the problem that cloud regions may significantly interfere with system noise estimation, the present invention introduces a cloud mask matrix J into the data fidelity term. i , effectively suppressing the interference of high reflectivity characteristics of cloud areas on system noise estimation. i is defined as follows:

[0058] J i is a diagonal matrix defined as follows:

[0059]

[0060] Among them, cloud area represents the cloud area, Indicates J i The nth diagonal element of , where n is the pixel number of the image.

[0061] In order to further reduce the interference of cloud area on system noise estimation, the present invention improves the image group matrix construction mechanism of KNN algorithm: only when the reference block When it is located in a non-cloudy area, a similar block search is performed in its local neighborhood to construct a cloud-free image group. This strategy significantly improves the consistency of system noise characteristics by shielding system noise outliers in cloud areas.

[0062] In summary, the SC-NCLR model proposed in the present invention is defined as:

[0063]

[0064] in Indicates that from y i -s to extract cloud-free image blocks and use the K nearest neighbor algorithm to construct a cloud-free image group, which is recorded as μ represents an adjustable parameter, y i represents the training image, y i -s indicates the image after removing system noise.

[0065] Step 2, solving the SC-LR model: The present invention preferably adopts the ADMM method to solve the SC-LR model. and s are separable during the optimization process, so they are converted into variables that are easy to calculate and s sub-problems, and according to the characteristics of each sub-problem, use simple and efficient methods to calculate them respectively.

[0066] Step 2.1, solve the auxiliary variable (symmetric semi-positive matrix ) Sub-problem

[0067] Given and s, The sub-optimization problem can be expressed as:

[0068]

[0069] The optimal solution is:

[0070]

[0071] Step 2.2, solve the low-rank approximate image group Subproblems

[0072] Fixed s and The sub-optimization problem can be expressed as:

[0073]

[0074] in is a cloud-free image group. The above formula is a bilaterally constrained weighted least squares regression problem, and its solution satisfies the following conditions:

[0075]

[0076] in I is the unit matrix. The above formula is a standard SE (Sylvester) equation. is a symmetric positive definite matrix, is a symmetric positive semidefinite matrix, so the equation has a unique solution.

[0077] because is a symmetric positive definite square matrix, so its characteristic can be decomposed into in, is The orthogonal matrix composed of the eigenvectors of is a diagonal matrix with diagonal elements The eigenvalues of express The transposed matrix of . Multiply both sides of formula (13) by The equation becomes:

[0078]

[0079] At this time, the above formula is converted to The SE equation of , the solution of the equation is:

[0080]

[0081] In the above formula, vec(·) represents matrix vectorization. Finally, The solution can be obtained by Get, where vec -1 (·) represents a vector matrix operation.

[0082] Step 2.3, solve the system noise s subproblem

[0083] fixed and The s-suboptimization problem can be expressed as:

[0084]

[0085] The closed-form solution of the above equation is:

[0086]

[0087] In the above formula, the matrix to be inverted is a diagonal matrix, so the subproblem s can be solved efficiently.

[0088] Step 3: Noise removal of area array imaging system

[0089] In practice, steps 1 and 2 can be performed offline, and the results from these steps can be used to remove system noise online. System noise in array satellite imagery primarily stems from the non-uniform response characteristics of CCD or CMOS detectors and their dark current effects. Furthermore, considering the inevitable random noise present throughout the imaging process, degraded images contaminated by these two types of noise can be modeled as follows:

[0090] y i =x i +n i =x i +s+r i (18)

[0091] Among them, r i represents random noise, which is assumed to be additive white Gaussian noise; s is system noise, whose characteristics are much more complex than additive white Gaussian noise and difficult to model with a simple analytical distribution; n i is the mixed noise of system noise and random noise, x i represents an ideal image without noise, y i Indicates the observed image.

[0092] In theory, system noise is sensor-specific, meaning all images captured by the same sensor contain the same system noise. Therefore, the system noise can be estimated using a small number of image samples and applied to all images from that sensor, achieving globally consistent noise removal. Once the system noise is solved offline using the ADMM method described above, its online removal process can be expressed as:

[0093]

[0094] in is the image after removing the system noise, It is important to note that, in the system noise removal stage, the present invention uses the observed value instead of the corrected value in the cloud area based on the cloud mask matrix to maintain the quality of the cloud area in the image.

[0095] For ease of implementation and reference, the following is a recommended workflow for noise removal in array satellite systems based on the SC-NCLR model:

[0096] Input: degraded image y i (i=1,...,M), cloud mask matrix J i , the noise weight matrix W i , image block size parameter p, number of similar image blocks parameter m, regularization parameters μ, α and λ, and maximum number of iterations T.

[0097] Initialization: Initial estimate of system noise

[0098] Loop iteration (t=1 to maximum number of iterations T):

[0099] Traverse the images (i = 1 to the maximum number of training images M):

[0100] 1) Image segmentation: Based on step 1, yi -s Split into overlapping reference image blocks

[0101] 2) Block matching: Based on step 1, for each image block Search for the m most similar image blocks in its local neighborhood and construct a group matrix

[0102] 3) Group matrix optimization: Based on step 2, the group matrix of each cloud-free image is optimized. Calculate its low-rank approximate image group matrix

[0103] 4) i=i+1

[0104] 5) Determine whether i>M satisfies

[0105] 6) If satisfied, exit the image traversal; if not satisfied, continue traversing the image

[0106] 7) Aggregation Matrix The block in the

[0107] 8) t=t+1

[0108] 9) Determine whether t>T satisfies

[0109] 10) If satisfied, exit the loop; if not satisfied, continue iterating

[0110] Output: Based on step 3, the image with system noise removed is obtained

[0111] To facilitate understanding of the effectiveness of the present invention's techniques, clean Gaofen-1 satellite images were used as the original reference images to verify the performance of different algorithms under independent and identically distributed Gaussian systematic noise of varying intensities. The experiment consisted of two independent phases: systematic noise extraction and systematic noise removal. To ensure no data leakage between the two phases, the two phases were conducted using independent data. Specifically, the systematic noise extraction phase used 100 Gaofen-1 satellite images, while the systematic noise removal phase used a separate set of 30 images, completely independent of the noise extraction phase. At the same time, the performance of the method provided in the embodiment of the present invention is compared with two other system noise removal methods, including Literature 1: Q.Xie et al., "Multispectral Images Denoising by Intrinsic Tensor SparsityRegularization," 2016IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016, pp.1692-1700, doi:10.1109 / CVPR.2016.187. Literature 2: J.Cheng, T.Liu and S.Tan, "Score Priors Guided DeepVariational Inference for Unsupervised Real-World Single Image Denoising," in 2023IEEE / CVF International Conference on Computer Vision (ICCV), Paris, France, 2023, pp.12891-12902, doi:10.1109 / ICCV51070.2023.01189.

[0112] like Figure 2 The visual comparison results of the noise removal of the displayed image system, where part (a) represents an image not contaminated by noise, part (b) represents an image contaminated by system noise and random noise, with the system noise variance being 60 and the random noise variance being 10, part (c) is the result after the system noise is removed in reference 1, part (d) is the result after the system noise is removed in reference 2, and part (e) is the result after the system noise is removed in the present invention. Figure 2 It can be seen that the method proposed in this invention performs better in terms of image clarity, detail preservation, and artifact suppression. Specifically, after removing system noise, the methods of Reference 1 and Reference 2 still have obvious artifacts and produce some erroneous texture information.

[0113] The corresponding quantitative evaluation results are shown in Table 1. The present invention uses peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) to evaluate the image quality after removing system noise. The best results are marked in bold. As can be seen from Table 1, at all noise levels, the system noise removal method based on SC-NCLR proposed in the present invention achieves higher PSNR results than other comparison methods. In terms of the SSIM indicator, it can be seen that the SC-NCLR method proposed in the present invention also achieves better performance than other comparison methods at all noise levels.

[0114] Table 1 Comparison of the average PSNR (dB) and SSIM results of each method under different system noise levels of GF-1 images

[0115]

[0116] In specific implementations, those skilled in the art can use software technology to automatically run the above process. Accordingly, if a noise removal solution for an area array imaging system based on self-constraint and non-convex low-rank approximation is provided, including a computer or server, and the above process is executed on the computer or server to perform noise removal for an area array imaging system based on self-constraint and non-convex low-rank approximation, it should also be within the scope of protection of the present invention.

[0117] In another embodiment, an electronic device is provided, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so as to enable the at least one processor to execute the above-mentioned method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation.

[0118] In another embodiment, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the above-mentioned method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation.

[0119] In another embodiment, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation is implemented.

[0120] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0121] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0122] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation, characterized by: Carry out the following process, A self-constrained non-convex low-rank approximate collaborative model is constructed. This involves directly imposing data fidelity constraints on the system noise term and introducing a noise weight matrix into the objective function to quantify the spatial distribution differences in noise intensity. The non-convex low-rank approximate model is combined with the non-local self-similarity of image patches and embedded with graph regularization to constrain the local structural smoothness of the image. Furthermore, a cloud mask matrix is introduced into the data fidelity term to block the interference of cloud coverage areas on the system noise estimation. The self-constrained non-convex low-rank approximate collaborative model is iteratively optimized using an alternating direction multiplier method to solve the system noise estimation value; The input image is denoised based on the system noise estimate, and the original observations are retained in the cloud region to maintain the image quality.

2. The method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation according to claim 1, characterized in that: The noise weight matrix is a diagonal matrix, and its diagonal elements are defined as the inverse of the noise standard deviation of each pixel.

3. The method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation according to claim 1, characterized in that: The non-convex low-rank approximation model is constructed by dividing the input image into partially overlapping reference image blocks, searching for similar image blocks and constructing an image group matrix, and applying a low-rank constraint to each image group matrix to constrain structural similarity.

4. The method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation according to claim 3, characterized in that: When constructing the image group matrix, similar image block search is performed only on the reference image blocks in the non-cloud area, thereby constructing the cloud-free image group.

5. The method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation according to claim 1, characterized in that: The cloud mask matrix is a diagonal matrix, and its diagonal elements are defined as values corresponding to pixels in cloud areas are 0, and values corresponding to pixels in non-cloud areas are 1.

6. The method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation according to claim 1, characterized in that: When the alternating direction multiplier method is used to iteratively optimize the self-constrained non-convex low-rank approximate collaborative model, the objective function is decomposed into a system noise subproblem, a low-rank approximate image group subproblem and an auxiliary variable subproblem, and they are solved separately.

7. The method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation according to claim 1, characterized in that: The noise removal stage adopts a sensor-specific correction strategy to apply the offline estimated system noise to all images captured by the same sensor, and realizes noise removal, in which the cloud region directly retains the observation value.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the noise removal method for an area array imaging system based on self-constraint and non-convex low-rank approximation as described in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation as claimed in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that: When the computer program is executed by a processor, the method for removing noise from an area array imaging system based on self-constraint and non-convex low-rank approximation as claimed in any one of claims 1 to 7 is implemented.

Images