A method for extracting remote sensing image stripe noise based on sparse optimization model

Abstract

The application provides a kind of remote sensing image stripe noise extraction method based on sparse optimization model, first, remote sensing image is carried out multi-scale wavelet decomposition, and the wavelet coefficient of each level is obtained;Second, based on the sparse characteristics of remote sensing stripe noise, the sparse optimization model is established for the wavelet coefficient of each level of remote sensing image respectively, and the sparse optimization model parameters are set;Alternating multiplier method is used to solve, solve the sparse optimization model, complete the stripe noise extraction corresponding to the wavelet coefficient of each level of remote sensing image;Finally, the processed wavelet coefficient is reconstructed by wavelet, and the stripe noise extraction of remote sensing image is completed.This method considers the sparse characteristics of stripe noise, and the model parameters are adaptively calculated, so that the parameters do not need to be set artificially in the extraction process of stripe noise, which not only avoids the error caused by manual parameter setting, but also solves the problem of huge workload in a large number of remote sensing image processing.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing, and specifically relates to a method for extracting stripe noise from remote sensing images based on a sparse optimization model. Background Technology

[0002] Remote sensing technology is a detection technique that acquires electromagnetic wave images of ground objects based on imaging technologies such as aerial photography, aerial scanning, and microwave radar, and obtains digital remote sensing images through analog-to-digital conversion. It has advantages such as unrestricted observation range, rapid information acquisition, and short update cycles, and has been widely used in smart city construction, land and forest cover monitoring, and other applications. However, inconsistencies in the responses of multiple detectors during remote sensing can lead to fringe noise along the scanning direction in the remote sensing images, severely affecting the detection results. To extract fringe noise from remote sensing images and improve the quality of the detection data, fringe noise extraction is necessary.

[0003] Chinese patent CN111145118A discloses "A method and apparatus for removing stripes from remote sensing images." This method first analyzes the cumulative sum of lunar irradiance grayscale values ​​for each individual detector element based on the lunar pixel grayscale values ​​and dark current count values. Second, it determines the grayscale correction coefficient for each detector element based on the cumulative sum of lunar irradiance grayscale values ​​for each individual detector element and the lunar irradiance information of a reference detector element. Finally, it removes stripes from the remote sensing image based on the individual detector element grayscale correction coefficients. However, this method requires both the lunar pixel grayscale values ​​and dark current count values ​​for each individual detector element to complete the stripe removal.

[0004] Zhang et al. (2009) designed a filter angle to remove periodic stripe noise using wavelet transform and other methods. Han et al. (2009), Ren et al. (2011), and Tan et al. (2013) proposed statistical methods based on moment matching and histogram matching to eliminate regular stripe noise. Li et al. (2021) estimated the stripe components in remote sensing images by establishing a minimum energy functional based on an L1 norm optimization model. However, the L1 norm is the sum of the absolute values ​​of the elements in the vector, and the existence of absolute values ​​leads to continuous oscillations when the error is close to 0. A sparse optimization model method that simultaneously considers the periodicity, non-periodicity, regularity, and irregularity of stripe noise in remote sensing images and avoids oscillations has yet to be found. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing a remote sensing image stripe noise extraction method based on a sparse optimization model, which solves the problem that traditional remote sensing image stripe noise extraction methods need to consider the characteristics of noise or the oscillation during processing.

[0006] To achieve the above objectives, a method for extracting stripe noise from remote sensing images based on a sparse optimization model includes the following steps:

[0007] a. Input the remote sensing image data to be processed. In order to preserve the characteristics of the data itself, there is no need to grid the image data.

[0008] b. Perform multi-scale wavelet decomposition on the remote sensing image to obtain wavelet coefficients at each level;

[0009] c. Based on the sparsity characteristics of remote sensing stripe noise, sparse optimization models are established for wavelet coefficients at each level of remote sensing images.

[0010] d. Set sparse optimization model parameters. The regularization coefficient in the model has a decisive effect on the stripe noise extraction effect. This key parameter is determined by spatial adaptive iteration to avoid introducing errors by manually selecting model parameters.

[0011] e. For the established sparse optimization model, the alternating multiplier method is used to solve it, and the stripe noise features corresponding to the wavelet coefficients at each level of the remote sensing image are obtained.

[0012] f. Set the iteration termination condition for the sparse optimization model. When the iteration termination condition is reached, complete the extraction of stripe noise corresponding to wavelet coefficients at each level of the remote sensing image.

[0013] g. Perform wavelet reconstruction on the processed wavelet coefficients to complete the stripe noise extraction of the remote sensing image.

[0014] h) Output the extracted stripe noise, and end the stripe noise extraction process of the remote sensing image.

[0015] Step a involves inputting the remote sensing image data to be processed. To preserve the inherent characteristics of the data, it is not necessary to grid the image data. Let the size of the remote sensing image data to be processed be... m*n Representing image data as There is no need to grid the data.

[0016] Step b involves performing multi-scale wavelet decomposition on the remote sensing image to obtain wavelet coefficients at each level. Since the remote sensing image contains periodic and aperiodic stripe noise (i.e., stripe noise at multiple frequencies), multi-scale wavelet decomposition is performed on the remote sensing image to ensure the completeness of stripe noise extraction.

[0017] (1)

[0018] In equation (1), This refers to the low-frequency portion of the remote sensing image. These represent the high-frequency components of different frequencies in a remote sensing image.

[0019] Step c describes establishing sparse optimization models for wavelet coefficients at each level of the remote sensing image based on the sparsity characteristics of remote sensing stripe noise. The wavelet coefficients at each level are divided into a superposition of stripe noise and signal data.

[0020] (2)

[0021] In equation (2), Low-frequency component of remote sensing image The stripe noise contained in it, Low-frequency component of remote sensing image The signal data contained therein.

[0022] Based on the sparsity characteristics of remote sensing stripe noise, a sparse optimization model for wavelet coefficients at various levels is established:

[0023] (3)

[0024] In equation (3), is the regularization coefficient, set in step d.

[0025] Low-frequency component of remote sensing image The second norm of medium fringe noise:

[0026] (4)

[0027] Low-frequency component of remote sensing image Total variation of signal data:

[0028] (5)

[0029] Following step c, sparse optimization models are established for wavelet coefficients at each level of the remote sensing image.

[0030] Step d describes setting the parameters of the sparse optimization model. The regularization coefficient in the model has a decisive effect on the stripe noise extraction effect. This key parameter is determined by a spatial adaptive iteration method to avoid introducing errors by manually selecting model parameters.

[0031] (6)

[0032] Step e describes solving the established sparse optimization model using the alternating multiplier method to obtain the stripe noise features corresponding to the wavelet coefficients at each level of the remote sensing image. The sparse model is then transformed into the corresponding scaled augmented Lagrangian function:

[0033] (7)

[0034] The alternating multiplier method is used to solve the sparse regularization constraint problem. The specific algorithm steps are as follows:

[0035] (1) Parameter initialization , , ( ), ;

[0036] (2) If the convergence condition is not met, perform iterative calculations using equations (3)-(5);

[0037] (3) ;

[0038] (4) ;

[0039] (5) ;

[0040] This continues until the decomposition process of the sparse model is completed.

[0041] Step f describes setting the iteration termination condition for the sparse optimization model. When the iteration termination condition is met, the stripe noise corresponding to the wavelet coefficients at each level of the remote sensing image is extracted. The iteration termination condition for solving the sparse model is:

[0042] (8)

[0043] Step g involves wavelet reconstruction of the processed wavelet coefficients to extract stripe noise from the remote sensing image. When reconstructing the decomposed wavelets, combining equations (1) and (2), the original remote sensing image is:

[0044] (9)

[0045] The stripe noise extracted from each wavelet layer is superimposed to obtain the stripe noise data of the remote sensing image:

[0046] (10)

[0047] The corresponding remote sensing image with stripe noise removed is :

[0048] (11)

[0049] Step h outputs the extracted stripe noise, ending the remote sensing image stripe noise extraction process. The remote sensing image stripe noise extraction result of this invention is... The remote sensing image with stripe noise removed is .

[0050] Compared with existing technologies, the beneficial effects of this invention are as follows: The remote sensing image stripe noise extraction method based on a sparse optimization model disclosed in this invention can quickly and effectively extract stripe noise from remote sensing images; the parameters used in this invention are all automatically iterated and do not require manual adjustment, making it a fully automatic stripe noise extraction method. This remote sensing image stripe noise extraction method not only avoids errors introduced by manually setting parameters in commonly used methods, but also solves the problem of the huge workload in processing large numbers of remote sensing images. Furthermore, from... Figure 1 and Figure 2 The comparison shows that, because this invention incorporates the sparse characteristics of stripe noise, the remote sensing image with stripe noise removed retains the original features of the remote sensing image, resulting in more complete stripe noise removal. The effectiveness of stripe noise removal can be further verified by comparing the row mean curves of the remote sensing images before and after stripe noise removal. Attached Figure Description

[0051] Figure 1 The image is a remote sensing image containing striped noise.

[0052] Figure 2 Remote sensing images for removing stripe noise;

[0053] Figure 3 Row mean curves of remote sensing images before and after stripe noise removal; Detailed Implementation

[0054] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0055] It should be noted that the terms "first terminal," "second terminal," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects. The terms "first" and "second" refer to the technical solutions disclosed in detail in this invention and are not necessarily performed sequentially. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0056] The following is a further detailed description with reference to the accompanying drawings and embodiments:

[0057] This invention discloses a method for extracting stripe noise from remote sensing images based on a sparse optimization model. Taking a remote sensing image containing stripe noise as an example, the image contains 500*500 pixels.

[0058] a. Input the remote sensing image data to be processed. To preserve the inherent characteristics of the data, there is no need to grid the image data. The size of the remote sensing image data to be processed is 500*500. Figure 1 As shown, the image data is represented as There is no need to grid the data.

[0059] b. Perform multi-scale wavelet decomposition on the remote sensing image to obtain wavelet coefficients at each level. Since remote sensing images contain periodic and aperiodic stripe noise (i.e., stripe noise at multiple frequencies), multi-scale wavelet decomposition is performed on the remote sensing image to ensure the completeness of stripe noise extraction.

[0060] (1)

[0061] In equation (1), This refers to the low-frequency portion of the remote sensing image. These represent the high-frequency components of different frequencies in a remote sensing image.

[0062] c. Based on the sparsity characteristics of remote sensing stripe noise, sparse optimization models are established for wavelet coefficients at each level of the remote sensing image. The wavelet coefficients at each level are divided into a superposition of stripe noise and signal data:

[0063] (2)

[0064] In equation (2), Low-frequency component of remote sensing image The stripe noise contained in it, Low-frequency component of remote sensing image The signal data contained therein.

[0065] Based on the sparsity characteristics of remote sensing stripe noise, a sparse optimization model for wavelet coefficients at various levels is established:

[0066] (3)

[0067] In equation (3), is the regularization coefficient, set in step d.

[0068] Low-frequency component of remote sensing image The second norm of medium fringe noise:

[0069] (4)

[0070] Low-frequency component of remote sensing image Total variation of signal data:

[0071] (5)

[0072] Following step c, sparse optimization models are established for wavelet coefficients at each level of the remote sensing image.

[0073] d. Set sparse optimization model parameters. The regularization coefficient in the model has a decisive effect on the stripe noise extraction effect. This key parameter is determined by spatial adaptive iteration to avoid introducing errors by manually selecting model parameters.

[0074] (6)

[0075] e. For the established sparse optimization model, the alternating multiplier method is used to solve it, obtaining the stripe noise features corresponding to the wavelet coefficients at each level of the remote sensing image. The sparse model is then transformed into the corresponding scaling format augmented Lagrangian function:

[0076] (7)

[0077] The alternating multiplier method is used to solve the sparse regularization constraint problem. The specific algorithm steps are as follows:

[0078] (1) Parameter initialization , , ( ), ;

[0079] (2) If the convergence condition is not met, perform iterative calculations using equations (3)-(5);

[0080] (3) ;

[0081] (4) ;

[0082] (5) ;

[0083] This continues until the decomposition process of the sparse model is completed.

[0084] f. Set the iteration termination condition for the sparse optimization model. When the iteration termination condition is reached, complete the extraction of fringe noise corresponding to wavelet coefficients at each level of the remote sensing image. The iteration termination condition for solving the sparse model is:

[0085] (8)

[0086] g. Perform wavelet reconstruction on the processed wavelet coefficients to complete the stripe noise extraction of the remote sensing image. When reconstructing the decomposed wavelets, combining equations (1) and (2), the remote sensing image before processing is:

[0087] (9)

[0088] The stripe noise extracted from each wavelet layer is superimposed to obtain the stripe noise data of the remote sensing image:

[0089] (10)

[0090] The corresponding remote sensing image with stripe noise removed is :

[0091] (11)

[0092] h) Output the extracted stripe noise, ending the remote sensing image stripe noise extraction process. The remote sensing image stripe noise extraction result of this invention is: The corresponding remote sensing image with stripe noise removed is ,like Figure 2 As shown, the effectiveness of stripe noise removal can be further verified by comparing the row mean curves of the remote sensing images before and after stripe noise removal, such as... Figure 3 As shown.

[0093] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for extracting stripe noise from remote sensing images based on a sparse optimization model, characterized in that, Includes the following steps: a. Input the remote sensing image data to be processed. In order to preserve the characteristics of the data itself, there is no need to grid the image data. b. Perform multi-scale wavelet decomposition on the remote sensing image to obtain wavelet coefficients at each level; c. Based on the sparsity characteristics of remote sensing stripe noise, sparse optimization models are established for wavelet coefficients at each level of remote sensing images. d. Set sparse optimization model parameters. The regularization coefficient in the model has a decisive effect on the stripe noise extraction effect. This key parameter is determined by spatial adaptive iteration to avoid introducing errors by manually selecting model parameters. e. For the established sparse optimization model, the alternating multiplier method is used to solve it, and the stripe noise features corresponding to the wavelet coefficients at each level of the remote sensing image are obtained. f. Set the iteration termination condition for the sparse optimization model. When the iteration termination condition is reached, complete the extraction of stripe noise corresponding to wavelet coefficients at each level of the remote sensing image. g. Perform wavelet reconstruction on the processed wavelet coefficients to complete the stripe noise extraction of the remote sensing image; h. Output the extracted stripe noise to end the stripe noise extraction process of the remote sensing image; In step b, the remote sensing image is subjected to multi-scale wavelet decomposition: D=A N +D1+D2+D3+…+D N (1) Among them, A N The low-frequency components of the remote sensing image are D1, D2, D3, ..., D. N High-frequency components of different frequencies in remote sensing images; In step c, the wavelet coefficients at each level are divided into a superposition of stripe noise and signal data: in, For the low-frequency component A of the remote sensing image N The stripe noise contained in it, For the low-frequency component A of the remote sensing image N The signal data contained therein; Based on the sparsity characteristics of remote sensing stripe noise, a sparse optimization model for wavelet coefficients at various levels is established: Where λ is the regularization coefficient, which is set in step d; For the low-frequency component A of the remote sensing image N The second norm of medium fringe noise: For the low-frequency component A of the remote sensing image N Total variation of signal data: Following step c, sparse optimization models are established for wavelet coefficients at each level of the remote sensing image.

2. The method for extracting stripe noise from remote sensing images based on a sparse optimization model as described in claim 1, characterized in that: In step a, let the size of the remote sensing image data to be processed be m*n, and represent the image data as D=[d 1,1 ,…,d m,1 ,…,d L,n ,…,d m,n There is no need to grid the data.

3. The method for extracting stripe noise from remote sensing images based on a sparse optimization model as described in claim 1, characterized in that: In step d, the sparse optimization model parameters are set:

4. The method for extracting stripe noise from remote sensing images based on a sparse optimization model as described in claim 1, characterized in that: In step e, the sparse model is transformed into the corresponding scaled augmented Lagrangian function: The alternating multiplier method is used to solve the sparse regularization constraint problem. The specific algorithm steps are as follows: (1) Initialize parameters u, v, λ (λ = 0), β; (2) If the convergence condition is not met, perform iterative calculations using equations (3)-(5); (3) (4) (5)l (k+1) =λ (k) -γβ(Au (k+1) +Bv (k+1) -b); This continues until the decomposition process of the sparse model is completed.

5. The method for extracting stripe noise from remote sensing images based on a sparse optimization model as described in claim 1, characterized in that: In step f, the iterative termination condition for solving the sparse model is:

6. The method for extracting stripe noise from remote sensing images based on a sparse optimization model as described in claim 1, characterized in that: In step g, when reconstructing the decomposed wavelets, combining the formulas from steps b and c, the remote sensing image before processing is: The stripe noise extracted from each wavelet layer is superimposed to obtain the stripe noise data of the remote sensing image: The corresponding remote sensing image with stripe noise removed is D signal :

Images