Satellite remote sensing image random stripe noise suppression method and system based on low-rank tensor approximation
By using a low-rank tensor approximation method, combined with one-way total variational UTV and low-rank tensor constraints, the problem of suppressing mixed noise of random stripes in satellite remote sensing images was solved, improving image quality and the accuracy of information analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-30
- Publication Date
- 2026-08-04
AI Technical Summary
Existing technologies struggle to effectively remove random stripe noise from satellite remote sensing images, leading to a decline in image quality and affecting the accuracy of subsequent interpretation and analysis.
A low-rank tensor approximation method is adopted to effectively suppress mixed noise in satellite images by establishing a random stripe mixed noise image model and combining one-way total variational UTV and low-rank tensor constraints.
It effectively suppresses mixed noise of random stripes, improves image quality, and enhances visual effects and the accuracy of information recognition and analysis.
Smart Images
Figure CN120543413B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of satellite remote sensing technology, and in particular to a method and system for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation. Background Technology
[0002] Due to differences in the performance of satellite CCD imaging systems and the influence of imaging environments such as illumination variations and atmospheric disturbances, different detectors exhibit inconsistent responses to the same radiant energy. This phenomenon inevitably introduces random noise and fringe noise into the imaging results, severely reducing the radiometric accuracy and spatial consistency of satellite images. Random noise is caused by sensor electronic noise or quantization errors and exhibits discrete distribution characteristics; fringe noise is generated by differences in the response of CCD detectors and manifests as periodic stripes along the imaging direction. Such noise leads to severe degradation of image quality, not only reducing visual appeal but also affecting the accurate identification and analysis of information by interpreters.
[0003] Existing research indicates that the key to stripe noise removal lies in finding a transform domain capable of achieving complete signal-noise separation. Current mainstream frequency domain methods, such as Fourier transform and wavelet transform, model stripe noise as spatially periodic noise and achieve denoising through the design of frequency domain filters. While these methods effectively suppress periodic stripes, they suffer from two inherent drawbacks: first, signal loss, where high-frequency ground feature information such as edges and textures may be mistakenly removed during the filtering process; and second, residual non-periodic noise, limiting their effectiveness in handling randomly distributed stripe noise such as inconsistencies in CCD detector response.
[0004] In the field of random noise removal from remote sensing images, existing research methods can be mainly divided into three categories: spatial domain filtering, transform domain sparse representation, and low-rank approximation methods. Spatial domain filtering methods use local filtering operators to weight the pixel neighborhood, with the core being the construction of filter masks to smooth noise. Typical methods include isotropic smoothing mean filtering, impulse noise suppression median filtering, morphological filtering based on geometric structure processing, and structure-oriented adaptive filtering that preserves local features. Transform domain sparse representation methods are based on the difference in signal and noise distribution in the transform domain. The effective signal of an image exhibits sparse characteristics under specific transform bases such as wavelet transform and curvelet transform, while random noise exhibits a non-sparse distribution. Noise separation can be achieved by thresholding the transform coefficients. Low-rank approximation methods utilize the inherent low-rank characteristics of clean images, employing matrix dimensionality reduction techniques to recover effective information from noisy data. These mainly include matrix low-rank approximation based on singular value decomposition, Hankel matrix reconstruction methods, and higher-dimensional tensor singular value decomposition (T-SVD). Although the above methods can also be used for stripe noise suppression, studies have shown that these traditional methods have significant limitations when dealing with mixed noise where random noise and stripe noise coexist. They are difficult to meet the removal requirements of different noise types at the same time and are prone to loss of effective information during the denoising process.
[0005] In summary, random stripe mixed noise in high-resolution satellite imagery significantly degrades image quality and affects the accuracy of subsequent interpretation and analysis. Therefore, it is urgent to study a method for removing random stripe mixed noise in satellite imagery that can effectively suppress mixed noise, thereby effectively reducing mixed noise and improving the quality of satellite remote sensing images. Summary of the Invention
[0006] The purpose of this invention is to provide a method and system for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation, which can effectively suppress mixed noise of random stripes in satellite images, reduce the impact of mixed noise, and improve image quality.
[0007] The technical solution to achieve the purpose of this invention is: a method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation, comprising the following steps:
[0008] Step 1: Establish a random stripe mixed noise image model;
[0009] Step 2: Transform the core task of satellite image denoising into an unconstrained optimization problem and determine the objective function;
[0010] Step 3: To suppress random noise in satellite imagery, perform LRA operation on the noisy data matrix / tensor.
[0011] Step 4: For stripe noise in satellite imagery, a stripe noise removal model based on unidirectional total variation UTV is used to separate stripe noise through directional selective regularization.
[0012] Step 5: Establish a UTV-LRTA hybrid denoising model, and suppress hybrid stripe noise in satellite imagery by combining low pipe rank tensor constraints with unidirectional total variation regularization.
[0013] Furthermore, the establishment of the random stripe mixed noise image model described in step 1 is as follows:
[0014] The Gaofen satellite uses a pushbroom imaging method, where each column of data in the image is acquired by the same detector element of the CCD sensor continuously scanning along the orbital direction. Pixels in the same column will be subject to the same degree of noise interference. Therefore, the stripe noise present in each column of the image is regarded as an additional noise component. Based on the assumption that the noise components are added rather than multiplied, a mixed noise image model is established, and the mathematical expression is as follows:
[0015] (1)
[0016] in, and These represent the noisy data tensor and the clean data tensor, respectively. Indicates stripe noise. Represents random noise, where These represent the column width, row width, and number of bands of the image, respectively.
[0017] Furthermore, step 2 transforms the core task of satellite image denoising into an unconstrained optimization problem, and the objective function is determined as follows:
[0018] The core task of satellite image denoising is to establish a model from noisy observation data. Restore to original cleaning data The inverse estimation process minimizes the objective function to achieve the desired result. The optimal estimate is given by the formula:
[0019] (2)
[0020] The denoising task in equation (2) is an ill-posed inverse problem. To address this problem, by introducing image tensor priors and stripe noise regularization constraints, the ill-posed inverse problem is transformed into an unconstrained optimization problem. The final objective function is expressed as a weighted combination of data fidelity terms and regularization terms, as shown in the formula:
[0021] (3)
[0022] The objective function in equation (3) consists of three key terms: the first term serves as the data fidelity term, ensuring the consistency between the reconstructed results and the original observation data; the second term... The first term is the image tensor regularization term, which eliminates unwanted artifacts in the reconstructed image by introducing prior constraints; the third term... This is a regularization term specifically designed for stripe noise, where the parameters... and The weight balance of these two regularization terms is controlled separately.
[0023] Furthermore, in step 3, the LRA operation is performed on the noisy data matrix / tensor to suppress random noise in the satellite imagery, as detailed below:
[0024] To address random noise in satellite imagery, random noise suppression is achieved by performing LRA operations on the noisy data matrix / tensor. This process can be formalized into the following optimization problem:
[0025] (4)
[0026] Furthermore, in step 4, a striated noise removal model based on unidirectional total variation UTV is used to remove striated noise in satellite imagery. This model achieves striated noise separation through direction-selective regularization, as detailed below:
[0027] Stripe noise in satellite imagery exhibits a linear pattern with a regular spatial distribution. This noise is directionally correlated and structurally consistent, showing a significant response in the horizontal derivative but almost no effect on the vertical derivative. Therefore, the UTV model achieves noise separation through directional selective regularization. For the stripe noise component, a smoothing constraint is applied along the stripe direction to promote noise separation; for the clean image component, structural integrity is maintained along the vertical direction. This process is formalized into the following optimization problem:
[0028] (5)
[0029] In the formula, the directional derivative operator and The gradients are calculated for the horizontal and vertical directions, respectively.
[0030] Furthermore, the establishment of the UTV-LRTA hybrid denoising model described in step 5, through low-rank tensor constraints combined with unidirectional total variation regularization, achieves the suppression of hybrid stripe noise in satellite imagery, as detailed below:
[0031] Step 5.1: Use the low-rank tensor approximation LRTA instead. The regularization term is replaced by a low-dimensional unidirectional total variation UTV constraint. The regularization term, through optimization of the fidelity term in equation (3), achieves the control of the fringe noise component. Explicit elimination, then regarding The final function is written as:
[0032] (6)
[0033] In the formula, , and This represents the regularization parameter used to adjust the weights of each item to balance the model;
[0034] Step 5.2, , , By substitution, the original unconstrained optimization problem (6) is transformed into an equivalent constrained optimization form, as shown in the formula:
[0035] (7)
[0036] Step 5.3: Based on the Bregman iterative algorithm, the constrained optimization problem (7) is transformed into the following equivalent unconstrained minimization form:
[0037] (8)
[0038] in, , and Indicates the Bregman penalty parameter; , , It is a Bregman tensor.
[0039] A system for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation is disclosed. This system implements the aforementioned method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation. The system comprises a first unit to a fifth unit, wherein:
[0040] The first unit is to establish a random striped mixed noise image model;
[0041] The second unit transforms the core task of satellite image denoising into an unconstrained optimization problem and determines the objective function.
[0042] The third unit addresses random noise in satellite imagery by performing LRA operations on the noisy data matrix / tensor to suppress random noise.
[0043] The fourth unit addresses stripe noise in satellite imagery by employing a stripe noise removal model based on unidirectional total variation UTV, achieving stripe noise separation through directional selective regularization.
[0044] The fifth unit establishes a UTV-LRTA hybrid denoising model, which uses low-rank tensor constraints combined with unidirectional total variation regularization to suppress hybrid stripe noise in satellite imagery.
[0045] A mobile terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation.
[0046] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation.
[0047] Compared with the prior art, the present invention has the following significant advantages: (1) It decomposes the noisy image into a non-striped noise image and a striped noise component by using unidirectional total variation UTV, thereby achieving effective separation of striped noise; it further suppresses random noise by using tensor kernel norm TNN regularization to constrain the low-rank characteristics of the image, thereby achieving effective suppression of random striped mixed noise; (2) It has strong denoising performance and high robustness, reduces the influence of random striped mixed noise, improves the image quality, improves the visual effect, and improves the accuracy of information recognition and analysis. Attached Figure Description
[0048] Figure 1 This is a flowchart of the method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation, as per the present invention.
[0049] Figure 2 This is a schematic diagram of the process for establishing a mixed noise image model in an embodiment of the present invention.
[0050] Figure 3 This is a schematic diagram of the directional characteristics of stripe noise in satellite imagery in an embodiment of the present invention.
[0051] Figure 4 This is a schematic diagram illustrating the principle of the UTV-LRTA hybrid denoising model in an embodiment of the present invention.
[0052] Figure 5 The diagram shows the noise suppression effects of various methods for synthesizing datasets in this embodiment of the invention.
[0053] Figure 6 The diagram shows the noise suppression effects of various methods on real satellite datasets in this embodiment of the invention. Detailed Implementation
[0054] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0055] Combination Figure 1 This invention discloses a method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation, comprising the following steps:
[0056] Step 1: Establish a random stripe mixed noise image model, as follows:
[0057] High-resolution satellites employ pushbroom imaging, where each column of data in the image is acquired by the same detector element of a CCD sensor continuously scanning along the orbital direction. Pixels in the same column are subject to the same level of noise interference. Therefore, the stripe noise present in each column of the image is considered an additional noise component. Based on the assumption that noise components are added rather than multiplied, a method is established as follows: Figure 2 The mathematical expression for the mixed noise image model shown is as follows:
[0058] (1)
[0059] in, and These represent the noisy data tensor and the clean data tensor, respectively. Indicates stripe noise. Represents random noise, where These represent the column width, row width, and number of bands of the image, respectively.
[0060] Step 2: Transform the core task of satellite image denoising into an unconstrained optimization problem and determine the objective function, as follows:
[0061] The core task of satellite image denoising is to establish a model from noisy observation data. Restore to original cleaning data The inverse estimation process minimizes the objective function to achieve the desired result. The optimal estimate is given by the formula:
[0062] (2)
[0063] The denoising task described by equation (1) is a typical ill-posed inverse problem. To address this problem, by introducing image tensor priors and stripe noise regularization constraints, it is transformed into an unconstrained optimization problem. The final objective function can be expressed as a weighted combination of data fidelity terms and regularization terms, as shown in the formula:
[0064] (3)
[0065] The objective function in equation (3) consists of three key terms: the first term serves as the data fidelity term, ensuring the consistency between the reconstructed results and the original observation data; the second term... The first term is the image tensor regularization term, which eliminates unwanted artifacts in the reconstructed image by introducing prior constraints; the third term... This is a regularization term specifically designed for stripe noise, where the parameter... and The weight balance of these two regularization terms is controlled separately.
[0066] Step 3: To suppress random noise in satellite imagery, perform LRA operation on the noisy data matrix / tensor, as follows:
[0067] Satellite image denoising methods based on low-rank tensor modeling have significant advantages. The low-rank approximation (LRA) method can effectively extract the low-rank characteristics of clean images by minimizing the objective function, demonstrating excellent performance in denoising applications. To address random noise in satellite images, random noise suppression is achieved by performing LRA operations on the noisy data matrix / tensor. This process can be formalized into the following optimization problem:
[0068] (4)
[0069] Step 4: For stripe noise in satellite imagery, a stripe noise removal model based on unidirectional total variation UTV is used. Stripe noise separation is achieved through direction-selective regularization, as detailed below:
[0070] The stripe noise removal model based on unidirectional total variation UTV has a clear physical basis. Stripe noise in satellite imagery exhibits a linear pattern with a regular spatial distribution, such as... Figure 3 As shown in (a), this noise exhibits strong directional correlation and significant structural consistency, such as Figure 3 As shown in (b) and (c), stripe noise exhibits a significant response in the horizontal derivative but has almost no effect on the vertical derivative. Therefore, the UTV model achieves noise separation through direction-selective regularization. For the stripe noise component, a smoothing constraint is applied along the stripe direction to promote noise separation; for the clean image component, structural integrity is maintained along the vertical direction. This process can be formalized into the following optimization problem:
[0071] (5)
[0072] In the formula, the directional derivative operator and The gradients are calculated for the horizontal and vertical directions, respectively.
[0073] Step 5: Establish a UTV-LRTA hybrid denoising model. Through low-rank tensor constraints combined with unidirectional total variational regularization, effective suppression of hybrid stripe noise in satellite imagery is achieved, such as... Figure 4 As shown, the details are as follows:
[0074] Step 5.1: Replace the traditional LRTA with a low-rank tensor approximation. The regularization term uses low-level unidirectional total variational UTV constraints instead of the conventional ones. The regularization term, through optimization of the fidelity term in formula (3), achieves the control of fringe noise components. Explicit elimination, then regarding The final function can be written as:
[0075] (6)
[0076] In the formula, , and This represents the regularization parameter used to adjust the weights of each item to balance the model;
[0077] Step 5.2, the core difficulty in solving formula (6) stems from The non-smoothness of the norm precludes the direct application of traditional gradient descent methods. Therefore, auxiliary variable techniques are used to reconstruct the original problem. , , By substitution, the original unconstrained optimization problem (6) is transformed into an equivalent constrained optimization form, thereby separating the non-smooth terms and maintaining the equivalence of the problem. The formula is as follows:
[0078] (7)
[0079] Step 5.3: Based on the Bregman iterative algorithm, the constrained optimization problem (7) is transformed into the following equivalent unconstrained minimization form:
[0080] (8)
[0081] in, , and Indicates the Bregman penalty parameter; , , It is a Bregman tensor.
[0082] This invention also provides a system for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation. This system is used to implement the aforementioned method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation. The system comprises a first unit to a fifth unit, wherein:
[0083] The first unit is to establish a random striped mixed noise image model;
[0084] The second unit transforms the core task of satellite image denoising into an unconstrained optimization problem and determines the objective function.
[0085] The third unit addresses random noise in satellite imagery by performing LRA operations on the noisy data matrix / tensor to suppress random noise.
[0086] The fourth unit addresses stripe noise in satellite imagery by employing a stripe noise removal model based on unidirectional total variation UTV, achieving stripe noise separation through directional selective regularization.
[0087] The fifth unit establishes a UTV-LRTA hybrid denoising model, which uses low-rank tensor constraints combined with unidirectional total variation regularization to suppress hybrid stripe noise in satellite imagery.
[0088] The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation.
[0089] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for suppressing random stripe noise in satellite remote sensing images based on low-rank tensor approximation.
[0090] Example
[0091] To comprehensively evaluate the performance of the proposed UTV-LRTA model, this embodiment employs a systematic experiment to validate the scheme. This system comprises two main parts: subjective evaluation, focusing on the effectiveness of stripe noise removal, the degree of detail preservation, and the improvement in overall image quality. On the synthetic dataset, the signal-to-noise ratio (SNR) is used as the core evaluation metric, defined as:
[0092] (9)
[0093] in, and These represent the clean image and the denoised image, respectively.
[0094] The advantage of the SNR metric is that it provides a numerical quality assessment standard, eliminates individual differences in subjective evaluation, and facilitates horizontal comparisons between different methods.
[0095] To fully verify the performance advantages of the UTV-LRTA hybrid model, this embodiment designed a systematic comparative experiment scheme, comparing the baseline model 1: the traditional low-rank tensor approximation (LRTA) model, the baseline model 2: the classical one-way total variation (UTV) model with the UTV-LRTA joint model proposed in this invention, and using publicly available remote sensing images, respectively, synthesized data with stripes and random mixed noise at different noise levels and real satellite image data.
[0096] 1. Composite image data
[0097] The UTV-LRTA model was systematically validated by constructing synthetic image data. The specific experimental design is as follows: Figure 5 Using the clean image shown in (d) as a baseline, random noise and stripe noise with specific parameters are manually added to generate an image like... Figure 5 The noisy test image shown in (a) is used as the model input. Based on this synthetic data, the signal-to-noise ratio (SNR) is used as the core quantitative indicator for evaluation; a higher SNR value indicates a better denoising effect. Table 1 shows the SNR values of each method on the synthetic dataset.
[0098] Table 1. Signal-to-noise ratio values of each method on the synthetic dataset.
[0099] method Synthetic dataset SNR UTV-LRTA 25.1551 Traditional UTV 24.1083 Traditional LRTA 17. 2677
[0100] Experimental results show that the traditional UTV and LRTA methods achieved SNR values of 24.1083 dB and 17.2677 dB, respectively, while the UTV-LRTA model proposed in this invention achieved 25.1551 dB, demonstrating a significant performance improvement. Figure 5 As shown, Figure 5 In the image, (a) represents the mixed noisy image, (b) represents the image denoised by the UTV-LRTV model, (c) represents the image denoised by the UTV-LRTV model, (d) represents the clean image, (e) represents the image denoised by the traditional UTV model, (f) represents the image denoised by the traditional UTV model, (g) represents the image denoised by the traditional LRTA model, and (h) represents the image denoised by the traditional LRTA model. Visual evaluation shows that the denoising result of UTV-LRTA is closest to the original clean image, and performs best in terms of edge sharpness and texture detail preservation; in contrast, traditional methods all show varying degrees of feature loss.
[0101] Based on a combination of quantitative indicators and qualitative analysis, the UTV-LRTA model demonstrates excellent denoising performance in synthetic data testing, validating the effectiveness of its method design.
[0102] 2. Real image data
[0103] Experimental verification results based on real satellite imagery data are as follows: Figure 6 As shown in the figure. The results show that there are significant differences in the performance of different denoising models. The denoising results of the traditional UTV model are as follows. Figure 6 As shown in (d), the stripe noise removal is incomplete, and noise residue remains. The denoising results of the traditional LRTA model are as follows... Figure 6 As shown in (f), a more pronounced stripe noise phenomenon exists. The noise residual plot analysis is as follows: Figure 6As shown in (e) and (g) above, both methods result in a loss of effective information, with the information loss in the LRTA model being particularly significant. Figure 6 As shown in (h) in the figure. The UTV-LRTA joint model exhibits comprehensive superiority as follows: Figure 6 As shown in (b) and (c), in terms of noise suppression, stripe noise was completely eliminated and random noise was effectively reduced; in terms of feature preservation, texture details were fully preserved and structural continuity was maintained. Experimental results show that a single model is difficult to achieve both denoising and feature preservation, and a joint modeling strategy is necessary. UTV-LRTA achieves the best overall performance.
[0104] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications 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 satellite remote sensing image random stripe noise suppression method based on low-rank tensor approximation, characterized in that, Includes the following steps: Step 1: Establish a random stripe mixed noise image model; Step 2: Transform the core task of satellite image denoising into an unconstrained optimization problem and determine the objective function; Step 3: To suppress random noise in satellite imagery, perform LRA operation on the noisy data matrix / tensor. Step 4: For stripe noise in satellite imagery, a stripe noise removal model based on unidirectional total variation UTV is used. Stripe noise separation is achieved through direction-selective regularization, as detailed below: Stripe noise in satellite imagery exhibits a linear pattern with a regular spatial distribution. This noise is directionally correlated and structurally consistent, showing a significant response in the horizontal derivative but almost no effect on the vertical derivative. Therefore, the UTV model achieves noise separation through directional selective regularization. Stripe noise components: Applying smoothing constraints along the stripe direction promotes noise separation. Cleaning image components: preserving structural integrity along the vertical direction; this process is formalized into the following optimization problem: （5） where the directional derivative operator and corresponds to the horizontal and vertical gradient computation, respectively; Step 5: Establish a UTV-LRTA hybrid denoising model. By combining low-rank tensor constraints with unidirectional total variation regularization, the hybrid stripe noise in satellite imagery is suppressed, as detailed below: Step 5.1, using low-rank tensor approximation LRTA instead regularization term, using low unidirectional total variation UTV constraint instead regularization term, explicit elimination of stripe noise components by optimizing the fidelity term of equation (3) The final function for is written as: （6） wherein, , and denote a regularization parameter for adjusting the weight of each term to balance the model; Step 5.2, replace the original unconstrained optimization problem (6) with the equivalent constrained optimization form, given by: , , Problem (6) is replaced with the equivalent constrained optimization form, given by: （7） Step 5.3: Based on the Bregman iterative algorithm, the constrained optimization problem (7) is transformed into the following equivalent unconstrained minimization form: （8） in, , and Indicates the Bregman penalty parameter; , , It is a Bregman tensor.
2. The method according to claim 1, wherein, The establishment of the random stripe mixed noise image model described in step 1 is as follows: The Gaofen satellite uses a pushbroom imaging method, where each column of data in the image is acquired by the same detector element of the CCD sensor continuously scanning along the orbital direction. Pixels in the same column will be subject to the same degree of noise interference. Therefore, the stripe noise present in each column of the image is regarded as an additional noise component. Based on the assumption that the noise components are added rather than multiplied, a mixed noise image model is established, and the mathematical expression is as follows: （1） wherein, and denote the noisy data tensor and the clean data tensor, respectively; denotes the stripe noise, denotes the random noise, wherein denote the column width, the row width, and the number of bands of the image, respectively.
3. The method according to claim 2, wherein, Step 2 transforms the core task of satellite image denoising into an unconstrained optimization problem, and determines the objective function as follows: The core task of satellite image denoising is to establish the inverse estimation process from the noisy observation data to the original clean data , which minimizes the objective function to achieve the optimal estimation of , which is given by （2） The denoising task in equation (2) is an ill-posed inverse problem. To address this problem, by introducing image tensor priors and stripe noise regularization constraints, the ill-posed inverse problem is transformed into an unconstrained optimization problem. The final objective function is expressed as a weighted combination of data fidelity terms and regularization terms, as shown in the formula: （3） The objective function in equation (3) consists of three key terms: the first term serves as the data fidelity term, ensuring the consistency between the reconstructed results and the original observation data; the second term... The first term is the image tensor regularization term, which eliminates unwanted artifacts in the reconstructed image by introducing prior constraints; the third term... This is a regularization term specifically designed for stripe noise, where the parameters... and The weight balance of these two regularization terms is controlled separately.
4. The method according to claim 3, wherein, Step 3 describes the implementation of LRA operations on the noisy data matrix / tensor to suppress random noise in satellite imagery, as detailed below: To address random noise in satellite imagery, random noise suppression is achieved by performing LRA operations on the noisy data matrix / tensor. This process can be formalized into the following optimization problem: （4）。 5. A satellite remote sensing image random stripe noise suppression system based on low-rank tensor approximation, characterized in that, This system is used to implement the satellite remote sensing image random stripe noise suppression method based on low-rank tensor approximation as described in any one of claims 1 to 4, the system comprising a first unit to a fifth unit, wherein: The first unit is to establish a random striped mixed noise image model; The second unit transforms the core task of satellite image denoising into an unconstrained optimization problem and determines the objective function. The third unit addresses random noise in satellite imagery by performing LRA operations on the noisy data matrix / tensor to suppress random noise. The fourth unit addresses stripe noise in satellite imagery by employing a stripe noise removal model based on unidirectional total variation UTV, achieving stripe noise separation through directional selective regularization. The fifth unit establishes a UTV-LRTA hybrid denoising model, which uses low-rank tensor constraints combined with unidirectional total variation regularization to suppress hybrid stripe noise in satellite imagery.
6. A mobile terminal comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the satellite remote sensing image random stripe noise suppression method based on low-rank tensor approximation as described in any one of claims 1 to 4.
7. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, it implements the steps in the satellite remote sensing image random stripe noise suppression method based on low-rank tensor approximation as described in any one of claims 1 to 4.