A method for detecting and removing stripe noise in spaceborne spectral images

Abstract

The present invention discloses a method for detecting and removing stripe noise in satellite-borne spectral images. The method comprises an image classification and characteristic analysis module, a stripe detection and analysis module (including a stripe detection unit, a characteristic analysis unit, and a degradation modeling unit), an image transformation module, and a soft threshold signal decomposition module. Based on independent analysis of the original space-spectral domain and the transform domain, the present invention comprehensively integrates the space-spectral domain analysis method and the transform domain analysis method to simultaneously detect and remove stripe noise. The method also fully analyzes the image characteristics and wavelet distribution characteristics, deploys an image decomposition method in the wavelet domain, effectively analyzes the stripe components, and retains image detail information. The present invention has the advantages of being adaptable to the complexity of spectral image sources and the diversity of application environments, taking into account the differences in different stripe noise distributions, and having strong generalization and robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing and image enhancement, and in particular to a method for detecting and removing stripe noise in satellite-borne spectral images, which is used to detect stripe noise in low-quality spectral images and improve the quality of spectral images obtained by satellite-borne imaging spectrometers that are interfered with by stripe noise. Background Art

[0002] Hyperspectral imagery boasts the advantages of "unified image and spectrum, nanometer-scale spectral resolution," enabling precise perception of the intrinsic discriminative properties of scene objects. It holds significant application value in national economic development, sustainable social development, and national defense security, and represents one of the technological high grounds for high-precision remote sensing. However, the noise problem in hyperspectral imagery has long been a key factor plaguing and hindering its application. Limited by imaging mechanisms, atmospheric effects, and component calibration, image degradation inevitably results in spatial information blurring, spectral absorption peak shifts, junk bands, and band distortion. These degradations lead to poor output quality and severe information loss, significantly limiting the accuracy and reliability of subsequent interpretation and analysis. Therefore, effectively eliminating the complex noise contained in images and reconstructing missing information is of great research significance for improving image interpretability and applicability. It can provide quality assurance and lay the data foundation for subsequent practical applications such as image interpretation and analysis.

[0003] The key to removing spectral image banding lies in the reasonable modeling of complex noise and the full mining and utilization of prior information in the image, so as to ensure that useful information in the image is reconstructed while removing noise and maintaining the discriminative spectral information of the image. In the early days, spectral image denoising methods borrowed from traditional two-dimensional image denoising modeling methods: representative technologies include methods based on filter design and statistical modeling, such as histogram matching or power spectrum filtering. This type of method is relatively mature in theory and has a simple model, but its adaptability is poor and it cannot handle more complex noise distribution situations; while the machine learning-based method analyzes and models the spectral image, introduces regularized priors such as total variation and low-rank sparsity, separates image information and banding noise, and can better retain and remove banding noise, but this type of method has poor ability to preserve the spatial-spectral eigenstructure.

[0004] On the other hand, because spectral image striping noise has complex sources and diverse distribution patterns, determining the distribution and situation of the striping is also crucial, allowing for the precise and effective deployment of de-striping methods. Furthermore, when designing de-striping methods, it is necessary not only to effectively analyze the distribution characteristics of the striping but also to consider the spatial-spectral characteristics of the spectral image. Therefore, it is necessary to design a striping noise detection and removal method that is broadly applicable and effectively preserves image information. Summary of the Invention

[0005] In view of the defects of the prior art, the present invention provides a method for detecting and removing stripe noise of satellite-borne spectral images, which solves the defects of the prior art.

[0006] The present invention fully considers the spatial-spectral characteristics and stripe distribution differences of spectral images, takes into account the noise removal ability and information retention ability of the de-striping method, and proposes a spectral image stripe noise detection and removal method based on transform domain analysis.

[0007] In order to achieve the above object of the invention, the technical solution adopted by the present invention is as follows:

[0008] A method for detecting and removing stripe noise in a spaceborne spectral image comprises the following steps:

[0009] Step 1: Construct a spectral image spatial-spectral characteristic analysis module. Classify the spectral image according to the imaging platform to determine the image prior and perform spatial-spectral characteristic analysis.

[0010] Step 2: Construct a band detection and analysis module and perform mathematical modeling of the image degradation containing bands. This section consists of three units: band detection unit, characteristic analysis unit, and degradation modeling unit.

[0011] Step 3: Construct an image transformation module, convert the image into the transformation domain space, extract the strip components and image information components, and select different transformation methods according to different spatial-spectral characteristic data and strip characteristic distribution. The present invention adopts wavelet transformation.

[0012] Step 4: Construct a noise removal module based on the transformation results, analyze the differences between each sub-band signal and strip of the wavelet transform, and design a wavelet domain soft threshold signal decomposition method to strip off the strip components and retain the useful information of the image. Then perform inverse wavelet transform and output the reconstructed image.

[0013] Furthermore, step 1 includes the following sub-steps:

[0014] Step 1.1: Classify spectral images: From the spectral dimension, images are classified based on the number and continuity of bands. That is, the spectral resolution is divided from low to high into: single-band panchromatic images, four-band multispectral images, eight-band multispectral images, other medium-resolution spectral images, and hyperspectral images. From the spatial dimension, images are classified based on spatial resolution and the richness of spatial information. Specifically, they are divided into: meters and below, 10-30 meters, 30 meters, and above 30 meters to hundreds of kilometers, with the richness of spatial detail information decreasing in order.

[0015] Step 1.2: Spatial-spectral characteristic analysis: Based on the imaging mechanism of spectral images, there are constraints on the spatial resolution and inter-spectral resolution of spectral images. Based on the image classification results, images with relatively low spectral resolution have richer spatial detail information. Therefore, when designing a de-banding method, particular consideration should be given to preserving spatial detail information, without having to consider issues such as spectral correlation and spectral continuity. For hyperspectral images, however, considerations such as spectral correlation, non-local similarity in the spatial dimension, and the distribution characteristics of the joint spatial-spectral dimension are necessary. Spatial-spectral characteristic analysis of spectral images is the foundation for designing de-banding methods, helping to obtain appropriate prior information during the algorithm design process, thereby achieving the goal of removing stripe noise and preserving intrinsic information.

[0016] Furthermore, step 2 includes the following sub-steps:

[0017] Step 2.1: The stripe detection unit adopts a comprehensive detection method, including spatial spectrum dimension distribution detection and transform domain detection, without the need for reference images. Due to the large differences in the distribution of spectral image characteristics and the very complex stripe distribution, in order to improve the effectiveness of the detection method, a comprehensive detection mode is proposed based on the single detection mode. The spatial spectrum dimension distribution detection determines the direction and band of the stripe through differential calculation. That is, for the spectral image

[0018]

[0019] Where N, M, and D represent the number of rows, columns, and bands in the image; n, m, and d represent the current number of rows, columns, and bands in the image. By observing the calculation results, it is clear that the position and direction of the stripes are different from each other, as well as the differences between the bands.

[0020] Transform domain detection includes Fourier transform detection and wavelet transform detection. Fourier transform adopts two-dimensional transform mode, namely

[0021]

[0022] Where u and v represent frequency domain variables, and x and y represent spatial domain variables. By observing the Fourier transform results, we can detect whether there is stripe noise in the image. When stripe noise is present, the energy of the Fourier transform image is more concentrated.

[0023] Wavelet transform detection uses a two-dimensional discrete wavelet transform to decompose the image using wavelet basis functions. This is implemented using a filter bank to generate baseband, horizontal subbands, vertical subbands, and diagonal subbands. The horizontal subbands, vertical subbands, and diagonal subbands can capture horizontal, vertical, and diagonal stripe noise, respectively.

[0024] Step 2.2: The characteristic analysis unit combines prior information with statistical analysis to first determine the noise sources based on the imaging observation process: ① due to the imaging environment, and ② due to misalignment or damage of system components. It then determines the noise directional characteristics, horizontal stripe continuity, vertical stripe periodicity, and inter-band correlation.

[0025] Step 2.3: The degradation modeling unit performs mathematical correlation modeling on the image and noise based on the strip detection results and characteristic analysis results. Affected by stripe noise The output spectrum image after interference is The degradation process can be modeled as follows:

[0026]

[0027] In the formula represents additive noise, and f represents the degeneration function, which can be linear or nonlinear depending on the noise situation.

[0028] Furthermore, step 3 includes the following sub-steps:

[0029] Step 3.1: Input the image to be restored and adaptively select the transformation model: M = 1 defaults to a two-dimensional wavelet transform, and M = 2 indicates a three-dimensional wavelet transform. For panchromatic and multispectral images, select a two-dimensional wavelet transform, while for hyperspectral images, select a three-dimensional wavelet transform. Set the wavelet basis function type (the default is Haar wavelet),

[0030] T = 'Haar', 'Daubechies (dbN)', 'Mexican Hat (mexh)', 'Morlet', 'Meyer'. Determine whether to perform destriping in blocks and select the multi-scale decomposition level L based on the image size.

[0031] Step 3.2: Perform wavelet transform on the input image and downsample it. For a two-dimensional wavelet transform, the l-th level wavelet transform outputs four subbands: (L and H are low-pass and high-pass filters, respectively); for three-dimensional wavelet transform, eight sub-wavelet components are output: and When L>1, continue to baseband and The wavelet transform is repeatedly performed until l=L.

[0032] Furthermore, step 4 includes the following sub-steps:

[0033] Step 4.1: Based on the stripe direction detection results of step 2 and the wavelet transform results of step 3.2, select the wavelet subband with stripe noise distribution to prepare for noise removal. Set the stripe component and image information component control weight parameter λ>0.

[0034] Step 4.2: Assume that the selected wavelet subband is The designed signal decomposition method is as follows: Decomposed into low-rank strip component L and sparse signal component S:

[0035]

[0036] Where s n is the nth column of S, ω n =1 / ||f n ||2,||·|| * represents the nuclear norm, and λ is the regularization factor. The alternating direction multiplication method is introduced to solve the above test, and we get and Two sub-questions:

[0037] The subproblem is solved by the soft threshold shrinkage operator Obtain:

[0038]

[0039] Where τ is the contraction operator and x represents a variable.

[0040] The subproblem can be solved by using the singular value contraction operator Solution:

[0041]

[0042] Where X=UΣV H is the singular value decomposition of X,

[0043] Step 4.3: Replace the subband with the output S component Then perform inverse wavelet transform step by step and output the reconstructed image

[0044] Compared with the prior art, the advantages of the present invention are:

[0045] The complexity of spectral image sources and the diversity of application environments are fully considered, while taking into account the differences in the distribution of different stripe noise; based on the independent analysis of the original space-spectral domain and transform domain, the space-spectral domain analysis method and the transform domain analysis method are comprehensively integrated to achieve the purpose of removing stripe noise while detecting it; the image characteristics and wavelet distribution characteristics are fully analyzed, and the image decomposition method is deployed in the wavelet domain to effectively analyze the stripe components and retain the image detail information; a method for detecting and removing stripe noise in spaceborne spectral images is proposed, which is applicable to different imaging platforms and application environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of the method for detecting and removing stripe noise in spaceborne spectral images of the present invention;

[0047] Figure 2 It is a spectral image classification unit based on the imaging platform constructed by the present invention;

[0048] Figure 3 It is a complex strip noise detection, identification, analysis and modeling unit constructed by the present invention;

[0049] Figure 4 This is a flow chart of target detection and recognition based on an aerial image dataset using the airborne optoelectronic video target intelligent detection and recognition method proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.

[0051] like Figure 1 As shown, a method for detecting and removing stripe noise in a spaceborne spectral image comprises the following steps:

[0052] Step 1: Construct a spectral image spatial-spectral characteristic analysis module. Classify the spectral image according to the imaging platform to determine the image prior and perform spatial-spectral characteristic analysis.

[0053] Step 2: Construct a band detection and analysis module and perform mathematical modeling of the image degradation containing bands. This section consists of three units: band detection unit, characteristic analysis unit, and degradation modeling unit.

[0054] Step 3: Construct an image transformation module, convert the image into the transformation domain space, extract the strip components and image information components, and select different transformation methods according to different spatial-spectral characteristic data and strip characteristic distribution. The present invention adopts wavelet transformation.

[0055] Step 4: Construct a noise removal module based on the transformation results, analyze the differences between each sub-band signal and strip of the wavelet transform, and design a wavelet domain soft threshold signal decomposition method to strip off the strip components and retain the useful information of the image. Then perform inverse wavelet transform and output the reconstructed image.

[0056] Specifically, step 1 includes the following steps:

[0057] Step 1.1: If Figure 2 As shown in the figure, a spectral image classification unit based on the imaging platform is constructed. From the spectral dimension, images are classified based on the number and continuity of bands. That is, the spectral resolution is divided into the following categories from low to high: single-band panchromatic imagery, four-band multispectral imagery, eight-band multispectral imagery, other medium-resolution spectral imagery, and hyperspectral imagery. From the spatial dimension, images are classified based on spatial resolution and spatial information richness. Specifically, they are divided into the following categories: meter level and below, 10-30 meter level, 30 meter level, and above 30 meters to hundreds of kilometers level, with the richness of spatial detail information decreasing in descending order.

[0058] Step 1.2: Spatial-spectral characteristic analysis: Based on the imaging mechanism of spectral images, there are constraints on the spatial resolution and inter-spectral resolution of spectral images. Based on the image classification results, images with relatively low spectral resolution have richer spatial detail information. Therefore, when designing a de-banding method, particular consideration should be given to preserving spatial detail information, without having to consider issues such as spectral correlation and spectral continuity. For hyperspectral images, however, considerations such as spectral correlation, non-local similarity in the spatial dimension, and the distribution characteristics of the joint spatial-spectral dimension are necessary. Spatial-spectral characteristic analysis of spectral images is the foundation for designing de-banding methods, helping to obtain appropriate prior information during the algorithm design process, thereby achieving the goal of removing stripe noise and preserving intrinsic information.

[0059] Furthermore, if Figure 3 As shown, step 2 includes the following sub-steps:

[0060] Step 2.1: The stripe detection unit adopts a comprehensive detection method, including spatial spectrum dimension distribution detection and transform domain detection, without the need for reference images. Due to the large differences in the distribution of spectral image characteristics and the very complex stripe distribution, in order to improve the effectiveness of the detection method, a comprehensive detection mode is proposed based on the single detection mode. The spatial spectrum dimension distribution detection determines the direction and band of the stripe through differential calculation. That is, for the spectral image

[0061]

[0062] Where N, M, and D represent the number of rows, columns, and bands in the image; n, m, and d represent the current number of rows, columns, and bands in the image. By observing the calculation results, it is clear that the position and direction of the stripes are different from each other, as well as the differences between the bands.

[0063] Transform domain detection includes Fourier transform detection and wavelet transform detection. Fourier transform adopts two-dimensional transform mode, namely

[0064]

[0065] Where u and v represent frequency domain variables, and x and y represent spatial domain variables. By observing the Fourier transform results, we can detect whether there is stripe noise in the image. When stripe noise is present, the energy of the Fourier transform image is more concentrated.

[0066] Wavelet transform detection uses a two-dimensional discrete wavelet transform to decompose the image using wavelet basis functions. This is implemented using a filter bank to generate baseband, horizontal subbands, vertical subbands, and diagonal subbands. The horizontal subbands, vertical subbands, and diagonal subbands can capture horizontal, vertical, and diagonal stripe noise, respectively.

[0067] Step 2.2: The characteristic analysis unit combines prior information with statistical analysis to first determine the noise sources based on the imaging observation process: ① due to the imaging environment, and ② due to misalignment or damage of system components. It then determines the noise directional characteristics, horizontal stripe continuity, vertical stripe periodicity, and inter-band correlation.

[0068] Step 2.3: The degradation modeling unit performs mathematical correlation modeling on the image and noise based on the strip detection results and characteristic analysis results. Affected by stripe noise The output spectrum image after interference is The degradation process can be modeled as follows:

[0069]

[0070] In the formula represents additive noise, and f represents the degeneration function, which can be linear or nonlinear depending on the noise situation.

[0071] Furthermore, step 3 includes the following sub-steps:

[0072] Step 3.1: Input the image to be restored and adaptively select the transform model: M = 1 defaults to a 2D wavelet transform, while M = 2 indicates a 3D wavelet transform. For panchromatic and multispectral images, select a 2D wavelet transform, while for hyperspectral images, select a 3D wavelet transform. Set the wavelet basis function type (default is Haar wavelet): T = 'Haar', 'Daubechies (dbN)', 'Mexican Hat (mexh)', 'Morlet', or 'Meyer'. Determine whether to perform block destriping and select the multiscale decomposition level L based on the image size.

[0073] Step 3.2: Perform wavelet transform on the input image and downsample it. For a two-dimensional wavelet transform, the l-th level wavelet transform outputs four subbands: (L and H are low-pass and high-pass filters, respectively); for three-dimensional wavelet transform, eight sub-wavelet components are output: and When L>1, continue to baseband and The wavelet transform is repeatedly performed until l=L.

[0074] Furthermore, step 4 includes the following sub-steps: constructing a noise removal module based on the transformation results, analyzing the differences between the sub-band signals and strips of the wavelet transform, and designing a wavelet domain soft threshold signal decomposition method to strip off the strip components and retain the useful information of the image, and then performing an inverse wavelet transform to output the reconstructed image.

[0075] Step 4.1: Based on the stripe direction detection results of step 2 and the wavelet transform results of step 3.2, select the wavelet subband with stripe noise distribution to prepare for noise removal. Set the stripe component and image information component control weight parameter λ>0.

[0076] Step 4.2: Assume that the selected wavelet subband is The designed signal decomposition method is as follows: Decomposed into low-rank strip component L and sparse signal component S:

[0077]

[0078] Where s n is the nth column of S, ω n =1 / ||f n ||2,||·|| * represents the nuclear norm, and λ is the regularization factor. The alternating direction multiplication method is introduced to solve the above test, and we get and Two sub-questions:

[0079] The subproblem is solved by the soft threshold shrinkage operator:

[0080]

[0081] Where τ is the contraction operator and x represents a variable.

[0082] The subproblem can be solved by the singular value contraction operator:

[0083]

[0084] Where X=UΣV H is the singular value decomposition of X,

[0085] Step 4.3: Replace the subband with the output S component Then perform inverse wavelet transform step by step and output the reconstructed image

[0086] Example 1

[0087] The following is the implementation process of the image acquired by the Medium Resolution Spectroscopic Imager (MERSI) of Fenyun-3D satellite: Figure 4 shown.

[0088] Step 1: Input the atmospherically and geometrically corrected captured image and normalize it. Based on the imaging platform, the image is input into the constructed spectral image spatial-spectral characteristic analysis module. The spectral image is categorized based on the imaging platform. The image is a medium-resolution spectral image with 25 bands in the spectral dimension; there are 6 channels with a spatial resolution of 250 meters and 19 channels with a spatial resolution of 1000 meters. Inter-band continuity is weak, while spatial homogeneity is strong. Furthermore, because the image is used for dynamic monitoring of large-scale environments such as ocean, land, and atmosphere, detailed texture information is less emphasized.

[0089] Step 2: Use spatial-spectral domain difference calculation combined with wavelet domain transform and Fourier transform to determine the image striping situation, filter out the bands containing stripes, analyze the striping characteristics, and establish a mathematical model.

[0090] Step 3: Construct an image transformation model and set parameter values ​​based on the results of image spatial and spectral characteristics analysis and stripe noise analysis. Convert the output image to wavelet space and analyze which sub-bands contain stripe components. For sub-bands without stripe components, temporarily retain them; for sub-bands containing stripe components, proceed to step 4.

[0091] Step 4: Perform wavelet domain soft threshold signal decomposition on the sub-band containing strips, strip the strip components, and retain the useful information of the image. If the pixel value of the decomposed component is low, further enhancement can be performed. Then, the sub-band and image components retained in step 3 are input into the wavelet inverse transform module together for image reconstruction, and finally the reconstructed image is output.

[0092] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for detecting and removing stripe noise in satellite-borne spectral images, characterized in that: The following steps are involved: Step 1: Construct a spectral image spatial-spectral characteristic analysis module; classify the spectral image according to the imaging platform, thereby determining the image prior and performing spatial-spectral characteristic analysis; Step 2: Construct a band detection and analysis module and perform mathematical modeling of the degradation of images containing bands. This module consists of three units: a band detection unit, a characteristic analysis unit, and a degradation modeling unit. The following sub-steps are included: Step 2.1: The strip detection unit adopts a comprehensive detection method; the spatial spectrum distribution detection determines the direction and band of the strip by differential calculation; that is, for the spectral image Where N, M, and D represent the number of rows, columns, and bands of the image; n, m, and d represent the current number of rows, columns, and bands of the image; by observing the calculation results, the position, direction, and difference between bands of the stripes can be obtained; Transform domain detection includes Fourier transform detection and wavelet transform detection. Fourier transform adopts a two-dimensional transform method, namely: Where u and v represent frequency domain variables; x and y represent spatial domain variables. By observing the Fourier transform results, we can detect whether the image has stripe noise. When stripe noise exists, the energy of the Fourier transform image is concentrated. Wavelet transform detection uses two-dimensional discrete wavelet transform to decompose the image through wavelet basis functions. In practice, it is implemented through filter banks to generate baseband, horizontal subband, vertical subband and diagonal subband. Among them, the horizontal subband, vertical subband and diagonal subband capture the stripe noise in the horizontal direction, vertical direction and diagonal direction respectively. Step 2.2: The characteristic analysis unit combines prior information with statistical analysis to first determine the source of noise based on the imaging observation process: ① caused by the imaging environment, ② caused by misalignment or damage of system components. It then determines the noise directional characteristics, horizontal strip continuity, vertical strip periodicity, and inter-band correlation. Step 2.3: The degradation modeling unit performs mathematical correlation modeling on the image and noise based on the strip detection results and characteristic analysis results; that is, for the clean image Affected by stripe noise The output spectrum image after interference is: The degradation process is modeled as follows: In the formula represents additive noise, f represents the degeneration function, which is linear or nonlinear according to the noise situation; Step 3: Construct an image transformation module, convert the image into the transform domain space, extract the strip components and image information components according to different spatial-spectral characteristic data and strip characteristic distribution; Step 4: Construct a noise removal module based on the transformation results, analyze the differences between each sub-band signal and strip in the wavelet transform, design a wavelet domain soft threshold signal decomposition method, strip off the strip components, retain the useful information of the image, and then perform inverse wavelet transform to output the reconstructed image.

2. The method for detecting and removing stripe noise in satellite-borne spectral images according to claim 1, characterized in that: Step 1 includes the following sub-steps: Step 1.1: Classify the spectral image: From the spectral dimension, images are divided based on the number and continuity of bands. That is, the spectral resolution is divided from low to high into: single-band panchromatic images, four-band multispectral images, eight-band multispectral images, other medium-resolution spectral images, and hyperspectral images. From the perspective of spatial dimension, the image classification is based on spatial resolution and spatial information richness, specifically divided into: meter level and below, 10-30 meter level, 30 meter level and above 30 meters to hundreds of kilometers level, with the richness of spatial detail information decreasing in descending order; Step 1.2: Spatial-spectral characteristics analysis: When removing stripes, the spatial detail information is preserved without considering the correlation characteristics of the spectral dimension and the spectral continuity problem. For hyperspectral images, the correlation of the spectral dimension, the non-local similarity of the spatial dimension, and the distribution characteristics of the spatial-spectral joint dimension are considered. The de-striation method is designed by analyzing the spatial-spectral characteristics of spectral images.

3. The method for detecting and removing stripe noise in satellite-borne spectral images according to claim 1, characterized in that: Step 3 includes the following sub-steps: Step 3.1: Input the image to be restored and adaptively select the transformation model: M = 1 defaults to a two-dimensional wavelet transform, while M = 2 indicates a three-dimensional wavelet transform. For panchromatic and multispectral images, select a two-dimensional wavelet transform, while for hyperspectral images, select a three-dimensional wavelet transform. Set the wavelet basis function type, which defaults to Haar wavelet: T = 'Haar', 'Daubechies (dbN)', 'Mexican Hat (mexh)', 'Morlet', 'Meyer'; determine whether to perform destriping in blocks and select the multi-scale decomposition level L based on the image size; Step 3.2: Perform wavelet transform on the input image and downsample it. For a two-dimensional wavelet transform, the l-th level wavelet transform outputs four subbands: L and H are low-pass and high-pass filters respectively; for three-dimensional wavelet transform, eight sub-wavelet components are output: and When L>1, continue to baseband and The wavelet transform is repeatedly performed until l=L.

4. The method for detecting and removing stripe noise in satellite-borne spectral images according to claim 1, characterized in that: Step 4 includes the following sub-steps: Step 4.1: Based on the stripe direction detection result of step 2 and the wavelet transform result of step 3, select the wavelet subband with stripe noise distribution to prepare for noise removal; set the stripe component and image information component control weight parameter λ>0; Step 4.2: Assume that the selected wavelet subband is The designed signal decomposition method is as follows: Decomposed into low-rank strip component L and sparse signal component S: Where s n is the nth column of S, ω n =1 / ||f n ||2,||·|| * represents the nuclear norm, λ is the regularization factor; the alternating direction multiplication method is introduced to solve the above test, and we get and Two sub-questions: The subproblem is solved by the soft threshold shrinkage operator Obtain: Where τ is the contraction operator and x represents the variable; The subproblem is solved by the singular value contraction operator Solution: Where X=UΣV H is the singular value decomposition of X, Step 4.3: Replace the subband with the output S component Then perform inverse wavelet transform step by step and output the reconstructed image

Images