A remote sensing satellite image stripe radiation correction method based on noise intensity estimation

Abstract

The application discloses a remote sensing satellite image stripe radiation correction method based on noise intensity estimation, and belongs to the technical field of satellite remote sensing. Column mean values of a plurality of single spectrum images are calculated, and column mean values of the plurality of single spectrum images after preliminary denoising are calculated. The column mean values of the plurality of single spectrum images are subtracted from the column mean values of the plurality of single spectrum images after preliminary denoising to obtain preliminary corrected noise intensity. The preliminary corrected noise intensity is constrained to obtain noise intensity. The column mean values of the plurality of single spectrum images are subtracted from the noise intensity to obtain the plurality of single spectrum images after preliminary correction. The plurality of single spectrum images after preliminary correction are filtered, and the plurality of single spectrum images after preliminary correction are subtracted from the plurality of single spectrum images after preliminary correction after filtering to extract noise images. Column mean values and column median values of the noise images are calculated, and the absolute values of the column mean values and the column median values are compared to select noise intensity. The plurality of single spectrum images after preliminary correction are subtracted from the noise intensity to obtain denoised images.

Description

Technical Field

[0001] This invention relates to the field of satellite remote sensing technology, and more specifically to a method for stripe radiance correction of remote sensing satellite images based on noise intensity estimation. Background Technology

[0002] With technological advancements, remote sensing satellite imagery has been widely applied in various fields, including environmental monitoring, agriculture, urban planning and management, and disaster early warning and relief. Remote sensing satellite images capture data on the Earth's surface reflection and radiation through satellite sensors, which is then analyzed and transformed into visualized image data. However, due to limitations in satellite imaging system performance and the influence of the external environment during image capture, different sensors may exhibit inconsistent responses to the same radiation energy during scanning, resulting in stripe noise in the images. This severely impacts visual quality and subsequent image processing. Therefore, there is an urgent need to develop relative radiometric correction algorithms for stripe noise in remote sensing satellite images to improve the quality and usability of image data.

[0003] Traditional relative radiometric correction algorithms based on noise mechanisms can be categorized into three types: those based on scene statistical features, those based on filtering, and those based on optimization. Methods based on scene statistical features, such as moment matching and histogram matching, are highly sensitive to the statistical distribution of different sub-scenes within the same band. When the stripe noise is irregular and the scene response distribution is uneven, the correction effect becomes unstable. Filtering-based methods include spatial domain filtering and frequency domain filtering. These methods often fail to accurately identify ground features and noise characteristics during stripe correction, thus filtering out useful information of the same frequency components during denoising, resulting in significant loss of image detail in the corrected remote sensing image. Optimization-based correction methods describe the denoising process as an ill-posed optimization problem of recovering a clean image from a degraded image. They utilize specific priors, such as sparsity and low rank, to solve the optimization problem for stripe denoising. However, because these problems require accurate characterization of the noise's prior features, they have limitations when processing satellite images with complex noise characteristics, leading to under-correction or over-correction.

[0004] In recent years, with the development of artificial intelligence technology, relative radiometric correction algorithms based on deep learning have been widely used. However, deep learning methods require clean, noise-free images as samples, and it is difficult to obtain a large number of such satellite images in actual industrial applications, which limits the development of deep learning models. Some studies have used artificially constructed noisy sample sets for model training, but these sets cannot accurately simulate the characteristics of actual noise, resulting in networks that do not achieve optimal performance. Furthermore, the presence of convolutional kernels in deep learning network models can cause loss of detail and reduced resolution in the corrected images.

[0005] In summary, existing mechanism-based and deep learning-based methods are both unable to achieve accurate denoising of stripe noise without losing image details. Summary of the Invention

[0006] This invention solves the problem that existing stripe correction models cannot achieve accurate denoising of satellite image stripes without loss of image details.

[0007] The present invention discloses a method for correcting stripe radiometric patterns in remote sensing satellite images based on noise intensity estimation, comprising the following steps:

[0008] Step S1: Multiple spectral images are separated into several single-spectral-band images;

[0009] Step S2: Calculate the column mean of several single-spectral-band images, and perform preliminary denoising on the column mean of several single-spectral-band images to obtain the column mean of several single-spectral-band images after preliminary denoising.

[0010] Step S3: The difference between the column mean of several single-spectral-band images and the column mean of several single-spectral-band images after preliminary denoising is obtained to obtain the preliminary corrected noise intensity. The preliminary corrected noise intensity is constrained to obtain the noise intensity.

[0011] Step S4: The difference between the column mean of several single-spectral-band images and the noise intensity is used to obtain several preliminarily corrected single-spectral-band images;

[0012] Step S5: Filter the pre-corrected single-spectral-band images, and extract the noise image by subtracting the pre-corrected single-spectral-band images from the filtered pre-corrected single-spectral-band images.

[0013] Step S6: Calculate the column mean and column median of the noisy image respectively, and select the value with the largest absolute value as the noise intensity. Difference between the preliminarily corrected spectral images and the noise intensity to obtain the denoised image.

[0014] Furthermore, in one embodiment of the present invention, in step S2, the preliminary denoising of the column mean of several single-spectral-band images to obtain the column mean of several preliminarily denoised single-spectral-band images specifically involves:

[0015]

[0016] Among them, i smooth,k This represents the column mean of several single-spectral-band images after preliminary denoising. Here, m is the smoothing coefficient, and i is a constant. k Let be the column mean of several single-spectral-band images, k = 1, ..., N, where N is the image width.

[0017] Furthermore, in one embodiment of the present invention, step S3, which involves obtaining the preliminary corrected noise intensity by taking the difference between the column mean of several single-spectral-band images and the column mean of the several single-spectral-band images after preliminary denoising, specifically includes:

[0018] i noise,k =i k -i smooth,k ;

[0019] Among them, i noise,k For the initial correction of noise intensity, i k Let i be the column mean of a number of single-spectral-band images, k = 1, ..., N, where N is the image width and i is the column mean of the image. smooth,k This represents the column mean of several single-spectral-band images after initial denoising.

[0020] Furthermore, in one embodiment of the present invention, in step S3, the noise intensity is obtained by constraining the initially corrected noise intensity, specifically as follows:

[0021]

[0022] Among them, i thres,k Let i be the noise intensity, and threshold be the threshold value. noise,k The noise intensity is for preliminary correction.

[0023] Furthermore, in one embodiment of the present invention, in step S4, the difference between the column mean of the plurality of single-spectral-band images and the noise intensity is used to obtain the plurality of preliminarily corrected single-spectral-band images, specifically as follows:

[0024] I SG,k =i k -i thres,k ;

[0025] Among them, I SG For the preliminary corrected spectral band images, i k Let i be the column mean of a number of single-spectral-band images, k = 1, ..., N, where N is the image width and i is the column mean of the image. thres,k Noise intensity.

[0026] Furthermore, in one embodiment of the present invention, step S5, specifically filtering the preliminarily corrected several single-spectral-band images, specifically involves:

[0027]

[0028] Among them, I bif,p This is the pre-corrected single-spectral-band image at position p after filtering. S is the size of the kernel function's spatial domain, r is the pixel range domain, q represents all pixels in the kernel function except the center point, and I... SG,qW represents the center pixel of the current kernel function. bif,p For normalization function, and These are the spatial domain weighting function and the pixel range domain weighting function, respectively.

[0029] Furthermore, in one embodiment of the present invention, in step S5, the noise image is extracted by subtracting the pre-corrected plurality of single-spectral-band images from the filtered pre-corrected plurality of single-spectral-band images, specifically as follows:

[0030] I noise =I SG -I bif ;

[0031] Among them, I noise For noisy images, I SG For the preliminary corrected spectral band images, I bif These are several single-spectral images after preliminary correction following filtering and denoising.

[0032] Furthermore, in one embodiment of the present invention, in step S6, selecting the value with the largest absolute value as the noise intensity specifically means:

[0033]

[0034] Among them, X noise,k X represents noise intensity. mean,k X is the column mean of the noisy image. median,k The column values ​​represent the values ​​in the noisy image.

[0035] Furthermore, in one embodiment of the present invention, step S6, which involves taking the difference between the preliminarily corrected image values ​​of several spectral bands and the noise intensity to obtain the denoised image, specifically includes:

[0036] I clean,k =I SG,k -X noise,k ;

[0037] Among them, X noise,k For noise intensity, I SG For the preliminary corrected spectral band images, I clean,k This is the image after denoising.

[0038] This invention solves the problem that existing stripe correction models cannot achieve accurate denoising of satellite image stripes without loss of image detail. Specific beneficial effects include:

[0039] 1. The present invention provides a method for radiometric correction of stripes in remote sensing satellite images based on noise intensity estimation. Existing technologies, based on mechanisms and deep learning models, cannot achieve accurate denoising of satellite image stripes without loss of image detail. To address this problem, this embodiment, for stripe noise images with different characteristics, eliminates the need for prior estimation. Through coarse-to-fine noise intensity estimation and calibration steps, it achieves radiometric correction of remote sensing images with complex stripe noise characteristics. Furthermore, the correction process is performed using column mean values, preserving image detail and reducing computational complexity. To avoid overcorrection during radiometric correction, mechanistic constraints are added to the correction process, resulting in higher image quality after correction. This process effectively avoids the technical problem of stripe correction methods failing to accurately denoise satellite images.

[0040] 2. The fringe radiometric correction method for remote sensing satellite images based on noise intensity estimation described in this invention addresses the problem that existing statistical methods require accumulating image data for correction, which is time-consuming and resource-intensive. To solve this problem, this invention performs correction on a single image basis, analyzing the statistical characteristics of each image for fringe correction. This eliminates the need to accumulate historical image samples, significantly saving processing time and resources. Therefore, the method described in this invention can serve as a supplement to existing statistical methods for relative radiometric correction.

[0041] The present invention provides a method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation, which is used to process multispectral and panchromatic images of remote sensing satellites and is unaffected by the type of ground cover.

[0042] The present invention provides a method for stripe radiometric correction of remote sensing satellite images based on noise intensity estimation. It offers an efficient and accurate algorithm for relative radiometric correction of remote sensing images, with good applicability and robustness, and can handle stripe noise from different satellite data. Attached Figure Description

[0043] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0044] Figure 1 This is a flowchart of the remote sensing satellite image stripe radiance correction method based on noise intensity estimation as described in the specific implementation method;

[0045] Figure 2 This is a front view of the remote sensing image stripe noise relative radiometric correction described in the specific implementation method;

[0046] Figure 3 This is a rear view of the remote sensing image stripe noise relative radiometric correction described in the specific implementation method;

[0047] Figure 4This is a front view of the remote sensing image stripe noise relative radiometric correction described in the specific implementation method;

[0048] Figure 5 This is a rear view of the remote sensing image stripe noise relative radiometric correction described in the specific implementation method. Detailed Implementation

[0049] Various embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. The embodiments described with reference to the drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0050] The method for correcting stripe radiometric patterns in remote sensing satellite images based on noise intensity estimation, as described in this embodiment, is characterized by including the following steps:

[0051] Step S1: Multiple spectral images are separated into several single-spectral-band images;

[0052] Step S2: Calculate the column mean of several single-spectral-band images, and perform preliminary denoising on the column mean of several single-spectral-band images to obtain the column mean of several single-spectral-band images after preliminary denoising.

[0053] Step S3: The difference between the column mean of several single-spectral-band images and the column mean of several single-spectral-band images after preliminary denoising is obtained to obtain the preliminary corrected noise intensity. The preliminary corrected noise intensity is constrained to obtain the noise intensity.

[0054] Step S4: The difference between the column mean of several single-spectral-band images and the noise intensity is used to obtain several preliminarily corrected single-spectral-band images;

[0055] Step S5: Filter the pre-corrected single-spectral-band images, and extract the noise image by subtracting the pre-corrected single-spectral-band images from the filtered pre-corrected single-spectral-band images.

[0056] Step S6: Calculate the column mean and column median of the noisy image respectively, and select the value with the largest absolute value as the noise intensity. Difference between the preliminarily corrected spectral images and the noise intensity to obtain the denoised image.

[0057] In this embodiment, step S2, which involves obtaining the column mean of several single-spectral-band images after preliminary denoising by using the column mean of several single-spectral-band images, specifically involves:

[0058]

[0059] Among them, i smooth,k This represents the column mean of several single-spectral-band images after preliminary denoising. Here, m is the smoothing coefficient, and i is a constant. kLet be the column mean of several single-spectral-band images, k = 1, ..., N, where N is the image width.

[0060] In this embodiment, step S3, which involves obtaining the preliminary corrected noise intensity by taking the difference between the column mean of several single-spectral-band images and the column mean of the several single-spectral-band images after preliminary denoising, specifically includes:

[0061] i noise,k =i k -i smooth,k ;

[0062] Among them, i noise,k For the initial correction of noise intensity, i k Let i be the column mean of a number of single-spectral-band images, k = 1, ..., N, where N is the image width and i is the column mean of the image. smooth,k This represents the column mean of several single-spectral-band images after initial denoising.

[0063] In this embodiment, step S3, which involves obtaining the noise intensity through preliminary correction noise intensity constraint processing, specifically includes:

[0064]

[0065] Among them, i thres,k Let i be the noise intensity, and threshold be the threshold value. noise,k The noise intensity is for preliminary correction.

[0066] In this embodiment, step S4 involves taking the difference between the column mean of the plurality of single-spectral-band images and the noise intensity to obtain the plurality of preliminarily corrected single-spectral-band images, specifically as follows:

[0067] I SG,k =i k -i thres,k ;

[0068] Among them, I SG For the preliminary corrected spectral band images, i k Let i be the column mean of a number of single-spectral-band images, k = 1, ..., N, where N is the image width and i is the column mean of the image. thres,k Noise intensity.

[0069] In this embodiment, step S5, specifically filtering the pre-corrected single-spectral-band images, involves:

[0070]

[0071] Among them, I bif,p This is the pre-corrected single-spectral-band image at position p after filtering. S is the size of the kernel function's spatial domain, r is the pixel range domain, q represents all pixels in the kernel function except the center point, and I...SG,q W represents the center pixel of the current kernel function. bif,p For normalization function, and These are the spatial domain weighting function and the pixel range domain weighting function, respectively.

[0072] In this embodiment, step S5, where the noise image is extracted by subtracting the pre-corrected single-spectral-band images from the filtered pre-corrected single-spectral-band images, specifically involves:

[0073] I noise =I SG -I bif ;

[0074] Among them, I noise For noisy images, I SG For the preliminary corrected spectral band images, I bif These are several single-spectral images after preliminary correction following filtering and denoising.

[0075] In this embodiment, step S6, selecting the value with the largest absolute value as the noise intensity, specifically means:

[0076]

[0077] Among them, X noise,k X represents noise intensity. mean,k X is the column mean of the noisy image. median,k The column values ​​represent the values ​​in the noisy image.

[0078] In this embodiment, step S6, which involves taking the difference between the preliminarily corrected image values ​​of several spectral bands and the noise intensity to obtain the denoised image, specifically includes:

[0079] I clean,k =I SG,k -X noise,k ;

[0080] Among them, X noise,k For noise intensity, I SG For the preliminary corrected spectral band images, I clean,k This is the image after denoising.

[0081] In existing technologies, filtering and optimization-based correction methods lose details in remote sensing images, statistical methods are time-consuming, and deep learning-based methods are limited in effectiveness in practical applications due to the difficulty in constructing training samples. Therefore, for remote sensing images with complex noise characteristics, current fringe correction models struggle to achieve accurate fringe noise removal without losing image details.

[0082] To solve the above technical problems, such as Figure 1 As shown, this embodiment designs a method for correcting stripe radiometric patterns in remote sensing satellite images based on noise intensity estimation, including the following steps:

[0083] Step S1, Remote sensing satellite image processing:

[0084] High-resolution remote sensing satellite images include full-spectral and multispectral bands. When processing multiple spectral images, they need to be separated into several single-spectral band images, which are then used for subsequent relative radiometric correction.

[0085] Step S2, preliminary denoising of several single-spectral-band images:

[0086] Taking column-oriented stripe noise as an example, the difference in column means among several single-segment images is used as a measure of noise intensity. When stripe noise exists in several single-segment images, the column mean will differ from the mean of its surrounding columns. The column means of several single-segment images are calculated, and preliminary denoising is performed on the images through filtering. For several single-segment images I, the column mean i of the several single-segment images is obtained. k k = 1, ..., N, where N is the image width. An initial correction is performed using an SG filter with a sliding window polynomial fitting method, specifically:

[0087] First, set the sliding window length w = 2m + 1, where m is a constant. Then, set the polynomial degree and perform polynomial fitting on the column means of several single-spectral-segment images within the window. The original vector set within the sliding window is then {i... k-m i k-m+1 ,…,i k+m-1 i k+m}

[0088] Secondly, based on the fitted polynomial, the estimated value of the center point within the sliding window is obtained to get the column mean i of several single-spectral-band images after preliminary denoising. smooth,k ,Right now:

[0089]

[0090] in, The smoothing coefficients are obtained by fitting the polynomial using the least squares method. The above process is repeated by moving the sliding window to achieve column mean filtering. The original vector set is processed into a processed vector set after this process.

[0091] Then, the original vector set is subtracted from the processed vector set, and the difference is regarded as the initial corrected noise intensity i. noise,k ,Right now:

[0092] i noise,k =i k -ismooth,k ;

[0093] To avoid overcorrection, the noise intensity i in the initial correction is... noise,k Mechanism constraints are applied to ensure that the noise intensity does not exceed a threshold. The constrained noise intensity i thres,k for:

[0094]

[0095] Finally, for several single-spectral-band images, i k The noise intensity is obtained by subtracting the unit value, thus achieving preliminary correction of several spectral bands of the image. SG :

[0096] I SG,k =I k -i thres,k .

[0097] Step S3: Refine the stripe noise extraction for several spectral bands of the image:

[0098] Several single-spectral images after initial denoising contain residual stripe noise, requiring further refined extraction and processing. A bilateral filtering method is used to denoise these images. The filtered images are considered relatively clean, and their difference is calculated with the original images containing residual noise. The resulting difference images are then used as the noise images. Since bilateral filtering preserves the edges of the single-spectral images while denoising, it avoids misclassifying valuable object information as noise. Specifically:

[0099] First, perform bilateral filtering on several single-band images to obtain i bif ,Right now:

[0100]

[0101] Among them, I bif,p Let be the image response value at position p after bilateral filtering, S be the size of the kernel function's spatial domain, r be the pixel range domain, q be all pixels in the kernel function except for the center point, and I be the pixel response value at position p. sG,q W represents the center pixel of the current kernel function. bif,p For normalization function, and These are the spatial domain weighting function and the pixel range domain weighting function, respectively.

[0102] in,

[0103]

[0104] Finally, the difference between the images before and after filtering is calculated to extract the noisy image I. noise ,Right now:

[0105] I noise =I Sg -I bif .

[0106] Step S4, calibration of stripe noise intensity in several single-spectral-band images:

[0107] Noise intensity is calibrated based on column-direction features of the extracted single-spectral band image stripes. For remote sensing images with complex land cover types, such as urban, farmland, and forest images, the column mean of the noise image can represent the basic noise intensity. For images with large differences in column-direction land cover responses, using the column median of the noise image to calibrate noise intensity can reduce errors caused by response differences. Therefore, by calculating the column mean and column median of the noise image respectively, the feature value with the largest absolute value is selected based on the difference for noise intensity calibration, and the final image correction is performed. Specifically:

[0108] First, calculate the column mean X of the extracted noisy image. mean,k , that is, X mean,k =mean(I noise ), and calculate the column midpoint X of the noisy image. median,k =median(I noise ).

[0109] Then, taking the larger absolute value of the column mean and the column median as the noise intensity, column by column:

[0110]

[0111] Finally, the images after preliminary denoising are processed according to column I. SG,k Subtracting the calibrated noise intensity further denoises the noisy image.

[0112] I clean,k =I SG,k -X noise,k .

[0113] To better illustrate the method for correcting stripe radiometric patterns in remote sensing satellite images based on noise intensity estimation described in this embodiment, the following examples are provided in detail:

[0114] like Figure 2 As shown, this image is a remote sensing satellite image. It exhibits noticeable stripe noise. The image is then denoised using the method described in this embodiment. Figure 3 As shown, the denoised image has completely removed the stripe noise. Similarly, as... Figure 4As shown, the image also has obvious stripe noise. After denoising the image stripes using the method described in this embodiment, as shown... Figure 5 As shown, the stripes in the image have completely disappeared.

[0115] In summary, the method described in this embodiment does not require prior estimation for stripe noise images with different characteristics. Through noise intensity estimation and calibration steps from coarse to fine, it achieves radiometric correction of remote sensing images with complex stripe noise characteristics.

[0116] The above provides a detailed description of a remote sensing satellite image stripe radiation correction method based on noise intensity estimation proposed in this invention. Specific examples have been used to illustrate the principle and implementation of this invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for stripe radiance correction of remote sensing satellite images based on noise intensity estimation, characterized in that, Includes the following steps: Step S1: Multiple spectral images are separated into several single-spectral-band images; Step S2: Calculate the column mean of several single-spectral-band images, and perform preliminary denoising on the column mean of several single-spectral-band images to obtain the column mean of several single-spectral-band images after preliminary denoising. Step S3: The difference between the column mean of several single-spectral-band images and the column mean of several single-spectral-band images after preliminary denoising is obtained to obtain the preliminary corrected noise intensity. The preliminary corrected noise intensity is constrained to obtain the noise intensity. Step S4: The difference between the column mean of several single-spectral-band images and the noise intensity is used to obtain several preliminarily corrected single-spectral-band images; Step S5: Filter the pre-corrected single-spectral-band images, and extract the noise image by subtracting the pre-corrected single-spectral-band images from the filtered pre-corrected single-spectral-band images. Step S6: Calculate the column mean and column median of the noisy image respectively, and select the value with the largest absolute value as the noise intensity. Difference between the preliminarily corrected spectral images and the noise intensity to obtain the denoised image.

2. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S2, the preliminary denoising of the column mean of several single-spectral-band images to obtain the column mean of the preliminarily denoised single-spectral-band images specifically involves: Among them, i smooth,k This represents the column mean of several single-spectral-band images after preliminary denoising. Here, m is the smoothing coefficient, and i is a constant. k Let be the column mean of several single-spectral-band images, k = 1, ..., N, where N is the image width.

3. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S3, the step of obtaining the preliminary corrected noise intensity by taking the difference between the column mean of several single-spectral-band images and the column mean of the several single-spectral-band images after preliminary denoising is specifically as follows: i noise,k =i k -i smooth,k ; Among them, i noise,k For the initial correction of noise intensity, i k Let i be the column mean of a number of single-spectral-band images, k = 1, ..., N, where N is the image width and i is the column mean of the image. smooth,k This represents the column mean of several single-spectral-band images after initial denoising.

4. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S3, the noise intensity is obtained by constraining the noise intensity after preliminary correction, specifically as follows: Among them, i thres,k Let i be the noise intensity, and threshold be the threshold value. noise,k The noise intensity is for preliminary correction.

5. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S4, the difference between the column mean of the several single-spectral-band images and the noise intensity is used to obtain several preliminarily corrected single-spectral-band images, specifically: I SG,k =i k -i thres,k ; Among them, I sG For the preliminary corrected spectral band images, i k Let i be the column mean of a number of single-spectral-band images, k = 1, ..., N, where N is the image width and i is the column mean of the image. thres,k Noise intensity.

6. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S5, filtering the preliminarily corrected single-spectral-band images specifically involves: Among them, I bif,p This is the pre-corrected single-spectral-band image at position p after filtering. S is the size of the kernel function's spatial domain, r is the pixel range domain, q represents all pixels in the kernel function except the center point, and I... SG,q W represents the center pixel of the current kernel function. bif,p For normalization function, and These are the spatial domain weighting function and the pixel range domain weighting function, respectively.

7. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S5, the noise image is extracted by subtracting the pre-corrected single-spectral-band images from the filtered pre-corrected single-spectral-band images. Specifically: I noise =I SG -I bif ; Among them, I noise For noisy images, I SG For the preliminary corrected spectral band images, I bif These are several single-spectral images after preliminary correction following filtering and denoising.

8. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S6, selecting the value with the largest absolute value as the noise intensity specifically means: Among them, X noise,k X represents noise intensity. mean,k X is the column mean of the noisy image. median,k The column values ​​represent the values ​​in the noisy image.

9. The method for correcting stripe radiance in remote sensing satellite images based on noise intensity estimation according to claim 1, characterized in that, In step S6, the step of obtaining the denoised image by subtracting the noise intensity from the initially corrected image values ​​of several spectral bands specifically involves: I clean,k =I SG,k -X noise,k ; Among them, X noise,k For noise intensity, I SG For the preliminary corrected spectral band images, I clean,k This is the image after denoising.

Images