Detection-guided noctilucent remote sensing image irregular stripe noise removal method and device

Abstract

The invention discloses a detection-guided noctilucent remote sensing image irregular stripe noise removal method and device, and the method comprises the steps: carrying out the intense light source, non-illumination and to-be-processed region division and edge enhancement of an original image, and obtaining an enhanced image; performing stripe noise iterative detection according to the enhanced image in combination with frequency domain analysis and multi-constraint Hough transform to obtain a stripe mask; aligning the strip mask with the binary image of the original image based on morphological operation, and performing inter-band filtering to obtain a noise-free mask; obtaining a noise target area according to the noise-free mask and the original image; performing progressive noise correction on the noise target area based on a dynamic window and an adaptive correction algorithm to obtain an initial result; and performing residual error correction according to the initial result in combination with the strong light source region to obtain a stripe-removed image.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing, and in particular to a method and apparatus for removing irregular stripe noise from nighttime light remote sensing images guided by detection. Background Technology

[0002] Nighttime light remote sensing images can directly reflect the intensity and spatial distribution characteristics of human activities, and have become an important data source for socio-economic parameter inversion and sustainable development research. However, due to factors such as sensor anomalies and atmospheric scattering, nighttime light remote sensing images generally suffer from stripe noise, which interferes with the acquisition of true light source information, leading to biases in subsequent quantitative analysis.

[0003] Based on their removal principles, existing methods for destriping remote sensing images can be categorized into three types: traditional image processing methods, deep learning-based methods, and optimization-based methods. Traditional image processing methods, due to their simplicity and high computational efficiency, were widely used in early research, often employing mean filtering or median filtering. However, while removing stripe noise, these methods often affect the background areas of the image, leading to the loss of some image detail. Deep learning-based methods primarily rely on large amounts of labeled data to train neural networks, thereby obtaining optimal model parameters suitable for stripe noise removal tasks. These methods include convolutional neural networks and generative adversarial networks. However, these methods generally depend on large-scale, high-quality training samples, making their application in real-world scenarios challenging. Optimization-based methods construct an objective function that includes data consistency and regularization terms. They use mathematical models to model and constrain noise, and achieve image restoration by solving for the extrema of this objective function. These methods typically use optimization techniques to decompose complex problems into multiple sub-problems for iterative solving until convergence conditions are met. While possessing good theoretical interpretability and controllability, its efficiency is low when processing large-scale remote sensing images due to the complexity of the iterative process and heavy computational burden. In summary, existing remote sensing image destriping methods all have limitations when dealing with complex and unevenly distributed stripe noise in real nighttime light remote sensing images. Furthermore, most current methods are designed for regular stripe noise and are ineffective at removing irregularly shaped and highly discontinuous stripe noise, making it difficult to fully restore the true information of the image.

[0004] Therefore, a new technical solution is urgently needed to address the technical problem of how to detect and remove stripe noise in nighttime light remote sensing images. Summary of the Invention

[0005] This invention provides a method and apparatus for detecting and removing irregular stripe noise in nighttime light remote sensing images, in order to solve the technical problem of how to detect and remove stripe noise in nighttime light remote sensing images.

[0006] To achieve the above objectives, the present invention provides a method for removing irregular stripe noise from nighttime light remote sensing images guided by detection, comprising: The original image is divided into areas with strong light sources, no light, and areas to be processed, and edge enhancement is performed to obtain an enhanced image. Based on the enhanced image, frequency domain analysis and multi-constraint Hough transform are combined to perform iterative detection of strip noise and obtain a strip mask. After aligning the strip mask with the binary image of the original image based on morphological operations, inter-band filtering is performed to obtain a noise-free mask; the noisy target region is obtained based on the noise-free mask and the original image. A progressive noise correction is performed on the noisy target area based on a dynamic window and an adaptive correction algorithm to obtain an initial result; based on the initial result and the strong light source area, residual correction is performed to obtain a destriped image.

[0007] Preferably, the stripe noise is iteratively detected based on enhanced images combined with frequency domain analysis and multi-constraint Hough transform, resulting in a stripe mask including: The enhanced image is subjected to a two-dimensional Fourier transform and spectral centering to obtain a frequency domain matrix. The intensity information accumulation value of each angle is obtained from the direction spectrum of the frequency domain matrix, and the direction corresponding to the maximum intensity is extracted to obtain the strip noise direction. The enhanced image is binarized and subjected to Hough transform to obtain an accumulation matrix. Progressive strip detection is performed based on the strip noise direction and the accumulation matrix to obtain a strip mask.

[0008] Preferably, the enhanced image is subjected to a two-dimensional Fourier transform and spectral centering to obtain a frequency domain matrix; the intensity information accumulation value at each angle is obtained based on the direction spectrum of the frequency domain matrix, and the direction corresponding to the maximum intensity is extracted to obtain the strip noise direction, including: Enhance images Perform a two-dimensional Fourier transform and spectral centering to obtain the frequency domain matrix. : ; in, This represents the Fourier transform operator; and These represent enhanced images. Horizontal and vertical coordinates; and Representing the frequency domain matrix respectively Horizontal and vertical coordinates; The directional spectrum of the frequency domain matrix, i.e., each frequency domain pixel. corresponding direction angle , is represented as: ; in, and These represent the number of rows and columns of the image, respectively. It is the arctangent function in the fourth quadrant; Strip noise direction Represented as: ; in, This indicates that the intensity is summed by angle; This indicates the search for the angle that maximizes the cumulative value of the spectral intensity. .

[0009] Preferably, the enhanced image is binarized and subjected to Hough transform to obtain a cumulative matrix; progressive strip detection is performed based on the strip noise direction and the cumulative matrix to obtain a strip mask, including: Definition of the first In progressive strip detection, the enhanced image is binarized to obtain the image. Then for the image The cumulative matrix obtained by performing the Hough transformation Represented as: ; in, Represents the Hough transform function; According to the cumulative matrix Obtain the normalized cumulative matrix , is represented as: ; in, This represents the dot division operation; To and A matrix of all ones of the same size; Introducing angle tolerance For strip noise direction Make corrections to obtain candidate detection angles. , is represented as: ; Based on candidate detection perspective For the cumulative matrix respectively and normalized cumulative matrix Peak detection is performed to obtain a set of candidate peak points. and , is represented as: ; ; in, Represents the row index of the cumulative matrix; Describes the th element in the cumulative matrix of Hough space. The distance parameter corresponding to the row; Represents the candidate detection angle in the Hough space accumulation matrix Column index; For the neighborhood window used for nonmaximum suppression; Indicates in Find the row index within the matrix that causes the cumulative matrix to reach a local maximum. ; and For the first The detection threshold for the wheel satisfies: ; ; in, Indicates the first Proportional parameters in wheel detection; This represents a function used to calculate the maximum value in a matrix; Then the first Wheel and belt inspection combined results for: ; in, This represents taking the union of sets; Based on the strip detection merging results Updated images and proportional parameters Iteratively perform strip detection until the scaling parameter is reached. The loop ends when the value drops to a preset value, resulting in a stripe detection set. and strip mask : ; ; ; ; in, This indicates an element-by-element logical OR operation on multiple matrices; Indicates the total number of iterations; Indicates the first The results of a single-round strip masking iteration; Indicates the width tolerance of the straight line; This represents the Hadamard product, which is the element-wise multiplication of matrices. Indicates the first The distance parameter of the strip noise line detected by the round iteration.

[0010] Preferably, based on the strip detection merging results Updated images and proportional parameters And iteratively perform strip detection, including: When the strip detection results are merged In the image Remove the band information detected in this round to obtain the starting image for the next round. , is represented as: ; According to the images Open the first Round iteration; When the strip detection results are merged Then update the image. and proportional parameters , obtained the The starting image of the round iteration and proportional parameters : ; ; According to the images and proportional parameters Open the first Round iteration.

[0011] Preferably, after aligning the strip mask with the binary image of the original image based on morphological operations, inter-band filtering is performed to obtain a noise-free mask; the noisy target region is obtained based on the noise-free mask and the original image, including: Using 0 as the threshold for the original image Binarization is performed to obtain a binary image of the original image. ; Adjusting the strip mask based on morphological operations Binary image of the original image Calculate the Hadamard product and obtain the aligned result. : ; in, This represents performing an expansion operation; Based on the alignment results and original images Obtain images unaffected by stripes. : ; Image in band dimension After applying medium filtering, binarization is performed to obtain a noise-free mask. According to the noise-free mask and original images Obtain the noise target area : .

[0012] Preferably, progressive noise correction is performed on the noisy target region based on a dynamic window and an adaptive correction algorithm, yielding initial results including: In the original image Set up a sliding window Each window's pixels are divided into a noisy pixel set P and a noise-free pixel set H, where all pixels in the noisy pixel set P belong to the noisy target region. If the proportion of noise-free pixels within a window is less than 40% of the total pixels within the window, it is marked as an invalid window; otherwise, it is marked as a valid window. Based on an adaptive correction algorithm, using the information of noise-free pixels as a reference, the correction values ​​of noisy pixels are calculated and noise repair is performed within each valid window. When the current loop ends, the original image is... The repaired pixels are marked as noise-free pixels, and the noise target area is updated. Then proceed to the next cycle; iterate until the original image is obtained. All noisy pixels were repaired, and the initial result was obtained.

[0013] Preferably, based on an adaptive correction algorithm, using information from noise-free pixels as a reference, the correction values ​​for noisy pixels are calculated and noise repair is performed within each effective window, including: For the One effective window Calculate the statistical moments of the noiseless pixel set H and the noisy pixel set P, including: ; ; ; ; in, and The mean and variance of the noise-free pixel set H are represented. and Represents the mean and variance of the noisy pixel set P; For window The number of noise-free pixels; For window The number of pixels with medium noise; and Indicates by subscript and Index to a specific range of pixels; The statistical moments of the jointly constrained noiseless pixel set H and the noisy pixel set P are consistent, transforming the problem into an optimization problem, including: ; Solving the optimization problem yields a unique solution: ; Based on the unique solution, pair each noise pixel Perform the conversion to obtain the repair result. : ; If the same noise pixel If it exists in multiple valid windows, meaning it is repaired in multiple valid windows, then... The repair result is the average of multiple repair results. : ; in, Indicates including noisy pixels The total number of valid windows; This represents the total number of all valid windows; Indicates the noise pixels The repair results within multiple valid windows are summed.

[0014] Preferably, residual correction is performed based on the initial results and the strong light source area to obtain the destriped image, including: Calculate the initial result residual plot : ; Residual plot with 0 as the threshold Binarization is performed to obtain the image ; for images Use opening operations to obtain a mask for the region related to the normal structure of the image. : ; in, This represents a corrosion operation; Represents an expansion operation; Represents structural elements; According to the area of ​​strong light source Initial results Residual plot and mask Obtain destriped images : ; ; in, These are the pixel coordinates; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Represents a logical OR operation.

[0015] The present invention also provides a detection-guided device for removing irregular strip noise from nighttime light remote sensing images, used in the method of the present invention, the device comprising a first module, a second module, a third module and a fourth module; The first module is used to divide the original image into areas with strong light sources, no light, and areas to be processed, and to enhance the edges to obtain an enhanced image; based on the enhanced image, frequency domain analysis and multi-constraint Hough transform are combined to perform iterative detection of strip noise to obtain a strip mask; The second module is used to align the strip mask with the binary image of the original image based on morphological operations and then perform inter-band filtering to obtain a noise-free mask; the noisy target region is obtained based on the noise-free mask and the original image. The third module is used to perform progressive noise correction on the noise target region based on dynamic windows and adaptive correction algorithms to obtain initial results; The fourth module is used to perform residual correction based on the initial results and the strong light source area to obtain the destriped image.

[0016] The present invention has the following beneficial effects: This invention presents a detection-guided method for removing irregular strip noise from nighttime light remote sensing images. It achieves precise localization of irregular, fragmented strips through an iterative detection mechanism with multi-condition constraints in Hough space, utilizes a mask-guided local window for adaptive correction of noisy pixels, and incorporates a light source region protection strategy specific to nighttime light images to optimize the spectral fidelity of the results. Compared to traditional strip removal methods, this invention offers the following advantages and benefits: Firstly, it exhibits strong robustness to complex strip shapes such as tilted and discontinuous stripes, effectively preserving the structural and brightness characteristics of key features like urban light sources in real images. Secondly, by integrating local self-correction processing and a light source mask protection mechanism, it demonstrates significant advantages in removing strip noise, preserving image details, and improving visual quality in scenarios with uneven noise distribution and large differences in feature brightness, making it highly practical and worthy of engineering promotion. In summary, the method of this invention can fully utilize the distribution characteristics of strip noise in spatial structure, achieve accurate detection of complex strip noise through a multi-condition constraint mechanism, and further combine the idea of ​​local optimization to perform fine denoising operation in the detected area, effectively improving the accuracy and robustness of denoising, and can effectively handle complex and irregular strip noise in nighttime light remote sensing images.

[0017] The detection-guided noise removal device for irregular stripes in nighttime light remote sensing images of the present invention, when used in the method of the present invention, has the same beneficial effects as the method of the present invention.

[0018] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0019] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow of a preferred embodiment of the present invention.

[0020] Figure 2 This is the first original drawing of a preferred embodiment of the present invention.

[0021] Figure 3 This is the second original drawing of a preferred embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram of the first original image stripe noise detection result according to a preferred embodiment of the present invention.

[0023] Figure 5 This is a schematic diagram of the second original image strip noise detection result according to a preferred embodiment of the present invention.

[0024] Figure 6 This is a schematic diagram of the denoising result of the first original image in a preferred embodiment of the present invention.

[0025] Figure 7 This is a schematic diagram of the denoising result of the second original image in a preferred embodiment of the present invention. Detailed Implementation

[0026] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0027] See Figure 1 In a preferred embodiment of the present invention, a method for removing irregular stripe noise from nighttime light remote sensing images guided by detection is provided, comprising: Q1. Divide the original image into strong light source, no light source, and unprocessed areas, and enhance the edges to obtain an enhanced image; based on the enhanced image, combine frequency domain analysis and multi-constraint Hough transform to perform iterative detection of strip noise and obtain a strip mask.

[0028] In a preferred embodiment of the present invention, the original image is divided into areas with strong light source, no light source, and areas to be processed, and edge enhancement is performed to obtain an enhanced image, including: According to the original image The pixel value size will affect the original image The area is divided into three zones: a strong light source, no light, and a zone to be processed. For the original image If any pixel in the image has a pixel value exceeding a set threshold (255) in any RGB band, then the pixel is determined to belong to a strong light source area. : ; in, These are the pixel coordinates; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Represents a logical OR operation.

[0029] If the pixel value of all three RGB bands is 0, then the pixel is determined to belong to an area without light.

[0030] Pixels other than those in areas with strong light sources and areas without light are considered as areas to be processed.

[0031] For the original image Perform edge enhancement processing: A guided filtering algorithm is used to obtain the original image. As a guide image, the filtering result is calculated. Edge information is then extracted through interpolation and enhanced by applying a gain factor. The enhanced image is shown below. include: ; in, This represents guided filtering of the input image; gain coefficient. Recommended setting: 5; Enhance image To enhance Nighttime light remote sensing image with edge information after multiplication.

[0032] In a preferred embodiment of the present invention, the stripe mask obtained by iterative detection of stripe noise based on enhanced image combined with frequency domain analysis and multi-constraint Hough transform includes: The enhanced image is subjected to a two-dimensional Fourier transform and spectral centering to obtain a frequency domain matrix. The intensity information accumulation value of each angle is obtained from the direction spectrum of the frequency domain matrix, and the direction corresponding to the maximum intensity is extracted to obtain the strip noise direction. The enhanced image is binarized and subjected to Hough transform to obtain an accumulation matrix. Progressive strip detection is performed based on the strip noise direction and the accumulation matrix to obtain a strip mask.

[0033] In a preferred embodiment of the present invention, the enhanced image is subjected to a two-dimensional Fourier transform and spectral centering to obtain a frequency domain matrix; the intensity information accumulation value of each angle is obtained based on the direction spectrum of the frequency domain matrix, and the direction corresponding to the maximum intensity is extracted to obtain the strip noise direction, including: Enhance images Perform a two-dimensional Fourier transform and spectral centering to obtain the frequency domain matrix. : ; in, This represents the Fourier transform operator; and These represent enhanced images. Horizontal and vertical coordinates; and Representing the frequency domain matrix respectively Horizontal and vertical coordinates; The directional spectrum of the frequency domain matrix, i.e., each frequency domain pixel. corresponding direction angle , is represented as: ; in, and These represent the number of rows and columns of the image, respectively. It is the arctangent function in the fourth quadrant; Strip noise direction Represented as: ; in, This indicates that the intensity is summed by angle; This indicates the search for the angle that maximizes the cumulative value of the spectral intensity. .

[0034] In a preferred embodiment of the present invention, the enhanced image is binarized and subjected to Hough transform to obtain an accumulation matrix; progressive strip detection is performed based on the strip noise direction and the accumulation matrix to obtain a strip mask, including: Definition of the first In progressive strip detection, the enhanced image is binarized to obtain the image. Then for the image The cumulative matrix obtained by performing the Hough transformation Represented as: ; in, Represents the Hough transform function; To capture the tilted stripes distributed at the edges of the image, a normalized cumulative matrix is ​​established to characterize the proportion of the largest straight line. According to the cumulative matrix Obtain the normalized cumulative matrix , is represented as: ; in, This represents the dot division operation; To and A matrix of all ones of the same size; Considering that strip noise in real nighttime light remote sensing images may exhibit bending, an angle tolerance is introduced during strip extraction. For strip noise direction Make corrections to obtain candidate detection angles. , is represented as: ; Based on candidate detection perspective For the cumulative matrix respectively and normalized cumulative matrix Peak detection is performed to obtain a set of candidate peak points. and , is represented as: ; ; in, Represents the row index of the cumulative matrix; Describes the th element in the cumulative matrix of Hough space. The distance parameter (polar radius value) corresponding to the row; Represents the candidate detection angle in the Hough space accumulation matrix Column index; For the neighborhood window used for nonmaximum suppression; Indicates in Find the row index within the matrix that causes the cumulative matrix to reach a local maximum. ; and For the first The detection threshold for the wheel satisfies: ; ; in, Indicates the first The scaling parameter in the round detection (initially set to 0.5, indicating that at least 50% of the maximum possible points are collinear); This represents a function used to calculate the maximum value in a matrix; Then the first Wheel and belt inspection combined results for: ; in, This represents taking the union of sets; Based on the strip detection merging results Updated images and proportional parameters Iteratively perform strip detection until the scaling parameter is reached. The loop ends when the value drops to a preset value (e.g., 0.3), resulting in a stripe detection set. and strip mask : ; ; ; ; in, This indicates an element-by-element logical OR operation on multiple matrices; Indicates the total number of iterations; Indicates the first The results of a single-round strip masking iteration; This indicates the width tolerance of the line; a setting of 1.5 pixels is recommended. This represents the Hadamard product, which is the element-wise multiplication of matrices. Indicates the first The distance parameter of the strip noise line detected by the round iteration.

[0035] In a preferred embodiment of the present invention, based on the strip detection merging results Updated images and proportional parameters And iteratively perform strip detection, including: When the strip detection results are merged In the image Remove the band information detected in this round to obtain the starting image for the next round. , is represented as: ; According to the images Open the first Round iteration; When the strip detection results are merged Then update the image. and proportional parameters , obtained the The starting image of the round iteration and proportional parameters : ; ; According to the images and proportional parameters Open the first Round iteration.

[0036] In a preferred embodiment of the present invention, progressive strip detection is performed based on the edge-oriented binarized image of Hough space. The feature is that the strip detection is performed under multiple constraints. The corresponding pixels of the strips detected in each round are removed from the edge binarized image so that broken or weak strips can be progressively captured in subsequent rounds.

[0037] Q2. Based on morphological operations, the strip mask is aligned with the binary image of the original image, and then inter-band filtering is performed to obtain a noise-free mask; the noisy target region is obtained based on the noise-free mask and the original image. Q2 specifically includes: To reduce the subtle spatial deviation between the strip mask and the original image, a threshold of 0 is used for the original image. Binarization is performed to obtain a binary image of the original image. ; Adjusting the strip mask based on morphological operations Binary image of the original image Calculate the Hadamard product and obtain the aligned result. : ; in, This represents performing an expansion operation; Based on the alignment results and original images Obtain images unaffected by stripes. : ; Image in band dimension After applying medium-level filtering, binarization is performed to remove residual random noise, resulting in a noise-free mask. According to the noise-free mask and original images Obtain the noise target area : ; In a preferred embodiment of the present invention, by solving for the noise target region, it is ensured that the denoising process targets only the real noise pixels, so as to better protect the effective information in the image.

[0038] Q3. Based on dynamic window and adaptive correction algorithm, progressive noise correction is performed on the noisy target area to obtain the initial result; based on the initial result and combined with the strong light source area, residual correction is performed to obtain the destriped image.

[0039] In a preferred embodiment of the present invention, progressive noise correction is performed on the noisy target region based on a dynamic window and an adaptive correction algorithm to obtain initial results including: In the original image Set up a sliding window The initial window size is set to (With radius r = 1), the pixels within each window are divided into a noise pixel set P and a noise-free pixel set H, where all pixels in the noise pixel set P belong to the noise target region. If the proportion of noise-free pixels within a window is less than 40% of the total pixels within the window, it is marked as an invalid window; otherwise, it is marked as a valid window. Based on an adaptive correction algorithm, using the information of noise-free pixels as a reference, the correction values ​​of noisy pixels are calculated and noise repair is performed within each valid window. When the current loop ends, the original image is... The repaired pixels are marked as noise-free pixels, and the noise target area is updated. Then proceed to the next cycle; iterate until the original image is obtained. All noisy pixels were repaired, and the initial result was obtained.

[0040] In a preferred embodiment of the present invention, if the proportion of repaired noise pixels is less than 10% of the total noise before the repair round after one round of repair, the sliding window radius is adaptively increased (recommended to be set to...). Then proceed to the next round of repairs to address different noise distributions in real-world scenarios.

[0041] In a preferred embodiment of the present invention, based on an adaptive correction algorithm, using information from noise-free pixels as a reference, the correction values ​​for noisy pixels are calculated within each effective window, and noise repair is performed, including: For the One effective window Calculate the statistical moments of the noiseless pixel set H and the noisy pixel set P, including: ; ; ; ; in, and The mean and variance of the noise-free pixel set H are represented. and Represents the mean and variance of the noisy pixel set P; For window The number of noise-free pixels; For window The number of pixels with medium noise; and Indicates by subscript and Index to a specific range of pixels; To minimize the data distribution difference between noisy pixels and healthy pixels, the statistical moments of the noiseless pixel set H and the noisy pixel set P are jointly constrained to be consistent, transforming the problem into an optimization problem, including: ; Solving the optimization problem yields a unique solution: ; Based on the unique solution, pair each noise pixel Perform the conversion to obtain the repair result. : ; If the same noise pixel If it exists in multiple valid windows, meaning it is repaired in multiple valid windows, then... The repair result is the average of multiple repair results. : ; in, Indicates including noisy pixels The total number of valid windows; This represents the total number of all valid windows; Indicates the noise pixels The repair results within multiple valid windows are summed.

[0042] In a preferred embodiment of the present invention, residual correction is performed based on the initial results and the strong light source region to obtain the destriped image, including: To prevent the loss of some details in the image during strip correction, correction processing is performed based on the residual map of the initial results. The initial results are calculated. residual plot : ; Residual plot with 0 as the threshold Binarization is performed to obtain the image ; for images Use opening operations to obtain a mask for the region related to the normal structure of the image. : ; in, This represents a corrosion operation; Represents an expansion operation; Represents structural elements; According to the area of ​​strong light source Initial results Residual plot and mask Obtain destriped images : ; ; in, These are the pixel coordinates; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Represents a logical OR operation.

[0043] The method of this invention integrates directional fault tolerance judgment, window statistical feature matching, and spectral consistency constraints between bands, effectively solving the problem of difficulty in accurately separating irregular strip noise and bright ground object signals in nighttime light remote sensing images.

[0044] This invention fully considers the irregularity and uneven brightness of strip noise in nighttime light remote sensing images, and constructs a priori detection framework that integrates frequency domain priors and structurally progressive priors. By estimating the principal direction of the strips through Fourier transform, the constrained strip extraction process in Hough space is guided, achieving stable detection of strip structures. Furthermore, by combining the maximum straight line ratio information to optimize the detection results, the overall accuracy and adaptability of the detection are further improved.

[0045] This invention fully considers the issues of local discontinuities and structural consistency in strip noise restoration. It distinguishes between noise-free and noise-free regions within a sliding window and uses statistical feature matching to generate correction coefficients for strip restoration. Local smoothing filtering is combined to alleviate the window fragmentation effect, improving the continuity of the restored structure and edge preservation. Furthermore, an iterative update mechanism and an adaptive window expansion strategy are designed to enhance the robustness of residual noise processing. Finally, highlight region information from the nighttime light image is fused to guide brightness optimization of the result, thereby improving the overall denoising quality and visual consistency.

[0046] In a preferred embodiment of the present invention, a device for removing irregular strip noise from nighttime light remote sensing images guided by detection is also provided, which is used in the method of the present invention. The device includes a first module, a second module, a third module and a fourth module. The first module is used to divide the original image into areas with strong light sources, no light, and areas to be processed, and to enhance the edges to obtain an enhanced image; based on the enhanced image, frequency domain analysis and multi-constraint Hough transform are combined to perform iterative detection of strip noise to obtain a strip mask; The second module is used to align the strip mask with the binary image of the original image based on morphological operations and then perform inter-band filtering to obtain a noise-free mask; the noisy target region is obtained based on the noise-free mask and the original image. The third module is used to perform progressive noise correction on the noise target region based on dynamic windows and adaptive correction algorithms to obtain initial results; The fourth module is used to perform residual correction based on the initial results and the strong light source area to obtain the destriped image.

[0047] The detection-guided noise removal device for irregular stripes in nighttime light remote sensing images of the present invention, when used in the method of the present invention, has the same beneficial effects as the method of the present invention.

[0048] Verification section: To verify the effectiveness of this invention in real-world complex scenarios, see [link / reference]. Figures 2 to 3 In this embodiment, two representative nighttime light remote sensing images were selected for strip noise detection and removal experiments, which were named the first original image and the second original image, respectively. Figures 4 to 7 The noise detection and denoising repair results based on the present invention are presented.

[0049] See Figures 4 to 5 During the detection phase, this invention demonstrated excellent feature extraction capabilities against both the distorted nonlinear color bands in the first original image and the strong background interference penetrating the bright urban clusters in the second original image. It accurately located the noise, and no falsely detected pixels belonging to ground feature lights appeared in the detection results, proving the algorithm's high robustness. See also Figures 6 to 7 In the denoising stage, the visual effect of the restored image is significantly improved. Comparative results show that the color bands originally covering the image are completely suppressed, and the background is restored to purity. Simultaneously, the method of this invention achieves an excellent balance between "denoising" and "edge preservation," ensuring that even the high-density urban textures and subtle edge details in the second original image are fully reconstructed without the common issues of over-smoothing or brightness attenuation. In summary, the experimental results demonstrate that this invention can accurately remove complex band noise while preserving the radiometric information and spatial structure of the original image to the greatest extent possible.

[0050] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for removing irregular stripe noise from nighttime light remote sensing images guided by detection, characterized in that, include: The original image is divided into areas with strong light sources, no light, and areas to be processed, and edge enhancement is performed to obtain an enhanced image. Based on the enhanced image, frequency domain analysis and multi-constraint Hough transform are combined to perform iterative detection of strip noise, and a strip mask is obtained. After aligning the strip mask with the binary image of the original image based on morphological operations, inter-band filtering is performed to obtain a noise-free mask; the noisy target region is obtained based on the noise-free mask and the original image. The noise target region is progressively rectified using a dynamic window and adaptive correction algorithm to obtain an initial result; based on the initial result and the strong light source region, residual correction is performed to obtain a destriped image.

2. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 1, characterized in that, Based on the enhanced image, frequency domain analysis, and multi-constraint Hough transform, iterative detection of strip noise is performed to obtain a strip mask including: The enhanced image is subjected to a two-dimensional Fourier transform and spectral centering to obtain a frequency domain matrix; the intensity information accumulation value of each angle is obtained according to the direction spectrum of the frequency domain matrix, and the direction corresponding to the maximum intensity is extracted to obtain the strip noise direction; the enhanced image is binarized and subjected to Hough transform to obtain an accumulation matrix; progressive strip detection is performed according to the strip noise direction and the accumulation matrix to obtain a strip mask.

3. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 2, characterized in that, The enhanced image is subjected to a two-dimensional Fourier transform and spectral centering to obtain a frequency domain matrix; The intensity information accumulation value at each angle is obtained based on the directional spectrum of the frequency domain matrix, and the direction corresponding to the maximum intensity is extracted to obtain the strip noise direction, including: The enhanced image Perform a two-dimensional Fourier transform and spectral centering to obtain the frequency domain matrix. : ； in, This represents the Fourier transform operator; and These represent enhanced images. Horizontal and vertical coordinates; and Representing the frequency domain matrix respectively Horizontal and vertical coordinates; The directional spectrum of the frequency domain matrix, i.e., each frequency domain pixel corresponding direction angle , is represented as: ； in, and These represent the number of rows and columns of the image, respectively. It is the arctangent function in the fourth quadrant; The direction of the strip noise Represented as: ； in, This indicates that the intensity is summed by angle; This indicates the search for the angle that maximizes the cumulative value of the spectral intensity. .

4. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 3, characterized in that, The enhanced image is binarized and subjected to Hough transform to obtain the cumulative matrix; Progressive strip detection is performed based on the strip noise direction and the accumulation matrix to obtain a strip mask, including: Definition of the first In progressive strip detection, the enhanced image is binarized to obtain the image. Then for the image The cumulative matrix obtained by performing the Hough transformation Represented as: ； in, Represents the Hough transform function; According to the cumulative matrix Obtain the normalized cumulative matrix , is represented as: ； in, This represents the dot division operation; To and A matrix of all ones of the same size; Introducing angle tolerance Regarding the direction of the strip noise Make corrections to obtain candidate detection angles. , is represented as: ； Based on the candidate detection angle For the cumulative matrix respectively and normalized cumulative matrix Peak detection is performed to obtain a set of candidate peak points. and , is represented as: ； ； in, Represents the row index of the cumulative matrix; Describes the th element in the cumulative matrix of Hough space. The distance parameter corresponding to the row; Represents the candidate detection angle in the Hough space accumulation matrix Column index; For the neighborhood window used for nonmaximum suppression; Indicates in Find the row index within the matrix that causes the cumulative matrix to reach a local maximum. ; and For the first The detection threshold for the wheel satisfies: ； ； in, Indicates the first Proportional parameters in wheel detection; This represents a function used to calculate the maximum value in a matrix; Then the first Wheel and belt inspection combined results for: ； in, This represents taking the union of sets; Based on the strip detection merging results Updated images and proportional parameters Iteratively perform strip detection until the scaling parameter is reached. The loop ends when the value drops to a preset value, resulting in a stripe detection set. and the strip mask : ； ； ； ； in, This indicates an element-by-element logical OR operation on multiple matrices; Indicates the total number of iterations; Indicates the first The results of a single-round strip masking iteration; Indicates the width tolerance of the straight line; This represents the Hadamard product, which is the element-wise multiplication of matrices. Indicates the first The distance parameter of the strip noise line detected by the round iteration.

5. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 4, characterized in that, Based on the strip detection merging results Updated images and proportional parameters And iteratively perform strip detection, including: When the strip detection results are merged In the image Remove the band information detected in this round to obtain the starting image for the next round. , is represented as: ； According to the images Open the first Round iteration; When the strip detection results are merged Then update the image. and proportional parameters , obtained the The starting image of the round iteration and proportional parameters : ； ； According to the images and proportional parameters Open the first Round iteration.

6. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 5, characterized in that, After aligning the strip mask with the binary image of the original image based on morphological operations, inter-band filtering is performed to obtain a noise-free mask. Based on the noise-free mask and the original image, the noise target region includes: Using 0 as the threshold for the original image Binarization is performed to obtain a binary image of the original image. ; The strip mask is adjusted based on morphological operations. The binary image after comparison with the original image Calculate the Hadamard product and obtain the aligned result. : ； in, This represents performing an expansion operation; According to the alignment result and original images Obtain images unaffected by stripes. : ； Image in band dimension After applying medium filtering, binarization is performed to obtain a noise-free mask. According to the noise-free mask and original images Obtain the noise target area : 。 7. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 6, characterized in that, Based on a dynamic window and adaptive correction algorithm, progressive noise correction is performed on the noise target region to obtain the following initial results: In the original image Set up a sliding window Each pixel within a window is divided into a noise pixel set P and a noise-free pixel set H, where all pixels in the noise pixel set P belong to the noise target region. If the proportion of noise-free pixels within a window is less than 40% of the total pixels within the window, it is marked as an invalid window; otherwise, it is marked as a valid window. Based on an adaptive correction algorithm, using the information of noise-free pixels as a reference, the correction value of noisy pixels is calculated and noise repair is performed within each valid window. When the current loop ends, the original image is... The repaired pixels are marked as noise-free pixels, and the noise target region is updated. Then proceed to the next cycle; iterate until the original image is obtained. All noisy pixels were repaired, resulting in the initial result.

8. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 7, characterized in that, Based on an adaptive correction algorithm, using information from noise-free pixels as a reference, the correction values ​​for noisy pixels are calculated within each effective window, and noise repair is performed, including: For the One effective window Calculate the statistical moments of the noiseless pixel set H and the noisy pixel set P, including: ； ； ； ； in, and The mean and variance of the noise-free pixel set H are represented. and Represents the mean and variance of the noisy pixel set P; For window The number of noise-free pixels; For window The number of pixels with medium noise; and Indicates by subscript and Index to a specific range of pixels; The statistical moments of the jointly constrained noiseless pixel set H and the noisy pixel set P are consistent, transforming the problem into an optimization problem, including: ； Solving the optimization problem yields a unique solution: ； Based on the unique solution, each noise pixel... Perform the conversion to obtain the repair result. : ； If the same noise pixel If it exists in multiple valid windows, meaning it is repaired in multiple valid windows, then... The repair result is the average of multiple repair results. : ； in, Indicates including noisy pixels The total number of valid windows; This represents the total number of all valid windows; Indicates the noise pixels The repair results within multiple valid windows are summed.

9. The method for removing irregular stripe noise from nighttime light remote sensing images guided by detection according to claim 8, characterized in that, Based on the initial results and residual correction performed on the strong light source region, the destriped image is obtained, including: Calculate the initial result residual plot : ； Residual plot with 0 as the threshold Binarization is performed to obtain the image ; for images Use opening operations to obtain a mask for the region related to the normal structure of the image. : ； in, This represents a corrosion operation; Represents an expansion operation; Represents structural elements; According to the area of ​​strong light source The initial result Residual plot and the mask Obtain destriped images : ； ； in, These are the pixel coordinates; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Original image of Pixel values ​​for the band; Represents a logical OR operation.

10. A detection-guided device for removing irregular stripe noise from nighttime light remote sensing images, used in the method described in any one of claims 1 to 9, characterized in that, The device includes a first module, a second module, a third module, and a fourth module; The first module is used to divide the original image into strong light source, no light source, and unprocessed areas and enhance the edges to obtain an enhanced image; based on the enhanced image, frequency domain analysis and multi-constraint Hough transform are combined to perform strip noise iterative detection to obtain a strip mask; The second module is used to align the strip mask with the binary image of the original image based on morphological operations and then perform inter-band filtering to obtain a noise-free mask; and to obtain the noisy target region based on the noise-free mask and the original image. The third module is used to perform progressive noise correction on the noise target region based on a dynamic window and an adaptive correction algorithm to obtain an initial result; The fourth module is used to perform residual correction based on the initial results and the strong light source area to obtain a destriped image.

Images