Remote sensing image strip noise removal method and device, equipment and storage medium

Abstract

The invention discloses a remote sensing image strip noise removal method, device and equipment and a storage medium, and relates to the technical field of remote sensing image processing, and the method comprises the steps: carrying out the filtering processing of an original image containing strip noise, obtaining a residual image, and obtaining an estimated strip noise column; dividing the original image along a vertical strip noise direction to obtain feature regions, and respectively clustering to obtain each strip noise column; determining constraint conditions, and constructing to obtain a stripe noise constraint model; and calculating the stripe noise constraint model, obtaining a noise gray value of each stripe noise column, obtaining a stripe noise image, removing the stripe noise image, and outputting a de-noised image. According to the method and the device, the defect of low precision caused by the fact that the position of the stripe noise is determined only through a residual image in the prior art is overcome, the influence of residual detail information in the residual image on the stripe noise can be overcome, and accurate positioning of all stripe noise columns or non-stripe noise columns is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a remote sensing image strip noise removal method, device, equipment and storage medium. BACKGROUND

[0002] In recent years, with the development of remote sensing technology, remote sensing images have been widely used in the fields of geography, environmental monitoring and urban planning. However, due to the working mode of the sensor and the interference in the data acquisition process, some specific frequency noise is often introduced, which causes obvious strip structure on the remote sensing image, that is, strip noise. These strip noises limit the application of remote sensing images in image interpretation and inversion, and therefore, appropriate methods are needed to remove the strip noise.

[0003] At present, the strip noise can be removed by the following four types of methods, including: optimization-based method, statistical-based method, filtering-based method and deep learning-based method. Among them, the optimization-based method removes the strip noise by establishing an energy functional and finding the minimum value, but the optimization process is over-smoothed, which causes image blur and artifacts, and the method is dependent on parameter adjustment and lacks robustness; the statistical-based method (such as matrix matching) corrects the noise by adjusting the mean and standard deviation of the distribution, but this method is only suitable for regular strip noise in uniform scenes, and cannot accurately separate the strip noise when facing more complex strip noise, which easily leads to residual noise or over-smoothing; the filtering-based method (including wavelet filtering, adaptive filtering, etc.) uses frequency domain analysis or spatial filtering to separate and remove the strip noise, but this method cannot accurately distinguish non-periodic strip noise from image content, which easily loses details or introduces ringing artifacts; the deep learning-based method can directly learn the mapping relationship from the noisy image to the clean image, so that the strip noise can be quickly and accurately estimated, but due to the lack of real strip noise data, the model needs to use artificially generated simulated strip noise when training, which makes it difficult to achieve the expected effect when removing real strip noise.

[0004] Therefore, there is currently a lack of a method that can accurately and efficiently remove the strip noise of remote sensing images. SUMMARY

[0005] The present application provides a remote sensing image strip noise removal method, device, equipment and storage medium to solve the defects of the above related technologies. The technical solution is as follows: In a first aspect, the present application provides a remote sensing image strip noise removal method, characterized in that it comprises: filtering an original image containing strip noise to obtain a residual image; estimate each estimated strip noise column of the original image based on the residual image; divide each estimated strip noise column of the original image into a feature region along a vertical strip noise direction; cluster each feature region to obtain each strip noise column of the original image; construct a strip noise constraint model based on the L1 norm of the difference between the image gray value at the strip noise column of the original image, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column closest to the strip noise column as a constraint condition; calculate the strip noise constraint model to obtain the noise gray value of each strip noise column and obtain a strip noise image; remove the strip noise image from the original image to output a denoised image.

[0006] In an optional implementation of the first aspect, the estimating each estimated strip noise column of the original image based on the residual image comprises: divide the residual image into multiple two-dimensional residual image blocks of the same size along the strip noise direction, stack each two-dimensional residual image block to obtain a three-dimensional residual image block, and divide the residual image multiple times to obtain a three-dimensional residual image block group composed of multiple three-dimensional residual image blocks, each three-dimensional residual image block having a different number of two-dimensional residual image blocks; extract features from each three-dimensional residual image block in the three-dimensional residual image block group to obtain image features of each three-dimensional residual image block, and determine the strip noise column in the corresponding three-dimensional residual image block according to the image features of each three-dimensional residual image block; for each column in the residual image, if the proportion of the number of image features of the corresponding column that are strip noise columns to the total number of image features of all three-dimensional residual image blocks is greater than or equal to a first preset threshold, the corresponding column is determined to be an estimated strip noise column; obtain each estimated strip noise column of the original image.

[0007] In an optional implementation of the first aspect, after estimating each estimated strip noise column of the original image based on the residual image, determining the type of the strip noise according to the proportion of all estimated strip noise columns in the original image comprises: calculate the proportion of the total number of all estimated strip noise columns to the total number of columns of the original image, and if the proportion is greater than or equal to a second preset threshold, determine that the strip noise of the original image is dense strip noise; otherwise, determine that the strip noise of the original image is sparse strip noise.

[0008] In an optional implementation of the first aspect, the feature region is divided along the vertical strip noise direction at each estimated strip noise column of the original image, and the feature region includes: The neighboring estimated strip noise column on one side or both sides of each estimated strip noise column is determined along the vertical strip noise direction, respectively, with each estimated strip noise column as a starting point. The region between each estimated strip noise column and the neighboring estimated strip noise column on one side or both sides is extracted from the original image, to obtain a feature region corresponding to the estimated strip noise column.

[0009] In an optional implementation of the first aspect, the strip noise column of the original image is clustered based on each feature region, respectively, and the method includes: The feature value mean of each column in each feature region is extracted, respectively, and clustering is performed based on the feature value mean of each column in each feature region, to extract a minimum cluster corresponding to each specific region. In the case where the strip noise is sparse strip noise, if the feature value mean of the s-th column in the specific region is equal to the value corresponding to any point in the minimum cluster, the s-th column in the specific region is determined as a strip noise column, otherwise, as a non-strip noise column. In the case where the strip noise is dense strip noise, if the feature value mean of the s-th column in the specific region is equal to the value corresponding to any point in the minimum cluster, the s-th column in the specific region is determined as a non-strip noise column, otherwise, as a strip noise column.

[0010] In an optional implementation of the first aspect, the strip noise constraint model is constructed with the L1 norm of the difference between the image gray value at the strip noise column on the original image, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column closest to the strip noise column as a constraint condition, and the method includes: The original image is evenly divided into a plurality of two-dimensional image blocks of the same size along the strip noise direction, to obtain a three-dimensional image block stacked by all the two-dimensional image blocks. For each two-dimensional image block in the three-dimensional image block, the strip noise constraint model is constructed with the L1 norm of the difference between the image gray value at the strip noise column on each two-dimensional image block, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column closest to the strip noise column as a constraint condition.

[0011] In an optional implementation of the first aspect, in the case where the strip noise is sparse strip noise, the strip noise constraint model is calculated to obtain the noise gray value of each strip noise column, to obtain a strip noise image, and the method includes: The strip noise constraint model is calculated to obtain noise grayscale values of all strip noise columns in each two-dimensional image block, and to obtain a corresponding sub-block strip noise of each two-dimensional image block. The corresponding sub-block strip noises of all two-dimensional image blocks are spliced to obtain a strip noise image of the original image.

[0012] In an optional implementation of the first aspect, in the case where the strip noise is dense strip noise, the calculation of the strip noise constraint model to obtain noise grayscale values of each strip noise column and to obtain a strip noise image includes: Each non-strip noise column is taken as a reference column, and noise grayscale values of a strip noise column closest to each reference column in each two-dimensional image block are calculated by using the strip noise constraint model to obtain a corresponding sub-block strip noise of each two-dimensional image block. The corresponding sub-block strip noises of all two-dimensional image blocks are spliced to obtain an intermediate strip noise image of the original image. The original image is removed on the basis of the strip noise image to output a denoised image, including: The original image is removed on the basis of the intermediate strip noise image to output an intermediate denoised image. If the intermediate denoised image does not contain a strip noise column, the intermediate denoised image is output as the denoised image. Otherwise, a pixel column at a position of a removed strip noise column is updated as a reference column, and the calculation of noise grayscale values of a strip noise column closest to each reference column in each two-dimensional image block by using the strip noise constraint model and the subsequent steps are performed.

[0013] In a second aspect, the present application further provides a strip noise removal device for remote sensing images, including: An image processing unit is configured to perform filtering processing on an original image containing strip noise to obtain a residual image. A strip noise estimation unit is configured to estimate each estimated strip noise column of the original image based on the residual image. A strip noise calculation unit is configured to divide each estimated strip noise column of the original image to obtain a feature region along a vertical strip noise direction. The strip noise calculation unit is further configured to obtain each strip noise column of the original image based on clustering of each feature region. The strip noise calculation unit is further configured to construct a strip noise constraint model by taking a minimum L1 norm of a difference between an image grayscale value at a strip noise column of the original image, a noise grayscale value of the strip noise column, and a non-noise grayscale value of a non-strip noise column closest to the strip noise column as a constraint condition. The strip noise calculation unit is further configured to calculate the strip noise constraint model to obtain a noise gray value of each of the strip noise columns and to obtain a strip noise image. The denoising unit is configured to remove the strip noise image from the original image to output a denoised image.

[0014] In a third aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the program to implement the method of the first aspect or any of the implementation manners of the first aspect.

[0015] The technical scheme provided by some embodiments of the present application has at least the following beneficial effects: The strip noise removal method for remote sensing images provided by the present application obtains the estimated position of the strip noise, i.e., the estimated strip noise column, through the residual graph, and further divides the feature region on the basis of the estimated strip noise column to determine the accurate position of the strip noise, i.e., the strip noise column, through clustering analysis, thereby overcoming the low precision defect caused by the related art of determining the position of the strip noise only through the residual graph, and overcoming the influence of the residual detail information in the residual graph on the strip noise to achieve accurate positioning of all strip noise columns or non-strip noise columns.

[0016] On the other hand, the present application obtains the strip noise image by solving the strip noise constraint model under the constraint condition that the L1 norm of the difference between the image gray value at the strip noise column on the original image, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column closest to the strip noise column is minimum, which utilizes the information of the adjacent non-strip noise column to maximize the preservation of the detail information and structural information of the original image, does not damage the effective information of the image, and can achieve effective removal of the real strip noise. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical scheme in the present application or the related art, the following will briefly introduce the drawings needed to be used in the embodiments or the related art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without any creative effort.

[0018] Figure 1 is one of the flowcharts of the strip noise removal method for remote sensing images provided by the embodiments of the present application; Figure 2 is an image schematic diagram of the filtering process of the strip noise removal method for remote sensing images provided by the embodiments of the present application; Figure 2Fig. 2 is a schematic diagram of the noise image in part (a), Figure 2 Fig. 3 is a schematic diagram of the smoothed image in part (b), Figure 2 Fig. 4 is a schematic diagram of the residual image in part (c); Figure 3 Fig. 5 is a schematic diagram of image block processing of a remote sensing image strip noise removal method provided by an embodiment of the present application; Figure 3 Fig. 6 is a schematic diagram of the residual image processing result in part (d), Figure 3 Fig. 7 is a schematic diagram of a three-dimensional residual image block group in part (e), Figure 3 Fig. 8 is a schematic diagram of Gabor features in part (f); Figure 4 Fig. 9 is a schematic diagram of clustering processing of a remote sensing image strip noise removal method provided by an embodiment of the present application; Figure 5 Fig. 10 is a schematic diagram of a flow of a remote sensing image strip noise removal method provided by an embodiment of the present application; Figure 6 Fig. 11 is a schematic diagram of a to-be-processed image of a remote sensing image strip noise removal method provided by an embodiment of the present application; Figure 7 Fig. 12 is a schematic diagram of a denoising effect of a remote sensing image strip noise removal method provided by an embodiment of the present application; Figure 8 Fig. 13 is a schematic diagram of a denoising effect of a remote sensing image strip noise removal method provided by an embodiment of the present application; Figure 9 Fig. 14 is a schematic diagram of a structure of a remote sensing image strip noise removal device provided by an embodiment of the present application; Figure 10 Fig. 15 is a schematic diagram of a structure of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0020] The terms "comprise", "comprising", "include", "including", "have" and "having" and any variations thereof in the Specification and in the Claims herein are intended to cover both the express stated features or steps and also other features or steps not expressly stated or inherent to the described methods, processes, systems, products or devices. For example, a process, method, system, product or device that comprises a list of steps or modules is not limited to those steps or modules which are recited but instead can also include additional steps or modules not expressly listed or inherent to such process, method, system, product or device.

[0021] It should be noted that the terms "first" and "second" in the present application are merely used to distinguish similar objects, and do not represent a specific order of the objects. Understandably, the "first" and "second" can be interchanged in a specific order or sequence as permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those described or illustrated herein.

[0022] The present application will be described in detail below with reference to specific embodiments.

[0023] Next, a remote sensing image strip noise removal method provided by an embodiment of the present application will be described below. Figure 1 Figure 1 Figure 1 A flowchart of a remote sensing image strip noise removal method provided by an embodiment of the present application is shown. As shown in Figure 1 The method comprises the following steps: S101, performing filtering processing on an original image containing strip noise to obtain a residual image; S102, estimating each estimated strip noise column of the original image based on the residual image; S103, dividing each estimated strip noise column of the original image to obtain a feature region along a direction perpendicular to the strip noise; S104, respectively clustering each feature region to obtain each strip noise column of the original image; S105, constructing a strip noise constraint model with the L1 norm of the difference between the image gray value at the strip noise column on the original image, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column closest to the strip noise column as a constraint condition; S106, calculating the strip noise constraint model to obtain the noise gray value of each strip noise column and obtain a strip noise image; S107, removing the strip noise image from the original image to output a denoised image.

[0024] ​​In some embodiments, in S101, the input original image can be subjected to multiple Gaussian low-pass filtering processes to obtain a Gaussian-smoothed image, and a residual map can be generated. If we denote... Y is the original image after Gaussian filtering and R is the residual image. The process of generating the residual image through filtering can be represented as follows: ; The generated residual map R removes structural information and most detail from the original image, while retaining all stripe noise, thus allowing it to be used to estimate approximate values ​​of the stripe noise. By employing a Gaussian low-pass filter, ringing effects can be avoided.

[0025] For example, a noisy image containing striped noise with a size of 400×400 pixels is taken and subjected to Gaussian low-pass filtering. The value range of the noisy image is normalized to an interval. During the test, the remote sensing image was subjected to Gaussian filtering five times in a loop, with each iteration involving 800 Gaussian low-pass filtering cycles on the noisy image. The resulting residual image is shown below. Figure 2 The smoothed image shown in section (a) is as follows: Figure 2 As shown in section (b), the residual plot is as follows: Figure 2 As shown in section (c).

[0026] It should be noted that the strip noise can be horizontal or vertical. The distribution pattern of the strip noise can be easily determined directly from the original image. For ease of calculation, the image with horizontal strip noise can be rotated to convert it into vertical strip noise. The distribution patterns of horizontal or vertical strip noise can be converted between each other by rotating the image as needed. In the embodiments of this application, the form of vertical strip noise is described.

[0027] In some embodiments, in S102, each estimated strip noise column in the original image can be estimated based on the residual map, that is, the estimated position of the strip noise column is obtained.

[0028] It should be noted that related technologies directly use filters to extract strip noise features from the original image. However, the original structural and detail information in the original image can affect the extraction of strip noise features, and some structural, detail information and strip noise features may become difficult to distinguish. This application extracts strip noise features based on residual maps, which can avoid confusing structural, detail information and strip noise features, thereby overcoming the shortcomings of related technologies.

[0029] S102 specifically includes the following steps: S1021, divide the residual image into a plurality of two-dimensional residual image blocks of the same size along the direction of the strip noise, and obtain a three-dimensional residual image block by stacking each two-dimensional residual image block; and obtain a three-dimensional residual image block group composed of a plurality of three-dimensional residual image blocks obtained by multiple divisions, each three-dimensional residual image block having a different number of two-dimensional residual image blocks.

[0030] When the residual image is divided into blocks, we divide it into a three-dimensional residual image block group with block sizes (i.e., the number of rows of each block) of , respectively, wherein the number of rows of each two-dimensional residual image block in the three-dimensional residual image block with the corresponding serial number p is , respectively, i.e., the residual image is divided into a plurality of two-dimensional residual image blocks with a number of rows of , and the three-dimensional residual image block with the serial number p is obtained by stacking the plurality of two-dimensional residual image blocks with a number of rows of ; and the size of the residual image is ; Exemplarily, as shown in Figure 3 , the residual image shown in part (d) of Figure 3 is processed, and the obtained three-dimensional residual image block group is shown in part (e) of Figure 3 .

[0031] By dividing the residual image R into blocks, the influence of residual detail information can be greatly reduced, the direction and structure information of the strip noise can be effectively preserved, and the position information of the strip noise will not be damaged, which is conducive to improving the calculation accuracy of the strip noise.

[0032] S1022, respectively, for each three-dimensional residual image block in the three-dimensional residual image block group, feature extraction is performed to obtain image features of each three-dimensional residual image block, and the strip noise column in the corresponding three-dimensional residual image block is determined according to the image features of each three-dimensional residual image block.

[0033] Specifically, a group of dense Gabor filters with different directions and different scales can be constructed to perform feature extraction on each two-dimensional residual image block in each three-dimensional residual image block to obtain image features of each two-dimensional residual image block in each three-dimensional residual image block.

[0034] More specifically, in this way, image features of strip noise at different directions and scales can be obtained, that is, Gabor features extracted by dense Gabor filters. Based on the Gabor features, it is easy to determine which columns in the corresponding three-dimensional residual image block are strip noise columns.

[0035] For example, the Gabor features of the extracted three-dimensional residual image block group are as follows: Figure 3 As shown in section (f), it can be seen that in the residual plot After segmenting the 3D blocks and processing them with a dense Gabor filter, the Gabor features of the strip noise in all directions can be completely extracted, thus effectively overcoming the residual map. The impact of residual details on the extraction of Gabor features from striped noise.

[0036] For example, the direction of the constructed dense Gabor filter can be determined according to the step size. In scope The sampling is performed within the area, and the scale of the dense filter constructed (i.e., the parameters) , The size of the 3D block is positively correlated with the block size in the following text, and the constructed dense filter is shown in the following formula: ; ; ; ; in, Let be the ordinal number of the direction of the dense Gabor filter. The ordinal number of the 3D residual image block; For the first The scale of a 3D residual image patch; M is the number of rows in the residual map; The standard deviation of the initial wavelength, The standard deviation of the initial Gaussian function; This indicates rounding up; x and y represent the coordinates in the image plane. The wavelength of the sinusoidal plane wave determines the center frequency of the filter; the smaller the wavelength, the higher the frequency. It is the phase offset that determines the starting position of the cosine function; The standard deviation of the Gaussian function determines the spatial range of the filter. The larger the standard deviation, the wider the range of the filter and the lower the degree of spatial localization. It is the aspect ratio, used to control the Gaussian function in space. and Different degrees of expansion in different directions; and The coordinates after rotation are related to the original x and y coordinates as follows: , ; Indicates the direction of the filter, with a value range of 1. It can control the filter's response to textures in different directions.

[0037] Specifically, when using a dense Gabor filter to extract Gabor features of strip noise in each 3D residual image block, the number of Gabor features extracted for each 3D residual image block can be determined, as follows: ; Furthermore, the total number of Gabor features extracted from all 3D residual image blocks can be determined as follows: ; in, To extract from 3D residual image blocks The extracted Gabor feature number, This represents the total number of Gabor features; The angle sampling step size of the dense Gabor filter. Possible values However, the embodiments in this application do not limit this.

[0038] S1023, for each column in the residual map, if the proportion of the number of image features in the corresponding column that are determined to be strip noise columns to the total number of image features in all three-dimensional residual image blocks is greater than or equal to a first preset threshold, then the corresponding column is determined to be an estimated strip noise column.

[0039] Specifically, a voting method can be used to fuse the extracted Gabor features of the strip noise to obtain a coarse localization result of the strip noise. If the proportion of image features in the m-th column of the residual image that indicate a strip noise column to the total number of image features in all three-dimensional residual image blocks is greater than or equal to a first preset threshold, then the voting determines the approximate location of the m-th column as a strip noise column. For example, the first preset threshold can be set to 90%, and the voting process can be performed using the following formula: ; in, For the coarse localization result of the strip noise in the m-th column of the staggered graph R, if Then the m-th column is the estimated strip noise column. N represents the number of image features in each block of Gabor features that indicate the m-th column is a strip noise column, and N is the number of columns in the residual map R.

[0040] It should be noted that the residual image R is obtained based on the filtering of the original image, and the residual image R has the same number of rows and columns as the original image. Therefore, each estimated strip noise column obtained based on the residual estimation is also each estimated strip noise column of the original image, S1024, obtaining each estimated strip noise column of the original image.

[0041] In some embodiments, after determining each estimated strip noise column of the original image in S102, the type of the strip noise can also be determined according to the proportion of all estimated strip noise columns in the original image, which is specifically implemented as follows: Calculate the proportion of the total number of all estimated strip noise columns to the total number of columns of the original image. If the proportion is greater than or equal to a second preset threshold, it is determined that the strip noise of the original image is dense strip noise. Otherwise, it is determined that the strip noise of the original image is sparse strip noise.

[0042] For example, the following formula is applied: ; Wherein, represents the type of strip noise, represents dense strip noise, represents sparse strip noise.

[0043] For example, the second preset threshold can be 50%. If the proportion of the total number of all estimated strip noise columns to the total number of columns of the original image is greater than or equal to 50%, , it is determined that the type of the strip noise of the original image is dense strip noise. Otherwise, , it is determined that the type of the strip noise of the original image is sparse strip noise.

[0044] In some embodiments, the accurate strip noise position is further determined based on the estimated strip noise column, that is, the strip noise column of the original image is obtained through S103-S104, which specifically includes: S103, the original image is divided into a feature region along the vertical strip noise direction at each estimated strip noise column, including: Respectively taking each estimated strip noise column as a starting point, the adjacent estimated strip noise columns on one side or both sides of each estimated strip noise column are determined along the vertical strip noise direction.

[0045] Specifically, the position relationship between the estimated strip noise columns can be determined according to the positions of all estimated strip noise columns, and the estimated strip noise columns adjacent to the left and / or right sides of the estimated strip noise column can be determined along the vertical strip noise direction.

[0046] For example, when the estimated strip noise column is located at the left edge of the original image, there is no adjacent estimated strip noise column on the left side, and thus the adjacent estimated strip noise column on the right side can be determined; when the estimated strip noise column is located at the right edge of the original image, there is no adjacent estimated strip noise column on the right side, and thus the adjacent estimated strip noise column on the left side can be determined; when the estimated strip noise column is located at other positions of the original image, there are adjacent estimated strip noise columns on the left side and the right side, and thus the adjacent estimated strip noise columns on the left side and the right side can be determined.

[0047] Further, an area between each estimated strip noise column and the adjacent estimated strip noise column on one side or both sides in the original image is extracted, to obtain a feature area containing the corresponding estimated strip noise column, and a plurality of feature areas are constructed as shown in Figure 4

[0048] For example, when the estimated strip noise column is located at the left edge of the original image, the left edge of the original image can be selected as the left edge of the feature area, and any column in the area between the estimated strip noise column and the adjacent estimated strip noise column on the right side can be selected as the right edge of the feature area, so as to construct the feature area; when the estimated strip noise column is located at the right edge of the original image, the right edge of the original image can be selected as the right edge of the feature area, and any column in the area between the estimated strip noise column and the adjacent estimated strip noise column on the left side can be selected as the left edge of the feature area, so as to construct the feature area; when the estimated strip noise column is located at other positions of the original image, any column in the area between the estimated strip noise column and the adjacent estimated strip noise column on the right side can be selected as the right edge of the feature area, and any column in the area between the estimated strip noise column and the adjacent estimated strip noise column on the left side can be selected as the left edge of the feature area, so as to construct the feature area.

[0049] It should be noted that the embodiments of the present application do not limit this, and the feature area only needs to contain the estimated strip noise column as the starting point, and the plurality of feature areas can contain overlapping areas.

[0050] S104, each strip noise column of the original image is obtained based on clustering of each feature area, specifically including the following steps: S1041, the feature value mean of each column in each feature area is extracted, and clustering processing is performed based on the feature value mean of each column in the feature area, to extract the minimum cluster corresponding to each specific area.

[0051] Specifically, the feature value of each column is specifically the gray value of each pixel on the corresponding column, and the feature value mean can be understood as the mean of the gray values of all pixels on the corresponding column.

[0052] ​Specifically, the k-means clustering method can be selected for the clustering processing, and the only parameter to be determined when the k-means clustering method is executed is the number of clusters, denoted as k. In the embodiments of the present application, the column means are only two types of strip noise column means and non-strip noise column means, and thus k can be 2, and the minimum cluster and the maximum cluster can be obtained through the clustering processing. The formula is as follows: wherein L represents the clustering result, represents a function of the clustering processing, and represents the mean value of the feature value of each column in the specific region.

[0053] In some embodiments, in the case of sparse strip noise, if the mean value of the feature value of the s-th column in the specific region is equal to the value corresponding to any point in the minimum cluster , it is determined that the s-th column in the specific region is a strip noise column, otherwise it is a non-strip noise column; In the case of dense strip noise, if the mean value of the feature value of the s-th column in the specific region is equal to the value corresponding to any point in the minimum cluster, it is determined that the s-th column in the specific region is a non-strip noise column, otherwise it is a strip noise column.

[0054] Specifically, according to the values corresponding to the points in the minimum cluster , the strip noise column or the non-strip noise column can be determined, and the specific method is as follows: wherein is the strip noise positioning result of the sparse strip noise, is the non-strip noise positioning result of the dense strip noise, is the value corresponding to the r-th point in the minimum cluster, is the value corresponding to the t-th point in the maximum cluster, is the mean value of the feature value of the s-th column in the extracted feature region, is the number of elements of the minimum cluster, is the number of elements of the maximum cluster, , If Z feature regions are obtained in step S103, there are a total of columns in the z-th feature region, and Z is the total number of feature regions.

[0055] In this way, the clustering is performed on each feature region, and each strip noise column of the original image can be obtained to determine the accurate position of the strip noise. ​​​

[0056] It should be noted that for remote sensing images with strip noise , removing strip noise means decomposing clean image X and strip noise image S from remote sensing image Y, if the strip noise is additive strip noise, then remote sensing image Y, clean image X and strip noise image S satisfy the following relationship: If the strip noise is multiplicative strip noise, then: In the formula, indicates the Hadamard product of the matrix, and the value of the image is non-negative. The multiplicative strip noise can be converted into additive strip noise by using the logarithmic function, so that the strip noise can be treated as additive noise.

[0057] In some embodiments, in S105, a strip noise constraint model can be constructed by taking the L1 norm of the difference between the image gray value at the strip noise column of the original image, the noise gray value of the strip noise column, and the non-noise gray value of the nearest non-strip noise column to the strip noise column as the constraint condition.

[0058] In order to obtain more accurate results, and considering that the longitudinal strip noise component will affect the row direction gradient of the remote sensing image, but has good smoothness along the strip noise direction, that is, the noise intensity of a column of strip noise is the same in the local range along the strip noise direction, thus, the original image Y can also be processed by averaging and blocking along the strip noise direction, including the steps of: averaging the original image along the strip noise direction to divide it into a plurality of two-dimensional image blocks of the same size to obtain a three-dimensional image block stacked by all two-dimensional image blocks , each two-dimensional image block has the same size, that is, each two-dimensional image block has the same number of rows.

[0059] In this way, the strip noise in each two-dimensional image block can satisfy the characteristic of the same noise intensity. Further, using the obtained strip noise positioning result or non-strip noise positioning result and the local continuity of the bottom layer image, the steps of S105 are executed, including: ​​​​For each two-dimensional image block in the three-dimensional image block, a strip noise constraint model is constructed by minimizing the L1 norm of the difference between the image gray value at the strip noise column, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column that is closest to the strip noise column, as shown in the following formula: ; in, For the first Block 2D image block Upper The image grayscale value at the nth bar noise column can be directly obtained from the nth bar noise column of the two-dimensional image block. The image grayscale values ​​extracted at the positions of each strip of noise column; Let be the physical quantity to be determined, specifically the th Block 2D image block Upper The noise grayscale values ​​of each strip noise column; In order to be with the first Block 2D image block Upper The non-noise grayscale value of the nearest non-noise column to the bar noise column; Used to calculate the L1 norm of the inner element "*". Let be the function used to find the minimum value. For strip noise constrained models, Let n be the nth strip noise column to be solved.

[0060] It should be noted that when determining the non-strip noise column that is closest to the strip noise column, you can choose the adjacent non-strip noise column on either side of the strip noise column as the non-strip noise column that is closest to the strip noise column.

[0061] Further, step S106 is executed to solve the strip noise constraint model, obtain the noise grayscale value of each strip noise column, obtain the strip noise image S, and step S107 is executed to remove the strip noise image S from the original image Y, outputting the denoised image X. Specifically, the following embodiments are included: In some embodiments, when the strip noise is sparse strip noise, step S106 includes: The strip noise constraint model is calculated to obtain each two-dimensional image block. By analyzing the noise grayscale values ​​of all strip noise columns, each two-dimensional image block can be obtained. The corresponding segmented strip noise; stitch together all 2D image blocks The corresponding segmented strip noise is used to obtain the strip noise image S of the original image Y.

[0062] In some embodiments, in the case that the strip noise is dense strip noise, in the strip noise constraint model, the calculation process is as follows Figure 5 The steps of S106 specifically include the following steps as shown in the flow chart: S1061, taking each non-strip noise column as a reference column.

[0063] S1062, calculating the noise gray value of the strip noise column closest to each reference column in each two-dimensional image block through the strip noise constraint model to obtain the corresponding block strip noise of each two-dimensional image block.

[0064] Specifically, in the case that the strip noise is dense strip noise, in the first calculation process, based on the accurate position of the non-strip noise column obtained by clustering in S104, the strip noise column closest to each non-strip noise column can be determined, and thus the corresponding block strip noise is calculated in S1062.

[0065] S1063, splicing the corresponding block strip noise of all two-dimensional image blocks to obtain the intermediate strip noise image of the original image.

[0066] Further, the steps of S107 include the following steps: S1071, removing the intermediate strip noise image from the original image to output an intermediate denoised image.

[0067] S1072, judging whether the intermediate denoised image contains a strip noise column.

[0068] If the intermediate denoised image does not contain a strip noise column, the steps of S1073 are performed, including: S1073, outputting the intermediate denoised image as the denoised image.

[0069] Otherwise, the steps of S1074 are performed, including: S1074, updating the pixel column at the position of the removed strip noise column to a reference column.

[0070] Then, go to S1062 to perform the steps of calculating the noise gray value of the strip noise column closest to each reference column in each two-dimensional image block through the strip noise constraint model and the subsequent steps.

[0071] In some embodiments, the method of global search optimization is used to realize efficient solution of the strip noise image In some embodiments, the process of solving the strip noise constraint model includes the following steps: S601, based on each estimated strip noise column in each three-dimensional residual image block in the residual image R obtained in S101, the maximum value of the gray value of the strip noise can be determined and the minimum value .

[0072] determine the number of candidate solutions calculated , the maximum value of the number of iterations , the objective function : ; S602, according to the maximum value of the gray value and the minimum value generate a series of candidate solutions, candidate solution vector composed of candidate solutions ; Generate candidate solutions based on the following formula: ; where, is a random array with dimension D uniformly distributed in the interval , and and are the upper and lower bounds of the search space, respectively.

[0073] S603, determine the growth rate of candidate solution , including: ; ; where, is the growth rate, is the growth rate, is the correction coefficient of the growth rate, is the growth rate at time in the discrete time system, is the growth rate at time t+1 in the discrete time system, is a random number of random variables, and its probability density function is equal to ; denotes a random vector with dimension D, which is a random number in the standard Gaussian normal distribution; is the growth rate of candidate solution .

[0074] S603, according to the objective function corresponding to each candidate solution The numerical values ​​are used to sort the candidate solution vectors in ascending order of objective function values. Each candidate solution Sorting is performed, that is, arranging the members from best to worst. The sorting result is: ; ; in, For the reordered elements, This represents the sorted candidate solution vector. This refers to the candidate solution corresponding to the minimum value of the objective function among the sorted candidate solution vectors, which is also the current optimal solution. , The first candidate solution vector in the sorted list There are 10 candidate solutions.

[0075] Specifically, after S603, for candidate solutions If candidate solution Corresponding objective function value Less than Multiply by parameter Then, the steps in S604 are executed, including: S604, Determining candidate solutions closest neighboring candidate solution Apply the formula: ; Specifically, if Not the first element in the sorted result , For the sorted results, the first candidate solutions Then take the value that is first in the sort. Take adjacent candidate solutions .

[0076] Furthermore, based on neighboring candidate solutions and growth rate Optimize candidate solutions : ; ; Specifically, after S603, for candidate solutions If the objective function value Greater than or equal to Multiply by parameter The steps of S605 include: S605, based on optimal solution Continue optimizing candidate solutions until the optimal solution is found. The optimal solution of the surrounding traversal is more optimal in the first iteration, and the formula is applied: ; wherein, the new candidate solution obtained based on the original optimal solution of the surrounding traversal; Further, the new value of the growth rate of the current member is calculated by the following formula: ; Specifically, the steps of S604 or S605 are performed on each candidate solution in the candidate solution vector , so that each candidate solution in the candidate solution vector is optimized until each candidate solution in it is optimized, and a new candidate solution vector is obtained.

[0077] S606, the candidate solution vector of the previous iteration is combined with the newly generated candidate solution vector of the current iteration to obtain: ; Each element in is sorted to obtain a candidate solution vector , and the formula is applied: ; The top members in the vector are selected as the final members of the current candidate solution, and a screened candidate solution vector is obtained, that is: .

[0078] The selected final members of the current iteration are taken as the initial candidate solution vector of the next iteration, and .

[0079] If the current iteration number is less than the given maximum iteration number, jump to the step of S602.

[0080] If the current iteration number is equal to the maximum iteration number given in S601, execute S607, including: obtain the candidate solution vector output by the last iteration, and extract the optimal solution from the candidate solution vector output by the last iteration Thus, the corresponding noise gray value is calculated. Specifically, the noise gray value of the jth strip noise column in the ith block is calculated as follows: Block two-dimensional image block The noise gray value of the jth strip noise column in the ith block is calculated as follows:

[0081] Through the steps of S601-S607, each strip noise column can be calculated.

[0082] In some embodiments, after S107, the obtained denoised image is again subjected to strip noise detection to determine whether strip noise still exists. If strip noise still exists, the denoising process of steps S101-S107 is repeated until complete removal of strip noise is achieved.

[0083] Exemplarily, the noise reduction ratio (NR), the blind image spatial quality evaluation method (BRISQUE), and the deep convolutional neural network-based blind image quality evaluation method (DIQA) can be used as quantitative evaluation indexes of the denoised image, and the column average cross-track curve and the power spectrum curve can be used as qualitative evaluation indexes of the denoised image to evaluate the performance of the method of the present application in removing real strip noise. The specific calculation method of NR is as follows: ; Wherein, represents the power of the frequency component of the original image containing strip noise, represents the power of the frequency component of the denoised image.

[0084] It should be noted that the complete removal of strip noise in the above embodiments can be understood as that, after detecting the denoised image based on the quantitative evaluation index and the qualitative evaluation index, it is determined according to the obtained index value that there is no strip noise in the denoised image obtained by processing.

[0085] In a specific embodiment, three real noise images provided by the China Resource Satellite Center are used as shown in FIG. 1, Figure 6 from left to right in FIG. 1 are Example 1, Example 2, and Example 3, and the three real images are normalized to Figure 6 , to detect the noise reduction effect of the method of the present application.

[0086] ​Exemplarily, the de-noising effects of the method provided in the embodiments of the present application and the methods used in related technologies are compared respectively, the methods used in related technologies include but are not limited to: a reweighted block sparse regularization de-striping method (RBS), an ADMM-based optimization model (ADOM), a double low-rank Hodrick-Prescott matrix decomposition (LRHP) based strip noise removal method, an anisotropic variation regularization and double low-rank matrix decomposition (ATVDLR) based strip noise removal method, a deep iterative reconstructive network (DIRN) and a deep recursive convolutional neural network (DnRCNN) based de-striping method. The comparison results are shown in FIGS. 1 to 3. Figure 7 、 Figure 8

[0087] Among them, Figure 7 FIGS. 1 to 3 show the strip noise removal results of RBS, ADOM and LRHP on three real noisy images respectively. Figure 7 The first column from top to bottom in FIGS. 1 to 3 corresponds to example 1, example 2 and example 3 in the three real noisy images respectively. Figure 8 FIGS. 4 to 6 show the strip noise removal results of ATVDLR, DIRN, DnRCNN and the method of the embodiments of the present application on three real noisy images respectively.

[0088] Among them, compared with other methods, the de-noising effects of RBS, ADOM and LRHP are better than those of other related technologies, but these three methods achieve the removal of strip noise at the cost of sacrificing the details of the image, which is not conducive to the application of the de-noised image in high-level tasks. The de-noising effects of ATVDLR, DIRN and DnRCNN are poor, among which, DIRN and DnRCNN are both deep learning based methods, and these two methods use simulation data in the training process, which cannot effectively remove the real strip noise in real remote sensing images; among them, ATVDLR is a joint regularization and low-rank matrix decomposition strip noise removal method, which will cause gray distortion in the de-noised image, which seriously affects the application of the de-noised image in subsequent inversion. ​

[0089] Therefore, the method provided by the embodiment of the present application can effectively remove the strip noise, and does not cause problems such as gray distortion and sacrifice of detailed information of an image, and is superior to the denoising method provided by the related art.

[0090] The following is a device embodiment of the present application, which can be used to execute the method embodiment of the present application. For details not disclosed in the device embodiment of the present application, refer to the method embodiment of the present application.

[0091] Next, please refer to Figure 9 The structure diagram of a remote sensing image strip noise removal device provided by an exemplary embodiment of the present application is shown in FIG. 1. The device can be realized by software, hardware or a combination of the two to become all or part of a terminal, and can also be integrated as an independent module on a server. The remote sensing image strip noise removal device in the embodiment of the present application can be applied to a terminal or cloud. The device 90 includes an image processing unit 901, a strip noise estimation unit 902, a strip noise calculation unit 903 and a denoising unit 904, wherein: The image processing unit 901 is configured to perform filtering processing on an original image containing strip noise to obtain a residual image; The strip noise estimation unit 902 is configured to estimate each estimated strip noise column of the original image based on the residual image; The strip noise calculation unit 903 is configured to divide a feature region along a vertical strip noise direction at each estimated strip noise column of the original image; The strip noise calculation unit 903 is further configured to respectively cluster each feature region to obtain each strip noise column of the original image; The strip noise calculation unit 903 is further configured to construct a strip noise constraint model based on a minimum L1 norm of a difference between an image gray value at a strip noise column on the original image, a noise gray value of the strip noise column and a non-noise gray value of a non-strip noise column closest to the strip noise column as a constraint condition; The strip noise calculation unit 903 is further configured to calculate the strip noise constraint model to obtain the noise gray value of each strip noise column, and obtain a strip noise image; The denoising unit 904 is configured to remove the strip noise image on the basis of the original image to output a denoised image.

[0092] It should be noted that the device 90 provided by the above embodiment is only used as an example to illustrate the division of the above functional modules when the device 90 performs the remote sensing image strip noise removal method. In actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and the remote sensing image strip noise removal method provided by the above embodiment belong to the same concept, and the implementation process is detailed in the method embodiment. Therefore, the detailed description is omitted here.

[0093] The embodiment of the present application further provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the program to realize the steps of the method of any one of the above embodiments.

[0094] Please refer to Figure 10 The structure block diagram of an electronic device provided by the embodiment of the present application is shown in the figure.

[0095] As Figure 10 The electronic device 1000 includes a processor 1001 and a memory 1002.

[0096] In the embodiment of the present application, the processor 1001 is the control center of the computer system, which can be the processor of a physical machine or the processor of a virtual machine. The processor 1001 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1001 can be implemented in at least one of the hardware forms of a DSP (Digital Signal Processing), a FPGA (Field-Programmable Gate Array) and a PLA (Programmable Logic Array).

[0097] The processor 1001 can also include a main processor and a coprocessor. The main processor is a processor for processing data in the wake-up state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in the standby state.

[0098] The memory 1002 can include one or more computer-readable storage media. The computer-readable storage media can be non-transitory. The memory 1002 can also include high-speed random access memory and can include non-volatile memory, such as one or more magnetic disk storage devices, flash memory devices, or other non-volatile solid-state storage devices. In some embodiments of the present application, the non-transitory computer-readable storage medium in the memory 1002 is used to store at least one instruction for being executed by the processor 1001 to implement the method in the embodiments of the present application.

[0099] In some embodiments of the present application, the electronic device 1000 further includes a peripheral device interface 1003 and at least one peripheral device 1004. The processor 1001, the memory 1002 and the peripheral device interface 1003 can be connected through a bus or a signal line. Each peripheral device 1004 can be connected to the peripheral device interface 1003 through a bus, a signal line or a circuit board. Specifically, the peripheral device interface 1003 can be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 1001 and the memory 1002.

[0100] In some embodiments of the present application, the processor 1001, the memory 1002 and the peripheral device interface 1003 are integrated on the same chip or circuit board; in some other embodiments of the present application, any one or two of the processor 1001, the memory 1002 and the peripheral device interface 1003 can be implemented on a separate chip or circuit board. The embodiments of the present application do not make specific limitations in this regard.

[0101] The structural block diagram of the electronic device shown in the embodiments of the present application does not constitute a limitation on the electronic device 1000, and the electronic device 1000 can include more or fewer components than shown, or combine certain components, or adopt a different arrangement of components.

[0102] The embodiments of the present application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method of any of the preceding embodiments. The computer-readable storage medium can include, but is not limited to, any type of disk including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro-drive, and a magneto-optical disk, ROM, RAM, EPROM, EEPROM, DRAM, VRAM, flash memory devices, a magnetic or optical card, a nanosystem (including molecular memory IC), or any type of media or device suitable for storing instructions and / or data.

[0103] Those skilled in the art can clearly understand the implementation of the various embodiments by means of software and the necessary general hardware platform from the above description of the embodiments, and of course, the embodiments can also be implemented by hardware. Based on such understanding, the above technical solutions or the part that contributes to the related art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in the various embodiments or some parts of the embodiments.

[0104] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for removing stripe noise from remote sensing images, characterized in that, include: The residual map is obtained by filtering the original image containing strip noise; Based on the residual map, each estimated strip noise column of the original image is obtained; Feature regions are obtained by dividing the original image at each estimated strip noise column along the vertical strip noise direction; Each strip noise column of the original image is obtained by clustering based on each feature region; A strip noise constraint model is constructed by taking the minimum L1 norm of the difference between the image gray value at the strip noise column, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column that is closest to the strip noise column as the constraint condition. The strip noise constraint model is calculated to obtain the noise grayscale value of each strip noise column, thus obtaining the strip noise image; The striped noise image is removed from the original image to output a denoised image.

2. The method for removing strip noise from remote sensing images according to claim 1, characterized in that, The step of estimating each estimated strip noise column of the original image based on the residual map includes: The residual map is divided into multiple two-dimensional residual image blocks of the same scale along the direction of the strip noise. A three-dimensional residual image block is obtained by stacking each two-dimensional residual image block. A three-dimensional residual image block group is obtained by dividing multiple three-dimensional residual image blocks, and each three-dimensional residual image block has a different number of two-dimensional residual image blocks. Feature extraction is performed on each three-dimensional residual image block in the three-dimensional residual image block group to obtain the image features of each three-dimensional residual image block. Based on the image features of each three-dimensional residual image block, the strip noise column in the corresponding three-dimensional residual image block is determined. For each column in the residual image, if the proportion of the number of image features that are determined to be strip noise columns in the corresponding column is greater than or equal to the total number of image features in all three-dimensional residual image blocks, then the corresponding column is determined to be an estimated strip noise column. Each estimated strip noise column of the original image is obtained.

3. The method for removing strip noise from remote sensing images according to claim 1, characterized in that, After estimating each estimated strip noise column of the original image based on the residual map, the type of strip noise is determined according to the proportion of all estimated strip noise columns in the original image, including: Calculate the ratio of the total number of all estimated strip noise columns to the total number of columns in the original image. If the ratio is greater than or equal to a second preset threshold, determine that the strip noise in the original image is dense strip noise. Otherwise, the strip noise of the original image is determined to be sparse strip noise.

4. The method for removing strip noise from remote sensing images according to claim 1, characterized in that, The feature regions are obtained by dividing the original image at each estimated strip noise column along the vertical strip noise direction, including: Starting from each estimated strip noise column, determine the estimated strip noise columns adjacent to any one or both sides of each estimated strip noise column along the direction perpendicular to the strip noise; Extract the region between each estimated strip noise column and the estimated strip noise columns adjacent to any one or both sides of the original image to obtain a feature region containing the corresponding estimated strip noise column.

5. The method for removing strip noise from remote sensing images according to claim 3, characterized in that, The step of obtaining each strip noise column of the original image based on clustering of each feature region includes: Extract the mean feature value of each column in each feature region, and perform clustering based on the mean feature value of each column in each feature region to obtain the minimum cluster corresponding to each specific region. In the case that the strip noise is sparse strip noise, if the mean value of the eigenvalues ​​in the s-th column in a specific region is equal to the value corresponding to any point in the minimum cluster, then the s-th column in the specific region is determined to be a strip noise column; otherwise, it is a non-strip noise column. In the case where the strip noise is dense strip noise, if the mean value of the eigenvalues ​​in the s-th column of a specific region is equal to the value corresponding to any point in the minimum cluster, then the s-th column of the specific region is determined to be a non-strip noise column; otherwise, it is a strip noise column.

6. The method for removing strip noise from remote sensing images according to claim 5, characterized in that, The strip noise constraint model is constructed by using the minimum L1 norm of the difference between the image grayscale value at the strip noise column in the original image, the noise grayscale value of the strip noise column, and the non-noise grayscale value of the non-strip noise column closest to the strip noise column as a constraint condition, including: The original image is divided into multiple two-dimensional image blocks of the same scale along the direction of the strip noise, and a three-dimensional image block is obtained by stacking all the two-dimensional image blocks. For each two-dimensional image block in the three-dimensional image block, a strip noise constraint model is constructed by taking the minimum L1 norm of the difference between the image gray value at the strip noise column, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column that is closest to the strip noise column as the constraint condition.

7. The method for removing strip noise from remote sensing images according to claim 6, characterized in that, In the case where the strip noise is sparse strip noise, the step of calculating the strip noise constraint model to obtain the noise grayscale value of each strip noise column and thus obtaining the strip noise image includes: The noise grayscale values ​​of all strip noise columns in each two-dimensional image block are obtained by calculating the strip noise constraint model, and the block strip noise corresponding to each two-dimensional image block is obtained. By stitching together the segmented strip noise corresponding to all two-dimensional image blocks, the strip noise image of the original image is obtained.

8. The method for removing strip noise from remote sensing images according to claim 6, characterized in that, In the case where the strip noise is dense strip noise, the step of calculating the strip noise constraint model to obtain the noise grayscale value of each strip noise column and obtaining the strip noise image includes: Each non-strip noise column is used as a reference column. The noise grayscale value of the nearest strip noise column in each reference column in each two-dimensional image block is calculated through the strip noise constraint model to obtain the block strip noise corresponding to each two-dimensional image block. By stitching together the block strip noise corresponding to all two-dimensional image blocks, the intermediate strip noise image of the original image is obtained; The step of removing the striped noise image based on the original image and outputting a denoised image includes: The intermediate stripe noise image is removed from the original image to output the intermediate denoised image. If the intermediate denoised image does not contain strip noise columns, then the intermediate denoised image is output as the denoised image. Otherwise, the pixel column at the position of the removed strip noise column is updated as the reference column, and the noise grayscale value of the nearest strip noise column in each reference column in each two-dimensional image block is calculated by the strip noise constraint model, and the following steps are performed.

9. A remote sensing image stripe noise removal device, characterized in that, include: The image processing unit is used to filter the original image containing stripe noise to obtain a residual map; A strip noise estimation unit is used to estimate each estimated strip noise column of the original image based on the residual map; The strip noise calculation unit is used to divide the original image into feature regions at each estimated strip noise column along the direction perpendicular to the strip noise. The strip noise calculation unit is also used to obtain each strip noise column of the original image based on each feature region; The strip noise calculation unit is also used to construct a strip noise constraint model by using the minimum L1 norm of the difference between the image gray value at the strip noise column in the original image, the noise gray value of the strip noise column, and the non-noise gray value of the non-strip noise column that is closest to the strip noise column as the constraint condition. The strip noise calculation unit is also used to calculate the strip noise constraint model, obtain the noise grayscale value of each strip noise column, and obtain the strip noise image; The denoising unit is used to remove the striped noise image based on the original image and output a denoised image.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 8.

Images