A method for repairing information loss areas in coastal remote sensing images based on an improved Criminisi algorithm

Abstract

The present invention discloses a method for repairing information loss areas in coastal remote sensing images based on an improved Criminisi algorithm. The method comprises: preprocessing original remote sensing images, including radiation correction, orthorectification, image fusion, and land-sea separation processing; performing spectral analysis on the preprocessed images to extract spectral features of information loss areas and other element areas; calculating highlight components based on the dark channel principle to locate the information loss areas, and classifying the information loss areas into sea element information loss areas and water body information loss areas by correcting near-infrared bands; repairing the classified information loss areas using the improved Criminisi algorithm, including: priority calculation in combination with the second-order moment of the gradient vector modulus; adaptive sample block generation based on texture complexity; searching for the best matching block in combination with matching criteria of structural similarity, color difference, and spatial distance; and performing boundary smoothing on the repaired images to eliminate boundary effects.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing images, and in particular to a method for repairing coastal zone remote sensing image information loss areas based on an improved Criminisi algorithm. Background Art

[0002] During remote sensing imagery acquisition, data may contain information loss due to sensor defects or atmospheric conditions (e.g., light pollution, cloud contamination, and wave contamination). Furthermore, information loss can occur due to seasonal and climatic factors, as well as long satellite re-entry times. Coastal areas, particularly bays and estuaries, present complex coastal conditions. When natural terrain acts as a barrier against wind and waves, information loss is more widespread and severe due to waves and solar flares. This severely degrades image quality, causing some areas to appear overbright or oversaturated, impacting image usability and the accuracy of target feature interpretation. Restoring coastal remote sensing imagery can improve image quality by removing or mitigating the effects of waves, solar flares, and clouds, restoring the true nature of the image. This provides more accurate information on coastlines, aquaculture areas, and land reclamation, helping scientists and decision makers better understand coastal change, assess the impacts of human activities and natural processes on marine ecosystems, evaluate the effectiveness of restoration projects, and monitor environmental changes in restoration areas, providing precise data support for coastal management and planning.

[0003] In coastal areas, information loss areas are primarily caused by three types of pollution: waves, flares, and clouds. Waves are wind-generated surface waves, whose magnitude is affected by wind speed, time, and region. Imagery primarily reveals superimposed waves, longer swells, and shorebreaks and breakers caused by reflection and curvature of coastal features and seafloor shape. Flares are primarily caused by specular reflection of solar radiation from the water surface. Their magnitude is affected by the shooting orientation, time, and region. They primarily appear in various areas of the water surface, revealing numerous spectral saturation regions. Clouds are caused by condensation of water vapor and are primarily influenced by climate, topography, and wind speed. Thick clouds can completely obscure surface information, while thin clouds can partially obscure features. The excessive brightness of these information loss areas results in significant contrast across the entire image, obscuring texture and spectral information for a significant portion of marine features. These three conditions lead to the loss of spectral and texture information for a significant amount of marine features, resulting in information loss areas.

[0004] Current image restoration algorithms can be primarily categorized as traditional image restoration algorithms and deep learning-based image restoration algorithms. Traditional image restoration algorithms include diffusion-based methods and sample-based methods. Diffusion-based restoration methods simulate the manual restoration process of image restorers by inferring image information of unknown regions using boundary information of missing regions. Representative methods include the total variation model (TV) and partial differential equation (PDE) method. While computationally simple, they struggle to restore images with complex textures and structures. Sample-based restoration methods utilize information from known regions of an image to repair missing or damaged portions. Typical examples include the Criminisi algorithm and texture synthesis methods. While the restoration results are natural and coherent, they are time-consuming and rely on known regions in the image. Deep learning-based image restoration algorithms use deep learning to learn image features and repair lost portions. Typical examples include the GAN method and the Transformer method. They offer excellent detail restoration, but require high training data quality and diversity. Summary of the Invention

[0005] To achieve the above-mentioned and other related purposes, the present invention discloses a method for repairing coastal zone remote sensing image information loss areas based on an improved Criminisi algorithm, comprising:

[0006] S1: Preprocessing of original remote sensing images, including radiometric correction, orthorectification, image fusion and land-sea separation;

[0007] S2: Perform spectral analysis on the preprocessed image to extract the spectral characteristics of the information loss area and other element areas;

[0008] S3: Calculate the highlight component based on the dark channel principle, locate the information loss area, and classify the information loss area into the sea element information loss area and the water body information loss area by correcting the near-infrared band;

[0009] S4: Use the improved Criminisi algorithm to repair the information loss area after classification, including:

[0010] Combined with the priority calculation of the second-order moment of the gradient vector;

[0011] Adaptive sample block generation based on texture complexity;

[0012] The matching criteria of structural similarity, color difference and spatial distance are combined to search for the best matching block;

[0013] S5: Perform boundary smoothing on the restored image to eliminate boundary effects.

[0014] Furthermore, the preprocessing in step S1 specifically includes:

[0015] Perform radiometric calibration, FLAASH atmospheric correction, and RPC orthorectification on multispectral images;

[0016] Perform radiometric calibration and RPC orthorectification on panchromatic images;

[0017] NNDiffuse Pan Sharpening algorithm is used for image fusion;

[0018] Land and sea separation is performed using shore DEM data combined with tide information.

[0019] Furthermore, the spectrum analysis in step S2 specifically includes:

[0020] Select feature pixels in the information loss area and other feature areas;

[0021] The Savitzky-Golay filtering method is used to generate a smooth spectrum curve;

[0022] Analyze the differences in spectral reflectance of each element in the visible light to near-infrared band.

[0023] Furthermore, step S3 includes locating the information loss area and classifying the information loss area, wherein:

[0024] Information loss area positioning includes:

[0025] Calculate the highlight component:

[0026] ;

[0027] Where: R, G, B, NIR represent the red, green, blue, and near-infrared bands of the image; Indicates the minimum reflectance of pixel a in all bands;

[0028] Perform highlight component correction. To ensure the stability of image chromaticity, subtract a scalar value from the band:

[0029] ;

[0030] ;

[0031] Where: Represents the highlight component of pixel a after correction in the i-th band, is the mean value of the highlight component of all valid pixels in the image, Indicates the number of effective pixels;

[0032] According to the corrected highlight component, determine the information loss area:

[0033] ;

[0034] Where: is the image after the corrected highlight component is binarized. is the segmentation threshold of the corrected highlight component determined by the natural break point classification method;

[0035] Information loss area classification includes:

[0036] use Classify the information loss areas into sea use factor information loss areas and water body information loss areas;

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] Where: is the corrected near-infrared band binary image, is the segmentation threshold of the modified near-infrared band determined by the natural break point classification method, It is the area where the information of sea use elements is lost. is the water body information loss area, is the value of pixel a in the corrected near-infrared band, is the value of pixel a in the near-infrared band before correction.

[0042] Furthermore, the priority calculation in step S4 includes:

[0043] ;

[0044] in, and represent the confidence item and data item respectively, is the second-order moment of the gradient vector modulus of the central pixel p, is the priority of pixel p;

[0045] ;

[0046] Where, Represents the gradient value of the known pixel point in the block to be repaired, Represents the average gradient of known pixels in the block to be repaired, and Represent the gradients of the point a to be repaired in the x and y directions respectively, Represents a sample block formed with point p as the center, express The number of pixels in the sample block.

[0047] Furthermore, the adaptive sample block generation in step S4 includes:

[0048] ;

[0049] Among them, odd is a function that calculates the nearest odd number after rounding the value, and min is a function that calculates the minimum value of all values. is the size of the sample block centered at point p.

[0050] Furthermore, the matching criteria in step S4 include:

[0051] = + + ;

[0052] ;

[0053] ;

[0054] ;

[0055] in, is the matching criterion, Represents a sample block The structural similarity of Represents a sample block The spatial distance, Represents a sample block Color differences, and denote the mean and variance of the sample block respectively, represents the Euclidean distance between pixels p and q, represents the diagonal distance of image I, a and b are the pixels at the corresponding positions of the sample block and the matching block respectively, 、 is a constant.

[0056] Furthermore, the best matching block search in step S4 includes:

[0057] Corrected near-infrared images are used in areas with lost sea element information As a search base map;

[0058] The NDWI index image is used as the search base map in the water body information loss area.

[0059] Furthermore, in step S5, the boundary effect is eliminated by using mean filtering.

[0060] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program implements the above method when executed by a processor.

[0061] By adopting the above technical solution, by improving the priority calculation of the Criminisi algorithm, introducing adaptive sample blocks and comprehensive matching criteria, and combining spectral analysis with dark channel principles to accurately locate the classification information loss area, the restoration accuracy and efficiency are effectively improved, texture distortion and boundary effects are reduced, and the spectral texture characteristics of sea elements obscured by waves, flares and clouds are restored. The quality of coastal remote sensing images is significantly improved, providing more accurate data support for coastline monitoring, ecological assessment and sea use management. For the highlight information loss areas caused by waves, flares and clouds in coastal remote sensing images, precise spectral positioning and classification, adaptive sample matching and boundary optimization are used to effectively restore the detailed characteristics of sea elements obscured by highlight pollution, overcoming the technical defects of existing methods in complex coastal scenes, such as insufficient restoration accuracy and abrupt boundary transitions. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0063] Figure 1 is a flow chart of the present invention;

[0064] Figure 2 Spectral curves of different elements;

[0065] Figure 3 It is the information loss area and the corrected near-infrared image;

[0066] Figure 4 Before and after changes for priority improvements;

[0067] Figure 5 The following are the restoration results and comparison charts. DETAILED DESCRIPTION

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0069] Reference Figure 1The embodiment of the present invention provides a method for repairing coastal remote sensing image information loss areas based on an improved Criminisi algorithm, comprising the following steps:

[0070] S1: Preprocessing of original remote sensing images, including radiometric correction, orthorectification, image fusion and land-sea separation;

[0071] S2: Perform spectral analysis on the preprocessed image to extract the spectral characteristics of the information loss area and other element areas;

[0072] S3: Calculate the highlight component based on the dark channel principle, locate the information loss area, and classify the information loss area into the sea element information loss area and the water body information loss area by correcting the near-infrared band;

[0073] S4: Use the improved Criminisi algorithm to repair the information loss area after classification, including:

[0074] Combined with the priority calculation of the second-order moment of the gradient vector;

[0075] Adaptive sample block generation based on texture complexity;

[0076] The matching criteria of structural similarity, color difference and spatial distance are combined to search for the best matching block;

[0077] S5: Perform boundary smoothing on the restored image to eliminate boundary effects.

[0078] In this embodiment, the above steps specifically include:

[0079] 1: Data Preprocessing. Preprocess domestic high-resolution imagery, including radiometric correction, orthorectification, and fusion. Because marine features are distributed on the seaward side of the coastline, the imagery is separated from land and sea using the newly surveyed coastline to eliminate interference from other onshore features.

[0080] 1-1: Data Processing. Multispectral images are subjected to radiometric calibration, atmospheric correction, and orthorectification. Panchromatic images are subjected to radiometric calibration and orthorectification. Atmospheric correction uses the FLAASH atmospheric correction method, and orthorectification uses the RPC orthorectification method. Radiometric correction effectively removes "noise" caused by sensor or atmospheric conditions, improving image fidelity and enabling better alignment with the actual radiometric brightness of terrain features, resulting in clearer display of coastal features. Multispectral and panchromatic images are fused using the NNDiffuse Pan Sharpening algorithm.

[0081] 1-2: Separate land and sea using the newly surveyed coastline. Using the DEM data of the target area's shores and beach, combined with tidal information, the coastline is optimized and corrected. The corrected coastline is used to separate land and sea areas, eliminating interference from land elements.

[0082] 2: Image spectral analysis. Based on the hyperspectral satellite data of the same region and phase, the pixels in the information loss area and other pixels are spectrally analyzed to obtain the difference in the spectral curve. Figure 2 .

[0083] 2-1: Feature pixel selection. Select feature pixels in the information loss area and other element areas respectively, and extract the spectral reflectance of each pixel;

[0084] 2-2: Spectral curve generation. Calculate the average spectrum of all pixels in the information loss area and other elements, draw the spectrum curve, and use the Savitzky-Golay filter method to smooth the spectrum curve to obtain the processed spectrum curves of various elements;

[0085] 2-3: Spectral Feature Analysis. Analysis of the processed spectral curve reveals that areas of information loss, often caused by flares, waves, and clouds, appear as bright areas in the image. These areas have significantly higher remote sensing reflectance in each band than other elements, and their shapes and intensities vary. These bright areas obscure some element information, resulting in information loss.

[0086] 3: Localization and classification of information loss areas.

[0087] 3-1: Locating Information Loss Areas. Because information loss areas are brightly lit areas such as waves, flares, and clouds, their radiant energy is high, resulting in larger pixel values compared to other areas of near-water radiation with lower energy. Based on the dark channel principle, other natural features in the coastal zone have very low values in at least one color channel. Therefore, the minimum value of all bands in the image can be considered the highlight component. However, considering the overall image chromaticity, this highlight component requires further correction. Finally, the information loss areas are located based on this corrected highlight component.

[0088] (1) Calculate the highlight component:

[0089]

[0090] Where: R, G, B, NIR represent the red, green, blue, and near infrared bands of the image. Indicates the minimum reflectance of pixel a in all bands.

[0091] (2) Perform highlight component correction. To ensure the stability of image chromaticity, subtract a scalar value from the band:

[0092]

[0093]

[0094] Where: Represents the highlight component of pixel a after correction in the i-th band, is the mean value of the highlight component of all valid pixels in the image, Indicates the number of valid pixels.

[0095] (3) Determine the information loss area based on the corrected highlight component:

[0096]

[0097] Where: is the image after the corrected highlight component is binarized. is the segmentation threshold of the corrected highlight component determined by the natural break point classification method.

[0098] 3-2: Information loss classification. Figure 3 The image is least affected by bright areas such as flares and waves in the near-infrared band, so the near-infrared band is corrected. Due to the strong absorption of the near-infrared band by water, the residual radiation intensity in the near-infrared band after atmospheric correction can be considered as information loss and sea element areas. The corrected near-infrared band is calculated by the difference between the near-infrared band and the highlight component. Not affected by the highlighted area, use sea elements in The upper grayscale value is higher, so use The information loss areas are classified into sea use element information loss areas and water body information loss areas.

[0099]

[0100]

[0101] ;

[0102] ;

[0103] Where: is the corrected near-infrared band binary image, is the segmentation threshold of the modified near-infrared band determined by the natural break point classification method, It is the area where the information of sea use elements is lost. is the water body information loss area, is the value of pixel a in the corrected near-infrared band, is the value of pixel a in the near-infrared band before correction.

[0104] 4. Restoring Lost Areas. The Criminisi algorithm is a classic example-based image restoration algorithm. Its core concept is to find the best matching block within a known area of the image and fill it into the padded block at the edge of the unknown area, achieving simultaneous propagation of structure and texture. However, it suffers from problems such as unreasonable priority settings, a single matching criterion, and slow execution. To address these issues, we propose an improved Criminisi algorithm to repair lost areas. The restoration steps are as follows:

[0105] 4-1: Priority calculation. The original algorithm priority is calculated by calculating the confidence and data items. When applied to the reconstruction of information loss areas in remote sensing images, it is very easy to produce multiple optimal matching blocks with the same SSD due to problems such as multiple bands, information redundancy, and small spectral differences. The random selection of matching benchmarks can lead to mismatching problems and cause obvious texture distortion. In the image, the spectral and textural characteristics of water bodies and sea features are significantly different. Water bodies are smooth or rippled, while sea features have regular textures. This paper adds the second-order moment of the gradient vector modulus to calculate the priority. The modulus of the gradient vector can measure the texture complexity of the pixel block, and its second-order moment represents the regularity of the texture. The introduction of texture features increases the algorithm's sensitivity to image texture details during the restoration process. The calculation formula is as follows.

[0106]

[0107] Where, Represents the gradient value of the known pixel point in the block to be repaired, Represents the average gradient of known pixels in the block to be repaired, and Represent the gradients of the point a to be repaired in the x and y directions respectively, Represents a sample block formed with point p as the center, express The number of pixels in the sample block, is the second-order moment of the gradient vector modulus of the central pixel p.

[0108] The confidence term is expressed in the form of a power function to suppress the sharp drop in confidence. At the same time, the multiplication operation in the priority function is changed to an addition operation, which improves the calculation speed while avoiding the mutual influence of data items that may cause the repair order to be wrong.

[0109]

[0110] in, and Represent the confidence term and data term of pixel p respectively, is the priority of pixel p.

[0111] The results before and after improvement are as follows Figure 4shown.

[0112] 4-2: Adaptive Sample Block Generation. In the Criminisi algorithm, sample blocks are set to a fixed size. However, in remote sensing imagery, different features often have different local textures. Using a fixed size for inpainting results in poor continuity in areas with complex textures and a long inpainting time in areas with simple textures. Therefore, this paper uses the texture complexity of the pixel region to determine the sample block size, improving the efficiency and accuracy of the matching search. Different sizes are used to adapt to different pixel texture conditions.

[0113]

[0114] Among them, odd is a function that calculates the nearest odd number after rounding the value, and min is a function that calculates the minimum value of all values. is the size of the sample block centered at point p.

[0115] 4-3: Best Matching Block Search. When searching for the best matching block, the Criminisi algorithm only considers the color information between the matching block and the block to be repaired, which can easily lead to mismatches. Furthermore, the search is performed using a global search, which results in a slow algorithm. To improve search accuracy, different matching base maps are used to search and fill in different types of information loss areas. Furthermore, structural and distance information are incorporated into the existing matching criteria to improve matching efficiency and accuracy.

[0116] During the restoration, the sea-use element loss area is restored first, and then the water body loss area is restored. Since the sea-use element has obvious features and high grayscale values in the corrected near-infrared image, the sea-use element loss area is restored first. , with the corrected near-infrared image As the search base map, the best matching block is found; since NDWI is more sensitive to water elements and has high grayscale values, it is used for water loss areas. , using NDWI as the search base map to find the best matching block.

[0117] When searching, the structural similarity index (SSIM) and the Euclidean distance of the center point ( ) to search, SSIM measures the similarity between blocks by brightness, contrast, and structure. The larger the value, the higher the similarity; The spatial position between blocks can be measured, and combined with the spatial distribution of elements in remote sensing images, the closer the spatial distance, the greater the probability of the best matching block. for:

[0118]

[0119]

[0120]

[0121] = + +

[0122] in, Represents a sample block The structural similarity of Represents a sample block The spatial distance, Represents a sample block Color differences, and denote the mean and variance of the sample block respectively, represents the Euclidean distance between pixels p and q, represents the diagonal distance of image I, a and b are the pixels at the corresponding positions of the sample block and the matching block respectively, 、 is a constant.

[0123] 5. Boundary smoothing. After the restoration is complete, the image boundaries may experience uneven transitions because the repaired area is restored in the form of sample blocks. This can cause boundary effects in the repaired image. These effects will leave clear traces of the restoration, affecting subsequent feature recognition. These traces are usually caused by pixels at the boundary of the highlighted area having a higher brightness than the surrounding pixels (including known pixels and repaired pixels). By performing grayscale assessment on the pixels, we determine which pixels need to be smoothed, and then use mean filtering to eliminate boundary effects.

[0124]

[0125] Among them, m is a pixel on the outer boundary of the area to be filled, is the neighborhood centered at m.

[0126] Repair results and comparison Figure 5 shown.

[0127] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art in the art to which the present invention pertains. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with those in the context of the prior art and, unless specifically defined, will not be interpreted in an idealized or overly formal sense.

[0128] For simplicity of description, the method embodiments are described as a series of actions. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because certain steps can be performed in other orders or simultaneously according to the embodiments of the present invention. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0129] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for repairing coastal remote sensing image information loss areas based on the improved Criminisi algorithm is characterized by: include: S1: Preprocessing of original remote sensing images, including radiometric correction, orthorectification, image fusion and land-sea separation; S2: Perform spectral analysis on the preprocessed image to extract the spectral characteristics of the information loss area and other element areas; S3: Calculate the highlight component based on the dark channel principle, locate the information loss area, and classify the information loss area into the sea element information loss area and the water body information loss area by correcting the near-infrared band; Step S3 includes locating the information loss area and classifying the information loss area, wherein: Information loss area positioning includes: Calculate the highlight component: I min (a)=min[I R (a),I G (a),I B (a),I NIR (a),…]; Where: R, G, B, NIR represent the red, green, blue, and near-infrared bands of the image; Image I min (a) represents the minimum reflectance of pixel a in all bands; Perform highlight component correction. To ensure the stability of image chromaticity, subtract a scalar value from the band: Where: I' min (a) represents the highlight component of pixel a after correction in the i-th band, It is the mean of the highlight components of all valid pixels in the image, and #pixels represents the number of valid pixels; According to the corrected highlight component, determine the information loss area: Where: H is the image after the corrected highlight component is binarized, T h is the segmentation threshold of the corrected highlight component determined by the natural break point classification method; Information loss area classification includes: Utilize I ′ NIR Classify the information loss areas into sea use factor information loss areas and water body information loss areas; I ′ NIR (a)=I NIR (a)-I min (a); S t =N∩H; S w =H-S t ; Where: N is the corrected near-infrared band binary image, T n is the segmentation threshold of the modified near-infrared band determined by the natural break point classification method, S t The area with lost sea use information, S w is the water body information loss area, I ′ NIR (a) is the value of pixel a in the corrected near-infrared band, I NIR (a) is the value of pixel a in the near-infrared band before correction; S4: Use the improved Criminisi algorithm to repair the information loss area after classification, including: Combined with the priority calculation of the second-order moment of the gradient vector; Adaptive sample block generation based on texture complexity; The matching criteria of structural similarity, color difference and spatial distance are combined to search for the best matching block; S5: Perform boundary smoothing on the restored image to eliminate boundary effects.

2. The method according to claim 1, characterized in that The preprocessing in step S1 specifically includes: Perform radiometric calibration, FLAASH atmospheric correction, and RPC orthorectification on multispectral images; Perform radiometric calibration and RPC orthorectification on panchromatic images; NNDiffuse Pan Sharpening algorithm is used for image fusion; Land and sea separation is performed using shore DEM data combined with tide information.

3. The method according to claim 1, characterized in that The spectrum analysis in step S2 specifically includes: Select feature pixels in the information loss area and other feature areas; The Savitzky-Golay filtering method is used to generate a smooth spectrum curve; Analyze the differences in spectral reflectance of each element in the visible light to near-infrared band.

4. The method according to claim 1, wherein The priority calculation in step S4 includes: Where C(p) and D(p) represent the confidence term and data term respectively, F(p) is the second-order moment of the gradient vector modulus of the central pixel p, and P(p) is the priority of pixel p; Where M represents the gradient value of the known pixel in the block to be repaired, Represents the average gradient of known pixels in the block to be repaired, G x (a) and G y (a) represents the gradient of the point a to be repaired in the x and y directions, Ψ p represents a sample block centered at point p, |Ψ p | indicates Ψ p The number of pixels in the sample block.

5. The method according to claim 4, characterized in that The adaptive sample block generation in step S4 includes: Where odd is the function that calculates the nearest odd number after rounding the value, min is the function that calculates the minimum value of all values, and B(p) is the size of the sample block centered at point p.

6. The method according to claim 5, characterized in that The matching criteria in step S4 include: d(Ψ p ,P q )=CD(Ψ p ,P q )+ED(Ψ p ,P q )+SSIM -1 (P p ,P q ); Among them, d(Ψ p ,Ψ q ) is the matching criterion, SSIM(Ψ p ,Ψ q ) represents the sample block Ψ p ,Ψ q The structural similarity of ED(Ψ p ,Ψ q ) represents the sample block Ψ p ,Ψ q The spatial distance, CD(Ψ p ,Ψ q ) represents the sample block Ψ p ,Ψ q The color difference of μ represents the mean of the sample block, σ(Ψ p ) 2 、σ(Ψ q ) 2 represents the variance of the sample block, σ(Ψ p ,Ψ q ) represents the covariance of the sample block, ||I(p)-I(q)|| represents the Euclidean distance between pixels p and q, dis(I) represents the diagonal distance of image I, a and b are the pixels at the corresponding positions of the sample block and the matching block, respectively, and C1 and C2 are constants.

7. The method according to claim 1, characterized in that The best matching block search in step S4 includes: The corrected near infrared image I was used in the area with lost sea element information ′ NIR As a search base map; The NDWI index image is used as the search base map in the water body information loss area.

8. The method according to claim 1, characterized in that In step S5, the boundary effect is eliminated by using mean filtering.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images