A method for thin cloud correction of visible light remote sensing images

Abstract

The present invention discloses a method for correcting thin fog and cloud of visible light remote sensing images, comprising: dividing the visible light remote sensing images to be processed into several sub-areas, searching for dark targets in each sub-area according to the first visible light band; dividing all dark targets in turn according to whether they are all marked as dark targets in the first visible light band, obtaining absolute dark targets and relative dark targets; fitting the absolute dark targets according to the first visible light band, and obtaining the corresponding quantitative relationship of fog and cloud intensity; deriving fog and cloud intensity values ​​of the relative dark targets in the second visible light band in turn, and performing the operation with the marked dark targets in the second visible light band to obtain the first dark target; generating fog and cloud intensity distribution according to the first dark target, and correcting the visible light remote sensing images. The present invention can make full use of the strong representation of fog and cloud by dark targets and the numerical correlation between bands, reduce the dependence on external data, and improve the correction effect and applicability.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image processing, and in particular to a method for correcting thin cloud and fog of visible light remote sensing images. Background Art

[0002] Visible light remote sensing images are one of the most frequently used data sources in earth science research and are widely used in tasks such as target recognition, image fusion, and scene segmentation. However, due to the short wavelength of visible light, it is easily interfered by thin clouds and fog, resulting in radiation bias in the acquired data and failing to accurately reflect the true surface spectral characteristics. Therefore, how to effectively correct the interference of thin clouds and fog and restore its ground representation capabilities is an important technical means to improve data quality and availability.

[0003] In order to correct the thin fog in visible light remote sensing images, researchers at home and abroad have developed a series of methods, which can be divided into two categories: methods based on multiple images and single images. The methods based on multiple images need to introduce external data as an aid when performing fog correction, such as images from different phases or sensors. Commonly used techniques include wavelet transform, regression analysis and cloud intensity value replacement. This type of method usually produces good visual results due to the introduction of auxiliary data. The single-image-based method does not need to introduce external data. It realizes thin cloud correction by maximizing the mining of image information or optimizing training parameters. It mainly includes traditional and deep learning methods. Among them, the traditional correction method mainly uses the spatial and spectral characteristics of the image to be processed to realize cloud correction. Typical methods include spectral transformation method, domain conversion method and dark target method. The deep learning correction method can be divided into two categories according to the different frameworks used. Convolutional neural network (CNN) and generative adversarial network (GAN) are representative methods. Dehaze-Net, RSC-Net and CR-GAN-PM are representative methods. This type of method uses the nonlinear mapping ability of neural networks, introduces large-scale training data sets to learn the transformation relationship between sample label pairs, and finally realizes cloud correction and surface restoration.

[0004] However, the cloud correction method based on multiple images in the prior art has the following disadvantages: since the correction is completed by using the complementary information of the introduced auxiliary data, it also restricts the applicable scenarios of the method; in addition, it does not take into account that due to the translucent characteristics of thin clouds and fog, some surface information in the area to be corrected can still be used for correction; the method based on a single image in the prior art also has the following disadvantages: first, the traditional correction method relies on different spatial or spectral prior knowledge, which has stronger universality and stability, but still does not sufficiently explore the spatial and spectral characteristics of cloud images and does not fully consider the complementarity between bands, resulting in reduced effectiveness of prior knowledge in some scenarios, limiting the correction effect; secondly, deep learning methods can usually achieve better correction effects, but they are too dependent on data sets and network structures and require more preparatory work, which is not easy to be quickly applied to single images. Summary of the invention

[0005] The embodiment of the present invention provides a method for correcting thin clouds and fog in visible light remote sensing images, which can make full use of the strong representation of dark targets for clouds and fog and the numerical correlation between bands, thereby reducing dependence on external data and achieving better correction effects and greater universality.

[0006] In a first aspect, an embodiment of the present invention provides a method for correcting thin cloud and fog in a visible light remote sensing image, the method comprising:

[0007] The visible light remote sensing image to be processed is divided into K sub-regions, and the pixel with the minimum value in each sub-region is searched as a dark target according to the first visible light band; wherein K is a positive integer, and the pixel value is the same as the cloud intensity value;

[0008] According to whether they are all marked as dark targets in the first visible light band, all dark targets are divided in turn to obtain absolute dark targets and relatively dark targets;

[0009] Fitting the absolute dark target according to the first visible light band to obtain a corresponding quantitative relationship of cloud intensity;

[0010] According to the quantitative relationship of cloud intensity, the cloud intensity values ​​of the relatively dark target in the second visible light band are deduced in sequence, and the cloud intensity values ​​are combined with the marked dark targets in the second visible light band to obtain the first dark target;

[0011] According to the first dark target, a cloud intensity distribution with high spatial granularity is generated to correct the visible light remote sensing image.

[0012] The present invention divides dark targets and fits absolute dark targets on visible light bands to obtain a quantitative relationship between cloud and fog intensities between visible light bands. The cloud and fog intensity value of the relative dark target on the second visible light band can be derived. On the same visible light band, the derived dark target and the marked dark target in the second visible light band are complementary. This complementary relationship can be used to perform a merge operation to obtain a first dark target that includes the absolute dark target and the dark target that satisfies the quantitative relationship of cloud and fog intensity. The complementary relationship between dark targets can be mined to expand the number of samples, thereby reducing dependence on external data and improving scalability and practicality.

[0013] Furthermore, the step of dividing the visible light remote sensing image to be processed into K sub-regions includes:

[0014] By using a superpixel segmentation algorithm based on simple linear iterative clustering, the pixels in the visible light remote sensing image are aggregated to form K irregular sub-regions, and the visible light remote sensing image is divided into corresponding K sub-regions.

[0015] The present invention adopts a superpixel segmentation algorithm of simple linear iterative clustering, which can achieve more reasonable image sub-region division, so as to more conveniently search for dark targets by area, and can ensure that each dark target obtained is evenly distributed in the visible light remote sensing image, so that it can be corrected to the entire visible light remote sensing image, thereby providing better and explainable scientificity for the thin cloud correction method of optical remote sensing images.

[0016] Furthermore, searching the pixel with the minimum value in each sub-region as a dark target according to the first visible light band includes:

[0017] According to the K sub-regions, the central cloud intensity value in each block is used as the seed point of the superpixel;

[0018] The distances between all pixels and the seed point are calculated block by block, and the pixel with the smallest distance is selected as the dark target.

[0019] Furthermore, the distance between all pixels and the seed point is calculated block by block, and the specific calculation formula of the distance is:

[0020]

[0021] Among them, the seed point is S, the pixel is p, d c and d s are the color distance and spatial distance between the pixel and the seed point in the visible light remote sensing image, respectively, c is the compactness factor, N s is a variable related to the image size and the number of superpixels.

[0022] Furthermore, the cand d s are the color distance and spatial distance between the pixel in the visible light remote sensing image and the seed point, respectively, and the calculation formulas are as follows:

[0023] d C (p, S) = || F p -F S ||2,d s (p, S) = ||X p -X S ||2, (2)

[0024] Among them, F P =[l P , b P , a P ] T and F S =[l S , b S , a S ] T are the average values ​​of the three color components l, b, a of the pixel and the seed point in the CIELab color space, p =[x p ,y p ] T and X S =[x S ,y S ] T are the centroid positions of the pixel and the seed point respectively.

[0025] Specifically, the step of classifying all dark targets in turn according to whether they are all marked as dark targets in the first visible light band, and obtaining absolute dark targets as absolute dark targets and relatively dark targets as relatively dark targets, includes:

[0026] If the dark object is recorded as a dark object in all visible light bands in the first visible light band, it is classified as an absolutely dark object;

[0027] If the dark object is recorded as a dark object only in a part of the first visible light band, it is classified as a relatively dark object.

[0028] The present invention divides dark targets into absolute dark targets and relative dark targets. Absolute dark targets are detectable in all visible light bands, while relative dark targets are detectable only in some visible light bands. Dark targets are classified so that linear fitting can be performed on absolute dark targets that are detectable in all visible light bands, and then a quantitative relationship related to cloud intensities between different visible light bands is established. The relative dark targets are deduced in the second visible light band, and cloud intensity values ​​that meet the correlation of cloud intensities between different visible light bands can be generated, so as to make full use of limited samples and can be corrected without the help of external data, thereby improving scalability and practicality.

[0029] Further, the absolute dark target is fitted according to the first visible light band to obtain a corresponding quantitative relationship of cloud intensity, including:

[0030] The first visible light bands are combined with any two visible light bands, and fitting is performed according to the absolute dark target to obtain the quantitative relationship between the cloud and fog intensities of any two visible light bands.

[0031] Furthermore, the quantitative relationship of cloud intensity is:

[0032] DO α，i =a ij ·DO α，j +b ij , (3)

[0033] Among them, DO α，i and DO α，j is the cloud intensity value of an absolutely dark target α in visible light band i and visible light band j respectively, a ij and b ij are the gain and offset corresponding to the linear fit.

[0034] Furthermore, the derivation of the cloud intensity values ​​of the relatively dark target in the second visible light band in sequence according to the cloud intensity quantitative relationship includes:

[0035] According to the quantitative relationship between the cloud intensity and the relatively dark target, the cloud intensity value of the relatively dark target in the unmarked visible light can be calculated. The calculation formula is:

[0036] DO β′，i =a ij ·DO β，j +b ij , (4)

[0037] Among them, DO β′，i is the cloud intensity value derived from a relatively dark target β in the visible light band i, DO β，jis the cloud intensity value of the known relatively dark target β in the visible light band j, a ij and b ij are the gain and offset corresponding to the linear fit.

[0038] Furthermore, the correction of the visible light remote sensing image includes:

[0039] Correction is performed on the visible light remote sensing image to be processed according to the first visible light band, and the correction formula is:

[0040]

[0041] in, is the visible light remote sensing cloud image to be processed for visible light i, ρ c，i is the high spatial fine-grained cloud intensity distribution generated according to the first dark target, ρ i is the correction result;

[0042] Determine whether the calibrated visible light remote sensing image is cloud-free. If there are still clouds, re-calibrate; otherwise, end the calibration.

[0043] The present invention uses multiple corrections to process visible light remote sensing cloud and fog images until the visible light remote sensing images are cloudless, which can improve the precision and accuracy of the correction, thereby obtaining completely cloud-free visible light remote sensing images, providing data support for subsequent optical remote sensing monitoring, land change surveys, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the process of thin cloud correction of visible light remote sensing images provided by an embodiment of the present invention;

[0045] Figure 2 It is a schematic diagram of a process of thin cloud correction for visible light remote sensing images provided by an embodiment of the present invention;

[0046] Figure 3 is a schematic diagram of superpixel segmentation and dark target search provided by an embodiment of the present invention;

[0047] Figure 4 This is a schematic diagram of fitting the quantitative relationship between dark target classification and cloud intensity provided by an embodiment of the present invention;

[0048] Figure 5 It is a schematic diagram of inter-band dark target complementarity and high-granularity cloud and fog distribution estimation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 , is a schematic diagram of the process of thin cloud correction of visible light remote sensing images provided by an embodiment of the present invention, including steps S11 to S15, specifically:

[0051] Step S11, divide the visible light remote sensing image to be processed into K sub-areas, and search the pixel with the minimum value in each sub-area according to the first visible light band as a dark target; wherein K is a positive integer, and the pixel value is the same as the cloud intensity value. That is to say, the larger the pixel value of the dark target, the larger its cloud intensity value.

[0052] It is worth noting that the first visible light band is all visible light bands, which is not limited here. Exemplarily, the present invention selects three visible light bands of red, blue and green as the first visible light band.

[0053] Specifically, the visible light remote sensing image to be processed is divided into K sub-regions, including: using a superpixel segmentation algorithm based on simple linear iterative clustering (SLIC) to aggregate pixels in the visible light remote sensing image to form K irregular sub-regions, and dividing the visible light remote sensing image into corresponding K sub-regions; wherein, the visible light remote sensing image to be corrected is aggregated according to pixels with similar spectral, texture and other characteristics, the K sub-regions have the same size, and the number K of divided sub-regions mainly depends on the size of the image to be processed and the complexity of the surface.

[0054] The pixel with the minimum value in each sub-region is searched as a dark target according to the first visible light band, including: according to the K sub-regions, the central cloud intensity value in each block is used as the seed point of the superpixel; the distance between all pixels and the seed point is calculated block by block, and the pixel with the smallest distance is selected as the dark target; wherein, the pixel with the minimum value in each sub-region is searched as a dark target according to the first visible light band, including: according to the K sub-regions, the central cloud intensity value in each block is used as the seed point of the superpixel; the distance between all pixels and the seed point is calculated block by block, and the pixel with the smallest distance is selected as the dark target, and the specific calculation formula of the distance is:

[0055]

[0056] Among them, the seed point is S, the pixel is p, d c and d s are the color distance and spatial distance between the pixel and the seed point in the visible light remote sensing image, respectively, c is the compactness factor, N s is a variable related to the image size and the number of superpixels.

[0057] Preferably, the compactness factor is a constant and takes a value of 20.

[0058] The variables related to the image size and the number of superpixels can be expressed as:

[0059]

[0060] Where N is the image size and K is the number of superpixels. Since each pixel will be searched by multiple seed points, each pixel will have a distance from the surrounding seed points. The seed point corresponding to the minimum value is taken as the cluster center of the pixel, and the pre-segmented superpixels can be iteratively adjusted to form irregular segmentation results. In the present invention, the more surface types there are in the visible light remote sensing image, the larger the K value should be, so as to achieve a reasonable superpixel segmentation result. The dark target samples in each band can be expressed as:

[0061]

[0062] Among them, {r, g, b} is the set of three visible light bands: red, blue, and green. i is a dark target in a certain visible light band i, is the distance between all pixels in sub-region K and the seed point.

[0063] Preferably, the d c and d s are the color distance and spatial distance between the pixel in the visible light remote sensing image and the seed point, respectively, and the calculation formulas are as follows:

[0064] d C (p, S) = || F p -F S ||2,d s (p, S) = ||X p -X S ||2, (4)

[0065] Among them, F P =[l P , b P , a P ] T and F S =[l S , b S, a S ] T are the average values ​​of the three color components l, b, a of the pixel and the seed point in the CIELab color space, p =[x p ,y p ] T and X S =[x S ,y S ] T are the centroid positions of the pixel and the seed point respectively. Figure 3 , is a schematic diagram of superpixel segmentation and dark target search provided by an embodiment of the present invention. By way of example, three visible light bands, red, blue and green, are selected, and dark targets are searched in sub-regions on the three visible light bands respectively, so that a consistent number of dark target samples can be obtained on each visible light band.

[0066] Step S12: Classify all dark targets in turn according to whether they are all marked as dark targets in the first visible light band, and obtain absolutely dark targets and relatively dark targets.

[0067] Specifically, according to whether they are all marked as dark targets in the first visible light band, all dark targets are divided in turn to obtain absolute dark targets and relative dark targets, including: if the dark target is recorded as a dark target in all visible light bands in the first visible light band, it is classified as an absolute dark target; if the dark target is recorded as a dark target only in part of the visible light bands in the first visible light band, it is classified as a relative dark target.

[0068] For example, if three visible light bands of red, blue and green are selected, the absolutely dark target can be expressed as:

[0069] DO α =DO r ∩DO g ∩DO b , (5)

[0070] Among them, DO r ,DO g and DO b are the distribution of dark targets in each visible light band, ∩ is the intersection operation, DO α is the spatial distribution of absolutely dark targets.

[0071] The absolute dark target samples and relative dark target samples in each visible light band can be expressed as:

[0072] DO α，i , i∈{r,g,b},

[0073] Step S13: fitting the absolute dark target according to the first visible light band to obtain a corresponding quantitative relationship of cloud intensity.

[0074] Specifically, the fitting of the absolute dark target according to the first visible light band to obtain the corresponding quantitative relationship of cloud intensity includes: fitting any two visible light bands as a combination of the first visible light band according to the absolute dark target to obtain the quantitative relationship of cloud intensity between any two visible light bands.

[0075] Preferably, the quantitative relationship of cloud intensity is:

[0076] DO α，i =a ij ·DO α，j +b ij , (7)

[0077] Among them, DO α，i and DO α，j is the cloud intensity value of an absolutely dark target α in visible light band i and visible light band j respectively, a ij and b ij are the gain and offset corresponding to the linear fit.

[0078] It is worth noting that the quantitative relationship of cloud and fog intensity of the present invention is a robust linear fit of two different visible light bands, and can also be calculated by other nonlinear fitting methods, which are not limited here. Exemplarily, by selecting three visible light bands of red, blue and green, three quantitative relationships of cloud and fog intensity can be calculated, including: the quantitative relationship of cloud and fog intensity in the red and blue visible light bands, the quantitative relationship of cloud and fog intensity in the blue-green visible light bands, and the quantitative relationship of cloud and fog intensity in the red and green visible light bands; in addition, dark targets have a significant representation of clouds and fog, and their pixel values ​​can usually be considered as the cloud and fog intensity in the local area where the dark target is located. According to the definition of absolute dark targets, its pixel values ​​in each visible light band all represent the cloud and fog intensity; in addition, there are a certain number of absolute dark target samples in each band, so the quantitative relationship of cloud and fog intensity can be fitted by robust linear fitting.

[0079] Step S14: according to the quantitative relationship of the cloud intensity, derive the cloud intensity value of the relatively dark target in the second visible light band in turn, and perform a combination operation with the marked dark target in the second visible light band to obtain the first dark target.

[0080] The derivation of the cloud intensity value of the relatively dark target in the second visible light band is specifically as follows: according to the cloud intensity quantitative relationship and the relatively dark target, the cloud intensity value in the second visible light band can be calculated, and the calculation formula is:

[0081] DO β′，i=a ij ·DO β，j +b ij , (8)

[0082] Among them, DO β′，i is the cloud intensity value derived from a relatively dark target β in the visible light band i, DO β，j is the cloud intensity value of the known relatively dark target β in the visible light band j, a ij and b ij are the gain and offset corresponding to the linear fit.

[0083] It is worth noting that the second visible light band is a visible light band in which a relatively dark target is not marked in the selected multiple visible light bands. Exemplarily, if the cloud intensity value in the red visible light band is calculated, and a relatively dark target is not marked in the red visible light band, then the second visible light band of the relatively dark target is the red visible light band. According to the quantitative relationship of cloud intensity between the red and blue visible light bands and the quantitative relationship of cloud intensity between the red and green visible light bands fitted by the absolute dark target, the cloud intensity values ​​of the relatively dark targets in the blue visible light band and the green visible light band under red light can be derived, that is, the cloud intensity value of the relatively dark target that should be marked in the red visible light band can be obtained. See. Figure 4 , is a schematic diagram of fitting the quantitative relationship between dark target classification and cloud intensity provided by an embodiment of the present invention. In the figure, robust linear fitting is performed on pairwise visible light bands according to the absolute dark target; wherein the horizontal and vertical coordinate values ​​of the obtained quantitative relationship of cloud intensity are the brightness values ​​(Digital Number, DN) of the remote sensing image pixels, which are used to identify the grayscale value of the ground object and are unitless integers.

[0084] The step S14 of taking and operating the marked dark target in the second visible light band to obtain the first dark target includes: after deriving the cloud intensity value of the relatively dark target in the second visible light band, increasing the distribution density of the dark target in a single band by dark target complementarity to obtain the first dark target, which can be expressed as:

[0085] DO′ i =DO i ∪DO β′，i , (9)

[0086] Among them, DO′ i The spatial distribution of dark targets after complementary encryption in visible light band i.

[0087] It is worth noting that if all dark targets are absolutely dark targets after statistics, there is no need to perform robust linear fitting for the absolute dark targets because there are no relatively dark targets for deriving cloud intensity values.

[0088] Step S15: Generate a cloud intensity distribution with high spatial granularity according to the first dark target, and correct the visible light remote sensing image.

[0089] The correcting of the visible light remote sensing image includes: correcting the visible light remote sensing image to be processed according to the first visible light band, and the correction formula is:

[0090]

[0091] in, is the visible light remote sensing cloud image to be processed for visible light i, ρ c，i is the high spatial fine-grained cloud intensity distribution generated according to the first dark target, ρ i is the correction result.

[0092] Preferably, the first dark target is processed according to the Kriging interpolation function to obtain a high spatial fine-grained cloud intensity distribution, which can be expressed as:

[0093] ρ c，i =Kriging(DO′ i ), (11)

[0094] Where Kriging(·) is the Kriging interpolation function. Figure 5 , is a schematic diagram of inter-band dark target complementation and high-grained cloud and fog distribution estimation provided by an embodiment of the present invention. In the figure, taking the red visible light band as an example, the relatively dark targets in the blue visible light band and the relatively dark targets in the green visible light band are derived on the red visible light band, and are complemented with the marked dark targets in the red visible light band to increase the distribution density of dark targets in a single band, and then the Kriging interpolation function is used to construct the high spatial fine-grained cloud and fog intensity distribution in the red visible light band.

[0095] It is worth noting that after all bands have been calibrated, it is necessary to determine whether the calibrated visible light remote sensing image is cloud-free. If there are still clouds, re-calibration is performed, otherwise the calibration is terminated.

[0096] The present invention adopts a superpixel segmentation algorithm, which can achieve more reasonable image sub-region division and dark target search, establish screening rules to classify dark targets, and can accurately estimate the quantitative relationship between cloud and fog intensity between visible light bands, explore the complementarity of dark target bands, encrypt dark targets based on the quantitative relationship between cloud and fog intensity, and generate high-spatial-granularity cloud and fog radiation. Therefore, there is no need to rely on additional auxiliary information, and sufficient dark target samples can be obtained using only limited visible light bands. The correction process is simple, has low data requirements, and is easy to implement, so that completely cloud-free visible light remote sensing images can be obtained, providing data support for subsequent optical remote sensing monitoring, land change surveys, etc.

[0097] The present invention also provides a flowchart of visible light remote sensing image correction, see Figure 2 , is a schematic diagram of a process of thin cloud correction for a visible light remote sensing image provided by an embodiment of the present invention, including steps S201 to S211, specifically:

[0098] Step S201: Acquire visible light remote sensing images.

[0099] Step S202: perform superpixel segmentation.

[0100] Step S203: Count local dark targets band by band.

[0101] Step S204, determine whether the statistically dark target is an absolute dark target, if not, proceed to step S205; otherwise, if it is determined to be an absolute dark target, proceed to step S206.

[0102] Step S206, obtaining an absolutely dark target; if all the dark targets are absolutely dark targets after counting, there is no need to proceed to step S207, but directly proceed to step S208.

[0103] Step S207: Perform pairwise robust linear fitting of visible light bands for absolutely dark targets.

[0104] Step S205: According to the linear fitting result, the relatively dark target is deduced in the second visible light band, and a combination operation is performed with the marked dark target in the second visible light band, that is, complementary encryption is performed, and then the process goes to step S208.

[0105] Step S208: Obtain the dark target distribution after complementary encryption.

[0106] Step S209: performing Kriging interpolation for the dark target after complementary encryption.

[0107] Step S210, construct the cloud fog intensity distribution and perform correction. If there is no cloud after correction, proceed to step S211; otherwise, proceed to step S203.

[0108] Step S211: Obtain a cloud-free visible light remote sensing image.

[0109] The present invention divides dark targets, fits absolute dark targets on visible light bands, obtains quantitative relationships between cloud and fog intensities between visible light bands, and then calculates the cloud and fog intensity value of the relative dark target on the second visible light band, thereby deducing the cloud and fog intensity value of the relative dark target on an unmarked visible light band. In the same visible light band, the derived dark targets and the marked dark targets in the second visible light band are complementary, and this complementary relationship can be used to perform a merge operation, thereby obtaining a first dark target that includes the absolute dark target and the dark target that satisfies the quantitative relationship of cloud and fog intensity. The complementary relationship between dark targets can be mined to expand the number of samples, thereby reducing dependence on external data and improving scalability and practicality.

[0110] Through the description of the above implementation mode, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present invention, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0111] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for thin cloud correction of visible light remote sensing images, characterized in that: The method comprises: The visible light remote sensing image to be processed is divided into K sub-regions, and the pixel with the minimum value in each sub-region is searched as a dark target according to the first visible light band; wherein K is a positive integer, and the pixel value is the same as the cloud intensity value; According to whether they are all marked as dark targets in the first visible light band, all dark targets are divided in turn to obtain absolute dark targets and relatively dark targets; Fitting the absolute dark target according to the first visible light band to obtain a corresponding quantitative relationship of cloud intensity; According to the quantitative relationship of cloud intensity, the cloud intensity values ​​of the relatively dark target in the second visible light band are deduced in sequence, and the cloud intensity values ​​are combined with the marked dark target in the second visible light band to obtain the first dark target; According to the first dark target, a cloud intensity distribution with high spatial granularity is generated to correct the visible light remote sensing image; The method of classifying all dark targets in turn according to whether they are all marked as dark targets in the first visible light band to obtain absolute dark targets and relatively dark targets includes: If the dark object is recorded as a dark object in all visible light bands in the first visible light band, it is classified as an absolutely dark object; If the dark object is recorded as a dark object only in a part of the first visible light band, the dark object is classified as a relatively dark object; The second visible light band is a visible light band in which a relatively dark target is not marked among the selected multiple visible light bands; The fitting of the absolute dark target according to the first visible light band to obtain a corresponding quantitative relationship of cloud intensity includes: The first visible light band is combined with any two visible light bands, and fitting is performed according to the absolute dark target to obtain the quantitative relationship of cloud and fog intensity between any two visible light bands; The quantitative relationship of cloud intensity is: DO α,i =a ij ·DO α,j +b ij , Among them, DO α,i and DO α,j is the cloud intensity value of an absolutely dark target α in visible light band i and visible light band j respectively, a ij and b ij is the gain and offset corresponding to the linear fit; The step of sequentially deriving the cloud intensity values ​​of the relatively dark target in the second visible light band according to the cloud intensity quantitative relationship includes: According to the quantitative relationship of cloud intensity and the relatively dark target, the cloud intensity value in the second visible light band can be calculated, and the calculation formula is: DO β′,i =a ij ·DO β,j +b ij , Among them, DO β′,i is the cloud intensity value derived from a relatively dark target β in the visible light band i, DO β,j is the cloud intensity value of the known relatively dark target β in the visible light band j, a ij and b ij are the gain and offset corresponding to the linear fit.

2. The method for thin cloud and fog correction of visible light remote sensing images as claimed in claim 1, characterized in that: The process of dividing the visible light remote sensing image to be processed into K sub-areas includes: By using a superpixel segmentation algorithm based on simple linear iterative clustering, the pixels in the visible light remote sensing image are aggregated to form K irregular sub-regions, and the visible light remote sensing image is divided into corresponding K sub-regions.

3. The method for thin cloud and fog correction of visible light remote sensing images as claimed in claim 1, characterized in that: The step of searching the pixel with the minimum value in each sub-region as a dark target according to the first visible light band includes: According to the K sub-regions, the central cloud intensity value in each block is used as the seed point of the superpixel; The distances between all pixels and the seed point are calculated block by block, and the pixel with the smallest distance is selected as the dark target.

4. The method for correcting thin fog and cloud of visible light remote sensing images as claimed in claim 3, characterized in that: The distance between all pixels and the seed point is calculated block by block, and the specific calculation formula of the distance is: Among them, the seed point is S, the pixel is p, d c and d s are the color distance and spatial distance between the pixel and the seed point in the visible light remote sensing image, respectively, c is the compactness factor, N s is a variable related to the image size and the number of superpixels.

5. The method for correcting thin cloud and fog of visible light remote sensing images as claimed in claim 4, characterized in that: The d c and d s are the color distance and spatial distance between the pixel in the visible light remote sensing image and the seed point, respectively, and the calculation formulas are as follows: d C (p,S)=||F p -F S ||2,d s (p,S)=||X p -X S ||2, (2) Among them, F P =[l P ,b P ,a P ] T and F S =[l S ,b S ,a S ] T are the average values ​​of the three color components l, b, and a of the pixel and the seed point in the CIELab color space, respectively, p =[x p ,y p ] T and X S =[x S ,y S ] T are the centroid positions of the pixel and the seed point respectively.

6. The method for thin cloud and fog correction of visible light remote sensing images as claimed in claim 1, characterized in that: The correcting of the visible light remote sensing image comprises: Correction is performed on the visible light remote sensing image to be processed according to the first visible light band, and the correction formula is: in, is the visible light remote sensing cloud image to be processed for visible light i, ρ c,i is the high spatial fine-grained cloud intensity distribution generated according to the first dark target, ρ i is the correction result; Determine whether the calibrated visible light remote sensing image is cloud-free. If there are still clouds, re-calibrate; otherwise, end the calibration.

Images