CN101246545A - Possion method for removing cloud from optical remote sensing image - Google Patents

Abstract

The invention discloses an Poisson method for cloud removing of optical remote sensing image, which comprises the steps of: selecting basic base map image and alternative image which meet treatment of image removing cloud layer by using image selector; identifying and detecting the range of cloud layer coverage area through cloud area identification unit, making formation cloud area identification template respectively; using the base map image, the alternative image and cloud area identification plate by pyramid processing to construct multi-level similar unit; proceeding Cloud area pixel detection progressively and connected cloud area detection to each grade pyramid, and alternative image detection to connected cloud area of cloud identification template which is processing; seeking linear transformation parameter by using image gradient field regulation pixel to corresponding cloudless point department of the alternative image and basic base map image around connected cloud area; taking non-cloud megapixel point around connected cloud area of basic base map image as Dirichlet boundary condition, and linear transformation alternative area image gradient field as guide field, proceeding fusion processing to the image by using Poisson removing cloud processing algorithm.

Description

A kind of remote sensing image goes the Poisson method of cloud

Technical field

The present invention relates to the pretreated method of a kind of image in the remote sensing image processing field, relate in particular to the Poisson method that a kind of remote sensing image removes cloud.

Background technology

Because remote sensing image when obtaining, is difficult to avoid cloudy rainy weather to cause to be obtained image and is subjected to blocking of cloud, need to be in feature in the motion process all the time according to cloud, the cloud layer of multidate image series is carried out cloud remove processing.Therefore, various remote sensing image clouds occur with arising at the historic moment removed disposal route.Usually, when thin cloud or mist appear in image, can carry out various images to single image and recover to handle.But when image spissatus layer and overlay area occur when big, then any method that original image is recovered to handle all seems weary and unable.Therefore, people consider to utilize the complementary information of multidate image series to carry out cloud removal processing, that is: utilize the image of close areal of time that alternate process is carried out in the cloud sector, thereby reach the effect of cloud.

At present, the internal and international relevant remote sensing image clouds removal of the report disposal route that goes up all is that the position of supposing cloud layer is in the dynamic change, all there is the possibility that is not covered by cloud layer in any place, so it is combined and spliced that the pixel that can select cloudless layer to cover carries out, then can generate a combination image complete, that no cloud layer covers.Thereby concrete disposal route mainly contains two kinds:

A kind of is that combined treatment (Composing) method is carried out image cloud removal pre-service, it is by calculating NDVI (vegetation index) value, adopting maximum vegetation index (MAX NDVI) to eliminate the influence of sun altitude, satellite visual angle and cloud, is the NDVI value but it obtains, inextensive palinspastic map.

Another kind method is based on the method that simple inverting substitutes, and it was divided into for two steps in cloud removal process: cloud detection and cloud are rejected.Wherein, cloud detection is often based on threshold technology, that is: at visible light, near-infrared band, cloud layer has high reflectance with respect to different underlying surfaces such as vegetation, soil, waters; and have the characteristic of low bright temperature and produce at the thermal infrared wave band, it is the basis that cloud is picked out.Because the threshold technology that cloud detection is adopted among the present invention adopts conventional method, only discusses the existing pros and cons of cloud elimination method at this.

After detecting the image medium cloud and covering pixel, simple inverting alternative method is to the cloud layer image in changing relatively, adopts same area, phase picture rich in detail data are carried out inverting and substituted the cloud sector when close, promptly cloud layer covered pixel and rejects and repair.Because it is X that this method is provided with the original image of cloud layer, the substitute picture rich in detail is Y, ymax, ymin, n, xmax, xmin and m are respectively substitute image pixel maximal value, minimum value, image picture elements number and original image pixel maximal value, minimum value and the pixel number of (not comprising that cloud covers pixel), and the substitute value xi of cloud layer covering pixel is the linear transformation of corresponding substitute image picture elements yi in the original image so:

$x_i=\frac{y_i-\stackrel{\‾}{y}}{y_{\max }-y_{\min }}(x_{\max }-x_{\min })+\stackrel{\‾}{x},$ $\stackrel{\‾}{x}=\frac{\mathrm{\Σ}{x}_i}{m},$ $\stackrel{\‾}{y}=\frac{\mathrm{\Σ}{y}_i}{n}.$

The fragmentary fritter cloud layer that distributes in the image then can the adjacent non-cloud pixel value of same image be carried out interpolation rejects.Though, simple inverting alternative energy is recovered original image, but owing to have tone difference between the image of phase when close, tangible substitutes' area edge fit vestige can appear after simple inverting substituted, if will eliminate these vestiges, often need carry out man-machine interactive operation, utilize the image stretch technology to regulate image enhancement effect of visualization in twos, but still can have influence on further quantitative Application analysis.Therefore, need to carry out cloud and reject when recovering original image at the relatively large spissatus layer in overlay area, the sharp trace that image mosaic occurs, and can't reach problem such as seamless spliced effect, the research and development remote sensing images go the algorithm of cloud layer, complete method and treatment scheme.At present, this image cloud is removed preprocess method and be there is no suitable method and can be used in the practical application.Yet along with the increase of various remote sensing images (visible light, near infrared and thermal infrared wave band) acquiring way, it is to have significance and the practical value that promotes technical development that the image cloud is removed the optimization process method.

Summary of the invention

Shortcoming at prior art, the purpose of this invention is to provide and a kind ofly can in big overlay area, remove spissatus layer, image after cloud is rejected had both recovered the gray feature of original image, reach seamless spliced effect again, and form the Poisson method of removing cloud at the remote sensing image of visible light and near-infrared band.

For achieving the above object, technical scheme of the present invention is: a kind of remote sensing image goes the Poisson method of cloud, and it may further comprise the steps:

A, utilize image selector to select one group to satisfy image and go basic base map image A that cloud layer handles and be distributed with complementary set of diagrams picture as substitute image sets B1 with basic base map image cloud ... Bn;

B, base map image A and the substitute image B of selecting 1 detected through the identification of cloud sector recognition unit, extract the scope in cloud covered areas territory, and the cloud covered areas territory is made into cloud sector identification masterplate C respectively _A, C _B1Binary map;

C, with base map image A, substitute image B 1 and cloud sector identification column C _A, C _B1Handle structure classification similar units through pyramid, obtain A _i, B _1i, C _A ⁱ, C _B1 ⁱ, i=1,2, Λ, n, corresponding 1 to n serves as reasons thin yardstick to thick yardstick;

D, since the n level, carry out step by step the cloud sector pixel and detect, each grade pyramid pursue cloud is communicated with the detection of distinguishing, for the i level, utilize connectedness to identify C _A ⁱThe cloud sector of each connection is to C _A ⁱEach connection cloud sector of handling image detection of substituting in the cloud identification masterplate, whether also be cloud sector, if not, then be for further processing, if then detect the next stage image or abandon substitute to this zone if differentiating in the institute corresponding region;

E, utilize image gradient field control pixel to the substitute image B _1iWith basic base map image A _iCorrespondence around in current connection cloud sector does not have the cloud point place and asks the linear transformation parameter;

F, with basic base map A _iIn respectively be communicated with the cloud sector periphery non-cloud pixel as the Dirichlet boundary condition, be guide field with the gradient fields of the substitutes' area image after the linear transformation, utilize Poisson to go the cloud layer processing unit that image is carried out fusion treatment, make it satisfy following alignment system of equations:

p?∈Ω， $\left|N_p\right|f_p-\underset{q\∈N_p\∩\mathrm{\Ω}}{\mathrm{\Σ}}{f}_q=\underset{q\∈N_p\∩\∂\mathrm{\Ω}}{\mathrm{\Σ}}{h}_q+\underset{q\∈N_p}{\mathrm{\Σ}}(g_p-g_q)$

Wherein, h represents basic base map image, and g represents the image of substituting, f is a defined range, and Ω is the subclass in corresponding cloud sector, and its border represents that with  Ω Np represents that S goes up neighbours' territory point or the eight neighborhoods point on each pixel p, | Np| represents the number of the point on the neighborhood, and fp is the pixel value of this point.

Also comprise step g in the said method: another image B 2 and last cloud A repeating step b, c, d, e, f as a result of going of phase when close, the Poisson cloud that carries out Multiple Cycle is removed and is handled.

Also comprise step h in the said method: basic base map image and substitute imagery exploitation fractus removal processing unit are carried out the image interpolation processing.

In step e, image gradient field control unit does not have the cloud point place to the correspondence around the current connection cloud sector and adopts least square method to ask the linear transformation parameter.

In step b, the cloud sector recognition unit adopts threshold technology or mode identification method to detect the cloud sector.

Description of drawings

Fig. 1 removes the processing flow chart of the Poisson method of cloud for remote sensing image of the present invention.

Embodiment

The present invention is described in further detail below in conjunction with accompanying drawing.

Theoretical foundation of the present invention is the principle that Poisson (Poisson) image co-registration and inverting substitute.Based on these two principles, we propose the borderline image pixel of a kind of cloud detection with base image as boundary condition, with the gradient fields in corresponding district on the adjacent time-series image of other complementation as guide field, the method of interpolation is filled up in the base image cloud sector, that is: the Poisson of remote sensing image merges cloud and removes disposal route.

Specifically, as shown in Figure 1, remote sensing image proposed by the invention goes the Poisson method of cloud, comprises that the bulk cloud layer is removed processing and fractus is handled two kinds of situations.When filling up the zone at bigger cloud, concrete analyzing and processing process comprises the steps:

A, utilize image selector to select one group to satisfy image and go basic base map image A that cloud layer handles and be distributed with complementary set of diagrams picture as substitute image sets B1 with basic base map image cloud ... Bn.

B, base map image A and the substitute image B of selecting 1 detected through the identification of cloud sector recognition unit, extract the scope in cloud covered areas territory by pixel, and the cloud covered areas territory is made into cloud sector identification masterplate C respectively _A, C _B1Binary map.

C, with base map image A, substitute image B 1 and cloud sector identification column C _A, C _B1Carry out pyramid (multiple dimensioned) resolution process, make up the pyramid similar units, obtain A _i, B _1i, C _A ⁱ, C _B1 ⁱ, i=1,2, Λ, n, so that adopt pyramid structure to accelerate speed of convergence, corresponding 1 to n serves as reasons thin yardstick to thick yardstick.

D, since the n level, the detection that cloud is communicated with the district is pursued to each grade pyramid in cloud sector detecting unit step by step, as: for the i level, utilize connectedness to identify C _A ⁱThe cloud sector of each connection is detected C to each when the connection cloud sector of pre-treatment _B1 ⁱIn the substitute image in corresponding cloud sector, that is: to C _A ⁱThe substitute image is detected in each connection cloud sector of handling in the cloud identification masterplate, and whether differentiate in the institute corresponding region also is the cloud sector.If not, then be for further processing, otherwise seek down piece image or abandon substitute this zone.

E, for the local non-cloud layer zone of visible light, near-infrared band, the data owner of phase will not be subjected to the influence of atmosphere and solar radiation simultaneously, can think that only there is linear differences in both, then utilizes image gradient field control unit to substitute image B 1 _iWith basic base map image A _iThere is linear relationship in no cloud point around the current connection cloud sector.Therefore, respectively to substitute image B 1 _iWith basic base map image A _iCorrespondence around in current connection cloud sector does not have cloud point and asks the linear transformation parameter, can eliminate the linear differences of substitute image and basic base map image.Its existing transform method can adopt least square method to find the solution the linear transformation parameter, makes the image pixel value of basic base map and substitute figure reach the most approaching.

F, with basic base map A _iIn respectively be communicated with the cloud sector periphery non-cloud pixel as the Dirichlet boundary condition, be guide field with the gradient fields of the substitutes' area image after the linear transformation, utilize Poisson to go the cloud layer processing unit that image is carried out fusion treatment, make it satisfy following alignment system of equations:

p?∈Ω， $\left|N_p\right|f_p-\underset{q\∈N_p\∩\mathrm{\Ω}}{\mathrm{\Σ}}{f}_q=\underset{q\∈N_p\∩\∂\mathrm{\Ω}}{\mathrm{\Σ}}{h}_q+\underset{q\∈N_p}{\mathrm{\Σ}}(g_p-g_q)$

Wherein, h represents basic base map image, and g represents the image of substituting, f is a defined range, and Ω is the subclass in corresponding cloud sector, and its border represents that with  Ω Np represents that S goes up neighbours' territory point or the eight neighborhoods point on each pixel p, | Np| represents the number of the point on the neighborhood, and fp is the pixel value of this point.

Step g, when close another image B 2 and last cloud A repeating step b, c, d, e, f as a result of going of phase, the Poisson cloud that carries out Multiple Cycle is removed and is handled.

Step h, basic base map image and substitute imagery exploitation fractus are removed processing unit carry out image interpolation and handle.

In said method, the concrete operations of step a belong to the conventional method of remote sensing images optimized choice, its fundamental purpose is to utilize image selector to select one group to satisfy the time-series image data that image goes cloud layer to handle, make original image satisfy the required pacing items of processing procedure, that is: objective area cloud layer overlay capacity minimum, top-quality when some phase images and be distributed with complementary set of diagrams picture as the substitute image sets as primary image with the primary image cloud.

Step b belongs to the important step of image cloud information extraction, and its extracting method often adopts the cloud sector recognition unit to detect the scope of cloud covered areas, so that make cloud sector identification masterplate respectively.Concrete method is that the cloud sector recognition unit adopts threshold technology or mode identification method to detect the cloud sector state: various threshold technologies are based on cloud for the physical characteristics of different underlying surface such as vegetation, soil, waters in different wave spectrum sections, as: visible light and near-infrared band have high reflectance, and have the characteristic etc. of low bright temperature at the thermal infrared wave band, and produced the difference of cloud detection method of optic.

Step c is that cloud is removed the image of processing and their cloud sector template makes up pyramid structure to participating in, available gaussian pyramid, and conversion obtains Ai, B1i, C _A ⁱ, C _B1 ⁱ, i=1,2, Λ, n, corresponding 1 to n serves as reasons thin yardstick to thick yardstick; It is to make up reasonably the flow process that cloud removal is step by step handled that pyramid structure Poi sson merges cloud removal flow process purpose, thereby reaches the purpose of acceleration iteration speed, optimization process flow.

In said method, steps d mainly is the complementarity of judging between selected image, selects suitable substitute image.This process is from the n level of step pyramid structure that c defines, utilize the cloud sector step by step detecting unit each grade pyramid is pursued the detection that cloud is communicated with the district, whether promptly the substitute image is detected in the connection cloud sector in a certain cloud identification masterplate, differentiating in the institute corresponding region also is the cloud sector.If not, then be for further processing, otherwise seek down piece image or abandon substitute this zone.

In said method, step e utilizes image gradient field control unit that the difference of gradient fields between basic base map and substitute figure image is regulated, and this method adopts least square method to find the solution the linear transformation parameter, is summed up as following formula (1):

$F(a,b)=\stackrel{n}{\mathrm{\Σ}}{E_i}^2=\stackrel{n}{\mathrm{\Σ}}{(y_i-{\mathrm{ax}}_i-b)}^2---\left(1\right)$

Constant a and b among the y=ax+b make formula (1) for minimum.At this, establishing aforementioned basic base map Ai respectively is y, and substitute image B 1i is x, and the correspondence around in current connection cloud sector does not have cloud point and asks the linear transformation parameter, make to reach that pixel value reaches the most approaching between two images, eliminate the purpose of linear difference between substitute image and basic base map image.

In said method, step f utilizes Poisson to go the cloud layer Processing Algorithm to carry out cloud layer to remove the image co-registration processing, suppose: h represents the image of basic base map, g represents the image of substituting, S is a target composograph f defined range, Ω is the subclass in corresponding cloud sector, and its border represents that with  Ω corresponding basic base map Ai is communicated with the non-cloud pixel of cloud sector periphery.As the Dirichlet boundary condition, as guide field, then guide interpolation to be defined as with boundary pixel with the gradient fields of the substitutes' area image after the linear transformation:

$\underset{f}{\min }{\∫\∫}_{\mathrm{\Ω}}{\left|\right|\▿f-\▿g\left|\right|}^2\mathrm{withf}{|}_{\∂\mathrm{\Ω}}=h{|}_{\∂\mathrm{\Ω}},---\left(2\right)$

Wherein, $\▿=[\frac{\∂}{\∂x},\frac{\∂}{\∂y}]$ Be partial differential operator, its optimum solution is the unique solution of following Poisson equation with the Dirichlet boundary condition:

Δf＝Δg，with?f| _Ω＝h| _Ω， (3)

Wherein, $\mathrm{\Δ}=\frac{{\∂}^2}{{\∂x}^2}+\frac{{\∂}^2}{{\∂y}^2}$ It is the Laplacian operator.

Above-mentioned basic base map or substitute image are discrete digital picture, and then S is limited discrete lattice point, and Np represents that S goes up neighbours' territory point or the eight neighborhoods point on each pixel p, | Np| represents the number of the point on the neighborhood, and fp is the pixel value of this point.Utilize finite difference, variational problem (2) can disperse and turn to following double optimization problem:

$\underset{f{|}_{\mathrm{\Ω}}}{\min }\underset{p}{\mathrm{\Σ}}\underset{q\∈N_p}{\mathrm{\Σ}}{(f_p-f_q-g_p+g_q)}^2,\mathrm{with}{f|}_{\∂\mathrm{\Ω}}={h|}_{\∂\mathrm{\Ω}},---\left(4\right)$

Its optimum solution satisfies following system of linear equations:

p∈Ω， $\left|N_p\right|f_p-\underset{q\∈N_p\∩\mathrm{\Ω}}{\mathrm{\Σ}}{f}_q=\underset{q\∈N_p\∩\∂\mathrm{\Ω}}{\mathrm{\Σ}}{h}_q+\underset{q\∈N_p}{\mathrm{\Σ}}(g_p-g_q)---\left(5\right)$

(5) formula is the system of linear equations of classics, sparse, symmetry, can find the solution it with process of iteration (sequence overrelaxation Gauss-Saden that iteration Gauss-Seidel iteration with successiveoverrelaxation or V-cycle multigrid).After the fusion of i level was finished, its fusion results as the initial value (accelerating iteration speed whereby) of next stage (i-1 level), was communicated with Qu Quyun to next stage one by one after simple interpolations.Wherein the initial value of n layer is An itself.

In said method, step g be at needs several when close the time-series image (Bn) of phase carry out multiple cloud and remove situation about handling, as when close another image B 2 and the last cloud fusion results A repeating step b～f that goes of phase, the Poisson cloud that carries out Multiple Cycle is removed and is handled.At first, the 1st grade fusion results A2 is exactly A, and the fusion results of B1 image promptly utilizes B1 that A is carried out the result that cloud is rejected.Secondly, the image B 2 of phase merges cloud with the A2 as a result that back merges according to preceding method when utilizing another close again.At last, the circulation back up to draw satisfied fusion results or all when close the data of phase all utilized and finished, then obtain merging the multiple cloud result images that goes that cloud is removed optimization process flow through Poisson.

And, then must pursue wave band repeating step b～g for multiwave image, remove with the cloud that obtains different-waveband and handle the basic data of image as further graphical analysis.Aforementioned 7 steps, the cloud that is primarily aimed at bulk cloud layer district is removed processing, exists and still have some very little fractus on the Poisson fusion results image.

In said method, step h is the cloud removal method at fractus, it adopts the disposal route of interpolation, promptly the cloud-free area picture dot carries out interpolation around the basis, and concrete interpolating method need be analyzed this regional space entity distribution character and correlativity decision, as: can adopt golden interpolation (Kriging) method in the gram usually, that is: hypothesis is in enough small neighbourhoods, the characteristic or the attribute data height correlation of relevant atural object, variance is changed to zero, can think that then interpolated point and variation difference on every side are zero, promptly gradient is zero.Thereby, can this carry out the zonule interpolation, or adopt the Poisson fusion method to draw end product as guide field.

In sum, the Poisson method that the remote sensing image that the present invention proposes removes cloud, it be at areal, one group of time-series image that phase did not obtain simultaneously carries out Poisson and merges cloud and remove the method for handling.It adopts in twos cloud to remove strategy to carry out image and substitute or fill up, be before guide field carries out interpolation processing to image based on the substitute image, must carry out linear transformation based on basic base map image gradient field to the substitute image earlier, to construct rational gradient fields; Then, utilize the Poisson equation of finding the solution again, realize local gray level or color modification, seamless spliced, thereby reach the cloud layer that both can reject in the image, can guarantee the seamless spliced cloud effect of going that original image information is intact again with the Dirichlet boundary constraint.

Advantage of the present invention or effect are embodied in following aspect:

1, main innovation part of the present invention is: on (1) method: adopt the Poisson fusion method to carry out cloud and remove processing.(2) on the optimization process flow: not only make up the pyramid treatment scheme and reach the purpose of accelerating iteration speed, optimization process flow, also design, realized carrying out the automated process flow that multiple cloud is removed with several close time-series images (Bn).(3) fractus of leaving at image has replenished the interpolation method of removing fractus, so that obtain more distinct image in the optimization process flow among the present invention.

2, major advantage of the present invention is: (1) provides a kind of can remove spissatus layer in big overlay area, solved the cloud layer of both rejecting on the image, can recover Poisson (Poisson) cloud of the gray feature of original image again and remove method and the flow process of handling, and then reach the effect of seamless image splicing.This method is adapted to carry out remote sensing image clouds in the Visible-to-Near InfaRed scope and removes processing.(2) pyramid and multiple optimization process flow of removing cloud have been realized, not only can improve cloud and remove pretreated work efficiency that also having solved needs several close time-series images (Bn) to carry out cloud removal processing automated process flow.(3) remote sensing images are not had the special processing requirement, make that the applicability of this method is better.(4) it is simply effective that the image cloud is removed the data pre-service.

Claims (5)

1, a kind of remote sensing image goes the Poisson method of cloud, it is characterized in that may further comprise the steps:

A, utilize image selector to select one group to satisfy image and go basic base map image A that cloud layer handles and be distributed with complementary set of diagrams picture as substitute image sets B1 with basic base map image cloud ... Bn;

B, base map image A and the substitute image B of selecting 1 detected through the identification of cloud sector recognition unit, extract the scope in cloud covered areas territory, and the cloud covered areas territory is made into cloud sector identification masterplate C respectively _A, C _B1Binary map;

C, with base map image A, substitute image B 1 and cloud sector identification column C _A, C _B1Handle the structure similar units through pyramid, obtain A _i, B _1i, C _A ⁱ, C _B1 ⁱ, i=1,2, Λ, n, corresponding 1 to n serves as reasons thin yardstick to thick yardstick;

D, since the n level, carry out the cloud sector pixel step by step and detect and each grade pyramid pursue cloud be communicated with the detection of distinguishing, for the i level, utilize connectedness to identify C _A ⁱThe cloud sector of each connection is to C _A ⁱEach connection cloud sector of handling image detection of substituting in the cloud identification masterplate, whether also be cloud sector, if not, then be for further processing, if then detect the next stage image or abandon substitute to this zone if differentiating in the institute corresponding region;

E, utilize image gradient field control pixel to the substitute image B _1iWith basic base map image A _iCorrespondence around in current connection cloud sector does not have the cloud point place and asks the linear transformation parameter;

F, with basic base map A _iIn respectively be communicated with the cloud sector periphery non-cloud pixel as the Dirichlet boundary condition, be guide field with the gradient fields of the substitutes' area image after the linear transformation, utilize Poisson to go the cloud layer processing unit that image is carried out fusion treatment, make it satisfy following alignment system of equations:

p∈Ω， $\left|N_p\right|f_p-\underset{q\∈N_p\∩\mathrm{\Ω}}{\mathrm{\Σ}}=\underset{q\∈N_p\∩\∂\mathrm{\Ω}}{\mathrm{\Σ}}{h}_q+\underset{q\∈N_p}{\mathrm{\Σ}}(g_p-g_q)$

Wherein, h represents basic base map image, and g represents the image of substituting, f is a defined range, and Ω is the subclass in corresponding cloud sector, and its border represents that with  Ω Np represents that S goes up neighbours' territory point or the eight neighborhoods point on each pixel p, | Np| represents the number of the point on the neighborhood, and fp is the pixel value of this point.

2, remote sensing image according to claim 1 goes the Poisson method of cloud, it is characterized in that: also comprise step g, when close another image B 2 and last cloud A repeating step b, c, d, e, f as a result of going of phase, the Poisson cloud that carries out Multiple Cycle is removed and is handled.

3, remote sensing image according to claim 2 goes the Poisson method of cloud, it is characterized in that: also comprise step h, basic base map image and substitute imagery exploitation fractus removal processing unit are carried out the image interpolation processing.

4, remote sensing image according to claim 1 goes the Poisson method of cloud, it is characterized in that: in step e, image gradient field control unit does not have the cloud point place to the correspondence around the current connection cloud sector and adopts least square method to ask the linear transformation parameter.

5, remote sensing image according to claim 1 goes the Poisson method of cloud, it is characterized in that: in step b, the cloud sector recognition unit adopts threshold technology or mode identification method to detect the cloud sector.

Images