CN109063663A - A kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence - Google Patents
A kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence Download PDFInfo
- Publication number
- CN109063663A CN109063663A CN201810908851.4A CN201810908851A CN109063663A CN 109063663 A CN109063663 A CN 109063663A CN 201810908851 A CN201810908851 A CN 201810908851A CN 109063663 A CN109063663 A CN 109063663A
- Authority
- CN
- China
- Prior art keywords
- cloud
- image
- matrix
- pixel
- remote sensing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 31
- 238000005516 engineering process Methods 0.000 title claims abstract description 14
- 239000011159 matrix material Substances 0.000 claims abstract description 60
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 11
- 230000011218 segmentation Effects 0.000 claims abstract description 7
- 238000006243 chemical reaction Methods 0.000 claims abstract description 4
- 238000000513 principal component analysis Methods 0.000 claims abstract 3
- 238000000034 method Methods 0.000 claims description 23
- 238000000354 decomposition reaction Methods 0.000 claims description 14
- 230000008439 repair process Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 4
- 238000013459 approach Methods 0.000 claims description 3
- 230000003190 augmentative effect Effects 0.000 claims description 3
- 238000003709 image segmentation Methods 0.000 claims description 3
- 238000002360 preparation method Methods 0.000 claims description 3
- 238000012847 principal component analysis method Methods 0.000 claims description 3
- 238000000926 separation method Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000011160 research Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 12
- 230000008901 benefit Effects 0.000 description 4
- 238000012804 iterative process Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000008602 contraction Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
- G06T3/147—Transformations for image registration, e.g. adjusting or mapping for alignment of images using affine transformations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of spissatus detection of timing remote sensing image by slightly to essence and minimizing technologies.Image is pre-processed first, including super-pixel segmentation and conversion timing sequence image constitute matrix.Background (ideal cloudless earth's surface information) and prospect (cloud and its shade) are modeled respectively using low-rank theory and structural sparse theory, it using Robust Principal Component Analysis frame and introduces affine Transform Model timing image is separated into foreground and background, obtain cloud and the cloud shadow region of super-pixel grade.Different scale factors is arranged to cloud sector and non-cloud sector again, is decomposed using original Robust Principal Component Analysis, the spissatus removal of remote sensing sequential images is completed.The present invention is directed to the multi-temporal remote sensing image sequence not being registrated, substantially increases the precision and efficiency of the spissatus removal of remote sensing image, and produce high-precision cloud and cloud shadow Detection product, with high multi-temporal remote sensing image research and application value.
Description
Technical field
The invention belongs to Remote Sensing Image Processing Technology fields, are related to a kind of by slightly to the spissatus detection of timing remote sensing image of essence
With minimizing technology.
Background technique
Influenced by optical sensor imaging mechanism, inevitably there is cloud block in remote sensing image, cause the quality of image by
To seriously affecting.Spissatus to block so that earth's surface information lacks completely, subsidiary cloud shade also seriously changes spectral information, destroys
The global consistency of remote sensing image, seriously hinders the work such as terrain classification, differentiation.Therefore, remove remote sensing image medium cloud and
The pollution of cloud shade, repairs corresponding earth's surface information and is of great significance.
Traditional remote sensing image goes cloud method to be often directed to individual image or the progress cloud block information reparation of two phase images.
Method based on individual image reparation according to pixel cloudless around cloud sector to transmitting interpolation inside cloud sector, with the increase in cloud sector,
Uncertainty is continuously increased, and causes to repair accuracy degradation.Using the cloudless image of close phase as with reference to image into
It is the strategy mainly used at present that row, which is repaired,.Most basic method is the cloudless picture replaced with target image cloud sector with reference to image
Element, and the color difference filled up between image blocks and target image is eliminated using even color and fusion method, while eliminating splicing seams effect.
It is similar, using with reference to image gradient information, as guidance, preferable reparation can also be obtained by carrying out information clone to reference image
As a result.Another kind of method positions pixel progress cloud sector information reparation similar with cloud sector pixel, including target using with reference to image
Similar pixel on image on the similar pixel of itself and reference image, such methods can overcome seasonal change to a certain extent
Color difference caused by change.For these methods all to cloud sector magnitude, there are cloud sectors to repair precision and efficiency has a greatly reduced quality when excessive
Problem.In recent years, machine learning is also applied to the removal of remote sensing image cloud, for example is based on rarefaction representation, compressed sensing and depth
The method of study.However, there are blurring effects for the method reparation result of rarefaction representation at present, and dictionary learning time cost is very
It is high;The atural object classification that compressed sensing method covers cloud sector size and cloud sector is sensitive;Deep learning is limited to the type of training sample
And quantity.It is all target image to be assumed that based on the cloudless method with reference to image and (i.e. Pixel-level is matched with reference to image rigid registrations
It is quasi-), the practicability for resulting in such method is limited.Remote sensing image data amount is huge, and the same more phase covering images in place are sufficient, benefit
It there is no disclosed research and technology with the batch image cloud removal that the image sequence of multidate carries out high-accuracy high-efficiency rate.
Summary of the invention
The problem of present invention aim to address multi-temporal remote sensing image sequence batch de clouds, provides a kind of precise and high efficiency
Cloud sector earth's surface information reparation method, it is capable of handling the image sequence not being registrated, while the picture of degree of precision can be generated
Plain grade cloud and its shadow Detection product repair precision height to high-volume timing remote sensing sequential images containing cloud, high-efficient.
The present invention provides a kind of by slightly to the smart spissatus detection of timing remote sensing image and minimizing technology, comprising the following steps:
Step 1, data preparation, the multidate image containing cloud for choosing certain amount and the same interest region of covering constitute image
Sequence;
Step 2, super-pixel segmentation is carried out to every image, and using the super-pixel block after segmentation as cloud and shade coarse extraction
Processing unit;
Step 3, every image store is normalized into 0 to 1 at column vector and by pixel value, and image sequence is pressed acquisition time
It is arranged in a matrix, the pixel number of individual image of behavior is classified as total image number;
Step 4, using low-rank model and structural sparse model respectively to background (ideal cloudless earth's surface information) and before
Scape (cloud and its shade) is modeled, and affine Transform Model is introduced, and constructs matrix decomposition objective function, using robust principal component point
Analysis frame carries out solution separation to the matrix of the input composition of image sequence containing cloud, completes the registration and cloud and its shade of sequential images
The coarse extraction of region super-pixel grade;
Step 5, building is repaired objective function and is set using original Robust Principal Component Analysis method to cloud sector and cloud-free area
Different scale factors is set, is completed to the information reparation of sequential images cloud sector and Pixel-level cloud and its shadow Detection;
Step 6, the image containing cloud step 5 obtained repairs result and cloud and shadow detection result according to the inverse place of step 3
Reason, output sequence image go cloud result and cloud and its shadow detection result.
Further, the image quantity chosen in step 1 no less than guarantees that each pixel counterpart orientation is set and is at least seen
Primary minimum image number is measured, i.e., the cloudless pixel of all images containing cloud can at least be combined into a cloudless image.
Further, use SLIC partitioning algorithm by every Image Segmentation at several super-pixel block in step 2.
Further, the specific implementation of step 4 includes following sub-step,
Step 4.1, assumed according to background low-rank and foreground structures be sparse it is assumed that building matrix decomposition objective function:
minL,s||L||*+λψ(S)
Wherein, L is low-rank matrix, indicates background, and S is sparse matrix, indicates prospect, and D indicates input matrix;λ indicates flat
Weigh coefficient, and τ indicates affine transformation, i.e.,It indicates Vec () is indicated
Image is changed into column vector;N and K respectively indicates total image number and every image super-pixel block number,Indicate each super-pixel
The weight of block,Indicate each super-pixel;||·||*For nuclear norm, the sum of representing matrix characteristic value;||·||∞For infinite model
Number indicates the maximum value in one group of numerical value;
Step 4.2, the objective function of step 4.1 is converted using non-tight augmented vector approach:
Wherein, Y indicates Lagrange multiplier, and μ indicates a positive scalar,For square of Frobenius norm,
The quadratic sum of representing matrix all pixels;Solve includes executing following steps:
Step 4.2.1 initializes L, S, τ, Y, μ, ρ;
Step 4.2.2 calculates L, converts step 4.1 formula to about the subproblem for solving L:
Step 4.2.3 calculates S, converts step 4.1 formula to about the subproblem for solving S:
Step 4.2.4 updates τ=τ+Δ τ;
Step 4.2.5 updates
Step 4.2.6 updates μ=ρ μ;
Step 4.2.7 calculates iterated conditional, terminates if meeting termination condition, otherwise return to step 4.2.2, Zhi Daoshou
Termination iteration is held back, the image sequence containing cloud being registrated and corresponding cloud and its shadow region exposure mask are exported.
Further, the specific implementation of step 5 includes following sub-step,
Step 5.1, building cloud sector information repairs objective function:
minL,S||L||*+α||PΩ(S)||1+β||PΩ-(S)||1
Wherein, PΩIndicate cloud and cloud shadow mask, PΩIndicate that cloudless area mask, α, β indicate different scale factors,Indicate sequence video conversion after the obtained registration of step 4 at input matrix, | | | |1For l1Norm indicates that all elements are exhausted
To the sum of value;
Step 5.2, using 5.1 objective function of non-tight method of Lagrange multipliers solution procedure, including following steps are executed:
Step 5.2.1 initializes L, S, Y, μ, ρ;
Step 5.2.2 calculates L,
Step 5.2.3 calculates S,
Step 5.2.4 updates
Step 5.2.5 updates μ=ρ μ;
Step 5.2.6 calculates stopping criterion for iteration, terminates if meeting termination condition, otherwise return to step 5.2.2, directly
Iteration, output matrix L and S are terminated to convergence.
Further, the specific implementation of step 6 includes following sub-step,
Step 6.1, the s-matrix obtained according to step 5, given threshold T to S carry out binaryzation with obtain Pixel-level cloud and
Shadow detection result, wherein threshold value T is the standard deviation of each column (corresponding to each image) of matrix S;
Step 6.2, the low-rank matrix L that step 5 obtains and the testing result that step 6.1 obtains are carried out according to step 3 inverse
Processing, output finally eliminate spissatus and cloud shade sequential images and corresponding cloud and cloud shadow Detection product.
Compared with prior art, it the advantages and benefits of the present invention are: the image sequence not being registrated can be handled, and is not required to
Select cloudless reference image and special cloud and cloud shadow Detection algorithm, to high-volume timing remote sensing sequential images containing cloud into
Row processing is completed at the same time cloud and cloud shadow Detection and cloud sector information reparation, and processing accuracy is high, high-efficient.
Detailed description of the invention
Fig. 1 is the overview flow chart of the embodiment of the present invention;
Fig. 2 is the sequence diagram of remote sensing image containing cloud of the embodiment of the present invention;
Fig. 3 is the simulation affine transformation cloud and shadow detection result schematic diagram of the embodiment of the present invention;
Fig. 4 is that the sequential images cloud sector information of the embodiment of the present invention repairs result schematic diagram;
Fig. 5 is that the sequential images cloud sector information of the embodiment of the present invention repairs detailed schematic;
Specific embodiment
Understand for the ease of those of ordinary skill in the art and implement the present invention, with reference to the accompanying drawings and embodiments to this hair
It is bright to be described in further detail, it should be understood that implementation example described herein is merely to illustrate and explain the present invention, not
For limiting the present invention.
It is specific the present invention provides a kind of spissatus detection of timing remote sensing image by slightly to essence and minimizing technology such as Fig. 1
Implement the following steps are included:
Step 1: data preparation, as shown in Fig. 2, choosing the multidate shadow containing cloud of certain amount and the same interest region of covering
As constitute image sequence, image quantity generally requires no less than guarantee each pixel counterpart orientation set at least observed once
Minimum image number (the cloudless pixel of all images containing cloud can at least be combined into a cloudless image), and image quantity more hair
Bright precision and odds for effectiveness is more obvious.Data may include single or multiple channels;
Step 2: super-pixel segmentation is carried out to every image.Using SLIC partitioning algorithm by every Image Segmentation at several super
Block of pixels, as cloud and the processing unit of shade coarse extraction.SLIC is open source algorithm, and 2 parameters need to be arranged, respectively pre- score
The tightness of block number and each piece is cut, the segmentation block number that the present invention recommends is set as the smaller value of image ranks number, closely
Degree is set as 30, and those skilled in the art can voluntarily adjusting parameter;
Step 3: every image store is normalized into 0 to 1 at column vector and by pixel value, and image sequence is pressed acquisition time
It is arranged in a matrix, the pixel number of individual image of behavior is classified as total image number;
Step 4: using low-rank model and structural sparse model respectively to background (ideal cloudless earth's surface information) and before
Scape (cloud and its shade) is modeled, and affine Transform Model is introduced, and constructs matrix decomposition objective function, using robust principal component point
Analysis frame carries out solution separation to the matrix of the input composition of image sequence containing cloud, completes the registration and cloud and its shade of sequential images
The coarse extraction of region super-pixel grade.Image registration containing cloud and cloud and cloud shadow extraction result are as shown in Figure 3, wherein first row is
The original image sequence containing cloud of the affine transformation of simulation, second is classified as the image sequence after registration, and third is classified as cloud and its shade
Super-pixel grade testing result.
Step 4 specific implementation of embodiment includes the following steps (overstriking representing matrix):
Step 4.1: assumed according to background low-rank and foreground structures are sparse it is assumed that building matrix decomposition objective function:
minL,S||L||*+λψ(S)
Wherein, L is low-rank matrix, indicates background, and S is sparse matrix, indicates prospect, and D indicates input matrix.λ indicates flat
Weigh coefficient, and τ indicates affine transformation, i.e.,It indicates Vec () is indicated
Image is changed into column vector.N and K respectively indicates total image number and every image super-pixel block number,Indicate j-th of image
In i-th of super-pixel block weight,Indicate i-th of super-pixel in j-th of image.||·||*For nuclear norm, representing matrix spy
The sum of value indicative;||·||∞For Infinite Norm, the maximum value in one group of numerical value is indicated;
Step 4.2: the objective function of step 4.1 is converted using non-tight augmented vector approach:
Wherein, Y indicates Lagrange multiplier, and μ indicates a positive scalar,For square of Frobenius norm,
The quadratic sum of representing matrix all elements.Solve includes executing following steps:
Step 4.2.1: initialization L=0, S=0, Y=D/J (D), μ=2/ | | D | |2, ρ=1.5, whereinThe line number of m representing matrix D.The initialization of τ: film sequence is registrated in advance a certain
It opens on image, those skilled in the art can be voluntarily registrated using any quick method for registering in advance, and precision is without reaching sub-
Pixel-level.The present invention is recommended to use robust multi-scale estimation method to calculate the initial value of τ, and sequential images is registrated in advance
Intermediate image;
Step 4.2.2: calculating L, converts step 4.1 formula to about the subproblem for solving L:
According to lemma, for the optimization problem about nuclear norm of following form:
It is X=Θ that it, which is solved,α(Z), wherein ΘαSingular value after indicating singular value decomposition is shunk:
Θα=U ∑αVT
Wherein U, ∑, V respectively indicate the left eigenvector matrix after Singular Value Decomposition Using, characteristic value diagonal matrix and the right side
Eigenvectors matrix.Singular value contraction, which refers to, shrinks singular value using α as threshold value, first by characteristic value diagonal matrix sigma diagonal line
Element subtracts threshold alpha, then newer diagonal entry and 0 size then retain if more than 0;If being set to 0 less than 0.Cause
This, the solution of L are as follows:
Step 4.2.3: calculating S, converts step 4.1 formula to about the subproblem for solving S:
It converts the optimization problem to and each super-pixel block is individually solved:
Here s is indicatedH is indicatedω is indicatedWherein, λ indicates coefficient of balance,For
The corresponding weight of each super-pixel,It is set asHereIt indicates pseudo- Infinite Norm, indicates one group
The maximum value after several maximum values, i.e. cutoff value are eliminated in variable, it is 0.1 that the present invention, which recommends truncation ratio,.To each super
The solution of block of pixels is resolved using the library SPAMS.
Step 4.2.4: τ is updated;
Calculating more new capital about τ is to carry out on 2d, it is necessary first to convert sequential images matrix inverse operation
For two dimensional image.By nonlinear constraint conditionLinearly turn to HereTable
Show Jacobian matrix.In order to make it easy to understand, assuming affine variation are as follows:
Jacobian matrix is the matrix that first-order partial derivative is arranged in a certain way, thereforeIt can indicate are as follows:
The update of τ value can indicate are as follows:
Here Δ τ can be by carrying out least square method rapid solving to every image.
Step 4.2.5: Y is updated;
Gain is carried out according to the residual values of matrix decomposition in each iterative process, at the kth iteration, the formula of update is such as
Under:
Step 4.2.6: μ is updated;
It is updated in each iterative process with certain incremental value, at the kth iteration, the update of μ value is as follows:
μk+1=ρ μk
Wherein ρ is the scalar greater than 1, indicates the amplitude that μ updates, and those skilled in the art can be adjusted accordingly.The present invention
The value of recommendation is 1.5.
Step 4.2.7: iterated conditional is calculated, is terminated if meeting termination condition, otherwise returns to step 4.2.2, Zhi Daoshou
Termination iteration is held back, the image sequence containing cloud being registrated and corresponding cloud and its shadow region exposure mask are exported.
The condition setting of iteration ends isThreshold value can be adjusted voluntarily, and threshold value is got over
Minor matrix Decomposition Accuracy is higher, but time cost is higher;Threshold value is bigger, and matrix decomposition precision is lower, and time consumption is less.Cloud and
Its shadow mask is obtained by the direct binaryzation of sparse matrix S.According to the affine running parameter τ being calculated after iteration convergence to sequence
Column image and cloud exposure mask are corrected, and the image sequence containing cloud being registrated and corresponding cloud and its shadow region exposure mask are exported.
Step 5: building cloud sector information repairs objective function, using original Robust Principal Component Analysis method, to cloud sector and
Different scale factors is arranged in cloud-free area, completes to the information reparation of sequential images cloud sector and Pixel-level cloud and its shadow Detection.
Step 5.1, building cloud sector information repairs objective function:
minL,S||L||*+α||PΩ(S)||1+β||PΩ-(S)||1
Wherein, PΩIndicate cloud and cloud shadow mask, PΩIndicate that cloudless area mask, α, β indicate different scale factors,
Indicate sequence video conversion after the obtained registration of step 4 at input matrix, | | | |1For l1Norm indicates that all elements are absolute
The sum of value;
Step 5.2, using 5.1 objective function of non-tight method of Lagrange multipliers solution procedure, including following steps are executed:
Step 5.2.1: initialization L=0, S=0,ρ=1.6,β
=1, m representing matrixLine number.Wherein
Step 5.2.2: L is calculated:
Method for solving is identical as step 4.2.2;
Step 5.2.3: S is calculated:
According to lemma, for following form about | | | |1The optimization problem of norm:
It is Soft that it, which is solved,λ(W).Here Soft indicates soft-threshold contraction operator, is defined as follows:
Softλ(W)=max (W- λ, 0)+min (W+ λ, 0)
Therefore cloud sector and cloud-free area solve the formula in step 5.2.3 using soft-threshold contraction operator respectively
,
Step 5.2.4: Y is updated;
Gain is carried out according to the residual values of matrix decomposition in each iterative process, at the kth iteration, the formula of update is such as
Under:
Step 5.2.5: μ is updated;
It is updated in each iterative process with certain incremental value, at the kth iteration, the update of μ value is as follows:
μk+1=ρ μk
Wherein ρ is the scalar greater than 1, indicates the amplitude that μ updates, and those skilled in the art can be adjusted accordingly.The present invention
The value of recommendation is 1.6.
Step 5.2.6: stopping criterion for iteration is calculated, is terminated if meeting termination condition, otherwise returns to step 5.2.2, directly
Iteration, output matrix L and S are terminated to convergence.
The condition setting of iteration ends isThreshold value can be adjusted voluntarily, the smaller square of threshold value
Battle array Decomposition Accuracy is higher, but time cost is higher;Threshold value is bigger, and matrix decomposition precision is lower, and time consumption is less.
Step 6: rule of thumb property threshold value progress binaryzation obtains cloud and its shadow Detection knot to the matrix S obtained to step 5
Fruit, the threshold value that the present invention recommends is set as the standard deviation of each column (corresponding to each image) of matrix S, as shown in figure 4, first
It is classified as true image sequence containing cloud, second is classified as cloud as a result, third is classified as cloud and its shadow detection result.The two-value that will be obtained
The matrix L that change matrix and step 5 obtain carries out inversely processing according to step 3, and output sequence image goes cloud result and cloud and shade to examine
Survey result.Fig. 5 shows cloud sector information reparation and cloud and its shadow Detection detail view, wherein first is classified as original image containing cloud
Detail view, second is classified as cloud and its shadow detection result detail view, and third is classified as cloud sector information and repairs detail view.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this
The limitation of invention patent protection range, those skilled in the art under the inspiration of the present invention, are not departing from power of the present invention
Benefit requires to make replacement or deformation under protected ambit, fall within the scope of protection of the present invention, this hair
It is bright range is claimed to be determined by the appended claims.
Claims (6)
1. a kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence, which comprises the following steps:
Step 1, data preparation, the multidate image containing cloud for choosing certain amount and the same interest region of covering constitute image sequence
Column;
Step 2, super-pixel segmentation is carried out to every image, and using the super-pixel block after segmentation as cloud and the place of shade coarse extraction
Manage unit;
Step 3, every image store is normalized into 0 to 1 at column vector and by pixel value, and image sequence is arranged by acquisition time
At a matrix, the pixel number of individual image of behavior is classified as total image number;
Step 4, using low-rank model and structural sparse model respectively to background (ideal cloudless earth's surface information) and prospect (cloud
And its shade) modeled, affine Transform Model is introduced, matrix decomposition objective function is constructed, using Robust Principal Component Analysis frame
Frame carries out solution separation to the matrix of the input composition of image sequence containing cloud, completes registration and cloud and its shadow region of sequential images
The coarse extraction of super-pixel grade;
Step 5, objective function is repaired in building, and using original Robust Principal Component Analysis method, cloud sector and non-cloud sector are arranged not
Same scale factor is completed to the information reparation of sequential images cloud sector and Pixel-level cloud and its shadow Detection;
Step 6, the image containing cloud step 5 obtained repairs result and cloud and shadow detection result according to the inversely processing of step 3, defeated
Sequential images go cloud result and cloud and its shadow detection result out.
2. a kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence according to claim 1, feature
Be: the image quantity chosen in step 1 no less than guarantee each pixel counterpart orientation set at least observed it is primary most
Few image number, i.e., the cloudless pixel of all images containing cloud can at least be combined into a cloudless image.
3. a kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence according to claim 1, feature
It is: uses SLIC partitioning algorithm by every Image Segmentation at several super-pixel block in step 2.
4. a kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence according to claim 1 or 2 or 3,
It is characterized by: the specific implementation of step 4 includes following sub-step,
Step 4.1, assumed according to background low-rank and foreground structures be sparse it is assumed that building matrix decomposition objective function:
minL,S||L||*+λψ(S)
Wherein, L is low-rank matrix, indicates background, and S is sparse matrix, indicates prospect, and D indicates input matrix;λ indicates balance system
Number, τ indicate affine transformation, i.e.,It indicates Vec () is indicated shadow
As changing into column vector;N and K respectively indicates total image number and every image super-pixel block number,Indicate each super-pixel block
Weight,Indicate each super-pixel;||·||*For nuclear norm, the sum of representing matrix characteristic value;||·||∞For Infinite Norm, table
Show the maximum value in one group of numerical value;
Step 4.2, the objective function of step 4.1 is converted using non-tight augmented vector approach:
Wherein, Y indicates Lagrange multiplier, and μ indicates a positive scalar,For square of Frobenius norm, square is indicated
The quadratic sum of battle array all pixels;Solve includes executing following steps:
Step 4.2.1 initializes L, S, τ, Y, μ, ρ;
Step 4.2.2 calculates L, converts step 4.1 formula to about the subproblem for solving L:
Step 4.2.3 calculates S, converts step 4.1 formula to about the subproblem for solving S:
Step 4.2.4 updates τ=τ+Δ τ;
Step 4.2.5 updates
Step 4.2.6 updates μ=ρ μ;
Step 4.2.7 calculates iterated conditional, terminates if meeting termination condition, otherwise return to step 4.2.2, until convergence is whole
Only iteration exports the image sequence containing cloud being registrated and corresponding cloud and its shadow region exposure mask.
5. a kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence according to claim 4, feature
Be: the specific implementation of step 5 includes following sub-step,
Step 5.1, building cloud sector information repairs objective function:
minL,S||L||*+α||PΩ(S)||1+β||PΩ-(S)||1
Wherein, PΩIndicate cloud and cloud shadow mask, PΩIndicate that cloudless area mask, α, β indicate different scale factors,It indicates
After the registration that step 4 obtains sequence video conversion at input matrix, | | | |1For l1Norm indicates all elements absolute value
With;
Step 5.2, using 5.1 objective function of non-tight method of Lagrange multipliers solution procedure, including following steps are executed:
Step 5.2.1 initializes L, S, Y, μ, ρ;
Step 5.2.2 calculates L,
Step 5.2.3 calculates S,
Step 5.2.4 updates
Step 5.2.5 updates μ=ρ μ;
Step 5.2.6 calculates stopping criterion for iteration, terminates if meeting termination condition, otherwise return to step 5.2.2, Zhi Daoshou
Hold back termination iteration, output matrix L and S.
6. a kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence according to right 5, it is characterised in that:
The specific implementation of step 6 includes following sub-step,
Step 6.1, the s-matrix obtained according to step 5, given threshold T carry out binaryzation to S to obtain Pixel-level cloud and shade
Testing result, wherein threshold value T is the standard deviation of each column (corresponding to each image) of matrix S;
Step 6.2, the low-rank matrix L that step 5 obtains and the testing result that step 6.1 obtains are subjected to inversely processing according to step 3,
Output finally eliminates spissatus and cloud shade sequential images and corresponding cloud and cloud shadow Detection product.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810908851.4A CN109063663B (en) | 2018-08-10 | 2018-08-10 | Thick cloud detection and removal method for time sequence remote sensing image from coarse to fine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810908851.4A CN109063663B (en) | 2018-08-10 | 2018-08-10 | Thick cloud detection and removal method for time sequence remote sensing image from coarse to fine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109063663A true CN109063663A (en) | 2018-12-21 |
CN109063663B CN109063663B (en) | 2021-08-03 |
Family
ID=64683232
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810908851.4A Active CN109063663B (en) | 2018-08-10 | 2018-08-10 | Thick cloud detection and removal method for time sequence remote sensing image from coarse to fine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109063663B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111798461A (en) * | 2020-06-19 | 2020-10-20 | 武汉大学 | Pixel-level remote sensing image cloud area detection method for guiding deep learning by coarse-grained label |
CN112070787A (en) * | 2020-08-10 | 2020-12-11 | 武汉大学 | Aviation three-dimensional point cloud plane segmentation method based on opponent reasoning theory |
CN112419380A (en) * | 2020-11-25 | 2021-02-26 | 湖北工业大学 | High-precision registration method for static orbit satellite sequence images based on cloud mask |
CN112529789A (en) * | 2020-11-13 | 2021-03-19 | 北京航空航天大学 | Weak supervision method for removing shadow of urban visible light remote sensing image |
CN113838065A (en) * | 2021-09-23 | 2021-12-24 | 江苏天汇空间信息研究院有限公司 | Automatic cloud removing method based on image markers |
CN114049562A (en) * | 2021-11-30 | 2022-02-15 | 中国科学院地理科学与资源研究所 | Method for fusing and correcting land cover data |
CN114298945A (en) * | 2022-01-05 | 2022-04-08 | 首都师范大学 | Optical remote sensing image thick cloud removing method based on virtual image construction |
CN117408949A (en) * | 2023-09-20 | 2024-01-16 | 宁波大学 | Cloud and cloud shadow detection method and device for seasonal dynamic threshold |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110149117A1 (en) * | 2009-12-22 | 2011-06-23 | Canon Kabushiki Kaisha | Method of foreground/background separation |
CN103035001A (en) * | 2012-12-06 | 2013-04-10 | 中国科学院自动化研究所 | Foundation automatic cloud detection method based on superpixel division |
CN105956557A (en) * | 2016-05-04 | 2016-09-21 | 长江水利委员会长江科学院 | Object-oriented timing sequence remote sensing image cloud coverage area automatic detection method |
CN107704807A (en) * | 2017-09-05 | 2018-02-16 | 北京航空航天大学 | A kind of dynamic monitoring method based on multi-source remote sensing sequential images |
-
2018
- 2018-08-10 CN CN201810908851.4A patent/CN109063663B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110149117A1 (en) * | 2009-12-22 | 2011-06-23 | Canon Kabushiki Kaisha | Method of foreground/background separation |
CN103035001A (en) * | 2012-12-06 | 2013-04-10 | 中国科学院自动化研究所 | Foundation automatic cloud detection method based on superpixel division |
CN105956557A (en) * | 2016-05-04 | 2016-09-21 | 长江水利委员会长江科学院 | Object-oriented timing sequence remote sensing image cloud coverage area automatic detection method |
CN107704807A (en) * | 2017-09-05 | 2018-02-16 | 北京航空航天大学 | A kind of dynamic monitoring method based on multi-source remote sensing sequential images |
Non-Patent Citations (2)
Title |
---|
WU Z 等: "Superpixel-based unsupervised change detection using multi-dimensional change vector analysis and SVM-based classification", 《ISPRS ANN. PHOTOGRAMM. REMOTE SENS》 * |
黄超: "基于超像素分割与低秩表示的高光谱图像去噪算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111798461A (en) * | 2020-06-19 | 2020-10-20 | 武汉大学 | Pixel-level remote sensing image cloud area detection method for guiding deep learning by coarse-grained label |
CN111798461B (en) * | 2020-06-19 | 2022-04-01 | 武汉大学 | Pixel-level remote sensing image cloud area detection method for guiding deep learning by coarse-grained label |
CN112070787A (en) * | 2020-08-10 | 2020-12-11 | 武汉大学 | Aviation three-dimensional point cloud plane segmentation method based on opponent reasoning theory |
CN112529789A (en) * | 2020-11-13 | 2021-03-19 | 北京航空航天大学 | Weak supervision method for removing shadow of urban visible light remote sensing image |
CN112529789B (en) * | 2020-11-13 | 2022-08-19 | 北京航空航天大学 | Weak supervision method for removing shadow of urban visible light remote sensing image |
CN112419380A (en) * | 2020-11-25 | 2021-02-26 | 湖北工业大学 | High-precision registration method for static orbit satellite sequence images based on cloud mask |
CN112419380B (en) * | 2020-11-25 | 2023-08-15 | 湖北工业大学 | Cloud mask-based high-precision registration method for stationary orbit satellite sequence images |
CN113838065A (en) * | 2021-09-23 | 2021-12-24 | 江苏天汇空间信息研究院有限公司 | Automatic cloud removing method based on image markers |
CN114049562A (en) * | 2021-11-30 | 2022-02-15 | 中国科学院地理科学与资源研究所 | Method for fusing and correcting land cover data |
CN114049562B (en) * | 2021-11-30 | 2022-06-21 | 中国科学院地理科学与资源研究所 | Method for fusing and correcting land cover data |
CN114298945A (en) * | 2022-01-05 | 2022-04-08 | 首都师范大学 | Optical remote sensing image thick cloud removing method based on virtual image construction |
CN117408949A (en) * | 2023-09-20 | 2024-01-16 | 宁波大学 | Cloud and cloud shadow detection method and device for seasonal dynamic threshold |
Also Published As
Publication number | Publication date |
---|---|
CN109063663B (en) | 2021-08-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109063663A (en) | A kind of spissatus detection of timing remote sensing image and minimizing technology by slightly to essence | |
Wang et al. | Spatio-temporal fusion for daily Sentinel-2 images | |
CN108629835B (en) | Indoor reconstruction method and system based on hyperspectral, true color image and point cloud complementation | |
CN111369487B (en) | Hyperspectral and multispectral image fusion method, system and medium | |
KR101857472B1 (en) | A method of calibrating a camera and a system therefor | |
US8699790B2 (en) | Method for pan-sharpening panchromatic and multispectral images using wavelet dictionaries | |
CN103020912B (en) | The remote sensing image restored method of a kind of combination wave band cluster and sparse expression | |
CN111161199B (en) | Space spectrum fusion hyperspectral image mixed pixel low-rank sparse decomposition method | |
CN107316309B (en) | Hyperspectral image saliency target detection method based on matrix decomposition | |
CN109685743A (en) | Image mixed noise removing method based on noise learning neural network model | |
CN102156969A (en) | Processing method for correcting deviation of image | |
CN108171249B (en) | RGBD data-based local descriptor learning method | |
CN112929626B (en) | Three-dimensional information extraction method based on smartphone image | |
CN111932452A (en) | Infrared image convolution neural network super-resolution method based on visible image enhancement | |
CN109859153B (en) | Multispectral image fusion method based on adaptive spectrum-spatial gradient sparse regularization | |
CN112634184A (en) | RGB image spectral reflectivity reconstruction method based on fusion convolution neural network | |
CN112529777A (en) | Image super-resolution analysis method based on multi-mode learning convolution sparse coding network | |
CN113935917A (en) | Optical remote sensing image thin cloud removing method based on cloud picture operation and multi-scale generation countermeasure network | |
CN113869262A (en) | Prediction method and device of land information of target area based on Unet model | |
CN110738693A (en) | ground-based imaging radar multi-angle image registration method | |
CN110335197A (en) | Based on the intrinsic demosaicing methods of non local statistics | |
Shi et al. | Radial lens distortion correction by adding a weight layer with inverted foveal models to convolutional neural networks | |
CN110334581A (en) | A kind of multi-source Remote Sensing Images change detecting method | |
CN111815620B (en) | Fabric defect detection method based on convolution characteristic and low-rank representation | |
Liu et al. | A general relative radiometric correction method for vignetting and chromatic aberration of multiple CCDs: Take the Chinese series of Gaofen satellite Level-0 images for example |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20181221 Assignee: SUPERMAP SOFTWARE Co.,Ltd. Assignor: WUHAN University Contract record no.: X2022420000064 Denomination of invention: A coarse to fine thick cloud detection and removal method for time series remote sensing images Granted publication date: 20210803 License type: Common License Record date: 20220720 |
|
EE01 | Entry into force of recordation of patent licensing contract |