CN105631903B - The remote sensing images Clean water withdraw method, apparatus of algorithm is cut based on RGBW feature space figures - Google Patents
The remote sensing images Clean water withdraw method, apparatus of algorithm is cut based on RGBW feature space figures Download PDFInfo
- Publication number
- CN105631903B CN105631903B CN201510989654.6A CN201510989654A CN105631903B CN 105631903 B CN105631903 B CN 105631903B CN 201510989654 A CN201510989654 A CN 201510989654A CN 105631903 B CN105631903 B CN 105631903B
- Authority
- CN
- China
- Prior art keywords
- water body
- remote sensing
- pixel
- sensing images
- water
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of remote sensing images Clean water withdraw method that algorithm is cut based on RGBW feature space figures.Water body index method and image classification method are combined by the present invention, the preliminary extraction of water body is carried out first with water body index method, then the initial target of algorithm is cut using the water body main body tentatively extracted as figure, in the RGBW feature spaces constructed by the present invention, the fine extraction of algorithm progress water body is cut using figure；It is adaptive to carrying out process of refinement at water boundary finally further according to water body index and color characteristic information.The invention also discloses a kind of remote sensing images Clean water withdraw device that algorithm is cut based on RGBW feature space figures.The present invention can realize that the Water-Body Information complicated to surrounding enviroment carries out accurately extraction and good to the extraction effect of water boundary；In addition, need not artificially interfere during the Clean water withdraw of the present invention, full-automatic, full intellectuality can be achieved, be more beneficial for the automatic batchization processing of a large amount of remote sensing image datas.
Description
Technical field
The present invention relates to a kind of remote sensing images Clean water withdraw method, apparatus that algorithm is cut based on RGBW feature space figures, category
In the technical field that remote sensing technology and hydrotechnics intersect.
Background technology
Remote sensing technology has that observation scope is wide, contain much information, obtains that information is fast, the update cycle is short, uses manpower and material resources sparingly
Advantage, in remote sensing images water body information automation extraction technology have become Investigation of water resources, utilization, water body Monitoring on Dynamic Change,
The powerful measure of planning for water resources development assessment etc. research.The reflection to solar radiation such as water body and land, vegetation, absorption and transmission
Characteristic is different, and the difference on remote sensing image is more apparent, and this characteristic is advantageous to us and Water-Body Information in remote sensing images is carried out
Extraction.
The water body information method of remote sensing images has a lot, and conventional is divided into two classes：One kind is to be based on Spectrum Analysis
Method, another kind of is the method based on image classification.
, will based on the method for Spectrum Analysis by analyzing the difference of atural object in remote sensing images in the reflectance signature of different-waveband
The water body wave band maximum with the difference of other Object reflections increases this species diversity by modes such as mathematical operations, so as to logical
The methods of crossing Threshold Analysis extracts water-outlet body.In the method based on Spectrum Analysis, according to water body in Multi-spectral Remote Sensing Data
Spectral characteristic, conventional is water body index method：Water body index method using water body visible light wave range absorb less, reflectivity it is low,
A large amount of transmissions, almost absorb whole near-infrareds, the characteristic of short-wave infrared ripple suppresses vegetation information extraction water body.For example,
McFeeters et al. (1996) proposes normalization difference water body index NDWI (Normalize Difference Water
Index) (referring to [McFeeters S K.The use of normalized difference water index (NDWI)
in the delineation of open water features[J].International Journal of Remote
Sensing, 1996,17 (7):1425-1432.])；Xu Han autumns et al. (2005) proposition can preferably extract urban water-body information
Improvement normalization difference water body index (MNDWI) (referring to [and Xu Han autumns using it is improved normalization difference water body index
(MNDWI) research [J] remote sensing journals of Water-Body Information, 2005,9 (5) are extracted:589-595.])；Shen accounts for cutting edge of a knife or a sword et al. (2013) and existed
On the basis of normalizing difference aqua index (NDWI) calculating, propose using Gaussian normalization water body index (GNDWI) extraction river
The model of water body is (referring to [Shen Zhanfeng, Xia Liegang, Li Junli, waiting to realize remote sensing image river using Gaussian normalization water body index
Accurate extraction [J] Journal of Image and Graphics of stream, 2013,18 (4):421-428.])；Zhou Yi et al. (2014) is by returning
One changes the green wave band amendment in difference water body index NDWI, it is proposed that the pseudo- normalization difference water body independent of middle infrared band
Index FNDWI (False NDWI) is (referring to [Zhou Yi, Xie Guanglei, Wang Shixin, waiting using pseudo- normalization difference water body index extraction
Tiny river information [J] the Earth Information Science journals in cities and towns periphery, 2014,16 (1):102-107.])；Ding Jianli (2015) etc.
People proposes the decision tree water that single band threshold method is combined with the shade water body index SWI (Shadow Water Index) built
Body information extracting method is (referring to [Chen Wenqian, Ding Jianli, Li Yanhua, waiting Clean water withdraw sides of the based on domestic GF-1 remote sensing images
Method [J] resources sciences, 2015,37 (6):1166-1172.]).
Method based on image classification is mainly the reflectivity wave spectrum showed by analyzing different atural objects on different-waveband
Feature, water body is extracted come the method classified to water body and non-water body using the method for supervised classification or unsupervised classification.Point
Class method is generally up to criterion to classify to the pixel in image with feature variance within clusters minimum, inter-class variance.It is conventional
Maximum likelihood method, ISODATA clustering procedures (Iterative Self-organizing Data are had based on image classification method
Analysis Technique Algorithm), decision tree etc..
The water boundary pixel such as lake, reservoir, river is mainly made up of beach and water body, due to mankind's activity, life of swimming
The reasons such as thing growth, seasonal water-level fluctuation change, water boundary areas case complexity is various, increases the difficulty of Clean water withdraw
Degree.In same width image, different water body units or respective physicochemical characteristic or because surrounding enviroment influence difference can caused by its
Imaging features be able to might not keep in balance, and the water quality situation, sediment charge etc. inside water body with coastal waters differs greatly, single
It is pure preferable for the main part effect for extracting water body using water body index method or the method for image classification, but water body index be present
The difficult determination of threshold value, water boundary treatment effect is undesirable, shoal is divided into water body by mistake, misses the problems such as extracting.
The content of the invention
The technical problems to be solved by the invention are to overcome prior art insufficient, there is provided one kind is based on RGBW feature spaces
Figure cuts the remote sensing images Clean water withdraw method, apparatus of algorithm, and water body is believed under complex background in achievable high-resolution remote sensing image
The accurate extraction of breath, the identification to water boundary is more accurate, and whole extraction process can realize it is full intellectualized, it is not necessary to it is artificial
Intervene.
It is of the invention specifically to solve above-mentioned technical problem using following technical scheme：
The remote sensing images Clean water withdraw method of algorithm is cut based on RGBW feature space figures, is comprised the following steps：
Step 1, the preliminary extraction that using water body index method remote sensing images are carried out with water body, and in the water body tentatively to extract
Largest connected region as water body main body；
Step 2, each pixel for remote sensing images, with its tri- face of R, G, B in standard pseudo color coding hologram remote sensing images
RGBW feature of the normalization difference water body index NDWI values of chrominance channel value and the pixel as the pixel, it is distant so as to construct
Feel the RGBW feature spaces of image；
Step 3, in RGBW feature spaces, do not made with the boundary rectangle inner and outer portions of the water body main body obtained by step 1
For initial target and background, cut algorithm using figure and image segmentation is carried out to remote sensing images；And repeatedly changed according to what image was split
For result, the pixel of target will be judged in each iteration as water body pixel, so as to extract finer water body main body；
Step 4, the borderline region for the water body main body extracted to step 3 carry out process of refinement, specific as follows：To described
Each pixel in borderline region, judges whether its NDWI value is more than NDWI threshold values, and judges standard pseudo color coding hologram remote sensing images
Change to whether the b channel values of the pixel after CIE Lab color spaces are less than 0；The pixel of above-mentioned two condition will be met simultaneously
It is determined as water body pixel.
The remote sensing images Clean water withdraw device of algorithm is cut based on RGBW feature space figures, the device includes：
Preliminary water body main body extraction unit, for preliminary extraction of the water body index method to remote sensing images progress water body, and with
Largest connected region in the water body tentatively extracted is as water body main body；
RGBW feature extraction units, for each pixel for remote sensing images, with it in standard pseudo color coding hologram remote sensing images
In tri- color channel values of R, G, B and the pixel normalization difference water body index NDWI values it is special as the RGBW of the pixel
Sign, so as to construct the RGBW feature spaces of remote sensing images；
Image segmentation unit, for the water body in RGBW feature spaces, exported with preliminary water body main body extraction unit
The boundary rectangle inner and outer portions of main body as initial target and background, do not cut algorithm using figure and carry out image to remote sensing images
Segmentation；And the successive ignition result split according to image, the pixel of target will be judged in each iteration as water body pixel,
So as to extract finer water body main body；
Process of refinement unit, the borderline region of the water body main body for being extracted to image segmentation unit become more meticulous
Processing, it is specific as follows：To each pixel in the borderline region, judge whether its NDWI value is more than NDWI threshold values, and judge
Whether the b channel values of the pixel are less than 0 after standard pseudo color coding hologram remote sensing images are changed to CIE Lab color spaces；It will expire simultaneously
The pixel of sufficient above-mentioned two condition is determined as water body pixel.
Compared with prior art, the invention has the advantages that：
The present invention, which can realize, to the complicated Water-Body Information of surrounding enviroment accurately extract and carry water boundary
Take and work well；In addition, need not artificially interfere during the Clean water withdraw of the present invention, full-automatic, full intellectuality can be achieved, more
Be advantageous to the automatic batchization processing of a large amount of remote sensing image datas.
Brief description of the drawings
Fig. 1 is the structural representation for the remote sensing images Clean water withdraw device that the present invention cuts algorithm based on RGBW feature space figures
Figure；
Fig. 2 is Clean water withdraw effect contrast figure.
Embodiment
The present invention is in view of the shortcomings of the prior art, propose a kind of remote sensing images water body that algorithm is cut based on RGBW feature space figures
Extracting method, its general principle are：Water body index method and image classification method are combined, water body is carried out first with water body index method
Preliminary extraction, the initial target of (Graph Cut) algorithm is then cut using the water body main body tentatively extracted as figure, in the present invention
In constructed RGBW feature spaces, the fine extraction of algorithm progress water body is cut using figure；Finally further according to water body index and face
Color characteristic information, it is adaptive to carrying out process of refinement at water boundary.
For the ease of public understanding, technical scheme is described in detail below in conjunction with the accompanying drawings：
The present invention cuts the remote sensing images Clean water withdraw method of algorithm based on RGBW feature space figures, specifically includes following step
Suddenly:
Step 1, the preliminary extraction that using water body index method remote sensing images are carried out with water body, and in the water body tentatively to extract
Largest connected region as water body main body.
The preliminary extraction of water body, such as institute in background technology are carried out in this step using existing Different Waters index method
Normalization difference water body index NDWI, Gaussian normalization water body index GNDWI, the improvement normalization difference water body index mentioned
MNDWI, the pseudo- normalization difference water body index FNDWI independent of middle infrared band, the shade of single band threshold method and structure
Decision tree water body information method that water body index SWI is combined etc..
Remote sensing images generally include the remotely-sensed data of multiple different-wavebands, with the light more of high score No.1 satellite 16m resolution ratio
Exemplified by spectrum width covering WFV (wide field of view) camera, it provides 0.45 μm~0.89 μm totally four wave band (blue light ripple
0.45 μm~0.52 μm of section, 0.52 μm~0.59 μm of green light band, 0.63 μm~0.69 μm of red spectral band, near infrared band 0.77
μm~0.89 μm) remotely-sensed data.Multiple-spatial resolution remote sensing image represents remote sensing image picture element with DN (Digital Number) value
Brightness value, record the gray value of atural object.DN values are the integer values of no unit, the radiometric resolution of its value size and sensor,
Thing emissivity, atmospheric transmittance and scattered power etc. are relevant.
By classics normalization difference aqua index NDWI exemplified by, its Water indices model with atural object each wave band DN values
Based on build, its calculation formula is：
Wherein DNgreenRepresent green light band, DNNIRRepresent near infrared band.In the multi-spectrum remote sensing image of high score No.1
It corresponds to second and the 4th wave band respectively.
NDWI water body indexes suppress the information such as land vehicles and protrude Water-Body Information, and formula (1) is unified to NDWI numbers
Value is stretched, and can make different sensors, the image of different image-forming conditions is also obtained with comparable, close statistics spy
The NDWI image wave bands of property, are easy to subsequently establish unified information extraction model.
According to the NDWI values calculated, preliminary water body rapidly can be carried out to remote sensing images using simple threshold method and carried
Take.It is of the invention from the water body tentatively extracted in order to avoid the interference of zonule similar with water body in image and picture noise
It is middle to choose largest connected region as water body main body.The probable ranges of water body main body can be so determined roughly.
Step 2, each pixel for remote sensing images, with its tri- face of R, G, B in standard pseudo color coding hologram remote sensing images
RGBW feature of the normalization difference water body index NDWI values of chrominance channel value and the pixel as the pixel, it is distant so as to construct
Feel the RGBW feature spaces of image.
So-called standard pseudo color coding hologram remote sensing images, it is with radiant correction (Radiometric Calibration), air school
Just (FLAASH atmospheric Correction), ortho-rectification (RPC Orthorectification) etc. are to remote sensing number
According to pre-processing, the near infrared band, red spectral band and green light band of remotely-sensed data are then corresponded into red, green, blue three respectively
Passage synthetic standards pseudo color coding hologram digital picture.Standard pseudo color coding hologram remote sensing images can fully show the difference of various atural object image features
Not, different atural object characteristics can be distinguished well, interpret type of ground objects；But the image has some bright with the image that general camera is shot
Aobvious color distortion, such as on standard pseudo color coding hologram remote sensing images, dense vegetation is shown as cerise, clear water body is shown as
Navy blue, exposed soil are shown as dark gray etc..Each pixel in standard pseudo color coding hologram remote sensing images is in RGB color
It is respectively provided with tri- color channel values of R, G, B.For each pixel of remote sensing images, its NDWI value can be calculated with formula (1).Will
Tri- Color Channels of RGB of standard pseudo color coding hologram remote sensing images are combined with NDWI values, you can structure combines remotely-sensed data colour face
Four dimensional feature spaces of color characteristic and NDWI models, we term it RGBW features (W is NDWI water body indexes value).RGBW features
The color characteristic for considering water body and other atural objects in remotely-sensed data coloured image is distinguished, it is also considered that traditional water body index
Influence.
Step 3, in RGBW feature spaces, do not made with the boundary rectangle inner and outer portions of the water body main body obtained by step 1
For initial target and background, cut algorithm using figure and image segmentation is carried out to remote sensing images；And repeatedly changed according to what image was split
For result, the pixel of target will be judged in each iteration as water body pixel, so as to extract finer water body main body.
In recent years, the image segmentation algorithm based on markov random file (MRF) model be widely used in various targets with
The complicated extracting target from images of situations such as background border is unintelligible, and prospect background is similar, and obtain good effect.It is based on
The dividing method of MRF models is established on the basis of MRF models and bayesian theory (Bayesian Theory), according to statistics
Optiaml ciriterion in decision-making and estimation theory determines the object function of image segmentation problem, asks for meeting condition using optimized algorithm
MRF maximum possible distribution, so as to divide the image into problem be converted to solve MRF distribution optimization problem.
Image segmentation can regard image pixel mark problem as, and the image of N number of pixel can use array X={ X1, X2...,
XNRepresent, each pixel XiThere is a mark xi∈ { 0,1 } (0 represents background, and 1 represents prospect).Knot using X as MRF
The Gibbs energy functions put and define figure are：
Wherein, C be neighborhood territory pixel set number, x={ x1,x2,…,xNBe each pixel mark, z={ z1,
z2,…,zNRepresent pixel brightness value, θ is MRF model parameters, by the characteristic value of display foreground and background pixel (in the present invention
In be RGBW value) distribution determine.
In formula,It is that pixel is classified as target for region energy item (Regional Term)
Or the punishment of background, it is defined as pixel i and is divided into mark xiProbability negative logarithm.
Ri(x, θ, z)=- log P (zi|xi) (3)
Region energy item R can directly be schemed by gray scale or gauss hybrid models GMM (Gaussian Mixture Model) is calculated
Obtain.
For boundary energy item (Boundary Term), boundary energy item embodies adjacent
The discontinuous punishment of domain pixel i and j, in RGBW models, the similitude of two pixels is weighed using Euclidean distance：
BI, j (i ≠ j)(x, z)=λ B<I, j>* δ (i, j) (4)
Wherein,B<I, j>=exp (- β | | zi-zj||2), λ=50, β are by pair between image pixel
Determined than degree, β=1/2 (zi-zj)2。
After the Gibbs energy functions for defining MRF, the problem of image is split, which is converted into, seeks the Gibbs energy functions of figure most
Smallization x*The problem of：
x*=arg minxE (x, θ, z) (5)
Boykov et al. is (referring to [Boykov Y, Kolmogorov V.An experimental comparison of
min-cut/max-flow algorithms for energy minimization in vision[J].Pattern
Analysis and Machine Intelligence,IEEE Transactions on,2004,26(9):1124-
1137.], iterative overall situation MRF gibbs theoretical based on max-flow min-cut (max-flow/min-cut)) is proposed
(Gibbs) figure of energy minimization cuts (Graph Cut) algorithm, obtains faster splitting speed and good image segmentation knot
Fruit.
GMM model is established to image can be using multidimensional characteristic come zoning energy term R, the more accurate segmentation knot of acquisition
Fruit.Each GMM model has n Gaussian component, three parameters：Each weight π of Gaussian component, the average of each Gaussian component to
Measure μ and covariance matrix δ.The GMM of description prospect and the GMM of description background parameter pass through known to K-Means algorithms cluster
Foreground pixel and background pixel determine.
GMM Gaussian component vector representation is k={ k1,k2,..,kn, each pixel in image obeys Gauss point
Cloth：
P(zi|xi, θ, ki)=N (z, θ；μ(xi, ki), δ (xi, ki)) (6)
(3) formula turns to after introducing GMM model：
The probability that the pixel is belonging respectively to target GMM and background GMM is calculated by the RGBW value of each pixel, is calculated
Region energy item R.
Figure cuts (Graph Cut) algorithm and establishes graph model G=(V, E) to image first, and wherein V is image pixel node
Set, node vi∈ V represent image a pixel, E be image side set, (vi,vj) ∈ V expression connecting nodes viAnd vj's
Dissimilar degree.In image segmentation, the weights on side represent the measure of dissimilarity of two pixels, by pixel in the bright of RGBW spaces
Angle value structure calculates edge smoothing energy term B as dissimilarity measurementI, j (i ≠ j)For the weights on side.Then two are added on figure G
Terminal vertex S nodes (representing prospect) and T nodes (representing background), and all newly-built a line of all nodes on V is connected respectively
It is connected on S and T, the probability that the weights on side are belonged to prospect or background by the node determines, zoning energy term RiFor side
Weights.
When the weights on all sides determine, figure structure is completed.Assuming that the label (label of each pixel) of entire image is L
={ l1,l2,…ln, wherein liFor 0 (background) or 1 (target), then pass through max-flow min-cut (max-flow/min-
Cut) to find figure minimal cut L (set on weights and minimum side) in the boundary of target and background occurs for algorithm, now
The ENERGY E (L) of figure=R (L)+B (L) is minimum, and the side simultaneously switched off can be partitioned from target and background, realize target just
With the separation of background.
Some researchers are improved it on the basis of figure cuts algorithm, for example, the figure segmentation method based on tensor space
(Malcolm J,Rathi Y,Tannenbaum A.A graph cut approach to image segmentation in
tensor space.In:Proceedings of the IEEE Conference on Computer Vision and
Pattern Recognition.Minnesota,USA:IEEE, 2007.1-8), algorithm (Price B are cut based on geodesic figure
L,Morse B,Cohen S.Geodesic graph cut for interactive image segmentation.In:
Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition.San Francisco,USA:IEEE, 2010.3161-3168) and based on sequence protection mobile optimization side
Method (Liu X Q, Veksler O, Samarabandu J.Order preserving moves for graph-cut-based
optimization.IEEE Transactions on Pattern Analysis and Machine Intelligence,
2010,32(7):1182-1196), preferable image segmentation is achieved.The present invention cuts algorithm using various existing figures
Water body in remote sensing images is finely extracted.
However, either classical figure cuts the various improvement figures of algorithm still hereafter and cuts algorithm, it is known that foreground pixel and
Background pixel is required to be specified by way of man-machine interactively, in addition, being also required to select by manual type after successive ignition
Select out optimal iteration result, it is difficult to realize the automatic business processing of remotely-sensed data.On the one hand, initial prospect, background is manually specified
And the optimal iteration result of selection, be required to the professional ability that height relies on user, if selected prospect, background or it is optimal repeatedly
It is undesirable for result, then it can directly affect the extraction result of water body；On the other hand, for the processing of high-volume remotely-sensed data, lead to
The mode for crossing manpower intervention carries out data processing, it is clear that is unpractical.In order to realize automatic accurate image segmentation, this
Invention does not provide initial prospect, background by the way of man-machine interactively, but the water body main body obtained by step 1 is obtained
Its boundary rectangle, and the initial foreground target of algorithm is cut using the rectangle frame inside points as figure, and using outer rectangular frame part as
Initial background.Certainly, also other modes can be taken to provide initial prospect, background automatically, for example, it is also possible to by step 1 institute
The initial foreground target of algorithm is cut in the central area of obtained water body main body as figure, by water body main body boundary rectangle outer frame part
As initial background.
In addition, after default n iteration (currently preferred iterations is 5~8), the present invention is more using this
The statistics of secondary iteration result adaptively judges object pixel and background pixel, is specially：Split according to image more
Secondary iteration result, the pixel of target will be judged in each iteration as water body pixel, the pixel for not meeting the condition is
Background pixel.Improved more than passing through, you can save existing figure and cut manpower intervention required for algorithm, so as to realize the automatic of algorithm
Change operation.
Step 4, the borderline region for the water body main body extracted to step 3 carry out process of refinement, specific as follows：To described
Each pixel in borderline region, judges whether its NDWI value is more than NDWI threshold values, and judges standard pseudo color coding hologram remote sensing images
Change to whether the b channel values of the pixel after CIE Lab color spaces are less than 0；The pixel of above-mentioned two condition will be met simultaneously
It is determined as water body pixel.
Relatively accurate complete water body can have been obtained after successive ignition segmentation by step 3.In order to reduce artificial farming
The mistake extraction of area, shallow water beach and planktonic organism overlay area, obtains more accurate water body profile, and the present invention is further dividing
Process of refinement is done at the water boundary for cutting to obtain.The main purpose on process of refinement border is more accurately to extract water boundary
Coastal waters part in region, the actual conditions of the different zones of water boundary are different, it is necessary to different situations according to region
Do different disposal.The borderline region of water body main body can use following division methods：Take in remote sensing images with the water obtained by step 3
As borderline region, the predetermined width can be according to being actually needed for the region in the range of predetermined width centered on phosphor bodies edge
Flexibly choose, preferably 10 pixel wides.
The NDWI water body index values in the regions such as the artificial dam of water boundary, artificial buffer zone are smaller, and coastal waters
NDWI water body index values it is larger, therefore the present invention pass through the method to the NDWI water body index value given thresholds in borderline region
Remove the boundary members such as artificial dam, artificial buffer strip.The selection of NDWI threshold values can rule of thumb be preset, and the present invention is preferably
Determined using following methods：After the water body main body that segmentation obtains is removed into borderline region part, the subregional NDWI of remainder is sought
Water body index value averageAnd setFor NDWI water body index threshold values, screen out and water body spy is not met in borderline region
The pixel of sign, that is, the pixel that wherein NDWI values are less than or equal to the NDWI water body indexes threshold value is screened out, so as to go well
Except the borderline manually region such as dam, artificial buffer zone, while the coastal waters part of aperture Heshui body characteristicses.
The mud silted land band of water boundary, the NDWI water body indexes value and water body for propagating the regions such as area, shoal artificially
The NDWI water body indexes value of main part very close to by can not go well to the method for NDWI water body index value given thresholds
Remove.But found according to substantial amounts of data analysis, in the standard false color image of multiband remote sensing Data Synthesis, mud silted land
Band, artificial farming area, the b passage pixel value of the region in Lab color spaces such as shoal are typically larger than 0, and coastal waters area
The b passage pixel values of domain pixel are typically smaller than 0.Therefore the present invention makes a return journey except the mud in borderline region deposits according to this characteristic
The regions such as area, artificial farming area, shallow water beach, i.e., change standard pseudo color coding hologram remote sensing images to Lab color spaces, sentence
Whether the b channel values of each pixel are less than 0 in disconnected borderline region, in this way, then retain the pixel as water body pixel, otherwise, will
It is removed as mud silted land band, artificial farming area, shoal etc..
Fig. 1 shows that the present invention cuts the basic knot of the remote sensing images Clean water withdraw device of algorithm based on RGBW feature space figures
Structure, as illustrated, the device includes：
Preliminary water body main body extraction unit, for preliminary extraction of the water body index method to remote sensing images progress water body, and with
Largest connected region in the water body tentatively extracted is as water body main body；
RGBW feature extraction units, for each pixel for remote sensing images, with it in standard pseudo color coding hologram remote sensing images
In tri- color channel values of R, G, B and the pixel normalization difference water body index NDWI values it is special as the RGBW of the pixel
Sign, so as to construct the RGBW feature spaces of remote sensing images；
Image segmentation unit, for the water body in RGBW feature spaces, exported with preliminary water body main body extraction unit
The boundary rectangle inner and outer portions of main body as initial target and background, do not cut algorithm using figure and carry out image to remote sensing images
Segmentation；And the successive ignition result split according to image, the pixel of target will be judged in each iteration as water body pixel,
So as to extract finer water body main body；
Process of refinement unit, the borderline region of the water body main body for being extracted to image segmentation unit become more meticulous
Processing, it is specific as follows：To each pixel in the borderline region, judge whether its NDWI value is more than NDWI threshold values, and judge
Whether the b channel values of the pixel are less than 0 after standard pseudo color coding hologram remote sensing images are changed to CIE Lab color spaces；It will expire simultaneously
The pixel of sufficient above-mentioned two condition is determined as water body pixel.
To verify effectiveness of the invention, using the inventive method and other existing methods to stone beam River Reservoir remote sensing figure
As carrying out water body information and being contrasted, data source applies the high score No.1 resource of center offer in China Satecom's resource
The remotely-sensed data image of the 16m resolution ratio of satellite WFV cameras shooting, the remote sensing image of 2014 is chosen as experimental data.Its
In several existing methods be specially：Original graph cuts algorithm (document [Rother C, Kolmogorov V, Blake A.Grabcut:
Interactive foreground extraction using iterated graph cuts[J].ACM
Transactions on Graphics(TOG),2004,23(3):309-314.] disclosed in method, with " Graph Cut "
Simplify and represent), normalization water body index method (document [McFeeters S K.The use of normalized
difference water index(NDWI)in the delineation of open water features[J]
.International Journal of Remote Sensing, 1996,17 (7):1425-1432.] in disclosed side
Method, with " NDWI " simplify represent), (document [Chen Wenqian, Ding Jianli, Li Yanhua, waits to be based on domestic GF-1 remote sensing to shadow index method
Clean water withdraw method [J] resources sciences of image, 2015,37 (6):1166-1172.] disclosed in method, with " SWI " simplify
Represent.
Fig. 2 shows the contrast situation of Clean water withdraw effect, and the image on the wherein row left side of top one is 03 month 2014 08
The standard pseudo color coding hologram digital picture of the Jiangsu Province's stone beam river base area synthesized after day high score No.1 WFV2 remote sensing images pretreatment,
It can be seen that reservoir border is based on the buffer zone manually built and artificial dam, but reservoir left margin exist it is artificial
The paddy field of cultivation, shoal etc., mud alluvial on the upper left corner border of reservoir be present.Image among the row of top one is people
The stone beam River Reservoir standard water body figure of work visual interpretation.Image on the right of the row of top one is to be utilized in RGB color feature space
Figure cuts the extraction result that (Graph Cut) algorithm obtains, it can be seen that and it is simple based on color characteristic, water body feature is not considered
It is more serious that figure cuts mistake extraction phenomenon of the algorithm when extracting water body.The image on the following row left side is normalization water body index
NDWI gray level images take the Clean water withdraw result of the binaryzation of threshold value 199, it can be seen that the paddy field of water boundary, shoal etc., which exist, misses
Extract phenomenon.Image among a following row takes the Clean water withdraw of the binaryzation of threshold value 198 for shade water body index SWI gray level images
As a result, it can be seen that the paddy field of artificial farming, shoal etc. can cause reservoir there is also extracting phenomenon by mistake using global threshold
Surrounding enviroment are also extracted as water body by mistake.Image on the right of a following row is the extraction result of the inventive method, with standard water body
Figure contrast it can be seen that the inventive method extraction work well and the treatment effect of borderline region is also preferable.
In order to carry out accurate quantitatively evaluating, the evaluation index recall ratio, precision ratio, the F that are further introduced into information retrieval refer to
It is several that above-mentioned extraction result is evaluated.Wherein recall ratio (Recall) is as shown in formula (8), and precision ratio (Precision) is such as
Shown in formula (9),
R=(the real water body number of pixels of extraction/manually mark water body number of pixels) * 100% (8)
P=(the water body number of pixels of the real water body number of pixels of extraction/algorithm segmentation extraction) * 100% (9)
F evaluation numbers are the weighted harmonic mean values of precision ratio and recall ratio, herein using F evaluation numbers as the synthesis assessed
Target, its specific formula are：
Extract evaluation of result table such as table 1：
Table 1 extracts evaluation of result table
Resultant effect of the inventive method in four kinds of methods is best as can be seen from Table 1, it is often more important that, the present invention is to water
The extraction at body edge is more accurate, and can realize full-automatic execution, is easy to large-scale application.
Claims (10)
1. the remote sensing images Clean water withdraw method of algorithm is cut based on RGBW feature space figures, it is characterised in that comprise the following steps：
Step 1, the preliminary extraction that using water body index method remote sensing images are carried out with water body, and in the water body tentatively to extract most
Big connected region is as water body main body；
Step 2, each pixel for remote sensing images, led to its tri- color of R, G, B in standard pseudo color coding hologram remote sensing images
RGBW feature of the normalization difference water body index NDWI values of road value and the pixel as the pixel, so as to construct remote sensing figure
The RGBW feature spaces of picture；
Step 3, in RGBW feature spaces, using step 1 obtained by water body main body boundary rectangle inner and outer portions not as just
The target and background of beginning, cut algorithm using the figure based on Markov random field model and image segmentation is carried out to remote sensing images；And
The successive ignition result split according to image, the pixel of target will be judged in each iteration as water body pixel, so as to carry
Take out finer water body main body；
Step 4, the borderline region for the water body main body extracted to step 3 carry out process of refinement, specific as follows：To the border
Each pixel in region, judges whether its NDWI value is more than NDWI threshold values, and judges to change standard pseudo color coding hologram remote sensing images
Whether the b channel values of the pixel are less than 0 after to CIE Lab color spaces；The pixel for meeting above-mentioned two condition simultaneously is judged
For water body pixel.
2. remote sensing images Clean water withdraw method as claimed in claim 1, it is characterised in that the NDWI threshold values are by the following method
It is determined that：The NDWI averages of each pixel in region in water body main body obtained by calculation procedure 3 in addition to borderline region, and with this
0.8 times of NDWI averages is used as NDWI threshold values.
3. remote sensing images Clean water withdraw method as claimed in claim 1, it is characterised in that cut algorithm using figure and remote sensing images are entered
Iterations when row image is split is 5~8.
4. remote sensing images Clean water withdraw method as claimed in claim 1, it is characterised in that the borderline region tool of the water body main body
Body is：The region in the range of predetermined width centered on water body body rim.
5. remote sensing images Clean water withdraw method as claimed in claim 4, it is characterised in that the predetermined width is 10 pixels.
6. the remote sensing images Clean water withdraw device of algorithm is cut based on RGBW feature space figures, it is characterised in that the device includes：
Remote sensing images are carried out the preliminary extraction of water body for water body index method by preliminary water body main body extraction unit, and with preliminary
Largest connected region in the water body of extraction is as water body main body；
RGBW feature extraction units, for each pixel for remote sensing images, with it in standard pseudo color coding hologram remote sensing images
R, RGBW feature of the normalization difference water body index NDWI values of tri- color channel values of G, B and the pixel as the pixel,
So as to construct the RGBW feature spaces of remote sensing images；
Image segmentation unit, for the water body main body in RGBW feature spaces, exported with preliminary water body main body extraction unit
Boundary rectangle inner and outer portions not as initial target and background, cut calculation using the figure based on Markov random field model
Method carries out image segmentation to remote sensing images；And the successive ignition result split according to image, it will be judged to mesh in each iteration
Target pixel is as water body pixel, so as to extract finer water body main body；
Process of refinement unit, the borderline region of the water body main body for being extracted to image segmentation unit carry out the place that becomes more meticulous
Reason, it is specific as follows：To each pixel in the borderline region, judge whether its NDWI value is more than NDWI threshold values, and judge by
Whether the b channel values of the pixel are less than 0 after standard pseudo color coding hologram remote sensing images are changed to CIE Lab color spaces；It will meet simultaneously
The pixel of above-mentioned two condition is determined as water body pixel.
7. remote sensing images Clean water withdraw device as claimed in claim 6, it is characterised in that the NDWI threshold values are by the following method
It is determined that：Calculate each pixel in the region in the water body main body that preliminary water body main body extraction unit is exported in addition to borderline region
NDWI averages, and it is used as NDWI threshold values using 0.8 times of the NDWI averages.
8. remote sensing images Clean water withdraw device as claimed in claim 6, it is characterised in that cut algorithm using figure and remote sensing images are entered
Iterations when row image is split is 5~8.
9. remote sensing images Clean water withdraw device as claimed in claim 6, it is characterised in that the borderline region tool of the water body main body
Body is：The region in the range of predetermined width centered on water body body rim.
10. remote sensing images Clean water withdraw device as claimed in claim 9, it is characterised in that the predetermined width is 10 pixels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510989654.6A CN105631903B (en) | 2015-12-24 | 2015-12-24 | The remote sensing images Clean water withdraw method, apparatus of algorithm is cut based on RGBW feature space figures |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510989654.6A CN105631903B (en) | 2015-12-24 | 2015-12-24 | The remote sensing images Clean water withdraw method, apparatus of algorithm is cut based on RGBW feature space figures |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105631903A CN105631903A (en) | 2016-06-01 |
| CN105631903B true CN105631903B (en) | 2018-01-16 |
Family
ID=56046790
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510989654.6A Expired - Fee Related CN105631903B (en) | 2015-12-24 | 2015-12-24 | The remote sensing images Clean water withdraw method, apparatus of algorithm is cut based on RGBW feature space figures |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105631903B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106097331B (en) * | 2016-06-07 | 2018-11-13 | 浙江工业大学 | A kind of container localization method based on lockhole identification |
| CN106650810B (en) * | 2016-12-26 | 2019-09-03 | 河海大学 | Reservoir water body classification method and device based on spectral properties information and spatial information |
| CN107271367B (en) * | 2017-05-05 | 2019-06-28 | 北京师范大学 | A kind of identifying water boy method and device |
| CN108985307B (en) * | 2018-07-16 | 2021-08-10 | 中国科学院东北地理与农业生态研究所 | Water body extraction method and system based on remote sensing image |
| CN109447916A (en) * | 2018-10-30 | 2019-03-08 | 环境保护部华南环境科学研究所 | A kind of water quality quantitative approach based on high-resolution remote sensing image |
| CN109886193B (en) * | 2019-02-21 | 2020-11-20 | 电子科技大学 | Volume cloud detection method and device and computer readable storage medium |
| CN110189328B (en) * | 2019-06-11 | 2021-02-23 | 北华航天工业学院 | Satellite remote sensing image processing system and processing method thereof |
| CN110390277B (en) * | 2019-07-04 | 2021-07-06 | 中科卫星应用德清研究院 | Complex underlying surface water body identification method and black and odorous water body prediction method |
| CN110443821B (en) * | 2019-07-09 | 2023-05-05 | 成都理工大学 | Image-based water body detection method and device |
| CN111862138A (en) * | 2020-07-21 | 2020-10-30 | 北京吉威空间信息股份有限公司 | Semi-automatic water body extraction method for remote sensing image |
| CN112634295B (en) * | 2020-12-29 | 2022-05-20 | 中国人民解放军国防科技大学 | Star sensor star point segmentation method based on double gradient thresholds |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102054274A (en) * | 2010-12-01 | 2011-05-11 | 南京大学 | Method for full automatic extraction of water remote sensing information in coastal zone |
-
2015
- 2015-12-24 CN CN201510989654.6A patent/CN105631903B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102054274A (en) * | 2010-12-01 | 2011-05-11 | 南京大学 | Method for full automatic extraction of water remote sensing information in coastal zone |
Non-Patent Citations (4)
| Title |
|---|
| Developing Superfine Water Index (SWI) for Global Water Cover Mapping Using MODIS Data;Ram C. Sharma et al;《remote Sensing》;20151021;全文 * |
| Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery;HANQIU XU;《International Journal of Remote Sensing》;20060720;第27卷(第14期);全文 * |
| Water Feature Extraction and Change Detection Using Multitemporal Landsat Imagery;Komeil Rokni et al;《Remote Sensing》;20140505;全文 * |
| 图谱迭代反馈的自适应水体信息提取方法;胡晓东 等;《测绘学报》;20111031;第40卷(第5期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105631903A (en) | 2016-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105631903B (en) | The remote sensing images Clean water withdraw method, apparatus of algorithm is cut based on RGBW feature space figures | |
| CN108549891B (en) | Multi-scale diffusion well-marked target detection method based on background Yu target priori | |
| WO2020258899A1 (en) | Mountain landscape architecture extraction method and equipment based on high-resolution remote sensing image | |
| Cheng et al. | SeNet: Structured edge network for sea–land segmentation | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| CN108573276A (en) | A kind of change detecting method based on high-resolution remote sensing image | |
| CN109448016A (en) | It is a kind of based on object-oriented and its be subordinate to rule remote sensing image tidal saltmarsh method | |
| CN111797712B (en) | Remote sensing image cloud and cloud shadow detection method based on multi-scale feature fusion network | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| Liu et al. | A new segmentation method for very high resolution imagery using spectral and morphological information | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN111783523B (en) | Remote sensing image rotating target detection method | |
| CN108596103A (en) | High resolution ratio satellite remote-sensing image building extracting method based on optimal spectrum Index selection | |
| Aytekın et al. | Unsupervised building detection in complex urban environments from multispectral satellite imagery | |
| CN109558806A (en) | The detection method and system of high score Remote Sensing Imagery Change | |
| CN107123150A (en) | The method of global color Contrast Detection and segmentation notable figure | |
| CN104318051B (en) | The rule-based remote sensing of Water-Body Information on a large scale automatic extracting system and method | |
| CN106650812A (en) | City water body extraction method for satellite remote sensing image | |
| Sun et al. | Object-oriented land cover classification using HJ-1 remote sensing imagery | |
| CN107564016B (en) | A kind of the Multi-Band Remote Sensing Images segmentation and labeling method of integrally object light spectrum information | |
| CN106960182A (en) | A kind of pedestrian integrated based on multiple features recognition methods again | |
| Topouzelis et al. | Oil spill detection: SAR multiscale segmentation and object features evaluation | |
| CN108710862A (en) | A kind of high-resolution remote sensing image Clean water withdraw method | |
| CN107992856A (en) | High score remote sensing building effects detection method under City scenarios | |
| Liu et al. | Object-oriented detection of building shadow in TripleSat-2 remote sensing imagery |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180116 Termination date: 20181224 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |