CN103971364B - Remote sensing image variation detecting method on basis of weighted Gabor wavelet characteristics and two-stage clusters - Google Patents
Remote sensing image variation detecting method on basis of weighted Gabor wavelet characteristics and two-stage clusters Download PDFInfo
- Publication number
- CN103971364B CN103971364B CN201410134520.1A CN201410134520A CN103971364B CN 103971364 B CN103971364 B CN 103971364B CN 201410134520 A CN201410134520 A CN 201410134520A CN 103971364 B CN103971364 B CN 103971364B
- Authority
- CN
- China
- Prior art keywords
- change
- remote sensing
- class
- gabor wavelet
- gabor
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a remote sensing image variation detecting method on the basis of weighted Gabor wavelet characteristics and two-stage clusters. The processed objects include optical remote sensing images and SAR (synthetic aperture radar) images and the remote sensing image variation detecting method includes (1) generating difference images according to remote sensing image types; (2) subjecting the difference images to Gabor wavelet transform; (3) extracting multiscale and multidirectional characteristics of the Gabor wavelet transform of the difference images; (4) designing the weighting coefficient and acquiring the weighted Gabor wavelet characteristics; (5) clustering the weight Gabor wavelet characteristics by means of the two-stage cluster strategy; (6) acquiring variation detecting results. By the remote sensing image variation detecting method, loss of marginal information is reduced, stronger, weak and slight varying areas can be detected at the same time, the total mistake pixel number is decreased, more detail information is reserved and variation results can be effectively extracted.
Description
Technical field
The invention belongs to digital image processing techniques field, relate generally to Multitemporal Remote Sensing Images change-detection research side
To, the specifically method for detecting change of remote sensing image based on weighting gabor wavelet character and two-stage cluster, process object is same
When contain multidate remote sensing image and synthetic aperture radar (synthetic aperture radar, sar) image.Should
Method can be applicable to forest inventory investigation, land cover pattern/land utilization dynamic monitors, environmental hazard estimation, urban planning and cloth
Office, Hitting Effect Evaluation, particularly in many practical problems of the earth observation such as Natural Disaster monitoring and evaluation.
Background technology
Remote sensing technology is to perceive target reflection or the electromagnetic wave itself radiating, visible ray, infrared ray etc. from remote, to mesh
The technology that detected and identified of marking, has the broad range of data data that can obtain, and the speed obtaining information is fast, cycle is short and information
The features such as measure big.According to the difference of acquisition modes, optical remote sensing and microwave remote sensing two class can be classified as.Wherein, optical remote sensing
Corresponding equipment is simple, acquired image space geometry high resolution, and image is easy to interpret, but can only use on daytime, is subject to
Weather condition impact is more serious;Because microwave has good penetration capacity, microwave remote sensing (sar) can round-the-clock, round-the-clock right
Ground imaging, but coherent imaging characteristic makes its image there is the property taken advantage of coherent speckle noise.Both respectively have feature, are complementary to one another, in state
More and more important effect is just being played in people's economy and national defense construction.
With the continuous development of sensor technology, space technology and computer technology, numerous machines/space remote sensing platform puts into
Operation, new, the high resolution sensor being carried can carry out to earth surface uninterruptedly, persistently observe, obtaining substantial amounts of
Wide cut, high-definition remote sensing data.But in the face of growing acquisition capability, therewith unbefitting be to remote sensing image processing with
The research and development of Interpretation Technology relatively lag behind it is impossible to meet the urgent needss of practical application.Wherein, Multitemporal Remote Sensing Images change-detection
As one of the basis processing with interpretation and key technology, refer to two width or many by obtaining to the same area, different times
Width remote sensing images are compared analysis, and then obtain the change of interested atural object, scene or target according to difference between image
Change information, is limited by mankind's activity aggravation and natural disaster takes place frequently, and its research and development is increasingly paid attention to by people, is increasingly becoming distant
The study hotspot in sense field.
In recent years, Chinese scholars propose much effectively change detecting method, and summing up to be divided into has supervision to become
Change detection and unsupervised change-detection.It is limited to have ground real change classification sample needed for supervision change-detection to be difficult to obtain, mesh
Front work focuses primarily upon unsupervised change detection class, and it is roughly divided into: (1) multi-temporal remote sensing figure based on Cluster Distribution Divergence
As change-detection;(2) the Multitemporal Remote Sensing Images change-detection based on differential image analysis;And (3) melted based on Markov
The Multitemporal Remote Sensing Images change-detection closed.Particularly the research to Equations of The Second Kind algorithm is the most universal, and core concept is to change
Test problems are considered as the binary classification/segmentation problem of image, can be subdivided into cluster analyses, intelligent optimization, Threshold segmentation, limited again
The strategies such as mixed model, markov random file, active profile and level set.Wherein, cluster analyses simply, are effectively subject to by it
To generally approving.
T.celik is in document [t.celik, " unsupervised change detection in satellite
images using principal component analysis and k-means clustering,”ieee
Geoscience and remote sensing letters, vol.6, no.4, pp.772-776,2009.] in propose earliest
The method for detecting change of remote sensing image of cluster analyses.Specifically, first with principal component analysiss (principal
Component analysis, pca) extract the corresponding characteristic vector of each pixel, reuse k-means algorithm by characteristic vector
It is divided into two classes (change class and do not change class), obtain change-detection figure.Although the method is simply effective, fully not sharp
With the partial structurtes information of differential image, and it is vulnerable to noise jamming.For improving the effectiveness of detection further, m.volpi etc.
[m.volpi,d.tuia,g.camps-valls,and m.kanevski,“unsupervisedchange detection
with kernels,”ieee geoscience and remote sensing letters,vol.9,no.6,pp.1026-
1030,2012] the Nonlinear separability problem in original input space is converted into by linearly may be used in feature space by nuclear mapping
Divide problem, realize cluster using core k-means algorithm.Meanwhile, for reduction influence of noise, applicant [y.q.cheng, h.c.li,
t.celik,and f.zhang,“frft-based improved algorithm of unsupervised change
detection in sar images via pca and k-means clustering”,in proceeding of ieee
international geoscience and remote sensing symposium,melbourne,australia,
July2013:1952-1955.] and y.g.zheng etc. [y.g.zheng, x.r.zhang, b.hou, and g.c.liu,
“using combined difference image and k-means clustering for sar image change
detection,”ieee geoscience and remote sensing letters,vol.11,no.3,pp.691-695,
2014] it is utilized respectively low order time fractional order fourier conversion (fractional fourier transform, frft), ppb
(probabilistic patch-based) wave filter produces the differential image of noise suppressed, then carries out spatial domain cluster change inspection
Survey, satisfied testing result can be obtained, but all still do not take into full account the partial structurtes information of differential image.In addition with
The raising of remote sensing images resolution, change and the separability not changed between class reduce, and uncertain and ambiguity strengthens, and makes
Obtain the change information that existing change detecting method can not detect high-resolution remote sensing image effectively, there is higher flase drop and leakage
Inspection probability, becomes the bottleneck of restriction high-definition remote sensing change-detection application, is badly in need of more excellent change detecting method.
Content of the invention
The present invention utilizes gabor wavelet transformation and Fuzzy c-means Clustering (fuzzy c-means, fcm) to propose based on weighting
Gabor wavelet character and the Multitemporal Remote Sensing Images change-detection new method of two-stage cluster.The method is by gabor wavelet transformation
Multiple dimensioned and Orientation Features, can effectively portray the partial structurtes information of differential image, and two-stage cluster strategy can reduce and makes an uproar
Sound affects on testing result, and algorithm simply effectively disclosure satisfy that the needs of reality.
A kind of method for detecting change of remote sensing image based on weighting gabor wavelet character and two-stage cluster, walks including following
Rapid:
Step 1, using areal different time obtained two width sizes identical, mutual registration remote sensing images x0With
x1To produce differential image xd;
Step 2, to differential image xdCarry out gabor wavelet transformation;
Step 3, extracts the multiple dimensioned and Orientation Features of differential image gabor wavelet transformation;
Step 4, designs weights coefficient, obtains weighting gabor wavelet character;
Step 5, is clustered to weighting gabor wavelet character using the two-stage cluster strategy based on fcm algorithm, concrete point
For two steps it may be assumed that
The characteristic vector of a differential image pixel that previous step is obtained by () with fcm algorithm is divided into: changes class, does not change
Class and border class, obtain the cluster centre changing class and not changing class, use vcAnd vuRepresent change class ω respectivelycDo not change class
ωuCluster centre, ωbRepresent border class;
B () calculates border class ωbEach of characteristic vector and change class and the distance that do not change class center of birdsing of the same feather flock together, by it
It is assigned to the class minimum with cluster centre distance;Then the change class that obtains with (a) step and do not change class combination, obtains new
Change class and do not change class;
Step 6, change class location of pixels puts 1, does not change class location of pixels and sets to 0, and obtains final two-value change-detection knot
Really.
Wherein: the multiple dimensioned and Orientation Features of the extraction differential image gabor wavelet transformation described in step 3, specifically real
Shi Wei: from the gabor small echo of 5 yardsticks and 8 directions, v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 }, v represents gabor small echo
Scale parameter, u embodies the set direction of gabor small echo;Differential image is with the gabor small echo phase of 5 yardsticks and 8 directions
40 characteristic image g are obtained after convolutionV, u(z), wherein v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 };Use following symbol table for convenience
Show: jk(z)=gV, u(z), wherein k=u+8v, v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 }, the corresponding feature of each pixel
Vector can be expressed as v (i, j)=[j0(i, j) j1(i, j) ... j39(i, j)], wherein (i, j) is coordinate points, and v (i, j) is figure
As the pixel gabor wavelet-based attribute vector that upper coordinate is (i, j) represents.
In addition, the weights factor design to extracted gabor wavelet character described in step 4, it is embodied as: different scale
Frequency parameter can obtain for kv=kmax/fv;If v is ∈ { 0 ..., 4 }, this corresponding frequency parameter is k0, k1, k2, k3, k4, then
Calculate weights coefficient to be followed successively byAnd be summed to
Then it is normalized and be followed successively by Different chis
Degree v ∈ { 0 ..., 4 } weights coefficient set gradually for Carrying out upset to normalization weights coefficient according to yardstick order is exactly lower directive weights system of each yardstick
Number, and then obtain weighting gabor wavelet character.
For different image sources, differential image xdThere are different acquisition modes: for remote sensing image, disparity map
As xdFor input picture x0And x1The absolute value directly subtracting each other, xd=|x1-x0|;For synthetic aperture radar (sar) image, it is poor
Different image xdTake input picture x0And x1Make the absolute value of logarithm ratio,
Beneficial effect
1st, the present invention is by gabor wavelet transformation, by extract differential image multiple dimensioned and Orientation Features can comprehensively,
Effectively portray the partial structurtes information of differential image, and weight coefficient is designed to weaken noise and heterogeneous picture according to different frequency
Element impact.
2nd, the two-stage cluster strategy based on fcm algorithm proposed by the invention clusters to weighting gabor wavelet character, can
The cluster centre that separability preferably changes class and do not change class is obtained in first order cluster process, and then in second level cluster
During can more effectively distinguish the classification of edge class pixel.
3rd, the method can obtain stronger region of variation simultaneously complete, faint, small change region is detected, reduce
The loss of marginal information.False retrieval pixel count and missing inspection pixel count can be taken into account so that total erroneous pixel number reduces, remain more thin
Section information, efficiently extracts result of variations.
Brief description
The Multitemporal Remote Sensing Images change detecting method entirety frame based on weighting gabor wavelet character and two-stage cluster for the Fig. 1
Frame;
Feature extraction mode after Fig. 2 differential image gabor wavelet transformation;
First group of remote sensing image data collection bern that Fig. 3 emulation experiment uses, the ers2sar figure that (a) in April, 1999 obtains
Picture, the ers2sar image that (b) in May, 1999 obtains, (c) change-detection is with reference to figure;
Second group of remote sensing image data collection burn that Fig. 4 emulation experiment uses, the optics of (a) August in 1986 acquisition on the 5th is distant
Sense image, the remote sensing image of (b) August in 1992 acquisition on the 5th, (c) change-detection is with reference to figure;
The 3rd group of remote sensing image data collection lake that Fig. 5 emulation experiment uses, the optics of (a) August in 1986 acquisition on the 5th is distant
Sense image, the remote sensing image of (b) August in 1992 acquisition on the 5th;
Tri- groups of remote sensing image data collection of Fig. 6 corresponding differential image respectively: (a) bern data set differential image, (b) burn
Data set differential image, (c) lake data set differential image;
Fig. 7 present invention emulates the change-detection result of sar view data bern, the change of (a) control methods 1 with control methods
Change testing result, the change-detection result of (b) control methods 2, the change-detection result of (c) control methods 3, (d) present invention's
Change-detection result, (e) change-detection is with reference to figure;
Fig. 8 present invention emulates the change-detection result of remote sensing image data burn, (a) control methods with control methods
1 change-detection result, the change-detection result of (b) control methods 2, the change-detection result of (c) control methods 3, (d) this
Bright change-detection result, (e) change-detection is with reference to figure;
Fig. 9 present invention emulates the change-detection result of remote sensing image data lake, (a) control methods with control methods
1 change-detection result, the change-detection result of (b) control methods 2, the change-detection result of (c) control methods 3, (d) this
Bright change-detection result.
Specific embodiment
Embodiments of the invention are described with reference to the accompanying drawings.It should be appreciated that each embodiment of invention described herein is only
Merely to preferably explain principle and the concept of the present invention, rather than the present invention to be limited.After reading such description,
Those skilled in the art are easy to construct other modifications or replace, and such modification or replacement should be understood to fall into the present invention's
In scope.
Fig. 1 gives the flow process framework of the embodiment of the present invention, is embodied as comprising the steps of
(1) Multitemporal Remote Sensing Images are utilized to produce differential image.Two width sizes are h × w and mutually registering remote sensing images
x0={ x0(i, j) | 1≤i≤h, 1≤j≤w } and x1={ x1(i, j) | 1≤i≤h, 1≤j≤w }, they are in areal
Different time t0And t1The remotely-sensed data obtaining respectively.Type according to remote sensing images generates differential image xd.
For optical imagery (optical images), input picture x0And x1The absolute value directly subtracting each other is exactly disparity map
As xd,
xd=|x1-x0|
And for sar(synthetic aperture radar) image, differential image xdIt is input picture x0And x1Ratio
Value image, in order to improve the pixel value of low amplitude, we execute a logarithm operation by reduced value image,
Wherein log is natural logrithm operator, can make to change class ω in differential imagecDo not change class ωuDistribution is more equal
Even.Make x simultaneously0And x1Inherent residual the property taken advantage of (×) speckle interference be converted into additivity (+) interference.
(2) the differential image x first step being obtaineddCarry out gabor wavelet transformation.Two-dimentional gabor Wavelet Kernel Function (filtering
Device) form can be expressed as,
WhereinV represents the scale parameter of gabor small echo, and u embodies the set direction of gabor small echo, z=
One of (x, y) correspondence image pixel coordinate, | | | | represent European norm computing.
Gabor wavelet vectors parameter kV, uControl width, the wavelength of oscillating part and the direction of gaussian window, permissible
It is expressed as,
Wherein kv=kmax/fv(v=0,1 ..., v-1),(u=0,1 ..., u-1), generallyParameter kmaxIt is the peak frequency of gabor small echo, and f represents between adjacent gabor small echo at the interval of frequency domain
The factor.
Gabor wavelet transformation is carried out to gray level image i (x, y), is equivalent to image and gabor small echo makees convolution, can be with table
Show as follows,
gV, u(z)=i (z) * gV, u(z)
Wherein, one of z=(x, y) correspondence image pixel coordinate, " * " represents two-dimensional convolution operation symbol, and v represents
The scale parameter of gabor small echo, u embodies the set direction of gabor small echo, gV, uZ () is two-dimentional gabor wavelet function,
gV, uZ () is the gabor wavelet function convolution results of image and corresponding yardstick v, direction u, i.e. the gabor Wavelet representation for transient of image.
(3) the multiple dimensioned and Orientation Features of differential image gabor wavelet transformation are extracted.Generally, from 5 chis
Degree and the gabor small echo in 8 directions, v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 }.Differential image is with 5 yardsticks and 8 directions
40 characteristic image g are obtained after gabor small echo phase convolutionV, u(z), wherein v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 }.In order under
Literary composition is easy to use, and we are represented with following symbol, jk=gV, u(z), wherein k=u+8v, v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 },
Then 40 characteristic images can be expressed as j (z)={ jk(z), k=0,1 ..., 39 }.Each pixel of therefore image is relative
The characteristic vector composition answered can be expressed as v (i, j)=[j0(i, j) j1(i, j ... j39(i, j)], wherein (i, j) is coordinate points,
V (i, j) is that on image, coordinate is that the pixel gabor wavelet-based attribute vector of (i, j) represents, it is capable of the office of accurately picture engraving
Portion's structural information and multidirectional feature, have certain robustness to rotation, yardstick and illumination variation etc..Characteristic extraction procedure
As shown in Figure 2.
(4) weights are designed to extracted gabor wavelet character.Generally, from 5 yardsticks and 8 directions
Gabor small echo, v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 }.The gabor feature in lower 8 directions of same yardstick has identical weights,
Weights arrange only related to yardstick, independent of direction.The frequency parameter of different scale can obtain for kv=kmax/fv.If v is ∈
{ 0 ..., 4 }, this corresponding frequency parameter is k0, k1, k2, k3, k4, then calculate weights coefficient and be followed successively byAnd be summed to Then carry out normalizing
Change is followed successively by Different scale v ∈ 0 ...,
4 } weights coefficient set gradually forSo
Explanation readily appreciates.For the different directions setting identical weights under same yardstick, calculate weights normalization.To normalization
Weights according to yardstick order, to carry out upset be exactly lower directive weights coefficient of each yardstick, and then obtain and weight gabor
Wavelet character.Noise and the impact of heterogeneous pixel can be weakened by arranging weights, be conducive to complete extraction change information.
(5) using the two-stage based on fcm algorithm proposed by the present invention cluster strategy, the gabor feature of weighting is gathered
Class.
(5a) characteristic vector of the differential image pixel obtaining previous step with fcm algorithm is divided into: changes class, does not change
Class and border class, obtain the cluster centre changing class and not changing class, use vcAnd vuRepresent change class ω respectivelycDo not change class
ωuCluster centre, ωbRepresent border class.
(5b) calculate border class ωbEach of characteristic vector and change class and the distance that do not change class center of birdsing of the same feather flock together, general
It is assigned to the class minimum with cluster centre distance.Then the change class that obtains with (5a) step and do not change class combination, obtains
New change class and do not change class.
(6) obtain change-detection result.Two-value change-detection result is cm={ cm (i, j) | 1≤i≤h, 1≤j≤w },
Wherein " 1 " represents that corresponding picture position is change (that is, belonging to change class), and " 0 " represents corresponding figure
Image position is indeclinable (that is, belong to not changing class).Change class location of pixels is put 1 by us, does not change class picture
Plain position sets to 0, and obtains final change-detection result.
The effect of the present invention can be further illustrated by following experiment:
1st, emulation experiment condition
Method for detecting change of remote sensing image based on gabor small echo and two-stage cluster proposed by the present invention and control methods are all
It is to realize on identical experiment simulation platform.Experimental calculation machine is configured to amd athlon (tm) 7750 dual core processor, main
Frequency 2.70ghz, ram can use for 4.00gb(3.25gb), 32-bit operating system, software platform is matlab7.13.
2nd, emulation experiment data
The emulation experiment of the present invention uses real remotely-sensed data, including remote sensing image and sar image.
First group of remote sensing image data collection bern.It is two width sar view data and an amplitude variationization in Fig. 3 with reference to figure.This two
Width size identical sar image is by ers2(european remote sensing2) the sar sensor of satellite is in Switzerland bern
Obtain respectively in April, 1999 and in May, 1999 in the neighbourhood.Piece image is bern before in April, 1999 flood occurs
Earth's surface information in the neighbourhood, the second width image is the earth's surface information of the same area after in May, 1999 flood occurs, two width images
Size be 301 × 301, gray level be 256.3rd width image is to change with reference to figure, and it is to be believed by local real ground
Breath combines a width bianry image obtained from known view data, and in figure white pixel represents the situation of change on ground.
Second group of remote sensing image data collection burn, Fig. 4 are two width remote sensing images and an amplitude variationization with reference to figure.Front two width
The size of image is 200 × 200, respectively at August in 1986 5 days and August in 1992 5 days in lake tahoe, reno, nevada
Obtain.Because ground forest fire leads to a large amount of regions to change, the 3rd width image is change with reference to figure.
3rd group of remote sensing image data collection lake, Fig. 5 are two width remote sensing images.The size of two width images be 200 ×
200, respectively at August in 1986 5 days and August in 1992 5 days in lake tahoe, reno, nevada obtain, and mainly show lake
The change that dry up in face, land use and deforestation lead to.
3rd, evaluation of result index
Typically in the case of known to actual change, weighing change detection algorithm performance has following three index.
(1) false alarms (fa): error detection pixel count, represents that the pixel that reality does not change and detects into change is total
Number.
(2) missed alarms (ma): loss detection pixel count, represents that actual change detects into unchanged pixel
Sum.
(3) total errors (te): total erroneous pixel number, is error detection pixel count (fa) above and loss detection
The summation of pixel count (ma).This is to weigh the most important index of change detection algorithm, be the main judgement that proves that result is good and bad according to
According to.
4th, contrast simulation experiment
In order to effectiveness of the invention is described, the simulated effect of the present invention is compared by we with following three method:
Control methods 1, document [t.celik, " unsupervised change detection in satellite
images using principal component analysis and k-means clustering,”ieee
Geoscience and remote sensing letters, vol.6, no.4, pp.772-776,2009.] in utilize main one-tenth
Analyze (pca, principal component analysis) and k-means clustering algorithm realizes the change inspection of remote sensing images
Survey.The method extracts the corresponding characteristic vector of each pixel first with principal component analysiss, reuses k-means algorithm by feature
Vector is divided into two classes (change class and do not change class), obtains change-detection figure.
Control methods 2, document [y.q.cheng, h.c.li, t.celik, and f.zhang, " frft-based
improved algorithm of unsupervised change detection in sarimages via pca and
k-means clustering”,in proceeding of ieee international geoscience and remote
Sensing symposium, melbourne, australia, july2013:1952-1955.] it is the fraction secondary using low order
The method that rank fourier conversion proposes to document above improves.
Control methods 3, document [y.g.zheng, x.r.zhang, b.hou, and g.c.liu, " using combined
difference image and k-means clustering for sar image change detection,”ieee
Geoscience and remote sensing letters, vol.11, no.3, pp.691-695,2014] first have to many
Phase remote sensing images carry out ppb(probabilistic patch-based) filtering, then utilize two different operators difference
Produce differential image to be combined in conjunction with average and medium filtering, then target differential image is obtained by linear weighted function, finally use
K-means cluster obtains change-detection result.
5th, emulation experiment content
Experiment 1 calculating differential image
The emulation experiment of the present invention adopts three groups of remote sensing image data collection, including a pair of sar image and two pairs of optical remote sensings
Image.Type according to remote sensing images, according to different formula, generates differential image xd, respectively as shown in Fig. 6 (a)-(c).
Experiment 2sar view data bern
This emulation experiment uses first group of remote sensing image data collection bern.Fig. 7 is that the present invention is imitated with control methods
The change-detection result of true sar view data bern, wherein Fig. 7 (a) is the change-detection result of control methods 1, and Fig. 7 (b) is
The change-detection result of control methods 2, Fig. 7 (c) is the change-detection result of control methods 3, and Fig. 7 (d) is the change of the present invention
Testing result, Fig. 7 (e) is change-detection with reference to figure.
In addition to carry out qualitative contrast according to Visual Observations Observations, also carry out Quantitative Comparison.Table 1 lists emulation experiment output figure
The accuracy evaluation data of picture.
Table 1 present invention emulates bern data variation testing result accuracy evaluation data with control methods
Test 3 remote sensing image data burn
This emulation experiment uses first group of remote sensing image data collection burn.Fig. 8 is that the present invention is imitated with control methods
The change-detection result of true remote sensing image data burn, wherein Fig. 8 (a) is the change-detection result of control methods 1, Fig. 8
B () is the change-detection result of control methods 2, Fig. 8 (c) is the change-detection result of control methods 3, and Fig. 8 (d) is the present invention
Change-detection result, Fig. 8 (e) is change-detection with reference to figure.
In addition to carry out qualitative contrast according to Visual Observations Observations, also carry out Quantitative Comparison.Table 2 lists emulation experiment output figure
The accuracy evaluation data of picture.
Table 2 present invention emulates burn data variation testing result accuracy evaluation data with control methods
Test 4 remote sensing image data lake
This emulation experiment uses first group of remote sensing image data collection lake.Fig. 9 is that the present invention is imitated with control methods
The change-detection result of true remote sensing image data lake, wherein Fig. 9 (a) is the change-detection result of control methods 1, Fig. 9
B () is the change-detection result of control methods 2, Fig. 9 (c) is the change-detection result of control methods 3, and Fig. 9 (d) is the present invention
Change-detection result.Because remote sensing image lake lacks change-detection with reference to figure, qualitative comparative result will be provided.
6th, analysis of simulation result
Emulate change-detection result and the accuracy evaluation number of sar view data bern by analyzing the present invention and control methods
According to understanding, total erroneous pixel number of simulation result of the present invention is 296, has obtained preferable knot compared to other control methods
Really.Its error detection pixel count is 131 simultaneously, and loss detection pixel count 165 has admirably achieved a kind of compromise.
Emulate change-detection result and the precision of remote sensing image data burn by analyzing the present invention and control methods
Assessment data understands, total erroneous pixel number of simulation result of the present invention is 1108, control methods 2 for 1057, both
Method is all substantially better than other two kinds of control methods.And in simulation result of the present invention, loss detection pixel count is minimum is 79,
Faint region of variation can preferably be detected, be conducive to the holding of region of variation detailed information.
Emulate change-detection result and the image of remote sensing image data lake by analyzing the present invention and control methods
Data set information understands, this view data mainly shows the change that lake surface dries up, land use and deforestation lead to.Contrast
Method 3 can preferably detect that lake surface dries up the change leading to, but cannot effective detection land use and deforestation lead to
Change, lose faint region of variation in a large number.Simulation result of the present invention can not only be more fully compared to other control methods
Detection lake surface dries up the change leading to, and remains the change that land use and deforestation lead to simultaneously.
In sum, the present invention is simultaneously suitable for detecting multidate remote sensing image and sar image, little using gabor
Wave conversion can extract the multiple dimensioned Orientation Features fully and effectively portraying differential image partial structurtes information it is contemplated that avoiding
Noise and the impact of heterogeneous pixel, obtain weighting gabor wavelet character by design weights coefficient and represent.Utilize the present invention simultaneously
Proposed the cluster strategy of the two-stage based on fcm algorithm weighting gabor wavelet character is clustered, two-stage cluster can produce two
At a distance of farther cluster centre it is easier to class will be changed and do not change class separately, obtain more preferable Detection results.By qualitative and
Quantitative comparison, may certify that proposed by the invention becoming based on the remote sensing images of weighting gabor wavelet character and two-stage cluster
Change detection method, faint, small change region can be detected while complete acquisition stronger region of variation, reduce edge
The loss of information.And take into account false retrieval pixel count and missing inspection pixel count so that total erroneous pixel number reduces, remain more details
Information, efficiently extracts result of variations.
Those of ordinary skill in the art is obviously clear and understands, the above example that the inventive method is lifted is only used for
The inventive method is described, and is not limited to the inventive method.Although the present invention, the present invention are effectively described by embodiment
There is the spirit without deviating from the present invention for many changes.Without departing from the spirit and substance of the case in the method for the present invention, originally
Skilled person works as and can make various corresponding changes or deformation according to the inventive method, but these corresponding changes or deformation
Belong to the protection domain of the inventive method requirement.
Claims (3)
1. a kind of method for detecting change of remote sensing image based on weighting gabor wavelet character and two-stage cluster, comprises the steps:
Step 1, using areal different time obtained two width sizes identical, mutual registration remote sensing images x0And x1With
Produce differential image xd;
Step 2, to differential image xdCarry out gabor wavelet transformation;
Step 3, extracts the multiple dimensioned and Orientation Features of differential image gabor wavelet transformation;
Step 4, designs weights coefficient, obtains weighting gabor wavelet character;
Step 5, is clustered to weighting gabor wavelet character using the two-stage cluster strategy based on fcm algorithm, is specifically divided into two
Step it may be assumed that
The characteristic vector of a differential image pixel that previous step is obtained by () with fcm algorithm is divided into: change class, do not change class and
Border class, obtains the cluster centre changing class and not changing class, uses vcAnd vuRepresent change class ω respectivelycDo not change class ωu's
Cluster centre, ωbRepresent border class;
B () calculates border class ωbEach of characteristic vector and change class and the distance that do not change class center of birdsing of the same feather flock together, distributed
To the class minimum with cluster centre distance;Then the change class that obtains with (a) step and do not change class combination, obtains new change
Class and do not change class;
Step 6, change class location of pixels puts 1, does not change class location of pixels and sets to 0, obtains final two-value change-detection result;
Weights factor design wherein described in step 4, is embodied as: the frequency parameter of different scale can obtain for kv=
kmax/fv;If v is ∈ { 0 ..., 4 }, this corresponding frequency parameter is k0, k1, k2, k3, k4, then calculate weights coefficient and be followed successively byAnd be summed to
Then it is normalized and be followed successively by
The weights coefficient of different scale v ∈ { 0 ..., 4 } set gradually for Carrying out upset to normalization weights coefficient according to yardstick order is exactly each yardstick
Lower directive weights coefficient, and then obtain weighting gabor wavelet character;Parameter kmaxIt is the peak frequency of gabor small echo,
And f represents the interval factor in frequency domain between adjacent gabor small echo.
2. the Remote Sensing Imagery Change Detection side based on weighting gabor wavelet character and two-stage cluster according to claim 1
Method is it is characterised in that the multiple dimensioned and Orientation Features of extraction differential image gabor wavelet transformation described in step 3, specifically real
Shi Wei: from the gabor small echo of 5 yardsticks and 8 directions, v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 }, v represents gabor small echo
Scale parameter, u embodies the set direction of gabor small echo;Differential image is with the gabor small echo phase of 5 yardsticks and 8 directions
40 characteristic image g are obtained after convolutionv,u(z), wherein v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 };Use following symbol table for convenience
Show: jk(z)=gv,u(z), wherein k=u+8v, v ∈ { 0 ..., 4 }, u ∈ { 0 ..., 7 }, the corresponding feature of each pixel
Vector can be expressed as v (i, j)=[j0(i,j)j1(i,j)…j39(i, j)], wherein (i, j) is coordinate points, and v (i, j) is figure
As the pixel gabor wavelet-based attribute vector that upper coordinate is (i, j) represents.
3. the Remote Sensing Imagery Change Detection side based on weighting gabor wavelet character and two-stage cluster according to claim 1
Method is it is characterised in that described differential image xd: for remote sensing image, differential image xdFor input picture x0And x1Directly
The absolute value subtracting each other, xd=| x1-x0|;For synthetic aperture radar (sar) image, its differential image xdTake input picture x0And x1
Make the absolute value of logarithm ratio,
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410134520.1A CN103971364B (en) | 2014-04-04 | 2014-04-04 | Remote sensing image variation detecting method on basis of weighted Gabor wavelet characteristics and two-stage clusters |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410134520.1A CN103971364B (en) | 2014-04-04 | 2014-04-04 | Remote sensing image variation detecting method on basis of weighted Gabor wavelet characteristics and two-stage clusters |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103971364A CN103971364A (en) | 2014-08-06 |
CN103971364B true CN103971364B (en) | 2017-02-01 |
Family
ID=51240813
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410134520.1A Expired - Fee Related CN103971364B (en) | 2014-04-04 | 2014-04-04 | Remote sensing image variation detecting method on basis of weighted Gabor wavelet characteristics and two-stage clusters |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103971364B (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268875B (en) * | 2014-09-26 | 2017-02-15 | 西安电子科技大学 | SAR image registration method based on specific value correlation function |
CN104268879B (en) * | 2014-09-28 | 2017-03-01 | 民政部国家减灾中心 | Building quantity of goods produced Damage assessment method based on remote sensing multispectral image |
CN104700411A (en) * | 2015-03-15 | 2015-06-10 | 西安电子科技大学 | Sparse reconstruction-based dual-time phase remote-sensing image change detecting method |
CN105118065A (en) * | 2015-09-14 | 2015-12-02 | 中国民航大学 | Polari SAR (metric synthetic aperture radar) image variation detection method of wavelet domain polarization distance transformation |
CN107886103B (en) * | 2016-09-29 | 2023-12-08 | 日本电气株式会社 | Method, device and system for identifying behavior patterns |
CN107368781B (en) * | 2017-06-09 | 2019-08-20 | 陕西师范大学 | Synthetic Aperture Radar images change detecting method based on Subspace partition |
CN110119782A (en) * | 2019-05-14 | 2019-08-13 | 西安电子科技大学 | SAR image change detection based on FPGA |
CN111242864B (en) * | 2020-01-12 | 2023-05-05 | 杭州电子科技大学 | Finger vein image restoration method based on Gabor texture constraint |
CN111369457B (en) * | 2020-02-28 | 2022-05-17 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Remote sensing image denoising method for sparse discrimination tensor robustness PCA |
CN111814671A (en) * | 2020-07-08 | 2020-10-23 | 三亚中科遥感研究所 | Forest land dynamic change monitoring method based on remote sensing cooperation |
CN112131969A (en) * | 2020-09-01 | 2020-12-25 | 河海大学 | Remote sensing image change detection method based on full convolution neural network |
CN112131968A (en) * | 2020-09-01 | 2020-12-25 | 河海大学 | Double-time-phase remote sensing image change detection method based on DCNN |
CN113537063B (en) * | 2021-07-16 | 2022-09-13 | 海南省林业科学研究院(海南省红树林研究院) | Forest environment remote sensing monitoring method and terminal based on comprehensive remote sensing technology |
CN117746242B (en) * | 2023-12-20 | 2024-08-06 | 中国科学院空天信息创新研究院 | Remote sensing prospecting method, device, equipment and medium based on multi-metal deposit remote sensing prospecting model |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101510310A (en) * | 2009-02-19 | 2009-08-19 | 上海交通大学 | Method for segmentation of high resolution remote sensing image based on veins clustering constrain |
CN102629380A (en) * | 2012-03-03 | 2012-08-08 | 西安电子科技大学 | Remote sensing image change detection method based on multi-group filtering and dimension reduction |
CN103218823A (en) * | 2013-05-08 | 2013-07-24 | 西安电子科技大学 | Remote sensing image change detection method based on nuclear transmission |
CN103366365A (en) * | 2013-06-18 | 2013-10-23 | 西安电子科技大学 | SAR image varying detecting method based on artificial immunity multi-target clustering |
-
2014
- 2014-04-04 CN CN201410134520.1A patent/CN103971364B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101510310A (en) * | 2009-02-19 | 2009-08-19 | 上海交通大学 | Method for segmentation of high resolution remote sensing image based on veins clustering constrain |
CN102629380A (en) * | 2012-03-03 | 2012-08-08 | 西安电子科技大学 | Remote sensing image change detection method based on multi-group filtering and dimension reduction |
CN103218823A (en) * | 2013-05-08 | 2013-07-24 | 西安电子科技大学 | Remote sensing image change detection method based on nuclear transmission |
CN103366365A (en) * | 2013-06-18 | 2013-10-23 | 西安电子科技大学 | SAR image varying detecting method based on artificial immunity multi-target clustering |
Non-Patent Citations (3)
Title |
---|
A review on Gabor wavelets for face recognition;Linlin Shen等;《Pattern Anal Applic》;20060930;第273-292页 * |
Gabor滤波器在车辆检测和车型识别中的应用研究;赵英男;《中国博士学位论文全文数据库 信息科技辑》;20050615;正文第20-21页 * |
基于模糊C均值算法的遥感图像变化检测的研究;李敏;《中国优秀硕士学位论文全文数据库(信息科技辑)》;20100115;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN103971364A (en) | 2014-08-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103971364B (en) | Remote sensing image variation detecting method on basis of weighted Gabor wavelet characteristics and two-stage clusters | |
Chen et al. | A practical trial of landslide detection from single-temporal Landsat8 images using contour-based proposals and random forest: A case study of national Nepal | |
Zhan et al. | Log-based transformation feature learning for change detection in heterogeneous images | |
Chen et al. | Forested landslide detection using LiDAR data and the random forest algorithm: A case study of the Three Gorges, China | |
Im et al. | A change detection model based on neighborhood correlation image analysis and decision tree classification | |
Huang et al. | A new building extraction postprocessing framework for high-spatial-resolution remote-sensing imagery | |
Lu et al. | Land cover classification in a complex urban-rural landscape with QuickBird imagery | |
Lu et al. | A survey of image classification methods and techniques for improving classification performance | |
Zhang et al. | A comparison study of impervious surfaces estimation using optical and SAR remote sensing images | |
CN103955926B (en) | Method for detecting change of remote sensing image based on Semi-NMF | |
CN102609701B (en) | Remote sensing detection method based on optimal scale for high-resolution SAR (synthetic aperture radar) | |
CN103578110B (en) | Multiband high-resolution remote sensing image dividing method based on gray level co-occurrence matrixes | |
CN105930772A (en) | City impervious surface extraction method based on fusion of SAR image and optical remote sensing image | |
Jiang et al. | The potential of more accurate InSAR covariance matrix estimation for land cover mapping | |
CN109583293A (en) | Aircraft Targets detection and discrimination method in satellite-borne SAR image | |
CN105844279A (en) | Depth learning and SIFT feature-based SAR image change detection method | |
CN106250895A (en) | A kind of remote sensing image region of interest area detecting method | |
CN105718942A (en) | Hyperspectral image imbalance classification method based on mean value drifting and oversampling | |
Shaoqing et al. | The comparative study of three methods of remote sensing image change detection | |
CN104680184B (en) | Polarization SAR terrain classification method based on depth RPCA | |
CN105160648A (en) | Radar target and shadow segmentation method based on wavelet and constant false alarm rate | |
CN107689051A (en) | A kind of multitemporal SAR image change detecting method based on changed factor | |
CN105512622A (en) | Visible remote-sensing image sea-land segmentation method based on image segmentation and supervised learning | |
CN107392863A (en) | SAR image change detection based on affine matrix fusion Spectral Clustering | |
Wang et al. | The PAN and MS image fusion algorithm based on adaptive guided filtering and gradient information regulation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170201 Termination date: 20200404 |