CN103400383A - SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection - Google Patents
SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection Download PDFInfo
- Publication number
- CN103400383A CN103400383A CN2013103253981A CN201310325398A CN103400383A CN 103400383 A CN103400383 A CN 103400383A CN 2013103253981 A CN2013103253981 A CN 2013103253981A CN 201310325398 A CN201310325398 A CN 201310325398A CN 103400383 A CN103400383 A CN 103400383A
- Authority
- CN
- China
- Prior art keywords
- band
- frequency sub
- sub
- sar
- disparity map
- 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.)
- Pending
Links
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses an SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection. The method comprises the following steps that a difference graph is obtained on two SAR images via a logarithm ratio method, non-subsampled contourlet transformation is carried out on the difference graph, a low-frequency sub-band coefficient is maintained to be unchanged, after a high-frequency sub-band coefficient is subjected to hard threshold treatment, a treated difference graph is obtained according to inverse transformation, after neighboring features extracted from each pixel of the difference graph are pulled into rows, feature vectors are obtained by adopting the compressed projection to reduce a dimension, the feature vectors are divided into a changed category and a non-changed category with a K average value, the method is simple and easy to implement, and a detection result graph is obtained. According to the method, the problem of distortion on a profile edge on denoising of the difference graph is mainly solved. The noise of the obtained detection result graph is lower, better edges and profiles are maintained, the compressed projection is adopted to reduce the dimension, the complexity of calculation is relatively reduced, meanwhile a total error rate in change detection is also reduced, better visual effect is provided, and the method can be applied to the disaster detection of the SAR image.
Description
Technical field
The invention belongs to technical field of image processing, relate generally to synthetic-aperture radar (Synthetic Aperture Radar) SAR Image Change Detection, specifically a kind of multidate SAR image change detection method based on NSCT and compression projection.
Background technology
The SAR image is that synthetic-aperture radar is by transmitting and receiving the electromagnetic wave of particular polarization mode, obtain the high-resolution radar image of ground object target echo coefficient, what obtain that image reflects in essence is Electromagnetic Scattering Characteristics and the architectural characteristic of target, and imaging effect depends on radar parameter and region electromagnetic parameter to a great extent.The SAR image can be applicable to military affairs, agricultural, and navigation, the numerous areas such as geographical supervision, such as the identification of military target, detection, the condition of a disaster detection and the control etc. of mining deposits.Because the SAR image contains a large amount of coherent speckle noises, the variation of SAR image detects than the variation of optical imagery and detects and have more difficulty.But the advantages such as the round-the-clock that synthetic-aperture radar SAR has, round-the-clock, penetrability is strong, the accurate and effective more of utilizing the SAR image to detect on a surface target.And the SAR Image Change Detection is importance in object detection field.Change the comparative analysis that detects by to areal different times image, according to the variance analysis to image, obtain needed feature changes information.Change detection techniques is applied to a lot of aspects, for example to location and the Disaster Assessment of seismic region; Monitoring to the crop growth situation; Monitoring that urban land uses etc.Changing to detect at environment, agricultural, water conservancy and military national economy numerous areas all having a very wide range of applications, is one of important research direction of remote sensing technology.
At present, at first 2 o'clock phase SAR images are done respectively the logarithm ratio computing based on non-supervisory SAR image change detection method, spot noise in 2 o'clock phase SAR images is converted into additive noise, again 2 o'clock phasors are done minute ratio and will obtain the disparity map of 2 o'clock phasors, then this disparity map is carried out effective analyzing and processing and obtains " variation " and " non-variation " zone of image, for " variation " and " non-variation " differentiating and processing, a lot of methods are arranged:
Directly use threshold process, although algorithm simple computation complexity is low, but the result that obtains is unsatisfactory, because the disparity map that obtains contains a lot of noises, disparity map is directly utilized threshold process, the testing result figure Noise that obtains is more, and edge and profile are unintelligible, and index and visual effect are all poor.
Based on the maximum method of expectation, be to be under space-independent condition in the hypothesis disparity map, automatically select the threshold process disparity map to reduce total mistake, these class methods are due to the correlativity of not considering between the disparity map pixel, cause that edge contour is more coarse as a result, noise is more.
Based on the method for markov random file, considered the problem of disparity map neighborhood space correlativity, the method relies on parameter, this parameter influence disparity map spatial coherence information.The noise of the testing result figure that the method obtains is more, and the edge contour of testing result figure is still more coarse, and the method computation complexity is higher, and actual measurement SAR Image Change Detection is not reached through engineering approaches application index.
The noise of the testing result figure that above method obtains is larger, edge contour is unintelligible or computation complexity is too large,, in order to reach better denoising effect and edge contour clearly, need to propose better to reduce the new method of noise and keep the edge information profile to adapt to the pretreated fact of the larger needs of disparity map noise in the SAR Image Change Detection.
Summary of the invention
The object of the invention is to overcome the deficiency of above-mentioned prior art, proposed that a kind of image border profile retentivity is good, noise is less, reduced the multidate SAR image change detection method based on NSCT and compression projection of computation complexity, used the method to reduce the image detection total false rate of figure as a result.
Realize that the technology of the present invention purpose technical scheme is: use non-lower sampling contourlet transfer pair image noise reduction, simultaneously to the edge contour of image keep better, then utilize the compression projection to the Feature Dimension Reduction relative reduce computation complexity, effectively overcome the defect that in final variation diagram, noise is more, edge contour is more coarse, and relative reduce time complexity.Changing testing process comprises the steps:
(1) the two width SAR images of the identical region of different time of input carried out the computing of logarithm ratio method, obtain a width logarithm ratio difference figure X.
(2) disparity map X is carried out the contourlet conversion of non-lower sampling, the low frequency sub-band coefficient that obtains and the high-frequency sub-band coefficient of different scale different directions are carried out the threshold value noise reduction process, low frequency sub-band and high-frequency sub-band after processing are carried out non-lower sampling contourlet inverse transformation, obtain the disparity map X of noise reduction and maintenance contour edge
d
(3) to disparity map X
dEach pixel extraction neighborhood characteristics, neighborhood characteristics is pulled into and classifies column vector Ve as, utilize the compression projection with column vector Ve dimensionality reduction, the column vector after dimensionality reduction is proper vector v.
(4) utilize the K mean cluster that proper vector v is divided into two classes, a class is the vector that changes, and cluster centre is V
c, another kind of is unchanged vector, cluster centre is V
u, produce last testing result figure CM={cm (i, j) according to Euclidean distance | 1≤i≤I, 1≤j≤J}, the computing formula of testing result figure is
255 pixels that represent corresponding spatial position change wherein, and 0 represent the unchanged pixel in corresponding locus, v (i, j) is the proper vector of the pixel of (i, j) for the locus coordinate.
Non-lower sampling contourlet conversion is as a kind of novel multiple dimensioned, and the multiresolution analysis instrument possesses, multiresolution, locality threshold sampling characteristic, multidirectional and anisotropy.The present invention utilizes the limit singularity of non-lower sampling contourlet conversion to disparity map noise reduction and keep the edge information and profile, then utilizes the compression projection to carry out dimensionality reduction, has relatively reduced calculated amount, and is more obvious for the minimizing of getting the larger image calculation amount of neighborhood characteristics.
The present invention realizes also being: the process of in step 2, the high-frequency sub-band of the low frequency sub-band that obtains and different scale different directions being carried out the threshold value noise reduction process comprises:
(2a) for the low frequency sub-band coefficient, remain unchanged;
(2b) to the high-frequency sub-band of different scale, all adopt hard-threshold to process, the sub-band coefficients mould remains unchanged greater than the sub-band coefficients of threshold value, the sub-band coefficients mould is made as zero less than the sub-band coefficients of threshold value, and the different scale threshold value is different, to the same threshold value of high-frequency sub-band employing of same yardstick different directions.
High-frequency sub-band coefficient after adopting hard-threshold to the non-lower sampling contourlet conversion that obtains carries out hard-threshold to be processed, simple, the disparity map of the noise reduction that obtains and keep the edge information profile.
The present invention realizes also being: in step 3 to disparity map X
dEach pixel extraction neighborhood characteristics, pull into row with neighborhood characteristics, then utilizes the process of compression projection dimensionality reduction to comprise:
(3a) to disparity map X
dThe n of neighborhood * n piece around each pixel decimation of locus, n is odd number, then n 〉=3 pull into column vector Ve with each n * n piece;
(3b) utilize the compression projecting method to carry out dimensionality reduction to column vector Ve, the vector after dimensionality reduction is the column vector v of m dimension, m＜n
2, the measurement matrix that uses in the compression projecting method is star-like sampled form;
Star-like measurement matrix, in the frequency domain coordinate system, low frequency mainly concentrates on the initial point place, high frequency is away from far point, and for the most of concentration of energy of image at low frequency, therefore star-like matrix, by higher sampling rate, has lower sampling rate for radio-frequency component to low frequency part.
To the neighborhood characteristics dimensionality reduction that obtains, can the relative reduce calculated amount, get piece for neighborhood characteristics larger, reduce the calculated amount effect more obvious.
The present invention has the following advantages compared with prior art:
1, the present invention adopts the preliminary disparity map of non-lower sampling contourlet transfer pair to carry out pre-service, effect after processing has not only kept profile and marginal information, disparity map has also been done the processing of noise reduction, then the local message of utilization variance figure is classified, and obtains testing result figure.Experiment shows the present invention after the processing of noise reduction and keep the edge information profile, and the noise of the result that obtains is lower, and edge and profile keep better.
2, the present invention adopts non-lower sampling contourlet conversion, it has multiple dimensioned, multi-direction and rotational invariance, effectively extract the image outline edge feature, based on the local message dimensionality reduction of compression projection to disparity map, reduced computation complexity, utilize the simplest K average to classify, simple, effect is also fine.
3, simulation result shows, the variation that the present invention is applied to a few width SAR detects, and can access lower error rate, and better visual effect.
Description of drawings
Fig. 1 is the schematic flow sheet of the inventive method;
Fig. 2 is with area, Ottawa two width SAR striographs, standard drawing and the variation reference diagram that the present invention relates to, the geomorphology information in Fig. 2 (a) expression area, Ottawa in May, 1997; The geomorphology information in Fig. 2 (b) expression area, Ottawa in August, 1997; Fig. 2 (c) expression changes the standard results figure that detects;
Fig. 3 is the inventive method and the two kinds of control methodss experimental result picture to area, Ottawa SAR image, wherein, Fig. 3 (a) represents CS-KSVD methods and results figure, and Fig. 3 (b) represents PCA-Kmeans methods and results figure, and Fig. 3 (c) represents the figure as a result of the inventive method
Fig. 4 is with Bem area two width SAR striographs, standard drawing and the variation reference diagram that the present invention relates to, the geomorphology information in Fig. 2 (a) expression Bern in April, 1999 area; The geomorphology information in Fig. 2 (b) expression Bern in May, 1999 area; Fig. 2 (c) expression changes the standard results figure that detects;
Fig. 5 is the inventive method and the two kinds of control methodss experimental result picture to Bern area SAR image, wherein, Fig. 3 (a) represents CS-KSVD methods and results figure, and Fig. 3 (b) represents PCA-Kmeans methods and results figure, and Fig. 3 (c) represents the figure as a result of the inventive method
Embodiment
Embodiment 1
A kind of SAR image change detection method based on NSCT and compression projection of the present invention, synthetic-aperture radar SAR image is by transmitting and receiving a kind of electromagnetic wave of particular polarization mode, obtain the high-resolution radar image of the echo coefficient of ground object target, the advantages such as the round-the-clock that SAR has, round-the-clock, penetrability is strong, utilize the SAR image to change and detect more accurately to facilitate.the method of prior art is classified to disparity map, as direct use threshold value, expectation is maximum, the methods such as markov random file, the testing result figure noise that obtains is larger, edge contour is more coarse, the present invention has adopted non-lower sampling contourlet transfer pair image noise reduction, simultaneously to the edge contour of image keep better, then utilize the compression projection to Feature Dimension Reduction, relative reduce computation complexity, overcome testing result figure noise more, the defect that edge contour is more coarse, and relative reduce time complexity, make to change to detect and more be conducive to apply in engineering practice.
With reference to Fig. 1, specific implementation step of the present invention is as follows:
(1) the SAR image of the input two identical regions of width different time, be the geomorphology information in area, Ottawa in May, 1997 referring to Fig. 2 (a) and Fig. 2 (b) Fig. 2 (a); Fig. 2 (b) is the geomorphology information in area, Ottawa in August, 1997.To the SAR image of this identical region of two width different time carry out logarithm difference computing structural differences figure X (i, j)=| log (X
1(i, j)/X
2(i, j)) |, the logarithm ratio difference figure of X for generating wherein, X (i, j) wherein, X
1(i, j), X
2(i, j) is respectively logarithm ratio figure X, image X
1, image X
2Coordinate is the pixel value of the pixel of (i, j) in image.
(2) disparity map X is carried out the contourlet conversion of non-lower sampling, the low frequency sub-band that obtains and the high-frequency sub-band of different scale different directions are carried out the threshold value noise reduction process, low frequency sub-band and high-frequency sub-band after processing are carried out non-lower sampling contourlet inverse transformation, obtain the disparity map X of noise reduction and maintenance contour edge
d, concrete steps are as follows:
(2a) for low frequency sub-band, its coefficient Y
3{ 1} remains unchanged;
(2b) to the high-frequency sub-band of different scale, all adopt hard-threshold to process, the sub-band coefficients mould remains unchanged greater than the sub-band coefficients of threshold value, and the sub-band coefficients mould is made as zero less than the sub-band coefficients of threshold value.Y
N, tK} is the high-frequency sub-band coefficient of t the direction of N yardstick of disparity map X, and k=N+1, N is yardstick, N=1 in this example, 2,4.For high-frequency sub-band, the computing formula that hard-threshold is processed is:
λ is threshold value.
In the present invention, the sub-band coefficients of different scale is Y
N, t{ the k} threshold value is different, to same yardstick and the same threshold value of high-frequency sub-band coefficients by using of different directions.In sub-band coefficients to same yardstick different directions, the high-frequency sub-band that namely the identical t of N is different adopts same threshold value.
Threshold value adopts Donoho and Johnstone to unify threshold value, and the threshold calculations formula is
Wherein σ is the noise criteria variance, and L is size or the length of signal.In actual environment, the noise criteria variance in image is unknowable, therefore when selected threshold, determine the noise criteria variance with method of estimation.Wherein the computing formula of evaluation method commonly used is σ=Y
N, t{ k}/0.6745, wherein Y
N, t{ k} is that yardstick is the value that the sub-band coefficients amplitude of N is arranged interposition in order.In practical operation, after image is determined, its high-frequency sub-band coefficient Y
N, t{ k} is known, so the value that the sub-band coefficients amplitude is arranged interposition in order is also known, and then can calculate σ, and signal length L is also known, and threshold value λ can calculate.
(3) to disparity map X
dEach pixel extraction neighborhood characteristics, neighborhood characteristics is pulled into and classifies column vector Ve as, utilize the compression projection with column vector Ve dimensionality reduction, the column vector after dimensionality reduction is proper vector v, concrete steps are as follows:
(3a) to disparity map X
dThe n of neighborhood * n piece around each pixel decimation of locus, n is odd number, then n 〉=3 pull into column vector Ve with each n * n piece.
(3b) utilize the compression projecting method to carry out dimensionality reduction to column vector Ve, the vector after dimensionality reduction is the column vector v of m dimension, m＜n
2, the measurement matrix that uses in the compression projecting method is star-like sampled form.
Star-like measurement matrix, in the frequency domain coordinate system, low frequency mainly concentrates on the initial point place, high frequency is away from far point, and for the most of concentration of energy of image at low frequency, therefore star-like matrix, by higher sampling rate, has lower sampling rate for radio-frequency component to low frequency part.
(4) utilize the K mean cluster that proper vector v is divided into two classes, a class is the vector that changes, and cluster centre is V
c, another kind of is unchanged vector, cluster centre is V
u, produce last testing result figure CM={cm (i, j) according to Euclidean distance | 1≤i≤I, 1≤j≤J}, the computing formula of testing result figure is
255 pixels that represent corresponding spatial position change wherein, and 0 represent the unchanged pixel in corresponding locus, v (i, j) is the proper vector of the pixel of (i, j) for the locus coordinate.Last output detections is figure CM as a result.
, along with the appearance of compressed sensing and ripe day by day, will compress projection application in image processing field.The contourlet conversion of non-lower sampling (Nonsubsampled contourlet Tranform, NSCT) be contourlet conversion after a kind of improve, have the characteristics such as multiple dimensioned, multi-direction, shift invariant, and can better catch the inherent geometry of image.NSCT is the conversion of line singularity, and changing edge contour in detection figure is important information,, for denoising and the keep the edge information profile information employing NSCT conversion process of disparity map, can obtain the disparity map of denoising and keep the edge information profile.And traditional Lee wave filter is based on get neighborhood averaging deblurring effect at moving window, and the contour edge of image can partial loss.Adopt wavelet transformation to process disparity map, wavelet transformation is the conversion of a singularity, bad for the edge contour retentivity, with the wavelet coefficient threshold value or set up the coefficients statistics model and suppress noise, although than space-wise, obtain better result, but easily produce Gibbs' effect, so the present invention does not utilize wavelet transformation to carry out denoising.The present invention adopts the preliminary disparity map of non-lower sampling contourlet transfer pair to carry out pre-service, effect after processing has not only kept profile and marginal information, disparity map has also been done the processing of noise reduction, utilize compression projection dimensionality reduction, relative reduce computation complexity, then the local message of utilization variance figure is classified, the testing result figure of be maintained profile and marginal information.
Embodiment 2
Based on NSCT and the compression projection the SAR image change detection method with embodiment 1,
This example is elaborated to the contourlet conversion of the non-lower sampling in step 2
In step 2, disparity map X is carried out 3 layers of non-lower sampling contourlet conversion, obtain low frequency sub-band and high-frequency sub-band;
At ground floor, yardstick N=1, resolve into a low frequency sub-band image X with disparity map X
1{ 1} and the logical sub-band images X of band
1{ 2} will be with logical sub-band images X
1{ 2} carries out 2
NThe level Directional Decomposition, in this example, N=1, be 2 Directional Decompositions, obtains 2 high-frequency sub-band coefficient Y
1, t{ 2, t is the direction sequence number for 2}, t=1 wherein;
At the second layer, yardstick N=2, decompose disparity map X the low frequency sub-band image X that generates through ground floor
1{ 1} resolves into a low frequency sub-band image X
2{ 1} and the logical sub-band images X of band
2{ 3} obtains 4 high-frequency sub-band coefficient Y
2, t{ 3}, t=1 wherein, 2,3,4;
At the 3rd layer, yardstick N=4, decompose disparity map X the low frequency sub-band image X that generates through the second layer
2{ 1} resolves into a low frequency sub-band image X
3{ 1} and the logical image X of band
3{ 4} obtains a low frequency sub-band coefficient Y
3{ 1} and 16 high-frequency sub-band coefficient Y
3, t{ 4}, t=1 wherein, 2 ... 16.
For low frequency sub-band coefficient Y
3{ 1} remains unchanged, and for high-frequency sub-band coefficients by using hard-threshold, processes.
The present invention is to the same threshold process denoising of the high-frequency sub-band coefficients by using of same yardstick different directions, when calculating standard variance σ, although the high-frequency sub-band of each yardstick has multiple directions, because the high-frequency sub-band coefficients by using same threshold value of the present invention to same yardstick different directions, thus in this example to the high-frequency sub-band coefficient Y of each yardstick
N, t{ sub-band coefficients while all adopting t=1 in k}, namely adopt Y in yardstick N=1
1,1{ 2} uses Y in yardstick N=2
2,1{ 3} uses Y in yardstick N=4
3,1{ 4}, the standard variance σ of corresponding yardstick can be calculated.
Embodiment 3
Based on NSCT and the compression projection the SAR image change detection method with embodiment 1-2,
In this example, to disparity map X
dEach pixel extraction neighborhood characteristics, neighborhood characteristics is pulled into and classifies column vector Ve as, utilize the compression projection that the process of column vector Ve dimensionality reduction is specialized explanation, in step 3 to disparity map X
dThe piece of each pixel decimation 3 * 3 of locus is as neighborhood, this neighborhood characteristics is pulled into column vector x, the dimension of x is 9, utilize compression projection dimensionality reduction, adopt star-like sample mode in this example, the dimensionality reduction formula is y=A * x, and A is star-like measurement matrix, y is the proper vector after dimensionality reduction, and the dimension of y is 3.
Embodiment 4
Based on NSCT and the compression projection the SAR image change detection method with embodiment 1-2,
In this example, if in step 3 to disparity map X
dThe piece of each pixel decimation 5 * 5 of locus, pull into row x, utilizes compression projection dimensionality reduction, utilizes star-like sample mode in this example, and the dimensionality reduction formula is y=A * x, and A is star-like measurement matrix, and y is the proper vector after dimensionality reduction, and the dimension of y is 10.
The result that the present invention adopts star-like measurement matrix to obtain is stable, does not need multiple averaging to obtain end product.PCA is also dimensionality reduction method commonly used, but PCA only can separate the data point of linear dependence, but for highly uncorrelated data point, can not well distinguish, and therefore under the impact of noise, the PCA dimensionality reduction easily produces error-detecting.The present invention does not adopt the random measurement matrix, and the result of the each experiment of random measurement matrix is different, needs many experiments to be averaged, and has strengthened the Time Calculation complexity, and has been unfavorable for the use in engineering practice.The star-like matrix that the present invention adopts, by higher sampling rate, has lower sampling rate for radio-frequency component to low frequency part, and the result of each experiment is stable, does not need repeatedly to be averaged.
Embodiment 5
Based on NSCT and the compression projection the SAR image change detection method with embodiment 1-4,
Effect of the present invention can further illustrate by following emulation.
1, simulated conditions
Adopted two groups of SAR images commonly used in the SAR Image Change Detection, first group of data is areas, Ottawa, and the image size is 290 * 350, and it is the Radarsat image.Second group of data is areas, Bern, and the image size is 301 * 301, and it is the ERS-2 image.Choose the index that loss, false drop rate and total error rate detect as variation.Use respectively CS-KSVD, the PCA-Kmeans method, the present invention and above-mentioned two kinds of methods all extract 3 * 3 piece.Column vector dimension after the CS-KSVD dimensionality reduction is 6, and the column vector dimension after dimensionality reduction of the present invention is 3, and the present invention adopts non-lower sampling contourlet conversion, and decomposing the number of plies is 3, and the directional subband number of every layer is respectively 2,4,16.
2, analysis of simulation result
To simulation result such as Fig. 2 of Ottawa SAR image, wherein Fig. 2 (a), Fig. 2 (b) are the reference diagram of two width multidates, and Fig. 2 (c) is standards change figure.Fig. 3 (a) is the result after processing for the CS-KSVD method, and Fig. 3 (b) is the result after processing for PCA-Kmeans, the result of Fig. 3 (c) after for the present invention's processing.
To simulation result such as Fig. 3 of Bern SAR image, wherein Fig. 4 (a), Fig. 4 (b) are the reference diagram of two width multidates, and Fig. 4 (c) is standards change figure.Fig. 5 (a) is the result after processing for the CS-KSVD method, and Fig. 5 (b) is the result after processing for PCA-Kmeans, the result of Fig. 5 (c) after for the present invention's processing.
As can be seen from Figure 3, result after the present invention processes not only has visual effect preferably, and effectively filtering assorted point, it is also very good that profile and edge keep, reduced whole error rate, all made moderate progress than other two kinds of methods aspect picture quality and visual effect.Figure compares with standards change, and the upper right corner and right half part, without the obviously appearance of assorted point, obviously reduce at the assorted point of the left-hand component of image in Fig. 3 (c), and contour edge information do not lose, and visual effect is fine.Image 3 (a), Fig. 3 (b) obviously have a lot of assorted points, i.e. flase drop pixel, and the undetected pixel of upper left corner part of image is more than Fig. 3 (c).Experiment shows the present invention after the processing of noise reduction and keep the edge information profile, and the noise of the result that obtains is lower, and edge and profile keep better.
As can be seen from Figure 5, in result after the present invention processes, the relative control methods of noise of unchanged part is less, edge contour keeps better, the raising of visual effect and the improvement of picture quality two aspects all are better than the method that contrasts, and have again verified the validity of noise reduction of the present invention and keep the edge information.
Embodiment 6
Based on NSCT and the compression projection the SAR image change detection method with embodiment 1-4, the condition of emulation and content are with embodiment 5
The present invention and existing method are as shown in table 1 to the variation monitoring performance index contrast table of Ottawa SAR image:
Each method of table 1 is to Ottawa SAR Image Change Detection performance comparison
Can see that from table 1 value of cited method the inventive method is starkly lower than table on undetected pixel in shows that the inventive method keeps better contour edge.Total error rate of this method is lower than existing method, and figure is the most approaching with standards change, and visual effect is better, shows that this method effectively removes noise, and has well kept profile and edge; In sum, the inventive method has obtained to change preferably and has detected index and effect.
The present invention and existing method detect index contrast table such as table 2 to the variation of area, Bern SAR image
Each method of table 2 is to area, Bern SAR Image Change Detection performance comparison
As can be seen from Table 2, the inventive method has reduced total false rate, and better visual effect is arranged, and is more approaching with the variation diagram of standard, and the present invention has obtained the variation that is better than cited method and detected index, shows that the variation of the inventive method detects better effects if.
in a word, SAR image change detection method based on NSCT and compression projection of the present invention, implementation step is: to two width SAR images, use the logarithm ratioing technigue to obtain disparity map, disparity map is carried out non-lower sampling contourlet conversion, after the high-frequency sub-band coefficient is carried out the hard-threshold processing, the disparity map that inverse transformation obtains processing, each pixel extraction neighborhood characteristics to this disparity map, neighborhood characteristics is pulled into row utilizes compression projection dimensionality reduction to obtain proper vector, utilize the K average that proper vector is divided into two classes, respectively to change and unchanged two classes, simple, obtain testing result figure, the present invention carries out non-lower sampling contourlet conversion and the high-frequency sub-band coefficient is carried out the threshold value noise reduction process disparity map, the testing result figure noise that obtains is lower, edge and profile have been kept, then each pixel extraction feature, utilize compression projection dimensionality reduction, relative reduce computation complexity, the present invention has the noise of removing testing result figure, and the contour edge that keeps variation diagram, reduce and change total false rate in detection, has advantages of simultaneously better visual effect, the condition of a disaster that can be used for the SAR image detects.
The present invention is all making moderate progress than methods such as PCA aspect variation detection performance and visual effect, and effectively goes the detailed information such as impurity point and maintenance image outline edge.
Claims (3)
- One kind based on NSCT and the compression projection the SAR image change detection method, it is characterized in that: include following steps:(1) 2 o'clock phase SAR view data of input are carried out the logarithm ratio computing, obtain a width logarithm ratio difference figure X;(2) disparity map X is carried out the contourlet conversion of non-lower sampling, the low frequency sub-band coefficient that obtains and the high-frequency sub-band coefficient of different scale different directions are carried out the threshold value noise reduction process, low frequency sub-band coefficient and high-frequency sub-band coefficient after processing are carried out non-lower sampling contourlet inverse transformation, obtain the disparity map X of noise reduction and maintenance contour edge d(3) to disparity map X dEach pixel extraction neighborhood characteristics, neighborhood characteristics is pulled into and classifies column vector V as e, utilize the compression projection with column vector V eDimensionality reduction, the column vector after dimensionality reduction are proper vector v;(4) utilize the K mean cluster that proper vector v is divided into two classes, a class is the vector that changes, and cluster centre is V c, another kind of is unchanged vector, cluster centre is V u, produce last testing result figure CM={cm (i, j) according to Euclidean distance | 1≤i≤I, 1≤j≤J}, the computing formula of testing result figure is
255 pixels that represent corresponding spatial position change wherein, and 0 represent the unchanged pixel in corresponding locus, v (i, j) is the proper vector of the pixel of (i, j) for the locus coordinate. - 2. the SAR image change detection method based on NSCT and compression projection according to claim 1 is characterized in that: the process of in step 2, the high-frequency sub-band of the low frequency sub-band that obtains and different scale different directions being carried out the threshold value noise reduction process comprises:(2a) the low frequency sub-band coefficient is remained unchanged;(2b) to the high-frequency sub-band coefficient of different scale, all adopt hard-threshold to process, the sub-band coefficients mould remains unchanged greater than the sub-band coefficients of threshold value, the sub-band coefficients mould is made as zero less than the sub-band coefficients of threshold value, the sub-band coefficients threshold value of different scale is different, to the same threshold value of high-frequency sub-band coefficients by using of same yardstick different directions.
- According to claim 2 based on NSCT and the compression projection the SAR image change detection method, it is characterized in that: in step 3 to disparity map X dEach pixel extraction neighborhood characteristics, pull into row with neighborhood characteristics, then utilizes the process of compression projection dimensionality reduction to comprise:(3a) to disparity map X dThe n of neighborhood * n piece around each pixel decimation of locus, n is odd number, then n 〉=3 pull into column vector V with each n * n piece e(3b) utilize the compression projecting method to column vector V eCarry out dimensionality reduction, the vector after dimensionality reduction is the column vector v of m dimension, m＜n 2, the measurement matrix that uses in the compression projecting method is star-like sampled form.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013103253981A CN103400383A (en) | 2013-07-22 | 2013-07-22 | SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013103253981A CN103400383A (en) | 2013-07-22 | 2013-07-22 | SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103400383A true CN103400383A (en) | 2013-11-20 |
Family
ID=49563995
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013103253981A Pending CN103400383A (en) | 2013-07-22 | 2013-07-22 | SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103400383A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103761731A (en) * | 2014-01-02 | 2014-04-30 | 河南科技大学 | Small infrared aerial target detection method based on non-downsampling contourlet transformation |
| CN104240249A (en) * | 2014-09-12 | 2014-12-24 | 西安电子科技大学 | SAR image change detection method based on directional wavelet transformation and improved level set |
| CN104392442A (en) * | 2014-11-18 | 2015-03-04 | 西北工业大学 | Remote sensing image change detecting method based on non-subsample Contourlet transform and active contour |
| CN107358205A (en) * | 2017-07-13 | 2017-11-17 | 西安电子科技大学 | A kind of polarization SAR remote sensing imagery change detection method based on NSCT DBN |
| CN111091107A (en) * | 2019-12-20 | 2020-05-01 | 广州杰赛科技股份有限公司 | Face region edge detection method and device and storage medium |
| CN111199188A (en) * | 2019-12-18 | 2020-05-26 | 星际空间（天津）科技发展有限公司 | Pixel processing method and device for remote sensing image difference map, storage medium and equipment |
| CN112308025A (en) * | 2020-11-23 | 2021-02-02 | 内蒙古工业大学 | Image change recognition method and device, storage medium and electronic equipment |
| WO2021028978A1 (en) * | 2019-08-09 | 2021-02-18 | 三菱電機株式会社 | Image change detection device and image change detection method |
| CN114049568A (en) * | 2021-11-29 | 2022-02-15 | 中国平安财产保险股份有限公司 | Object shape change detection method, device, equipment and medium based on image comparison |
| CN116910489A (en) * | 2023-09-07 | 2023-10-20 | 中国水利水电第九工程局有限公司 | Wall seepage prevention detection method based on artificial intelligence and related device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080050021A1 (en) * | 2006-08-25 | 2008-02-28 | Plant James R | Method of difference sensing through optical coherent change detection |
| CN102096921A (en) * | 2011-01-10 | 2011-06-15 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image change detection method based on neighborhood logarithm specific value and anisotropic diffusion |
| US20110222781A1 (en) * | 2010-03-15 | 2011-09-15 | U.S. Government As Represented By The Secretary Of The Army | Method and system for image registration and change detection |
| CN102800074A (en) * | 2012-07-12 | 2012-11-28 | 西安电子科技大学 | Synthetic aperture radar (SAR) image change detection difference chart generation method based on contourlet transform |
| CN103198480A (en) * | 2013-04-02 | 2013-07-10 | 西安电子科技大学 | Remote sensing image change detection method based on area and Kmeans clustering |
-
2013
- 2013-07-22 CN CN2013103253981A patent/CN103400383A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080050021A1 (en) * | 2006-08-25 | 2008-02-28 | Plant James R | Method of difference sensing through optical coherent change detection |
| US20110222781A1 (en) * | 2010-03-15 | 2011-09-15 | U.S. Government As Represented By The Secretary Of The Army | Method and system for image registration and change detection |
| CN102096921A (en) * | 2011-01-10 | 2011-06-15 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image change detection method based on neighborhood logarithm specific value and anisotropic diffusion |
| CN102800074A (en) * | 2012-07-12 | 2012-11-28 | 西安电子科技大学 | Synthetic aperture radar (SAR) image change detection difference chart generation method based on contourlet transform |
| CN103198480A (en) * | 2013-04-02 | 2013-07-10 | 西安电子科技大学 | Remote sensing image change detection method based on area and Kmeans clustering |
Non-Patent Citations (2)
| Title |
|---|
| LEYUAN FANG ET AL.: "Multitemporal Image Change Detection with Compressed Sparse Representation", 《 2011 18TH IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING》 * |
| 慕娟等: "遥感图像的NSCT 自适应阈值去噪方法", 《信号与信息处理》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103761731A (en) * | 2014-01-02 | 2014-04-30 | 河南科技大学 | Small infrared aerial target detection method based on non-downsampling contourlet transformation |
| CN104240249A (en) * | 2014-09-12 | 2014-12-24 | 西安电子科技大学 | SAR image change detection method based on directional wavelet transformation and improved level set |
| CN104240249B (en) * | 2014-09-12 | 2017-03-15 | 西安电子科技大学 | Based on direction wave conversion and the SAR image change detection of improved level set |
| CN104392442A (en) * | 2014-11-18 | 2015-03-04 | 西北工业大学 | Remote sensing image change detecting method based on non-subsample Contourlet transform and active contour |
| CN107358205A (en) * | 2017-07-13 | 2017-11-17 | 西安电子科技大学 | A kind of polarization SAR remote sensing imagery change detection method based on NSCT DBN |
| JPWO2021028978A1 (en) * | 2019-08-09 | 2021-10-21 | 三菱電機株式会社 | Image change detection device and image change detection method |
| WO2021028978A1 (en) * | 2019-08-09 | 2021-02-18 | 三菱電機株式会社 | Image change detection device and image change detection method |
| CN111199188A (en) * | 2019-12-18 | 2020-05-26 | 星际空间（天津）科技发展有限公司 | Pixel processing method and device for remote sensing image difference map, storage medium and equipment |
| CN111199188B (en) * | 2019-12-18 | 2023-07-11 | 星际空间（天津）科技发展有限公司 | Pixel processing method, device, storage medium and equipment of remote sensing image difference map |
| CN111091107A (en) * | 2019-12-20 | 2020-05-01 | 广州杰赛科技股份有限公司 | Face region edge detection method and device and storage medium |
| CN112308025A (en) * | 2020-11-23 | 2021-02-02 | 内蒙古工业大学 | Image change recognition method and device, storage medium and electronic equipment |
| CN112308025B (en) * | 2020-11-23 | 2023-01-31 | 内蒙古工业大学 | Image change identification method and device, storage medium and electronic equipment |
| CN114049568A (en) * | 2021-11-29 | 2022-02-15 | 中国平安财产保险股份有限公司 | Object shape change detection method, device, equipment and medium based on image comparison |
| CN116910489A (en) * | 2023-09-07 | 2023-10-20 | 中国水利水电第九工程局有限公司 | Wall seepage prevention detection method based on artificial intelligence and related device |
| CN116910489B (en) * | 2023-09-07 | 2023-11-21 | 中国水利水电第九工程局有限公司 | Wall seepage prevention detection method based on artificial intelligence and related device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103400383A (en) | SAR (synthetic aperture radar) image change detection method based on NSCT (non-subsampled contourlet transform) and compressed projection | |
| CN102324021B (en) | Infrared dim-small target detection method based on shear wave conversion | |
| Deledalle et al. | Exploiting patch similarity for SAR image processing: The nonlocal paradigm | |
| CN101634709B (en) | Method for detecting changes of SAR images based on multi-scale product and principal component analysis | |
| CN103456018B (en) | Remote sensing image change detection method based on fusion and PCA kernel fuzzy clustering | |
| CN102156971B (en) | Speckle suppression method of synthetic aperture radar (SAR) image based on linear singularity information | |
| Zhou et al. | Linear feature detection in polarimetric SAR images | |
| CN106296655A (en) | Based on adaptive weight and the SAR image change detection of high frequency threshold value | |
| CN102521818A (en) | Fusion method of SAR (Synthetic Aperture Radar) images and visible light images on the basis of NSCT (Non Subsampled Contourlet Transform) | |
| CN102509263B (en) | K-SVD (K-means singular value decomposition) speckle inhibiting method based on SAR (synthetic aperture radar) image local statistic characteristic | |
| CN107507209B (en) | Printogram extraction method of polarized SAR image | |
| CN101685158B (en) | Hidden Markov tree model based method for de-noising SAR image | |
| CN103093432B (en) | Polarized synthetic aperture radar (SAR) image speckle reduction method based on polarization decomposition and image block similarity | |
| CN103020922A (en) | PCA (principal component analysis) transformation based SAR (synthetic aperture radar) image speckle suppression method | |
| CN105205484A (en) | Synthetic aperture radar target detection method based on curvelet transformation and Wiener filtering | |
| CN104299232A (en) | SAR image segmentation method based on self-adaptive window directionlet domain and improved FCM | |
| CN103077507B (en) | Beta algorithm-based multiscale SAR (Synthetic Aperture Radar) image denoising method | |
| Kumar et al. | A multi-objective enhanced fruit fly optimization (MO-EFOA) framework for despeckling SAR images using DTCWT based local adaptive thresholding | |
| CN108428221A (en) | A kind of neighborhood bivariate shrinkage function denoising method based on shearlet transformation | |
| Wang et al. | Desert seismic noise suppression based on multimodal residual convolutional neural network | |
| CN103426145A (en) | Synthetic aperture sonar speckle noise suppression method based on multiresolution analysis | |
| CN102201116A (en) | Synthetic aperture radar (SAR) image speckle suppression method by combining direction aggregation | |
| CN113205564A (en) | SAR intelligent target edge reconstruction method | |
| CN102693530B (en) | Synthetic aperture radar (SAR) image despeckle method based on target extraction and speckle reducing anisotropic diffusion (SRAD) algorithm | |
| CN104200472A (en) | Non-local wavelet information based remote sensing image change detection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20131120 |