CN105988113A - Polarmetric synthetic aperture radar (SAR) image change detection method - Google Patents

Abstract

The invention belongs to the technical field of SAR image processing. Through deep interpretation on polarimetric SAR data, a method of quickly and accurately extracting a two-time phase change area to realize polarimetric SAR image change detection based on a joint weighted polarization difference is researched. The polarmetric SAR image change detection method by the technical scheme adopted by the invention comprises the following steps: (1) pre-processing is carried out, and deorientation and speckle noise suppression are carried out on two already-registered time phase polarimetric SAR images; (2) feature vectors kAi and kBi for corresponding pixel points in the two time phase images are built; (3) a polarization scattering difference and a polarization power difference for the corresponding pixel points in the two time phase images are calculated; (4) corresponding weighted coefficients are allocated according to the relative size of the two differences, a joint weighted polarization difference is obtained after summation, and a difference image is built; and (5) threshold segmentation is carried out on the difference image to extract a change area. The method is mainly applied to an image processing situation.

Description

Polarization synthetic aperture radar image change detecting method

Technical field

The invention belongs to SAR image processing technology field, relate to a kind of based on the polarization SAR combining weighting polarization difference degree Image change detection method.

Background technology

Synthetic aperture radar (Synthetic Aperture Radar, SAR) is a kind of round-the-clock, round-the-clock high-resolution Rate microwave imaging system, because it uses microwave technology to carry out imaging, therefore the restriction that can effectively break away from the natural causes such as weather is real Existing real-time monitored, these features also make it have the excellent of uniqueness at aspects such as Natural calamity monitoring, oceanographic observation, military surveillances Gesture.Traditional single channel single polarization SAR image is only capable of obtaining target under the transmitting-receiving combination of a certain particular polarization for the ground scene and dissipates Penetrating characteristic, information content is extremely limited.In recent years, SAR system (the i.e. polarization SAR of ground object target polarization scattering characteristics can be obtained System) development, polarimetric synthetic aperture radar imaging technique also reaches its maturity.Compared with single polarization SAR image, many/full poles The information content that change SAR image is comprised is bigger, more can disclose the scattering mechanism of target complete and accurate, scheme for fully excavating Target information in Xiang, raising segmentation, classification, object detection and recognition performance provide data and support and guarantee.Due to polarization SAR system imaging mechanism is complicated and starts late, and rationally correctly interprets Polarimetric SAR Image not a duck soup, and this also makes polarization SAR image processes association area research and has very big development space.Based on polarimetric synthetic aperture radar (PolSAR) image The important branch that change detection techniques interprets as SAR image, crowds such as the condition of a disaster estimation, urban planning and military attacks Multi-field play very important effect.

In recent years, the suitable in-depth study of change detection techniques to SAR image for the Chinese scholars, and propose perhaps Multi-method, such as ratio method, differential technique, category method, vegetation indexing method etc..Change context of detection, state at multilevel configuration Although inside and outside scholar is achieved with certain progress, but every research is not perfect, ripe, does not forms unified system.2003 Knut Conradsen of international space research institute of year Denmark University of Science and Technology et al. proposes a kind of based on multiple Wishart distribution Polarimetric SAR Image change detecting method, the method is realized the change detection of Polarimetric SAR Image by constructing Likelihood ration test amount. Within 2004, Japanese scholars Muhtar Qong proposes a kind of Polarimetric SAR Image change detecting method based on polarized state construction, The method is improve the accuracy of change detection by constructing Optimal polarization state.Laurent Ferro-Famil profit in 2008 Achieve change detection with maximum likelihood ratio (Maximum Likelihood ratio, MLR) construction change detection characteristic quantity. Esra Erten in 2011 et al. utilizes KL distance to achieve the change detection of Polarimetric SAR Image as change detection characteristic quantity. Bhogendra Mishra in 2013 et al. proposes a kind of new change detection characteristic quantity normalization diversity ratio (Normalized Difference ratio, NDR), experiment shows, when utilizing this detection limit to be changed detection, is comprised Information content more more than simple ratio detection limit, Detection results is more preferable.

Content of the invention

For overcoming the deficiencies in the prior art, it is contemplated that by deeply understanding polarization SAR data, research is a kind of fast Speed, the method extracting two Temporal variation regions accurately, it is achieved based on the Polarimetric SAR Image change combining weighting polarization difference degree Detection method.The technical solution used in the present invention is, polarization synthetic aperture radar image change detecting method, and step is as follows:

(1) pre-process, carry out orientation and coherent speckle noise to the two phase polarization synthetic aperture radar images having registrated Suppression operation；

(2) the characteristic vector k of corresponding pixel points in phase images when constructing two_AiAnd k_Bi

k_Ai=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

k_Bi=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

Wherein, k_Ai、k_BiRepresent respectively pixel i feature in phase polarimetric synthetic aperture radar data for A, the B two to Amount, T_mnRepresenting the element in the coherence matrix T of polarization synthetic aperture radar image, subscript T represents transposition；

(3) the Polarization scattering diversity factor of phase images corresponding pixel points and polar power diversity factor when calculating two

If two clarification of objective vectors are respectively k_AiAnd k_Bi, corresponding antenna receives power and is respectively P_AiAnd P_Bi, definition Polarization scattering diversity factor between two targets and polar power diversity factor are respectively as follows:

$D_1=1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2}$

$D_2=1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}}$

Wherein, | | | |₂Representing 2 norms of vector, subscript H represents conjugation transposition；

(4) the corresponding weight coefficient of relative size distribution according to two species diversity degree, obtains after summation combining weighting polarization Diversity factor, constructs differential image, combines weighting polarization difference degree and is:

$d=a\·D_1+b\·D_2=a\·(1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2})+b\·(1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}})$

Wherein, a, b are weight coefficient and meet a+b=1, and a, b determine the tribute to overall diversity factor for the two species diversity types Offer size；

(5) enter row threshold division to differential image and extract region of variation: first with change detection characteristic quantity i.e. disparity map The average of picture and standard deviation determine candidate thresholds, then account for the ratio-dependent of all pixel numbers according to change point number final Segmentation threshold, finally utilize threshold segmentation method extract region of variation.

The feature of the present invention and providing the benefit that:

The present invention comprehensively utilizes the polarization information of Polarimetric SAR Image, to combine weighting polarization difference degree as change detection Tolerance, give a kind of based on combine weighting polarization difference degree Polarimetric SAR Image change detecting method, the method can be fast The region of variation of phase images when speed, effective detection two.

Brief description:

Fig. 1 gives two phase Pauli exploded views of farm field data.

Fig. 2 gives the differential image of the inventive method.

Fig. 3 give the inventive method Threshold segmentation after result.

Fig. 4 gives the flow chart that the present invention proposes method.

Detailed description of the invention

The present invention makes full use of the information of Polarimetric SAR Image, it is achieved that a kind of based on the pole combining weighting polarization difference degree Changing SAR image change detection, concrete technical scheme is divided into following steps:

(1) pre-process.Mainly include carrying out orientation to the two phase Polarimetric SAR Images having registrated and coherent speckle noise presses down System operation, the impact on change testing result with this random orientation reducing ground object target and coherent speckle noise.

(2) the characteristic vector k of corresponding pixel points in phase images when constructing two_AiAnd k_Bi。

k_Ai=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

k_Bi=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

Wherein, k_Ai、k_BiRepresent pixel i at characteristic vector in phase polarization SAR data for A, the B two, T respectively_mnRepresent pole Changing the element in the coherence matrix T of SAR image, subscript T represents transposition.

(3) the Polarization scattering diversity factor of phase images corresponding pixel points and polar power diversity factor when calculating two.

If two clarification of objective vectors are respectively k_AiAnd k_Bi, corresponding antenna receives power and is respectively P_AiAnd P_Bi, definition Polarization scattering diversity factor between two targets and polar power diversity factor are respectively as follows:

$D_1=1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2}$

$D_2=1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}}$

Wherein, | | | |₂Representing 2 norms of vector, subscript H represents conjugation transposition.

(4) the corresponding weight coefficient of relative size distribution according to two species diversity degree, obtains after summation combining weighting polarization Diversity factor, constructs differential image.Combining weighting polarization difference degree is:

$d=a\·D_1+b\·D_2=a\·(1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2})+b\·(1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}})$

Wherein, a, b are weight coefficient and meet a+b=1, and a, b determine the tribute to overall diversity factor for the two species diversity types Offer size.

(5) enter row threshold division to differential image and extract region of variation.First with change detection characteristic quantity (i.e. disparity map Picture) average and standard deviation determine candidate thresholds, then account for the ratio-dependent of all pixel numbers according to change point number Whole segmentation threshold, finally utilizes threshold segmentation method to extract region of variation.

The present invention gives a kind of new weight coefficient selection scheme:

Theoretically, in the case that image-forming condition is constant, what same ground object target showed in phase images when difference dissipates Penetrate characteristic and power characteristic should be identical.When two, the ground object target of phase images same position changes, its scattering Characteristic and power characteristic also should change accordingly, but intensity of variation is not necessarily identical, it is possible to scattering properties change is obvious, It is also possible to power characteristic change substantially.Therefore, should compromise according to the relative size of two species diversity when weights assigned coefficient a, b Consider.Based on above-mentioned consideration, the selection scheme of weight coefficient is as follows:

Work as D₁＞ D₂When, the principle of the weight coefficient of distribution is: a ＜ 0.5, b ＞ 0.5；

Work as D₁≤D₂When, the principle of the weight coefficient of distribution is: a ＞ 0.5, b ＜ 0.5.

The present invention proposes the automatic Selection of Image Threshold of a kind of integrated data histogram distribution and change point ratio, and it is main Want thought to be the average by differential image and standard deviation determines candidate thresholds, then change according to prior informations such as change point ratios In generation, determines final segmentation threshold.Specifically comprise the following steps that

(1) initiation parameter.If presetting change point ratio is M, iterations is

(2) mean μ and the standard deviation sigma of whole differential image are calculated；

(3) determine that the threshold value in this iterative cycles is

(4) row threshold division is entered to differential image, determine actual change ratio M '；

(5) if being unsatisfactory for stopping criterion for iteration M '≤M, then updateAnd repeat step (3-5), until meet M '≤ M, now exports segmentation threshold Th.

The present invention utilizes the difference degree combining corresponding target in phase images when weighting polarization difference degree describes two, this survey Degree can by adjust weight coefficient size realize difference type stress select.First the pole of corresponding target when two is calculated Changing scattering diversity factor and polar power diversity factor, the weight coefficient selection scheme then proposing according to the present invention determines weighting system Number, weights polarization difference degree by obtaining combining between corresponding target after two parts diversity factor weighted sum, finally utilizes threshold value to divide The technology of cutting extracts region of variation, it is achieved change detection.Present invention proposition is described below weights polarization difference degree based on combining The implementation process of Polarimetric SAR Image change detecting method.

1. pre-process.

The operation of this step mainly includes carrying out orientation and Speckle noise removal to the two phase Polarimetric SAR Images having registrated Operation, the impact on change testing result with this random orientation reducing ground object target and coherent speckle noise.It is known that polarization The coherent speckle noise of SAR image is inevitable, therefore, needed to carry out coherent spot to image and pressed down before being changed detection System, namely traditional denoising, to reduce the impact on experimental result for the coherent speckle noise.The present invention utilizes 2006 The fall spot algorithm based on scattering model that Lee proposes carries out Speckle reduction to image.Orientation is gone to refer to remove target orientation pair The impact of scattering and the substantive characteristics that highlights target, two different orientations and the duplicate Scattering Targets of other features, After this operation, its Polarization scattering information should be on all four.Orientation operation will be gone as change detection early stage pretreatment One of process, the impact on experimental result for the random orientation, the beneficially reliability of Enhancement test result can be prevented effectively from And accuracy.

2. the characteristic vector k of corresponding pixel points in phase images when constructing two_AiAnd k_Bi。

The gyromagnetic variable polarization effect of target generally uses the polarization coherence matrix T of all polarization information containing target to carry out table Levying, its expression-form is as follows:

$T=\left[\begin{array}{ccc}{T}_{11}& T_{12}& T_{13}\\ T_{21}& T_{22}& T_{23}\\ T_{31}& T_{32}& T_{33}\end{array}\right]=\left[\begin{array}{ccc}{T}_{11}& T_{12}& T_{13}\\ T_{12}^{*}& T_{22}& T_{23}\\ T_{13}^{*}& T_{23}^{*}& T_{33}\end{array}\right]$

Wherein, T_mnRepresenting the element in polarization coherence matrix T, subscript * represents conjugate operation.

T matrix vector is obtained:

K=[T₁₁ T₁₂ T₁₃ T₂₁ T₂₂ T₂₃ T₃₁ T₃₂ T₃₃]^T

Wherein, subscript T represents transposition.Hermite matrix in view of T matrix, i.e. the upper triangle element and lower three of matrix Angle element meets the relation being conjugated each other, therefore, it can only utilize the upper Order Triangular Elements usually structural feature vector of T matrix, its table Reach form as follows:

K=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

Obviously, vector K and vector k comprises identical target information, but k vector is six-vector, utilizes this vector to carry out Operand can be reduced during calculating, improve operation efficiency.Corresponding in two phase Polarimetric SAR Images, we are denoted as down Formula:

k_Ai=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

k_Bi=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

Wherein, k_Ai、k_BiRepresent pixel i in characteristic vector in phase polarization SAR data for A, the B two respectively.

3. calculate scattering diversity factor and power difference degree.

If the characteristic vector of two phase Polarimetric SAR Image respective pixel is respectively k_AiAnd k_Bi, corresponding antenna receives power It is respectively P_AiAnd P_Bi, define the Polarization scattering diversity factor between two targets and polar power diversity factor be respectively as follows:

$D_1=1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2}$

$D_2=1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}}$

Wherein, | | | |₂Representing 2 norms of vector, subscript H represents conjugation transposition computing.Polarization scattering difference describes Being the correlation of two coherence matrixes, what polar power difference described is the echo power difference of two targets.

4. calculate and combine weighting polarization difference degree.

The corresponding weight coefficient of relative size distribution according to two species diversity degree, obtains associating weighting polarization difference after summation Degree, constructs differential image.Combining weighting polarization difference degree is:

$d=a\·D_1+b\·D_2=a\·(1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2})+b\·(1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}})$

Wherein, a, b are weight coefficient and meet a+b=1, and a, b determine the tribute to overall diversity factor for the two species diversity types Offer size.What polarization difference degree described is the difference degree between two targets, diversity factor numerical value two targets of bigger explanation it Between difference bigger；Otherwise, diversity factor numerical value is less, then the difference degree of two targets is less.

The numerical characteristic of the scattering difference of statistical comparison multi-group data and power difference finds, and not all pixel is all deposited At the unified difference type being definitely dominant, therefore give all pixels and order the same weight coefficient and can lose a part of suboptimum The information of difference type.In consideration of it, the present invention proposes a kind of new weight coefficient allocative decision.Theoretically, slice is being become In the case that part is constant, scattering properties and power characteristic that same ground object target shows in phase images when difference should be identical 's.When two, the ground object target of phase images same position changes, its scattering properties and power characteristic also should be sent out accordingly Changing, but intensity of variation is not necessarily identical, it is possible to and scattering properties change is substantially, it is also possible to power characteristic change is substantially. Therefore, should consider according to the relative size compromise of two species diversity when weights assigned coefficient a, b.Based on above-mentioned consideration, we are true Fixed weight coefficient selection scheme is as follows:

Work as D₁＞ D₂When, the principle of the weight coefficient of distribution is: a ＜ 0.5, b ＞ 0.5；

Work as D₁≤D₂When, the principle of the weight coefficient of distribution is: a ＞ 0.5, b ＜ 0.5.

5. Threshold segmentation extracts region of variation.

Integrated data histogram distribution that this step proposes first with the present invention and the Automatic thresholding of change point ratio Method determines segmentation threshold, then utilizes this segmentation threshold, carries out to differential image splitting and obtains final change testing result.

The present invention proposes the automatic Selection of Image Threshold of a kind of integrated data histogram distribution and change point ratio, and it is main Want thought to be the average by differential image and standard deviation determines candidate thresholds, then change according to prior informations such as change point ratios In generation, determines final segmentation threshold.This threshold selection method is inspired by the confidential interval concept of Gaussian Profile, differential image equal Value characterizes the central distribution situation of whole measures of dispersion to be split, but the region of variation in experimental data is often less, directly profit Often cause substantial amounts of false-alarm by average as segmentation threshold, it is therefore desirable to add up on the basis of average standard deviation, as The standard deviation of cumulative several times properly can be determined according to change point ratio.Specifically comprise the following steps that

(1) initiation parameter.If presetting change point ratio is M, iterations is

(2) mean μ and the standard deviation sigma of whole differential image are calculated；

(3) determine that the threshold value in this iterative cycles is

(4) row threshold division is entered to differential image, determine actual change ratio M '；

(5) if being unsatisfactory for stopping criterion for iteration M '≤M, then updateAnd repeat step (3-5), until meet M '≤ M, now exports segmentation threshold Th.

Explanation experiment effect below in conjunction with the accompanying drawings:

The Pauli that Fig. 1 gives the two phase polarization SAR data gathering in California, USA Jin Si county overhead divides Xie Tu, left figure is collected on May 19th, 2011, and right figure is collected on May 20th, 2011.Due to the agriculture that mid-May is local just Busy saves, although therefore data only mutually every two days, there is also the obvious change that the irrigation due to crops, sowing and farming cause Change, figure is labelled with more obvious region of variation at 7.Fig. 2 and Fig. 3 gives the differential image of the inventive method and final Change testing result, it can be seen that the present invention can efficiently extract region of variation, and false-alarm is few, clear-cut.

Claims (1)

1. a polarization synthetic aperture radar image change detecting method, is characterized in that, step is as follows:

(1) pre-process, carry out orientation and Speckle noise removal to the two phase polarization synthetic aperture radar images having registrated Operation；

(2) the characteristic vector k of corresponding pixel points in phase images when constructing two_AiAnd k_Bi

k_Ai=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

k_Bi=[T₁₁ T₁₂ T₁₃ T₂₂ T₂₃ T₃₃]^T

Wherein, k_Ai、k_BiRepresent pixel i at characteristic vector in phase polarimetric synthetic aperture radar data for A, the B two, T respectively_mn Representing the element in the coherence matrix T of Polarimetric SAR Image, subscript T represents transposition；

(3) the Polarization scattering diversity factor of phase images corresponding pixel points and polar power diversity factor when calculating two

If two clarification of objective vectors are respectively k_AiAnd k_Bi, corresponding antenna receives power and is respectively P_AiAnd P_Bi, define two Polarization scattering diversity factor between target and polar power diversity factor are respectively as follows:

$D_1=1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2}$

$D_2=1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}}$

Wherein, | | | |₂Representing 2 norms of vector, subscript H represents conjugation transposition；

(4) the corresponding weight coefficient of relative size distribution according to two species diversity degree, obtains associating weighting polarization difference after summation Degree, constructs differential image, combines weighting polarization difference degree and is:

$d=a\·D_1+b\·D_2=a\·(1-\frac{\left|{k_{Ai}}^H{k}_{Bi}\right|}{\left|\right|k_{Ai}\left|{|}_2\right|\left|k_{Bi}\right|{|}_2})+b\·(1-\frac{2}{P_{Ai}/P_{Bi}+P_{Bi}/P_{Ai}})$

Wherein, a, b are weight coefficient and meet a+b=1, and it is big to the contribution of overall diversity factor that a, b determine two species diversity types Little；

(5) enter row threshold division to differential image and extract region of variation: first with the change detection i.e. differential image of characteristic quantity Average and standard deviation determine candidate thresholds, then account for final the dividing of ratio-dependent of all pixel numbers according to change point number Cut threshold value, finally utilize threshold segmentation method to extract region of variation.