CN106778884A - The Classification of Polarimetric SAR Image method of plane is decomposed based on scattering entropy and three-component - Google Patents

Abstract

The present invention provides a kind of Classification of Polarimetric SAR Image method that plane is decomposed based on scattering entropy and three-component, is related to radar image processing technology field.Scattering entropy and Freeman three-components are 9 classes to Target scalar initial division by the method, and calculate cluster centre according to 9 classifications, Target scalar cluster is arrived by desired number according to Wishart distances, atural object is divided into by entropy scattering atural object high, middle entropy scattering atural object and low entropy scattering atural object according to scattering entropy H first, 3 classifications are divided into 9 class atural objects by surface scattering, even scattering and volume scattering respectively again, preliminary classification is further divided with Wishart graders then.The present invention can carry out more accurately classification to Polarimetric SAR Image, without approximate in processing procedure, can preferably retain detailed information, greatly reduce the mistake classification of Urban Areas.

Description

The Classification of Polarimetric SAR Image method of plane is decomposed based on scattering entropy and three-component

Technical field

Plane is decomposed based on scattering entropy and three-component the present invention relates to radar image processing technology field, more particularly to one kind Classification of Polarimetric SAR Image method.

Background technology

Polarimetric synthetic aperture radar (Polarimetric synthetic aperture radar, abbreviation PolSAR or pole Change SAR) as a kind of space flight of active, air remote sensing means, the characteristics of with round-the-clock, all weather operations, environmental protection, The aspects such as disaster monitoring, oceanographic observation, resource exploration, precision agriculture, geological mapping, government's public decision making have widely should With.

Atural object and land use classes are polarimetric synthetic aperture radar applications main, for the supervision of terrain classification Classification emerges in an endless stream with unsupervised classification algorithm.Wherein unsupervised classification can be summarized as three classes, and first kind method is merely with SAR The statistical nature of data；Equations of The Second Kind method is that SAR data is classified using inherent physical scatterers characteristic, this method Advantage there is provided type identification information, but classification chart picture general lack of detailed information；3rd class method combines the system of data Meter feature and physical scatterers feature, therefore most effectively SAR data can be classified, current 3rd class method is that hot topic is ground Study carefully.

In recent years, a large amount of Classification of Polarimetric SAR Image methods are suggested.If Jong-Sen Lee are in Unsupervised The texts of TerrainClassification Preserving Polarimetric Scattering Characteristics mono- The sorting algorithm mentioned, it is three kinds of scattering types, Ran Hou that the algorithm is decomposed picture breakdown first with Freeman-Durden Using Wishart grader Iterative classifications.Although the algorithm remains the main scattering properties and classification results of terrestrial object information Convergence stability, but still have the following disadvantages：First, axial symmetry decomposes Freeman-Durden has invariable rotary Property, mistake may be caused to classify relative to the change of radar observation direction target orientation；Next, rough surface is mistaken for body and dissipates Penetrate；Again, vegetation region is mistaken for surface scattering；Finally, mistake classification caused by mobile target.

The content of the invention

For the defect of prior art, the present invention provides a kind of polarization SAR that plane is decomposed based on scattering entropy and three-component Image classification method, further according to scattered power exhaustive division on the basis of being divided according to scattering entropy, improves nicety of grading, energy Preferably retain detailed information, greatly reduce the mistake classification of Urban Areas.

A kind of Classification of Polarimetric SAR Image method that plane is decomposed based on scattering entropy and three-component, is comprised the following steps：

Step 1：Input polarization SAR data to be sorted, and Lee filtering process is carried out, obtain the data after denoising；

Step 2：The coherence matrix of each pixel after calculating denoising, in the case of singly station back scattering mechanism, meets mutual Easy condition, coherence matrix is three-dimensional matrice T₃；

When vector quantization is carried out to collision matrix S using Pauli bases, target scattering vector k_3PFor

The polarization coherence matrix T of target₃It is shown below：

Wherein, S_hhAnd S_vvIt is same polarization passage echo power, S_hhRepresent horizontal polarization passage echo power, S_vvRepresent and hang down Straight POLARIZATION CHANNEL echo power, S_hvAnd S_vhIt is cross polarization passage echo power, H represents conjugate transposition, and * is to seek conjugation；

Step 3：To the coherence matrix T after denoising₃Cloude decomposition is carried out, 3 non-negative characteristic values and corresponding spy is obtained Vector is levied, so as to obtain the scattering entropy H of corresponding pixel points；

Step 3.1：By solving coherence matrix T₃Characteristic value and characteristic vector, by coherence matrix T₃Resolve into three solely Vertical coherence matrix T_0i(i=1,2,3) sum, obtains 3 orthogonal targets, and its decomposition formula is shown below：

Wherein, λ_iAnd e_iCharacteristic value and characteristic vector are represented respectively, and H represents conjugate transposition, e_iRepresentation such as following formula institute Show：

Wherein,Represent the absolute phase information of target, α_i、β_i、δ_iAnd γ_iIt is Polarization scattering parameter；

Step 3.2：The characteristic value and characteristic vector obtained according to decomposition, obtain the computing formula such as following formula institute of scattering entropy H Show：

Wherein, P_iRepresent and correspond to by eigenvalue λ_iThe pseudo- probability of acquisition, P_i=λ_i/(λ₁+λ₂+λ₃), the value of scattering entropy H Scope is 0≤H≤1；

Step 4：Difference according to scattering entropy divides an image into three kinds of entropy region high, middle entropy region and low entropy region, Wherein H≤0.5 is low entropy region, and 0.5 ＜ H≤0.9 is middle entropy region, and H ＞ 0.9 are entropy region high；

Step 5：Solve Terrain Scattering power in ready-portioned three kinds of area types more than respectively, and judge that its scattering is accounted for Excellent type；According to main scattering mechanism (surface scattering power P_s, even scattered power P_d, volume scattering power P_vIn the maximum) Image is further divided into 9 scattering classifications, entropy surface scattering respectively high, entropy high on the basis of step 4 has divided 3 classes Even scattering, entropy volume scattering high, middle entropy surface scattering, the scattering of middle entropy even, middle entropy volume scattering, low entropy surface scattering, low entropy idol Secondary scattering, low entropy volume scattering；

Wherein, each scattered power computing formula is shown below：

Wherein, P_s、P_d、P_vSurface scattering power, even scattered power, volume scattering power, f are represented respectively_s、f_d、f_vRespectively The contribution of surface scattering component, even scattering component, volume scattering component to general power is represented, β represents the mould of bragg scatterings Type coefficient, α represents the amplitude fading and phase place change parameter of the vertically polarized wave and horizontal polarized wave during radar return；

Step 6：Every kind of scattering classification is a kind of cluster, and the average coherence matrix of above-mentioned 9 kinds of clusters is calculated respectively, and makees It is corresponding cluster centre；

Average coherence matrix is shown below：

Wherein, Z represents average coherence matrix, and n represents that vector u (k) is k-th sample of haplopia coherence matrix regarding number；

Step 7：9 scattering classifications of preliminary classification are clustered again according to Wishart distance metrics formula, to reach more It is preferable classification results, calculates each pixel in image it the distance between with all kinds of centers, and the pixel is drawn Assign in that minimum class；

Wishart distance metric formula are shown below：

Wherein, ω_mRepresent class, C_mRepresent such corresponding covariance matrix；

Step 8：Each classification after for clustering again, the scattering classification marked according to it is coloured.

As shown from the above technical solution, the beneficial effects of the present invention are：The present invention provide one kind be based on scattering entropy and Three-component decomposes the Classification of Polarimetric SAR Image method of plane, more accurately classification can be carried out to Polarimetric SAR Image, at place Without approximate during reason.It is 9 classes to Target scalar initial division by scattering entropy and Freeman three-components, and according to 9 classifications Cluster centre is calculated, Target scalar cluster is arrived by desired number according to Wishart distances.First according to scattering entropy H by atural object It is divided into entropy scattering atural object high, middle scattering atural object and low entropy scattering atural object, then respectively by 3 classifications by surface scattering, even scattering 9 class atural objects are divided into volume scattering, preliminary classification is further divided with Wishart graders then, can be preferably Reservation detailed information, greatly reduce Urban Areas mistake classification.

Brief description of the drawings

Fig. 1 is original polarization SAR image power diagram provided in an embodiment of the present invention；

Fig. 2 is the Classification of Polarimetric SAR Image side that plane is decomposed based on scattering entropy and three-component provided in an embodiment of the present invention Method flow chart；

Fig. 3 carries out the filtered SAR images of Lee to Fig. 1 for provided in an embodiment of the present invention；

Fig. 4 is the simulation result figure to Fig. 3 preliminary classifications provided in an embodiment of the present invention；

Fig. 5 is the simulation result figure to Fig. 3 final classifications provided in an embodiment of the present invention.

Specific embodiment

With reference to the accompanying drawings and examples, specific embodiment of the invention is described in further detail.Hereinafter implement Example is not limited to the scope of the present invention for illustrating the present invention.

As shown in figure 1, being the original polarization SAR image power diagram of the present embodiment, scattering entropy and three points are based on using one kind The SAR image sorting technique that amount decomposes plane is processed the image, and its flow is as shown in Fig. 2 specific method is as described below.

Step 1：Input polarization SAR data to be sorted, and Lee filtering process is carried out, obtain the data after denoising.This reality In applying example, the original image of Fig. 1 is carried out after denoising as shown in Figure 3.

Step 2：The coherence matrix of each pixel after calculating denoising.

Vector quantization is carried out to collision matrix S using Pauli bases, the advantage for carrying out matrix vector using Pauli bases is The element of resulting Scattering of Vector is convenient to explain physical scatterers mechanism.In the situation of singly station back scattering mechanism Under, meet reciprocity condition, now coherence matrix is three-dimensional matrice T₃, target scattering vector k_3PFor

It can thus be concluded that the polarization coherence matrix T of target₃It is shown below：

In formula, S_hhAnd S_vvIt is same polarization passage echo power, S_hhRepresent horizontal polarization passage echo power, S_vvRepresent and hang down Straight POLARIZATION CHANNEL echo power, S_hvAnd S_vhIt is cross polarization passage echo power, H represents conjugate transposition, and * is to seek conjugation.

Step 3：To the coherence matrix T after denoising₃Cloude decomposition is carried out, 3 non-negative characteristic values and corresponding spy is obtained Vector is levied, so as to obtain the scattering entropy H of corresponding pixel points；

Step 3.1：By solving coherence matrix T₃Characteristic value and characteristic vector, by coherence matrix T₃Resolve into three solely Vertical coherence matrix T_0i(i=1,2,3) sum, obtains 3 orthogonal targets, and its decomposition formula is shown below：

Wherein, λ_iAnd e_iCharacteristic value and characteristic vector are represented respectively, and H represents conjugate transposition, e_iRepresentation such as following formula institute Show：

Wherein,Represent the absolute phase information of target, α_i、β_i、δ_iAnd γ_iIt is Polarization scattering parameter；

Step 3.2：The characteristic value and characteristic vector obtained according to decomposition, obtain the computing formula such as following formula institute of scattering entropy H Show：

Wherein, P_iRepresent and correspond to by eigenvalue λ_iThe pseudo- probability of acquisition, P_i=λ_i/(λ₁+λ₂+λ₃), the value of scattering entropy H Scope is 0≤H≤1.

Step 4：Difference according to scattering entropy divides an image into three kinds of entropy region high, middle entropy region and low entropy region, Wherein H≤0.5 is low entropy region, and 0.5 ＜ H≤0.9 is middle entropy region, and H ＞ 0.9 are entropy region high.

Step 5：Solve Terrain Scattering power in ready-portioned three kinds of area types more than respectively, and judge that its scattering is accounted for Excellent type (referring to step 4 ready-portioned 3 kinds of regions).According to main scattering mechanism (surface scattering power P_s, even scattered power P_d, volume scattering power P_vIn the maximum) by image Preliminary division be 9 scattering classifications, entropy surface scattering respectively high, entropy high Even scattering, entropy volume scattering high, middle entropy surface scattering, the scattering of middle entropy even, middle entropy volume scattering, low entropy surface scattering, low entropy idol Secondary scattering, low entropy volume scattering.

In the present embodiment, result after preliminary classification as shown in figure 4, with Fig. 1 and Freeman-Durden classification results pair Than understanding：The appearance of negative power value is avoided after the classification of the present embodiment, the classifying quality of water area is better than Freeman-Durden classification results, but the two precision is attained by more than 90%；Due to the difference of building orientation, Freeman-Durden classification is more serious to wrong point of phenomenon of building, and the present embodiment method makes moderate progress, but still has identification not Place out；In road area, road and bridge are considered as a class by the present embodiment；Freeman-Durden sorting techniques are basic On road and other atural objects are not distinguished, mainly road mistake is divided into the classification of vegetation and building, this implementation Example road k-path partition is correct, and the wrong point of phenomenon of leakage point is not occurred；In vegetation area, the present embodiment nicety of grading also has and preferably changes It is kind, but vegetation and the other profile information of the species of building two be not obvious.

Wherein, the computing formula of each scattered power is shown below：

Wherein, P_s、P_d、P_vSurface scattering power, even scattered power, volume scattering power, f are represented respectively_s、f_d、f_vRespectively The contribution of surface scattering component, even scattering component, volume scattering component to general power is represented, β represents the mould of bragg scatterings Type coefficient, α represents the amplitude fading and phase place change parameter of the vertically polarized wave and horizontal polarized wave during radar return；

Because each pixel of image can obtain P_s、P_d、P_vThree values, take the maximum and are occurred as the pixel Scattering type.

Step 6：Every kind of scattering classification is a kind of cluster, and the average coherence matrix of above-mentioned 9 kinds of clusters is calculated respectively, and makees It is corresponding class center；

The result being averaging processing to coherence matrix is equivalent to the result that Stoke matrixes are averaging processing, is concerned with Matrix obeys multiple Wishart distributions, and polarization SAR multiple look processing is that some incoherent haplopia coherence matrixes are carried out averagely, to put down Equal coherence matrix is shown below：

In formula, Z represents average coherence matrix, and n is represented depending on number (being assigned to the pixel count of a certain cluster classification), vector u K () is k-th sample of haplopia coherence matrix.

Step 7：9 scattering classifications of preliminary classification are clustered again according to Wishart distance metrics formula, to reach more It is preferable classification results.It is calculated each pixel in image the distance between with all kinds of centers, and the pixel is drawn Assign in that minimum class, then the purpose of subseries is that the data of misclassifications are entered individually in above-mentioned preliminary classification Row is classified again.

N is shown below depending on the distance measure formula that PolSAR data are classified：

In formula, ω_mRepresent class, P (ω_m) represent ω_mThe prior probability of class, C_mRepresent such corresponding covariance matrix； For the polarization SAR data all unknown to all kinds of prior probabilities, it is assumed that P (ω_m) it is identical, distance metric now with regard number n without Close, obtain Wishart distance metric formula and be shown below：

Step 8：Each classification after for clustering again, the scattering classification marked according to it is coloured.Final classification Rear image as shown in figure 5, compared with Fig. 4, water area and classification of road the DeGrain change after Wishart iteration, Vegetation and the unconspicuous situation of construction zone profile have clear improvement, and each atural object nicety of grading is as shown in table 1.

Each atural object nicety of grading of table 1

Project	Water body	Road	Building	Vegetation
					Freeman-Durden	90.14%	41.21%	44.27%	58.31%
Preliminary classification result of the present invention	92.09%	76.98%	58.35%	80.71%
					Iterative classification result of the present invention	92.47%	77.85%	68.74%	88.12%

The classification decomposed based on Freeman three-components can keep atural object physical message, but lack detailed information, especially exist The classification of Urban Areas.Because downtown areas are influenceed by building and vegetation scattering type difference, Freeman tri- is directly utilized Scattering mechanism classification in component decomposition carries out classification and has following difficulty to Urban Areas：(1) different type of ground objects performances Be scattering mechanism of the same race, such as smooth smooth road of clearing the way of building object plane peace, water surface atural object show as surface scattering, wall and Even scattering between ground and trunk and trunk；(2) type of ground objects of the same race shows as different scattering mechanisms, such as different directions Angle scattering mechanism between building such as direction of building and radar bearing between can show as even scattering and Volume scattering, the foliage portion of vegetation can occur surface scattering, even scattering occur, between leaf and limb between limb and ground Generation volume scattering.Therefore, the classification to complicated downtown areas can not only consider the scattering mechanism of target, should also be with statistics priori Knowledge is combined, and improves classification capacity.

Scattering entropy and Freeman three-components are 9 classes to Target scalar initial division by the present invention, and according to 9 classification meters Cluster centre is calculated, Target scalar cluster is arrived by desired number according to Wishart distances.Atural object is divided according to scattering entropy H first Be entropy high scattering atural object, middle scattering atural object and low entropy scattering atural object, respectively by 3 classifications by surface scattering, even scattering and Volume scattering is divided into 9 class atural objects altogether, and preliminary classification is further divided with Wishart graders then.Can be preferably Reservation detailed information, greatly reduce Urban Areas mistake classification, with more practice significance.

Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although The present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those within the art that：It still may be used Modified with to the technical scheme described in previous embodiment, or which part or all technical characteristic are equal to Replace；And these modifications or replacement, the essence of appropriate technical solution is departed from the model that the claims in the present invention are limited Enclose.

Claims (1)

1. a kind of Classification of Polarimetric SAR Image method that plane is decomposed based on scattering entropy and three-component, it is characterised in that：The method bag Include following steps：

Step 1：Input polarization SAR data to be sorted, and Lee filtering process is carried out, obtain the data after denoising；

Step 2：The coherence matrix of each pixel after calculating denoising, in the case of singly station back scattering mechanism, meets reciprocity bar Part, coherence matrix is three-dimensional matrice T₃；

When vector quantization is carried out to collision matrix S using Pauli bases, target scattering vector k_3PFor

$k_{3P}=\frac{1}{\sqrt{2}}{\left[\begin{array}{ccc}{S}_{hh}+S_{vv}& S_{hh}-S_{vv}& 2S_{hv}\end{array}\right]}^T;$

The polarization coherence matrix T of target₃It is shown below：

$T_3=k_{3P}{k}_{3P}^H=\frac{1}{2}\left[\begin{array}{ccc}|S_{hh}+S_{vv}{|}^2& (S_{hh}+S_{vv}){(S_{hh}-S_{vv})}^{*}& 2(S_{hh}+S_{vv})S_{hv}^{*}\\ (S_{hh}-S_{vv}){(S_{hh}+S_{vv})}^{*}& |S_{hh}-S_{vv}{|}^2& 2(S_{hh}-S_{vv})S_{hv}^{*}\\ 2{(S_{hh}+S_{vv})}^{*}{S}_{hv}& 2S_{hv}{(S_{hh}-S_{vv})}^{*}& 4|S_{hv}{|}^2\end{array}\right];$

Wherein, S_hhAnd S_vvIt is same polarization passage echo power, S_hhRepresent horizontal polarization passage echo power, S_vvRepresent vertical pole Change passage echo power, S_hvAnd S_vhIt is cross polarization passage echo power, H represents conjugate transposition, and * is to seek conjugation；

Step 3：To the coherence matrix T after denoising₃Cloude decomposition is carried out, 3 non-negative characteristic values and corresponding Characteristic Vectors are obtained Amount, so as to obtain the scattering entropy H of corresponding pixel points；

Step 3.1：By solving coherence matrix T₃Characteristic value and characteristic vector, by coherence matrix T₃Resolve into three it is independent Coherence matrix T_0i(i=1,2,3) sum, obtains 3 orthogonal targets, and its decomposition formula is shown below：

$T_3=\underset{i=1}{\overset{3}{\Σ}}{\mathrm{\λ}}_i{e}_i\·e_i^H=T_{01}+T_{02}+T_{03};$

Wherein, λ_iAnd e_iCharacteristic value and characteristic vector are represented respectively, and H represents conjugate transposition, e_iRepresentation be shown below：

Wherein,Represent the absolute phase information of target, α_i、β_i、δ_iAnd γ_iIt is Polarization scattering parameter；

Step 3.2：The characteristic value and characteristic vector obtained according to decomposition, the computing formula for obtaining scattering entropy H are shown below：

$H=\underset{i=1}{\overset{3}{\Σ}}(-P_i{\log }_3{P}_i);$

Wherein, P_iRepresent and correspond to by eigenvalue λ_iThe pseudo- probability of acquisition, P_i=λ_i/(λ₁+λ₂+λ₃), the span of scattering entropy H It is 0≤H≤1；

Step 4：Difference according to scattering entropy divides an image into three kinds of entropy region high, middle entropy region and low entropy region, wherein H ≤ 0.5 is low entropy region, and 0.5 ＜ H≤0.9 is middle entropy region, and H ＞ 0.9 are entropy region high；

Step 5：Solve Terrain Scattering power in ready-portioned three kinds of area types more than respectively, and judge that it scatters the class that is dominant Type；According to main scattering mechanism (surface scattering power P_s, even scattered power P_d, volume scattering power P_vIn the maximum) will figure 9 scattering classifications, entropy surface scattering respectively high, entropy even high are further divided on the basis of as dividing 3 classes in step 4 Scattering, entropy volume scattering high, middle entropy surface scattering, the scattering of middle entropy even, middle entropy volume scattering, low entropy surface scattering, low entropy even dissipate Penetrate, low entropy volume scattering；

Wherein, each scattered power computing formula is shown below：

$\left\{\begin{array}{c}{P}_s=f_s(1+|\mathrm{\β}{|}^2)\\ P_d=f_d(1+|\mathrm{\α}{|}^2)\\ P_v=f_v\end{array}\right.;$

Wherein, P_s、P_d、P_vSurface scattering power, even scattered power, volume scattering power, f are represented respectively_s、f_d、f_vRepresent respectively , to the contribution of general power, β represents the model system of bragg scatterings for surface scattering component, even scattering component, volume scattering component Number, α represents the amplitude fading and phase place change parameter of the vertically polarized wave and horizontal polarized wave during radar return；

Step 6：Every kind of scattering classification is a kind of cluster, and the average coherence matrix of above-mentioned 9 kinds of clusters is calculated respectively, and as phase The cluster centre answered；

Average coherence matrix is shown below：

$Z=\frac{1}{n}\underset{k=1}{\overset{n}{\Σ}}u\left(k\right)u{\left(k\right)}^H;$

Wherein, Z represents average coherence matrix, and n represents that vector u (k) is k-th sample of haplopia coherence matrix regarding number；

Step 7：9 scattering classifications of preliminary classification are clustered again according to Wishart distance metrics formula, is more managed with reaching The classification results thought, calculate each pixel in image it the distance between with all kinds of centers, and the pixel is divided into In that minimum class of distance；

Wishart distance metric formula are shown below：

$d_2(Z,{\mathrm{\ω}}_m)=ln\left|C_m\right|+Tr\left(C_m^{-1}Z\right);$

Wherein, ω_mRepresent class, C_mRepresent such corresponding covariance matrix；

Step 8：Each classification after for clustering again, the scattering classification marked according to it is coloured.