CN101464956A - Method for classifying polarization synthetic aperture radar image based on sub-aperture analysis - Google Patents

Abstract

A method for classifying radar images of polarization synthetic apertures based on sub-aperture analysis belongs to the technical field of radar imagery, which relates to a polarization synthetic aperture radar (PolSAR) imagery processing technology. The classifying method is characterized in that the initial classification of SAR images polarized in the entire resolution ratio is primarily performed by adopting an H/Alpha plane; then the sub-aperture decomposition of SAR images polarized in the entire resolution ratio is performed; then a center of an initial category Vm is calculated through the coherent matrix of sub-aperture images; the distance measure of all pixels and each class center in all the sub-aperture images is calculated, and the pixels are categorized to the class with the minimum distance measure; finally, whether the iteration terminal conditions are not met or not, if the terminal conditions are met, the iteration is terminated; if the terminal conditions are not met, the iteration is continuously performed by returning to step 2. The invention can synthesize scattering properties of the target and the change of the scattering properties at different visual angles, and the classification precision is improved, so as to bring better classifying effect.

Description

A kind of polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture

Technical field

The invention belongs to the radar imagery technical field, relate to polarimetric synthetic aperture radar (PolSAR) image processing techniques, particularly high precision polarization SAR image classification technology.

Background technology

Full polarimetric SAR is for the information processing in the radar image and obtain the target information that provides abundant more.Utilize complete polarization SAR echo data, can classify the target scattering type.At present the full polarimetric SAR sorting technique is mainly divided two big classes by given data type: based on the classification of scattering coefficient matrix with based on the classification of coherence matrix or covariance matrix.What haplopia SAR data generally provided is the scattering coefficient matrix, can under level and vertical polarization base, utilize its element to carry out the Pauli decomposition, image is classified, the decomposition method that also can utilize Krogager and cameron to propose decomposes under the circular polarisation base, in most cases uses Pauli to decompose.Look data more and generally provide coherence matrix or covariance matrix, coherence matrix is carried out svd, become series of parameters with eigenwert with combination of eigenvectors, these parameters and scattering physical mechanism are closely related, interval according to parameter can be classified to the target on the image, as the H/ alpha taxonomy.Behind the H/ α preliminary classification,, carry out the Wishart iteration, can obtain classification results more completely according to the statistical property of Polarization scattering data.

Based on full resolution polarization SAR classification of Data method: promptly the main technical schemes of H/ α/Wishart sorting technique is as follows.

Coherence matrix is carried out feature decomposition to be got:

$T={\mathrm{<figure-callout\; id="3"\; label="U"\; filenames="A200910058210E00101.png"\; state="\{\{state\}\}">U</figure-callout>}}_3\left[\begin{array}{ccc}{\mathrm{\&lambda;}}_1& 0& 0\\ 0& {\mathrm{\&lambda;}}_2& 0\\ 0& 0& {\mathrm{\&lambda;}}_3\end{array}\right]U_3^H---\left(1\right)$

Wherein H represents conjugate transpose, matrix U ₃Can parametrization be expressed as:

$U_3=\left[\begin{array}{ccc}\cos \left({\mathrm{\&alpha;}}_1\right)e^{{\mathrm{i\&phi;}}_1}& \cos \left({\mathrm{\&alpha;}}_2\right)e^{{\mathrm{i\&phi;}}_2}& \cos \left({\mathrm{\&alpha;}}_3\right)e^{{\mathrm{i\&phi;}}_3}\\ \sin \left({\mathrm{\&alpha;}}_1\right)\cos \left({\mathrm{\&beta;}}_1\right)e^{{\mathrm{i\&delta;}}_1}& \sin \left({\mathrm{\&alpha;}}_2\right)\cos \left({\mathrm{\&beta;}}_2\right)e^{{\mathrm{i\&delta;}}_2}& \sin \left({\mathrm{\&alpha;}}_3\right)\cos \left({\mathrm{\&beta;}}_3\right)e^{{\mathrm{i\&delta;}}_3}\\ \sin \left({\mathrm{\&alpha;}}_1\right)\sin \left({\mathrm{\&beta;}}_1\right)e^{{\mathrm{i\&gamma;}}_1}& \sin \left({\mathrm{\&alpha;}}_2\right)\sin \left({\mathrm{\&beta;}}_2\right)e^{{\mathrm{i\&gamma;}}_2}& \sin \left({\mathrm{\&alpha;}}_3\right)\sin \left({\mathrm{\&beta;}}_3\right)e^{{\mathrm{i\&gamma;}}_3}\end{array}\right]---\left(2\right)$

In the above-mentioned decomposition, λ ₁〉=λ ₂〉=λ ₃Be the real character value of T matrix, matrix U ₃Be unitary matrix, its column vector promptly is the proper vector of coherence matrix T.The probability P that each vector occurs _iExpression, P _iCan calculate with following formula:

$P_i=\frac{{\mathrm{\&lambda;}}_i}{\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_j}---\left(3\right)$

For the degree of the scattering mechanism that characterizes target chaotic on statistics, can introduce entropy H, it is defined as follows:

$H=-\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}{P}_i{\log }_3\left(P_i\right)---\left(4\right)$

If entropy is smaller, illustrate that depolarisation effect in the scattering process a little less than, this moment, eigenvalue of maximum characteristic of correspondence vector can be used as the scattering matrix of prevailing scattering mechanism, other little eigenwert characteristic of correspondence vector can be ignored.If entropy H is bigger, then the target depolarisation effect is stronger, and can not think to have only a kind of equivalent scattering mechanism this moment again, must consider full feature value spectrum.Along with the increase of entropy, fewer and feweri by the scattering classification number that polarization data can distinguish, limiting case is that polarization information becomes 0 when H=1, and target scattering becomes a pure random noise process.

Though the α angle is to occur with angle form, it with target towards haveing nothing to do, what its was represented is the degree of freedom of scattering target internal, is used for representing the type of scattering mechanism.

When carrying out the polarization SAR image classification, the angle of average alpha often of usefulness.That is:

α＝P ₁α ₁+P ₂α ₂+P ₃α ₃ (5)

According to the scope at H and α place, can be on H-α plane be 9 category regions with the atural object Preliminary division, as shown in Figure 3:

(1) regional Z9: low entropy area scattering

Appearance is the α angle less than 42.5 ° low entropy scattering in this zone.Wherein comprise geometrical optics area scattering and physical optics area scattering, the Bragg area scattering and do not relate to HH and the VV passage between the special scattering phenomenon of 180 ° of phase reversals.The water surface of L-band and pattern-band, the ice face of L-band, and very smooth land face all belongs to this class.

(2) regional Z8: low entropy dipole scattering

What this zone occurred is to have the very strong correlation scattering mechanism of large unbalance on HH and VV channel amplitude.As dipole scattering mechanism independently, and strong correlation towards the vegetation scattering of anisotropic scattering unit.This regional width is by the measurement capability decision of radar to POLARIZATION CHANNEL ratio HH/VV.

(3) regional Z7: low entropy repeatedly reflects

This zone is corresponding to the scattering events of or odd number secondary reflection inferior through even number.As isolated scattering insulation or metallicity dihedral angle scattering thing.α in this zone〉47.5 °.

(4) regional Z6: middle constant entropy area scattering

The increase of the entropy that is caused by the scattering surface roughness has been reflected in this zone.In the area scattering theory, the entropy of low frequency scattering such as Bragg scattering is 0, and the entropy of high frequency scattering such as geometrical optics scattering also is 0.Yet, between this two classes scattering limit, because physical characteristics and scattering mechanism that secondary wave is propagated, can cause the increase of entropy, therefore when surfaceness changed, the scattering entropy can increase, as the scattering entropy that comprises the face of flattened ball scattering unit can reach 0.6～0.7.

(5) regional Z5: middle constant entropy vegetation scattering

This zone also is a median size entropy zone, but its main scattering mechanism is the dipole scattering.The increase of this regional internal entropy is because towards the center at angle statistical distribution.The scattering that this zone comprises is a vegetated surface, includes anisotropic scattering unit in the scattering surface, has medium scattering unit towards the angular dependence (-dance) degree.

(6) regional Z4: repeatedly scattering of middle constant entropy

This zone mainly is the dihedral angle scattering with medium scattering entropy.A kind of is in forest, after the electromagnetic wave of pattern-band or L-band passes tree crown, the bireflection scattering mechanism will take place, and wherein the effect of tree crown is exactly the size that has increased the scattering entropy.Another kind is in the urban district, because there is intensive repeatedly scattering center in the urban district.

(7) regional Z3: high entropy area scattering

This class scattering is not in the reasonable zone on H-α plane, because we can not distinguish entropy H〉0.9 area scattering, this is because along with the increase of entropy, classification capacity is worse and worse.This has illustrated that also polarization radar is more effective to low entropy problem.

(8) regional Z2: high entropy vegetation scattering

When α=45 °, in the time of H=0.95, volume scattering (volume scattering) will appear.This class scattering thing can be an anisotropic acicular particles, also can be a low-loss symmetry particle.The scattering entropy of both of these case is all more than 0.9.When entropy continuation increase, then the reasonable zone on the H-α plane is more and more littler, and the scattering of forest tree crown just drops in this zone, and some vegetation cover, and has the scattering surface of the scattering unit formation of height anisotropy behavior, also can drop in this zone.

(9) regional Z1: repeatedly scattering of high entropy

The entropy H of this type of scattering〉0.9, though the scattering entropy than higher, but still can be distinguished the bireflection scattering mechanism.This class scattering can occur in the forest, also can occur in limb and tree crown and grow complete vegetation area.

In the regional Z1-Z9 of Fig. 3, regional Z3 is the classification that impossible occur, and therefore can tentatively target be divided into 8 classes with this classification plane.Above-mentioned classification results is carried out the Wishart iteration, and step is as follows:

(1) with the average of the coherence matrix of each class among the H/ alpha taxonomy result as initial value T _m ⁰(i=0)

(2) with distance measurement method view picture polarization SAR image is classified

(3) calculate T with the sorted pixel that obtains in the 2nd step for each class _m ^I+1

(4) got back to for second step, continue iteration, reach up to user-defined end condition

Can obtain classification results more accurately after the Wishart iteration.

Traditional H/ α/Wishart sorting technique is based on full aperture data (full resolution image), the orientation of not considering target scattering characteristics is to variation, this analysis for target scattering characteristics is coarse, for the variation of the scattering properties of evaluating objects more accurately along with the visual angle, orientation, whole radar aperture can be divided into several sections, to areal imaging respectively on these littler apertures, then one-tenth's image is analyzed, Here it is, and sub-aperture is analyzed.

Summary of the invention

The invention provides a kind of polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture, these method zygote aperture data, the polarization SAR image is classified, can the scattering properties of integration objective under different visual angles, and the variation of scattering properties, improve nicety of grading, bring better classifying quality.

In synthetic process, a target is by the diverse location irradiation along the radar track in the orientation.Therefore, each scatterer is observed under a series of visual angles by SAR.These a series of perspective definition are sub-aperture.Complex target has anisotropic geometry, when from the irradiation of different position, may show different electromagnetism behaviors.Some natural medium has periodic structure (as the crops zone), when perhaps the strong scattering body is linearly aligned, also all shows the polarization characteristic of variation easily.

Detailed technology scheme of the present invention is as follows:

A kind of polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture comprises following concrete steps:

Step 1: adopt H/ α plane that full resolution polarization SAR image is carried out preliminary classification, be divided into into 8 classes (as shown in Figure 2), promptly low entropy area scattering zone Z9, low entropy dipole scattering region Z8, low entropy be reflector space Z7, middle constant entropy area scattering zone Z6, middle constant entropy vegetation scattering region Z5, middle constant entropy scattering region Z4, high entropy vegetation scattering region Z2 and high entropy scattering region Z1 repeatedly repeatedly repeatedly.

Step 2: full resolution polarization SAR image is carried out sub-aperture decompose.The method of decomposing is:

At first, the full aperture view data is transformed into the orientation spectral domain by one dimension Fourier conversion; The then sub-aperture that the full aperture view data of orientation spectral domain is resolved into different numbers according to concrete needs, each sub-aperture comprises certain azimuth coverage; At last, the sub-subaperture image after all sub-aperture data are transformed into time domain and obtain decomposing by anti-Fourier conversion.

Because the orientation to wave number (being exactly the meaning of frequency in fact) and orientation to the pass at visual angle be:

$k_x=\frac{2\mathrm{\&omega;}}{c}\sin \mathrm{\&phi;}$

(ω is a carrier frequency, c is the light velocity), so according to the scope of the corresponding visual angle φ of different wave-number range (Doppler frequency scope), as shown in Figure 2, also the full aperture image can be decomposed on different son spectrums (sub-aperture), obtain a plurality of images of low resolution.

Step 3: according to each preliminary classification of step 1, with the coherence matrix calculating initial category center V of all sub-subaperture images _m:

$V_m=\frac{1}{n_{m}R}\underset{i=1}{\overset{R}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n_m}{\mathrm{\&Sigma;}}}{T}_{\mathrm{ij}}---\left(6\right)$

N wherein _mThe pixel number of representing the m class, R are represented sub-aperture number.

Step 4: the distance measure that calculates each pixel and center of all categories in all sub-subaperture images:

$d(T,V_m)=\ln \left|V_m\right|+\mathrm{tr}\left(V_m^{-1}T\right)---\left(7\right)$

Pixel is ranged that class of distance measure minimum.

Step 5: check whether stopping criterion for iteration reaches, if reach termination of iterations; If do not reach, return step 2 and continue iteration.(the iteration termination condition can be stipulated iterations by the user, also can stipulate the cluster centre that twice iteration obtains

With

Between distance less than certain threshold value)

The invention has the beneficial effects as follows:

A kind of polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture provided by the invention, promptly based on the sorting technique of the H/ α/Wishart in sub-aperture, can the scattering properties of integration objective under different visual angles, and the variation of scattering properties, improve nicety of grading, bring better classifying quality.Compare simple H/ α/Wishart sorting technique based on full resolution SAR image, classification results of the present invention shows the merging better effects if in similar zone, and the boundary in all kinds of zones is more clear; Vegetation area is clearer, and the urban district presents more details.

Description of drawings

The sub-aperture of Fig. 1 SAR radar system analysis principle synoptic diagram.

The corresponding relation figure at Fig. 2 Doppler frequency scope and visual angle.

The H/ α planimetric map of Fig. 3 H/ α/Wishart sorting technique.

The full aperture image of one width of cloth polarization SAR image of Fig. 4 specific embodiment of the invention and three sub-subaperture images.

The classification results of Fig. 5 specific embodiment of the invention.

Fig. 6 tradition is based on the classification results of full resolution image H/ α/Wishart sorting technique to areal among the embodiment.

Embodiment

One width of cloth polarization SAR image is carried out sub-aperture decompose, the power diagram picture before and after the decomposing intensity of each POLARIZATION CHANNEL (be and) as shown in Figure 4.

Use based on the H/ α/Wishart sorting technique of sub-aperture data the polarization SAR image in Foulum area is classified, decompose the decline of the resolution of bringing, only former data decomposition is become three sub-apertures in order to reduce sub-aperture as far as possible.Classification results (iteration 2 times) as shown in Figure 5.

For the match stop effect, we are with traditional classifying to the Foulum area with H/ α/Wishart sorting technique based on full resolution image, and the result as shown in Figure 6.

The classification results of comparison diagram 5 and Fig. 6 as can be seen, a kind of polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture provided by the invention, i.e. branch analogy based on the H/ α/Wishart in sub-aperture has better effect based on H/ α/Wishart classification of full resolution SAR image merely: the merging better effects if in similar zone, and the boundary in all kinds of zones is more clear; Vegetation area is clearer, and the urban district presents more details.

Claims (5)

1, a kind of polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture comprises following concrete steps:

Step 1: adopt H/ α plane that full resolution polarization SAR image is carried out preliminary classification, be divided into into 8 classes, promptly low entropy area scattering zone Z9, low entropy dipole scattering region Z8, low entropy be reflector space Z7, middle constant entropy area scattering zone Z6, middle constant entropy vegetation scattering region Z5, middle constant entropy scattering region Z4, high entropy vegetation scattering region Z2 and high entropy scattering region Z1 repeatedly repeatedly repeatedly:

Step 2: full resolution polarization SAR image is carried out sub-aperture decompose;

Step 3: according to each preliminary classification of step 1, with the coherence matrix calculating initial category center V of all sub-subaperture images _m

$V_m=\frac{1}{n_{m}R}\underset{i=1}{\overset{R}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n_m}{\mathrm{\&Sigma;}}}{T}_{\mathrm{ij}}---\left(6\right)$

N wherein _mThe pixel number of representing the m class, R are represented sub-aperture number;

Step 4: the distance measure that calculates each pixel and center of all categories in all sub-subaperture images:

$d(T,V_m)=\ln \left|V_m\right|+\mathrm{tr}\left(V_m^{-1}T\right)---\left(7\right)$

Pixel is ranged that class of distance measure minimum;

Step 5: check whether stopping criterion for iteration reaches, if reach termination of iterations; If do not reach, return step 2 and continue iteration.

2, the polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture according to claim 1 is characterized in that, the method that step 2 pair full resolution polarization SAR image carries out the concrete decomposition of sub-aperture decomposition is:

At first, the full aperture view data is transformed into the orientation spectral domain by one dimension Fourier conversion; The then sub-aperture that the full aperture view data of orientation spectral domain is resolved into different numbers according to concrete needs, each sub-aperture comprises certain azimuth coverage; At last, the sub-subaperture image after all sub-aperture data are transformed into time domain and obtain decomposing by anti-Fourier conversion.

3, the polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture according to claim 1 is characterized in that, the method that step 2 pair full resolution polarization SAR image carries out the concrete decomposition of sub-aperture decomposition is:

According to the scope of the visual angle φ of different Doppler frequency scope correspondences, the full aperture image is decomposed on the sub-aperture of difference, obtain a plurality of images of low resolution.

4, the polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture according to claim 1 is characterized in that stopping criterion for iteration described in the step 5 is user-defined iterations.

5, the polarization synthetic aperture radar image sorting technique of analyzing based on sub-aperture according to claim 1 is characterized in that stopping criterion for iteration described in the step 5 is: the cluster centre that twice iteration obtains

With

Between distance less than certain threshold value.

Images