CN101339245A - Multi- baseline interference synthetic aperture radar interference phase unwrapping method - Google Patents

Abstract

The invention discloses an expansion method of an interference phase of multi-baseline interference synthetic aperture radar. The process is that a SAR imaging processing is carried out respectively to M echo data received by a satellite; any one of the images after the SAR imaging processing is selected as a main image; by using the main image as a reference, a rough registration is carried out respectively to other images through a correlation method so as to obtain M-1 roughly-registered SAR images; A roughly-registered SAR image is used for constructing an optimum weighted joint data vector jx (i, w<opt>); a covariance matrix Cov<jx> (i, w<opt>) is estimated according to the optimum weighted joint data vector and resolved characteristically so as to obtain a cost function; the interference phase phi<i> corresponding to the minimum value of the output power of the cost function is used as the deployment result of the interference phase; the above-mentioned relevant process is repeated respectively to each of pixels to obtain the expansion result of the interference phase of the entire image. The expansion method disclosed by the invention has the advantages of small calculating amount and accurate expansion of the interference phase when the SAR image registration accuracy is bad, thereby being used for the reconstruction of a ground three-dimensional terrain with high resolution and high precision and the detection of a ground moving target.

Description

Multi-baseline interference synthetic aperture radar interference phase unwrapping method

Technical field

The present invention relates to the The radar exploration technique field, be a kind of based on the multi-baseline interference synthetic aperture radar interference phase unwrapping method of optimizing echo data, can under the very poor condition of synthetic-aperture radar SAR image registration accuracy, launch the interferometric phase between respective pixel exactly, can realize ground high resolving power, high-precision three-dimensional terrain reconstruction and ground moving target are detected.

Background technology:

Interference synthetic aperture radar InSAR is a kind of important remote sensing technology, can realize round-the-clock to ground high spatial resolution, high-precision three-dimensional mapping and ground moving target detection.Interference phase unwrapping is one of important step during InSAR handles, the conventional InSAR interference phase unwrapping method, as least square LS method, region growing method and branch blanking method, all require the precision of image registration to reach 1/10～1/100 resolution element, otherwise can influence subsequent treatment, thereby cause topographical height measurement precision and reliability to reduce.And therefore traditional single baseline InSAR system has limited it to complex-terrain, as the high precision mapping ability of city, mountain valley and steep cliff etc. owing to be subjected to that the fuzzy and elevation of interferometric phase is stacked etc. to be had a strong impact on.Many baselines InSAR system can effectively overcome these shortcomings.Many baselines InSAR system can make full use of the reliability that information that its length baseline obtains improves phase unwrapping, and can not reduce measurement of higher degree precision, have the more dimensional topography re-configurability of high precision and Geng Gao robustness than traditional single baseline InSAR system.Therefore, research has strong sane multi-baseline interference phase developing method to the image registration error, has important practical value.

At present, mainly contain following several to the sane interference phase unwrapping method of image registration error:

1.D.C.Ghiglia enclosed in " Interferometric Synthetic Aperture Radar Terrain Elevation Mapping fromMultiple Observations " literary composition of delivering in the meeting of border at the 6th Digital Signal Processing of IEEE Deng 1994, proposed to comprise in many baselines phase developing method phase developing method based on minimum variance LS and maximum likelihood ML.In " Phase Unwrapping of Multibaseline Interferometry UsingKalman Filtering " article that M.G.Kim etc. delivered in the 7th Image Processing international conference of IEE in 1999, utilize Kalman filtering to carry out many baselines phase unwrapping.G.Fornaro etc. 2006 are at IET Radar, Sonar ﹠amp; In " Maximum Likelihood Multi-Baseline SARInterferometry " literary composition of delivering on the Navig, utilize the statistical property of interferometric phase to carry out many baselines phase unwrapping.In " Multibaseline InSARDEM Reconstruction:The Wavelet Approach " that F.Alessandro etc. delivered on IEEE Trans.on GRS in 1999, utilize small wave converting method to carry out digital elevation figure and recover.In " Phase-unwrapping ofSAR Interferogram with Multi-frequency or Multi-baseline " that W.Xu etc. delivered in IEEE GRS international conference in 1994, utilize many baselines phase developing method to comprise Chinese remainder theorem, projection and linear combination method.These methods can recover to bring difficulty to interference phase unwrapping and landform altitude under the very big situation of image registration error.

In " ImageAuto-Coregistration and InSAR Interferogram Estimation Using Joint SubspaceProjection " article that people such as Li Zhenfang delivered in 2007, utilized the coherence messages and the space projection technology of neighbor on IEEE Trans.on GRS.Though can when having registration error, obtain satisfied interference phase unwrapping result, when launching interferometric phase, at first to determine the noise subspace dimension, if the noise subspace dimension is estimated to be forbidden, must influence the expansion result of interferometric phase.

2. people such as hair will outstanding person " based on the InSAR interferometric phase robustness estimation of associating pixel model " delivered at systems engineering and electronics in 2008, in this article, proposed a kind of projection vector and determined the noise subspace dimension based on interferometric phase and proper vector structure, and then the method for estimation InSAR interferometric phase, this method is when there is registration error in image, not only can estimate the dimension of noise subspace exactly, strengthen the practicality and the robustness of associating pixel method estimation interferometric phase simultaneously.But, when the estimating noise subspace, need artificial setting threshold to judge." the estimating " that Mao Zhijie etc. deliver at the electric wave science journal based on the adapting to image registration InSAR interferometric phase of optimizing echo data, the interferometric phase method of estimation of single pixel model is united in proposition based on weighting, though when estimating interferometric phase, do not need to determine the noise subspace dimension, but when estimating interferometric phase, need at first to determine the registration error direction, and need search when determining optimum weights, so calculated amount is bigger.

Summary of the invention

The objective of the invention is to overcome the deficiency of above-mentioned prior art, provide a kind of calculated amount little, and need not many baselines InSAR interference phase unwrapping method that setting threshold is judged based on the optimization echo data, to solve the interferometric phase that under the very poor condition of SAR image registration accuracy, can accurately launch between respective pixel, realize accurate reconstruct to dimensional topography.

For achieving the above object, method of the present invention comprises following process:

A. the M width of cloth echo data that satellite is received carries out SAR imaging processing, M 〉=3 respectively;

B. selecting any piece image after the SAR imaging processing as master image, serves as with reference to correlation method or other method for registering images with this master image, and the image to other carries out thick registration respectively, obtains the thick registration SAR of M-1 width of cloth image;

C. utilize the optimum weighting associating of SAR image configuration data vector jx (i, w behind the thick registration _Opt);

$\mathrm{jx}(i,w_{\mathrm{opt}})={[x_1\left(i\right),{\hat{x}}_2(i,w_{\mathrm{opt}}),{\hat{x}}_3(i,w_{\mathrm{opt}}),\&CenterDot;\&CenterDot;\&CenterDot;,{\hat{x}}_M(i,w_{\mathrm{opt}})]}^T$

In the formula, x ₁(i) be master image, ${\hat{x}}_2(i,w_{\mathrm{opt}}),{\hat{x}}_3(i,w_{\mathrm{opt}}),\&CenterDot;\&CenterDot;\&CenterDot;,{\hat{x}}_M(i,w_{\mathrm{opt}})$ After optimizing respectively the 2nd, the 3 ..., M width of cloth SAR complex pattern data, w _OptBe optimum weighing vector, i represents the current master image that will calculate and thick registering images pixel label, and subscript T represents matrix transpose operation;

D. according to optimum weighting associating data vector jx (i, w _Opt) the estimate covariance matrix is:

Cov _jx(i，w _opt)＝E[jx(i，w _opt)jx(i，w _opt) ^H]

In the formula, E[] expression statistical average, subscript H represents conjugate transpose;

E. to covariance matrix Cov _Jx(i, w _Opt) carry out feature decomposition, draw cost function:

In the formula, Open signal subspace, { β into covariance matrix _n ^(l)} _L=2 ^MOpen into noise subspace;

F. carry out interference phase unwrapping by cost function, be about to the pairing interferometric phase of output power minimum value of cost function

As the interference phase unwrapping result.

The present invention is because the optimum weighting associating of the SAR image configuration after utilizing thick registration data vector carries out the adapting to image registration process to echo data, and make full use of the coherence messages of neighbor, not only do not need search, reduce operand greatly, no matter the direction of image registration error estimates with the dimension that size can accurately obtain noise subspace, thereby can under the very poor condition of SAR image registration accuracy, launch interferometric phase between respective pixel exactly.

Simulation result shows, not only can carry out the self-adaptation registration, and, be under the condition of a pixel in the image registration error without estimating noise subspace dimension to the SAR image, can launch the interferometric phase between respective pixel exactly, thereby accurately realize dimensional topography reconstruct.

Can describe in detail by following accompanying drawing and l-G simulation test purpose of the present invention, feature, advantage.

Description of drawings

Fig. 1 is a method flow diagram of the present invention;

Fig. 2 is master image and the thick registering images synoptic diagram that the present invention constructs optimum weighting associating data vector;

Fig. 3 be in the SAR image of using in the emulation of the present invention pixel to the location drawing:

Fig. 4 is the eigenvalue distribution figure of the covariance matrix of emulation of the present invention;

Fig. 5 is the histogram of the data-optimized front and back in the emulation of the present invention;

Fig. 6 is the root-mean-square error figure that the interferometric phase in the emulation of the present invention changes along with registration error;

Fig. 7 is the root-mean-square error figure that the interferometric phase of emulation of the present invention changes along with SNR;

Fig. 8 is the landform altitude figure that uses in the emulation of the present invention;

Fig. 9 is the simulation result figure of the present invention when the accurate registration of image;

Figure 10 is the simulation result figure of the present invention when registration error is [0.5,0.5];

Figure 11 is the simulation result figure of the present invention when registration error is [1.0,1.0].

Embodiment

With reference to Fig. 1, realize that step of the present invention is as follows

Step 1, the M width of cloth echo data that satellite is received carries out the SAR imaging processing respectively.

According to the characteristic and the satellite parametric reduction of satellite reception data, utilize distance-Doppler's R-D algorithm or linear modulation mark CS algorithm or frequency change mark FS algorithm that the echo data of reception is carried out the SAR imaging processing, its raw data is become the SAR view data.

Step 2, the thick registration of image.

Selecting any piece image after the SAR imaging processing as master image, serves as with reference to correlation method or other method for registering images with this master image, and the image to other carries out thick registration respectively, obtains the thick registration SAR of M-1 width of cloth image.In the thick registration process of this image, do not require and as traditional interferometric phase method of estimation, require the precision of image registration must reach sub-pixel, be that registration accuracy will reach 1/10 to 1/100 pixel, only require that image registration accuracy reaches Pixel-level, it is just much of that to that is to say that registration accuracy allows to reach a resolution element, has alleviated the difficulty of image registration so greatly.

Step 3 is calculated optimal weight vector.

1. calculate master image and thick registration m (m=2,3 ..., M) the associating data vector x of width of cloth SAR image _{1, m}(i).

The structure of master image is shown in Fig. 2 (a), the structure of thick registering images is shown in Fig. 2 (b) Fig. 2 (c), i is for will launch the pixel of interferometric phase among Fig. 2, according to the master image of Fig. 2 and the structure of thick registering images, can get the 1st width of cloth and m (m=2,3 ..., M) the associating data vector x of width of cloth SAR image _{1, m}(i) be:

x _1，m(i)＝[x ₁(i)，x _m(i)] ^T

(1)

＝[x ₁(i)，x _m(i-4)，…，x _m(i)，…，x _m(i+4)] ^T

In the formula, x ₁(i), x ₂(i), x ₃(i) ..., x _M(i) be respectively complex pattern data behind master image and the thick registration, i-4, i-3, i-2, i-1, i, i+1, i+2, i+3, i+4 represent m width of cloth echo data pixel i and contiguous pixel cell thereof, and subscript T represents matrix transpose operation;

2. by associating data vector x _{1, m}(i) calculate its covariance matrix

(i), promptly

${\mathrm{Cov}}_{x_{1,m}}\left(i\right)=E\left[x_{1,m}\left(i\right)x_{1,m}^H\left(i\right)\right];---\left(2\right)$

In the formula, E[] expression statistical average, subscript H represents conjugate transpose;

3 under minimum mean square error criterion, calculates optimal weight vector by covariance matrix:

$w=\mathrm{inv}\left({\mathrm{Cov}}_{x_{1,m}}\left(i\right)\right)\&CenterDot;A---\left(3\right)$

In the formula, inv () represents matrix inversion operation, $A={[\mathrm{1,0},\&CenterDot;\&CenterDot;\&CenterDot;,0]}_{10x1}^T,$ Make w=[w ₁, w ₂] ^TWherein, w ₁=w (1), w ₂=[w (2) ..., w (10)] ^TCan obtain based on the optimal weight vector of optimum weighting associating data vector model shown in Figure 2 by (3) formula be:

$w_{\mathrm{opt}}^m=\mathrm{abs}({\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="A20081015058100151.png,A20081015058100152.png"\; state="\{\{state\}\}">w</figure-callout>}}_2/w_1),m=\mathrm{2,3},\&CenterDot;\&CenterDot;\&CenterDot;,M.---\left(4\right)$

Step 4 is constructed optimum weighting associating data vector.

1. with resulting optimal weight vector w _Opt ^m, m=2,3 ..., M is weighted thick registration SAR image respectively, the m width of cloth SAR data after being optimized:

${\hat{x}}_m(i,w_{\mathrm{opt}}^{\left(m\right)})=w_{\mathrm{opt}}^{\left(m\right)T}{x}_m\left(i\right),m=\mathrm{2,3},\&CenterDot;\&CenterDot;\&CenterDot;,M---\left(5\right)$

In the formula, $w_{\mathrm{opt}}^{\left(m\right)}={[w_{i-4}^{\left(m\right)},\&CenterDot;\&CenterDot;\&CenterDot;,w_i^{\left(m\right)},\&CenterDot;\&CenterDot;\&CenterDot;,w_{i+4}^{\left(m\right)}]}^T$ Be the optimal weight vector of master image with respect to m width of cloth image, x _m(i)=[x _m(i-4) ..., x _m(i) ..., x _m(i+4)] ^TBe m (m=2,3 ..., M) width of cloth view data;

2. draw optimum weighting associating data vector by the M-1 width of cloth SAR view data after master image and the optimization:

$\mathrm{jx}(i,w_{\mathrm{opt}})={[x_1\left(i\right),{\hat{x}}_2(i,w_{\mathrm{opt}}),{\hat{x}}_3(i,w_{\mathrm{opt}}),\&CenterDot;\&CenterDot;\&CenterDot;,{\hat{x}}_M(i,w_{\mathrm{opt}})]}^T---\left(6\right)$

Step 5, the estimate covariance matrix.

1. with optimum weighted data vector jx (i, w _Opt) the corresponding covariance matrix Cov of estimation _Jx(i, w _Opt), formula is as follows:

Cov _jx(i，w _opt)＝E[jx(i，w _opt)jx(i，w _opt) ^H] (7)

2. to covariance matrix Cov _Jx(i, w _Opt) carry out feature decomposition, obtain the signal subspace of covariance matrix respectively

And noise subspace { β _n ^(l)} _L=2 ^M, promptly

In the formula,

Be called steering vector, I is a unit matrix, σ _x ²(i) echo power of remarked pixel i, σ _n ²The expression noise power, R _Js(i, w _Opt) be called the optimization coherence function matrix of pixel to i, R _Js(i, w _Opt) can obtain by the Amplitude Estimation of covariance matrix

$R_{\mathrm{js}}(i,w_{\mathrm{opt}})=|{\mathrm{Cov}}_{\mathrm{jx}}(i,w_{\mathrm{opt}})-{\mathrm{\&sigma;}}_n^{2}I|---\left(9\right)$

To R _Js(i, w _Opt) carry out feature decomposition and obtain big eigenvalue respectively _rAnd characteristic of correspondence vector and β _Js, thus, Open into covariance matrix Cov _Jx(i, w _Opt) signal subspace, { β _n ^(l)} _L=2 ^MOpen noise subspace into covariance matrix;

3. because covariance matrix Cov _Jx(i, w _Opt) signal subspace

Be orthogonal to its noise subspace { β _n ^(l)} _L=2 ^M, the signal subspace of covariance matrix to the noise subspace projection, is obtained cost function:

Step 6 utilizes cost function that interferometric phase is launched.

1. be step-length with 0.01 radian, with φ _iIn each value substitution cost function in given range, calculate its result respectively;

2. institute's result calculated is arranged, find out minimum value wherein, the phase place of this minimum value correspondence

As the result of interference phase unwrapping, carry out the dimensional topography elevation map with resulting expansion interferometric phase result and recover.

Step 7 is obtained the interferometric phase image of whole landform.

Each pixel of master image and thick registration SAR image is carried out the operation of above-mentioned steps 3～step 6 respectively, just can obtain the interferometric phase image of whole landform, can obtain the elevation map of whole landform.

Effect of the present invention can further specify by following simulation result.

1. simulated conditions

According to the Hill equation, be example with 3 Cartwheel satellite configurations, choose some orbital positions three satellites constantly and carry out emulation.Corresponding effectively vertical course base length is respectively: the vertical parallax B between the satellite 1 and 2 ₁₂Vertical parallax B between=63.8 meters, satellite 1 and 3 ₁₃=345.3 meters.The downwards angle of visibility of radar is 40 °, utilizes the echo power of width of cloth actual measurement SAR image as each SAR pixel of ground scene, generates three width of cloth SAR images.Radar is operated in X-band.First satellite emission signal, all satellites receive echoed signal.The ground echo signal that 3 satellites are received carries out the SAR imaging processing, and the signal to noise ratio snr of SAR image is 20dB.Its coefficient of coherence is by vertical parallax length, local terrain slope and SNR decision.Artificially generated terrain be many mountain regions shape as shown in Figure 8, its maximum height is 326.7 meters, minimum point is-75 meters.

2. simulation result

(1) relation of optimal weight vector and image registration error is as Fig. 3, Fig. 4 and shown in Figure 5.

Among Fig. 3, each circle is represented a pixel cell, and i is exactly the center pixel unit of requirement interferometric phase.4 curves among Fig. 4 are respectively the distribution plan of the feature under the different registration error situations.

When the correlativity of complete registration of image and supposition image was higher, covariance matrix promptly had a big eigenwert and other two little eigenwerts to go to zero, and its structure is shown in Fig. 3 (a), and eigenvalue distribution is shown in the curve a among Fig. 4.

When image orientation or apart from when registration error is 0.5 pixel, as Fig. 3 (b), by obtaining optimal weight vector w _OptPixel i and neighbor are weighted processing, and the coherence messages of the feasible echo data of optimizing can not lost, thereby can accurately obtain the dimension of noise subspace, and its eigenvalue distribution is shown in the curve b among Fig. 4.

When existing distance during to registration error such as Fig. 3 (c), still can obtain good treatment effect by optimum weighted to the orientation simultaneously, its result is shown in the curve c among Fig. 4.

When having a pixel error such as Fig. 3 (d), still can accurately obtain the dimension of noise subspace by optimum weighted, its result is shown in the curve d among Fig. 4.

For example, the accurate registration of image between satellite 1 and 2, between the satellite 1 and 3 distance to or the orientation to registration error be 0.5 pixel, simultaneously exist distance to the orientation be [0.5 to registration error, 0.6] individual pixel and when having the registration error of 1 pixel, obtain one group of optimal weight vector w by processing to echo data _Opt ¹, w _Opt ², w _Opt ³And w _Opt ⁴, unite data vector by constructing optimum weighting, thereby can realize satellite is received the optimization process of echo data.

Fig. 5 is that the promptly optimum power of [0.5,0.6] individual pixel is for registration error $w=w_{\mathrm{opt}}^3$ The time histogram of echo data before and after optimizing, wherein, Fig. 5 (a) and 5 (b) represent respectively to optimize the preceding orientation of data to distance to the statistics histogram, Fig. 5 (c) and 5 (d) represent the histogram after the optimization respectively.

From Fig. 3 and Fig. 4 as seen, in optimum weighting joint observation vector, work and mainly be and the relevant pixel of registration error size direction, by echo data being carried out optimization weighting Combined Treatment, and make full use of the pixel i and the information of neighbor on every side, no matter the size of registration error and direction can accurately obtain the dimension of noise subspace.Simultaneously as shown in Figure 5, by the weighting Combined Treatment, realized SAR image adaptive registration operation.

(2) probatio inspectionem pecuoarem the inventive method is to the robustness of image registration error, shown in Fig. 6-7 and Fig. 9～11.

Figure 6 shows that the accurate registration of Fig. 2 (a) and Fig. 2 (b), Fig. 2 (a) is [0 with the registration error of Fig. 2 (c), 0.2,0.4,0.6,0.8,1] and during individual pixel, the root-mean-square error figure that the interferometric phase that launches changes along with registration error, three interferometric phase square error curves are estimated to obtain by classic method, existing shape method and the present invention respectively.

Fig. 7 describes is registration error between Fig. 2 (a) and Fig. 2 (b) and Fig. 2 (a) and Fig. 2 (c) when being [0,0.5,1] individual pixel, SNR by zero when changing to 25dB root-mean-square error along with SNR variation relation figure.Having provided simultaneously the root-mean-square error curve in Fig. 7 compares with card Carsten Ramelow lower bound CRLB.

Figure 9 shows that the accurately result figure during registration of Fig. 2 (a), Fig. 2 (b) and Fig. 2 (c).Fig. 9 (a) and 9 (b) are respectively the interferometric phase image that utilizes classic method to obtain between satellite 1 and satellite 2 and satellite 1 and the satellite 3, Fig. 9 (c) is when accurate registration, the expansion interferometric phase that Combined Treatment by the inventive method obtains and then the landform altitude figure of recovery, 9 (d) are by the elevation map of Fig. 9 (c) reconstruct landform and the Error Graph of artificially generated terrain elevation map.

Figure 10 be Fig. 2 (a) respectively and the as a result figure of the registration error between Fig. 2 (b) and Fig. 2 (c) when being [0.5,0.5] individual pixel.Figure 10 (a) and 10 (b) are respectively the interferometric phase image that utilizes classic method to obtain between satellite 1 and satellite 2 and satellite 1 and the satellite 3, Figure 10 (c) for Fig. 2 (a) respectively and the registration error between Fig. 2 (b) and Fig. 2 (c) be [0.5,0.5] during individual pixel, by the landform altitude figure that the inventive method is recovered, 10 (d) recover the elevation map of landform and the Error Graph of artificially generated terrain elevation map by Figure 10 (c).

Figure 11 shows that Fig. 2 (a) respectively and the result figure of the registration error between Fig. 2 (b) and Fig. 2 (c) when being [1.0,1.0] individual pixel.Figure 11 (a) and 11 (b) are respectively the interferometric phase image that utilizes classic method to obtain between satellite 1 satellite 2 and satellite 1 and the satellite 3, Figure 11 (c) be Fig. 2 (a) respectively and the registration error between Fig. 2 (b) and Fig. 2 (c) be [1.0,1.0] during individual pixel, carry out the expansion interferometric phase and then the reconstruct three-dimensional land map that obtain after the Combined Treatment by the inventive method; 11 (d) recover the elevation map of landform and the Error Graph of artificially generated terrain elevation map by Figure 11 (c).

From as can be known as the simulation result Fig. 6-7 and Fig. 9～11, carry out interference phase unwrapping or landform altitude when recovering with the inventive method, the present invention has very strong robustness and practicality to image registration sum of errors signal to noise ratio (S/N ratio).

To sum up, the inventive method make full use of respective pixel to and the coherence messages of neighbor, make corresponding SAR image carry out the self-adaptation registration process by constructing the optimum weight vectors of many baselines, no matter the size of image registration error and direction can accurately obtain the interference phase unwrapping result sane to the image registration error.

Claims (4)

1. multi-baseline interference synthetic aperture radar interference phase unwrapping method comprises following process:

A. the M width of cloth echo data that satellite is received carries out SAR imaging processing, M 〉=3 respectively;

B. selecting any piece image after the SAR imaging processing as master image, serves as with reference to using correlation method with this master image, and the image to other carries out thick registration respectively, obtains the thick registration SAR of M-1 width of cloth image;

C. utilize the optimum weighting associating of SAR image configuration data vector jx (i, w behind the thick registration _Opt);

$\mathrm{jx}(i,w_{\mathrm{opt}})={[x_1\left(i\right),{\hat{x}}_2(i,w_{\mathrm{opt}}),{\hat{x}}_3(i,w_{\mathrm{opt}}),\&CenterDot;\&CenterDot;\&CenterDot;,{\hat{x}}_M(i,w_{\mathrm{opt}})]}^T$

In the formula, x ₁(i) be master image,

After optimizing respectively the 2nd, the 3 ..., M width of cloth SAR complex pattern data, w _OptBe optimum weighing vector, i represents the current master image that will calculate and thick registering images pixel label, and subscript T represents matrix transpose operation;

D. according to optimum weighting associating data vector jx (i, w _Opt) estimate covariance matrix and covariance matrix carried out feature decomposition,

${\mathrm{Cov}}_{\mathrm{jx}}(i,w_{\mathrm{opt}})=E\left[\mathrm{jx}(i,w_{\mathrm{opt}})\mathrm{jx}{(i,w_{\mathrm{opt}})}^H\right]$

In the formula, E[] expression statistical average, subscript H represents conjugate transpose;

E. to covariance matrix Cov _Jx(i, w _Opt) carry out feature decomposition, draw cost function:

In the formula, Open signal subspace, { β into covariance matrix _n ^(l)} _L=2 ^MOpen into noise subspace;

F. carry out interference phase unwrapping by cost function, be about to the pairing interferometric phase of output power minimum value of cost function

As the interference phase unwrapping result;

G. to each pixel implementation C～F operation respectively of master image and thick registration SAR image, obtain the interference phase unwrapping result of entire image.

2. interference phase unwrapping method according to claim 1 is characterized in that the optimum weighting associating of the structure described in step C data vector jx (i, w _Opt), carry out according to the following procedure:

Ca. calculate master image and thick registration m (m=2,3 ..., M) the associating data vector x of width of cloth SAR image _{1, m}(i)

$x_{1,m}\left(i\right)={[x_1\left(i\right),x_m\left(i\right)]}^T$

$={[x_1\left(i\right),x_m(i-4),\&CenterDot;\&CenterDot;\&CenterDot;,x_m\left(i\right),\&CenterDot;\&CenterDot;\&CenterDot;,x_m(i+4)]}^T$

In the formula, x ₁(i), x ₂(i), x ₃(i) ..., x _M(i) be respectively complex pattern data behind master image and the thick registration, i-4, i-3, i-2, i-1, i, i+1, i+2, i+3, i+4 represent the pixel i and the contiguous pixel cell thereof of m width of cloth view data, and subscript T represents matrix transpose operation;

Cb. by associating data vector x _{1, m}(i) calculate its covariance matrix

Promptly

${\mathrm{Cov}}_{x_{1,m}}\left(i\right)=E\left[x_{1,m}\left(i\right)x_{1,m}^H\left(i\right)\right];$

Cc. under minimum mean square error criterion, calculate optimal weight vector by covariance matrix:

$w=\mathrm{inv}\left({\mathrm{Cov}}_{x_{1,m}}\left(i\right)\right)\&CenterDot;A$

In the formula, inv () represents matrix inversion operation, $A={[\mathrm{1,0},\&CenterDot;\&CenterDot;\&CenterDot;,0]}_{10\&times;1}^T,$ Make w=[w ₁, w ₂] ^T(w ₁=w (1),

w ₂=[w (2) ..., w (10)] ^T), then optimal weight vector is:

$w_{\mathrm{opt}}^m=\mathrm{abs}(w_2/w_1),m=\mathrm{2,3},\&CenterDot;\&CenterDot;\&CenterDot;,M$

Cd. use resulting optimal weight vector w _Opt ^mRespectively to m (m=2,3 ..., M) the thick registration SAR of width of cloth image is weighted, and obtains m width of cloth SAR view data, promptly

${\hat{x}}_m(i,w_{\mathrm{opt}}^{\left(m\right)})=w_{\mathrm{opt}}^{\left(m\right)T}{x}_m\left(i\right),m=\mathrm{2,3},\&CenterDot;\&CenterDot;\&CenterDot;,M$

In the formula, $w_{\mathrm{opt}}^{\left(m\right)}={[w_{i-4}^{\left(m\right)},\&CenterDot;\&CenterDot;\&CenterDot;,w_i^{\left(m\right)},\&CenterDot;\&CenterDot;\&CenterDot;,w_{i+4}^{\left(m\right)}]}^T$ Be the optimal weight vector of m width of cloth image with respect to master image, x _m(i)=[x _m(i-4) ..., x _m(i) ..., x _m(i+4)] ^TIt is the SAR view data behind the thick registration of the m width of cloth;

Ce. draw optimum weighting associating data vector by the SAR view data after master image and the optimization:

$\mathrm{jx}(i,w_{\mathrm{opt}})={[x_1\left(i\right),{\hat{x}}_2(i,w_{\mathrm{opt}}),{\hat{x}}_3(i,w_{\mathrm{opt}}),\&CenterDot;\&CenterDot;\&CenterDot;{\hat{x}}_M(i,w_{\mathrm{opt}})]}^T.$

3. interference phase unwrapping method according to claim 1 is characterized in that step e carries out according to the following procedure:

Ea. to covariance matrix Cov _Jx(i, w _Opt) carry out feature decomposition, obtain the signal subspace of covariance matrix respectively

And noise subspace { β _n ^(l)} _L=2 ^M, promptly

In the formula,

Be called steering vector, I is a unit matrix, σ _x ²(i) echo power of remarked pixel i, σ _n ²The expression noise power, R _Js(i, w _Opt) be called the optimization coherence function matrix of pixel to i, R _Js(i, w _Opt) can obtain by the Amplitude Estimation of covariance matrix, promptly

$R_{\mathrm{js}}(i,w_{\mathrm{opt}})=|{\mathrm{Cov}}_{\mathrm{jx}}(i,w_{\mathrm{opt}})-{\mathrm{\&sigma;}}_n^{2}I|$

Eb. to R _Js(i, w _Opt) carry out feature decomposition and obtain big eigenvalue respectively _rAnd characteristic of correspondence vector and β _Js, thus,

Open into covariance matrix Cov _Jx(i, w _Opt) signal subspace, { β _n ^(l)} _L=2 ^MOpen noise subspace into covariance matrix;

Ec. with the signal subspace of covariance matrix to the noise subspace projection, obtain cost function:

4. interference phase unwrapping method according to claim 1 is characterized in that the pairing interferometric phase of output power minimum value with cost function described in the step F

As the interference phase unwrapping result, carry out according to the following procedure:

Fa. be step-length with 0.01 radian, with φ _iIn each value substitution cost function in given range, calculate its result respectively;

Fb. institute's result calculated is arranged, find out minimum value wherein, the phase place of this minimum value correspondence Result as interference phase unwrapping.

Images