CN105005048A - Saliency-map-based Laplacian cooperation compression radar imaging method - Google Patents

Abstract

Disclosed in the invention is a saliency-map-based Laplacian cooperation compression radar imaging method. The method comprises the following steps: step 1, carrying out processing by using a range Doppler algorithm to obtain a low-resolution ISAR image; step 2, constructing a sparse dictionary; step 3, carrying out PTC conversion based on a formula on a preliminary ISAR imaging result S^to obtain a saliency map; step 4, constructing a Laplacian matrix of the saliency map; and step 5, in order to obtain a high-resolution ISAR image, carrying out processing by using a basis pursuit algorithm and an analytical method alternately to solve an optimum theta and T. According to the invention, on the basis of the regional homogeneity priori assumption of image data, the background clutter is reduced and noises are suppressed; and the imaging quality is effectively improved.

Description

Based on the Laplacian collaborative compression radar imaging method of remarkable figure

Technical field

The present invention relates to a kind of synthetic aperture radar image-forming technology, particularly relate to a kind of Laplacian collaborative compression radar imaging method based on remarkable figure.

Background technology

Inverse synthetic aperture radar (ISAR) (ISAR) is a kind of high-resolution imaging radar being different from conventional radar, there is the ability of moving target (as aircraft, naval vessel and guided missile etc.) being carried out to imaging and identification, can the precise image of round-the-clock, remote acquisition target, there is important dual-use value.In order to give full play to ISAR high-resolution imaging ability, and making ISAR better adapt to the application scenarios of actual demand and constantly change, being necessary to proceed deep discussion to ISAR technology.

For high-resolution imaging, the response of target scattering can be superposed by a series of reflection separately and obtain, its echoed signal of ISAR imaging is sparse in time domain, according to this priori, utilize the compressed sensing of develop rapidly in recent years (CS) theory can obtain high-resolution radar image from little umber of pulse.But the process that radar image recovers can see the sparse solution of a searching underdetermined equation as, and this is a np hard problem.For solving this problem, scholar proposes a lot of algorithm as greedy tracing algorithm, relaxed algorithm etc., and wherein tracing algorithm calculated amount is very large, and speed of convergence is slower.

Summary of the invention

The object of the present invention is to provide a kind of Laplacian collaborative compression radar imaging method based on remarkable figure, utilize the locally coherence a priori assumption of view data to reduce background clutter, restraint speckle, effectively improve the quality of imaging.

For achieving the above object, technical scheme of the present invention is a kind of Laplacian collaborative compression radar imaging method based on remarkable figure of design, comprises the steps:

Step 1, adopts range Doppler algorithm (RDA) to obtain the ISAR image of low resolution;

Step 2, structure sparse dictionary:

A) hypothesis does not get over parasang migration, and introduce noise, l parasang comprises K orientation to different scattering centers, then the echoed signal s of l parasang _l(t) be:

$s_l\left(t\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}{B}_k\·\mathrm{rect}\left(\frac{t}{T_a}\right)\exp (-j2\mathrm{\π}{f}_k\·t)+n_l$

Wherein f _kfor Doppler frequency, B _kfor the reflection amplitude of a kth scattering point, K is the number of scattering point, n _lit is the additive noise of l parasang;

B) definition time sequence t:[1:N] ^tΔ t, wherein N=T _a/ Δ t is umber of pulse, Δ t=1/f _rfor the time interval, f _rfor the pulse repetition rate of radar, the number of definition Doppler is Q, then corresponding DOPPLER RESOLUTION is Δ f _d=f _r/ Q, discrete Doppler sequence is f _d: [1:Q] ^tΔ f _d-(f _r/ 2), the Q of setting should be greater than umber of pulse N, constructs sparse dictionary thus as follows:

Wherein, 0≤q≤Q, then the matrix form of the echoed signal of l range unit is:

s _l＝Ψθ _l+n _l

Wherein vectorial θ _lthe l that (l=1,2...L) is image array arranges, θ _lthe complex amplitude at corresponding K the strong scattering center of nonzero element in (l=1,2...L);

Step 3, to preliminary ISAR imaging results the PCT conversion done based on following formula is significantly schemed I:

$P=\mathrm{sign}\left(\mathrm{CT}\left(\hat{S}\right)\right);F=\left|{\mathrm{CT}}^{-1}\left(P\right)\right|;I=G*F^2$

Wherein CT, CT ^-1be respectively profile transformation and inverse transformation, G is a dimensional Gaussian low-pass filter, and * is convolution operation, and sign () is sign function:

$\mathrm{sign}\left(t\right)=\left\{\begin{array}{c}+1,t\&GreaterEqual;0\\ -1,t<0\end{array};\right.$

Step 4, constructs the Laplacian matrix of remarkable figure:

L＝D-G

Similar matrix G is based on remarkable figure, is constructed and produce by gaussian kernel function, and its i-th row jth column element computing method are as follows:

Wherein I _ifor the orientation of i-th parasang in the remarkable figure that preliminary imaging low Resolution Radar ISAR image obtains is to coefficient vector, for k the neighbour nearest apart from i-th range unit,

D is degree (degree) matrix of figure, represents the connection weights of each node and other nodes, and on diagonal line, the value of each element is

The form of the Laplacian Matrix of the remarkable figure after normalization is

Step 5, for obtaining high-resolution ISAR image, being namely used alternatingly Θ and T of base tracing algorithm and Analytic Method optimum, namely separating optimization problem:

Wherein W _l=diag{w _il, calculated by the weight matrix of remarkable figure:

1) initialization λ, initialization

2) following algorithm is repeated, until Θ does not change:

2.1) base tracing algorithm is utilized to solve following formula to upgrade Θ:

$\left\{\begin{array}{cc}\underset{{\mathrm{\&theta;}}_l}{\min }& \underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\left({\left|\right|W_l{\mathrm{\&theta;}}_l\left|\right|}_1\right)\\ s.t& \underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\left({\left|\right|\left[\begin{array}{c}\mathrm{\&lambda;}{s}_l\\ r{t}_l\end{array}\right]-\left[\begin{array}{c}\mathrm{\&lambda;\&Psi;}\\ \mathrm{rI}\end{array}\right]{\mathrm{\&theta;}}_l\left|\right|}_2^2\right)\&le;\mathrm{\&epsiv;}\end{array}\right.;$

2.2) following formula is adopted to upgrade T:

2.3) following formula is adopted to upgrade λ:

$\mathrm{\&lambda;}=\mathrm{\&lambda;}+\mathrm{\&mu;}\stackrel{L}{\mathrm{\&Sigma;}}{\left|\right|s_l-\mathrm{\&Psi;}{\mathrm{\&theta;}}_l\left|\right|}_2^2.$

The present invention is according to the locally coherence a priori assumption of radar data, spatially close data point has similarity, Laplacian regular terms is introduced in optimization problem, solution procedure adopts alternating direction multiplier method (ADMM) solving-optimizing problem, by combining the mode that each parasang solves, reduce noise and the background clutter of Radar Signal Transmission process, improve image quality.

Accompanying drawing explanation

Fig. 1 is overall flow schematic diagram of the present invention;

Fig. 2 (a) is the exemplary plot of significantly figure I in the present invention;

Fig. 2 (b) is the contour map of significantly figure I in the present invention;

Fig. 2 (c) is the remarkable figure I exemplary plot in the present invention after two dimension median filter and morphological image closed operation;

Fig. 2 (d) is the exemplary plot of marking area in the present invention;

Fig. 3 (a) is the ISAR image of 256 pulse Yak-42 in conventional RD algorithm;

Fig. 3 (b) is the ISAR image of 64 pulse Yak-42 in conventional RD algorithm;

Fig. 3 (c) is the ISAR contour map of 256 pulse Yak-42 in conventional RD algorithm;

Fig. 3 (d) is the ISAR contour map of 64 pulse Yak-42 in conventional RD algorithm;

Fig. 4 (a) is emulation experiment 1 signal to noise ratio snr of the present invention when being 2dB, the ISAR imaging results figure of 32 pulses;

Fig. 4 (b) is emulation experiment 1 signal to noise ratio snr of the present invention when being 2dB, the ISAR imaging results figure of 64 pulses;

Fig. 4 (c) is emulation experiment 1 signal to noise ratio snr of the present invention when being 2dB, the ISAR imaging results figure of 96 pulses;

Fig. 4 (d) is emulation experiment 1 signal to noise ratio snr of the present invention when being 2dB, the ISAR imaging results contour map of 32 pulses;

Fig. 4 (e) is emulation experiment 1 signal to noise ratio snr of the present invention when being 2dB, the ISAR imaging results contour map of 64 pulses;

Fig. 4 (f) is emulation experiment 1 signal to noise ratio snr of the present invention when being 2dB, the ISAR imaging results contour map of 96 pulses.

Embodiment

Below in conjunction with drawings and Examples, the specific embodiment of the present invention is further described.Following examples only for technical scheme of the present invention is clearly described, and can not limit the scope of the invention with this.

The technical scheme that the present invention specifically implements is:

As shown in Figure 1, Figure 2 shown in (a) ~ Fig. 2 (d), a kind of Laplacian collaborative compression radar imaging method based on remarkable figure, comprises the steps:

Step 1, adopts range Doppler algorithm (RDA) to obtain the ISAR image of low resolution;

Step 2, structure sparse dictionary:

A) hypothesis does not get over parasang migration, and introduce noise, l parasang comprises K orientation to different scattering centers, then the echoed signal s of l parasang _l(t) be:

$s_l\left(t\right)=\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{B}_k\&CenterDot;\mathrm{rect}\left(\frac{t}{T_a}\right)\exp (-j2\mathrm{\&pi;}{f}_k\&CenterDot;t)+n_l$

Wherein f _kfor Doppler frequency, B _kfor the reflection amplitude of a kth scattering point, K is the number of scattering point, n _lit is the additive noise of l parasang;

B) definition time sequence t:[1:N] ^tΔ t, wherein N=T _a/ Δ t is umber of pulse, Δ t=1/f _rfor the time interval, f _rfor the pulse repetition rate of radar, the number of definition Doppler is Q, then corresponding DOPPLER RESOLUTION is Δ f _d=f _r/ Q, discrete Doppler sequence is f _d: [1:Q] ^tΔ f _d-(f _r/ 2), the Q of setting should be greater than umber of pulse N, constructs sparse dictionary thus as follows:

Wherein, 0≤q≤Q, then the matrix form of the echoed signal of l range unit is:

s _l＝Ψθ _l+n _l

Wherein vectorial θ _lthe l that (l=1,2...L) is image array arranges, θ _lthe complex amplitude at corresponding K the strong scattering center of nonzero element in (l=1,2...L);

Step 3, to preliminary ISAR imaging results the PCT conversion done based on following formula is significantly schemed I:

$P=\mathrm{sign}\left(\mathrm{CT}\left(\hat{S}\right)\right);F=\left|{\mathrm{CT}}^{-1}\left(P\right)\right|;I=G*F^2$

Wherein CT, CT ^-1be respectively profile transformation and inverse transformation, G is a dimensional Gaussian low-pass filter, and * is convolution operation, and sign () is sign function:

$\mathrm{sign}\left(t\right)=\left\{\begin{array}{c}+1,t\&GreaterEqual;0\\ -1,t<0\end{array};\right.$

Step 4, constructs the Laplacian matrix of remarkable figure:

L＝D-G

Similar matrix G is based on remarkable figure, is constructed and produce by gaussian kernel function, and its i-th row jth column element computing method are as follows:

Wherein I _ifor the orientation of i-th parasang in the remarkable figure that preliminary imaging low Resolution Radar ISAR image obtains is to coefficient vector, for k the neighbour nearest apart from i-th range unit,

D is degree (degree) matrix of figure, represents the connection weights of each node and other nodes, and on diagonal line, the value of each element is

The form of the Laplacian Matrix of the remarkable figure after normalization is

Step 5, for obtaining high-resolution ISAR image, being namely used alternatingly Θ and T of base tracing algorithm and Analytic Method optimum, namely separating optimization problem:

Wherein W _l=diag{w _il, calculated by the weight matrix of remarkable figure:

1) initialization λ, initialization

2) following algorithm is repeated, until Θ does not change:

2.1) base tracing algorithm is utilized to solve following formula to upgrade Θ:

$\left\{\begin{array}{cc}\underset{{\mathrm{\&theta;}}_l}{\min }& \underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\left({\left|\right|W_l{\mathrm{\&theta;}}_l\left|\right|}_1\right)\\ s.t& \underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\left({\left|\right|\left[\begin{array}{c}\mathrm{\&lambda;}{s}_l\\ r{t}_l\end{array}\right]-\left[\begin{array}{c}\mathrm{\&lambda;\&Psi;}\\ \mathrm{rI}\end{array}\right]{\mathrm{\&theta;}}_l\left|\right|}_2^2\right)\&le;\mathrm{\&epsiv;}\end{array}\right.;$

2.2) following formula is adopted to upgrade T:

2.3) following formula is adopted to upgrade λ:

$\mathrm{\&lambda;}=\mathrm{\&lambda;}+\mathrm{\&mu;}\stackrel{L}{\mathrm{\&Sigma;}}{\left|\right|s_l-\mathrm{\&Psi;}{\mathrm{\&theta;}}_l\left|\right|}_2^2.$

Effect of the present invention can be further illustrated by following emulation experiment.

Emulation experiment 1

Experiment condition: the Yak-42 airplane data using the collection of C-band (5.52-GHz) ISAR radar is experimental subjects, the pulse width that radar system launches linear FM signal is 25.6-μ s, range resolution 0.375m, centre frequency 5.52-GHZ, overall pulse number is 256; This experiment is Intel (R) Corei3 at CPU, and dominant frequency is 2.2GHz, and the Win7 system inside saving as 4G adopts software MATLAB R2010a to emulate.

Utilize method of the present invention and range Doppler algorithm (RDA) to carry out contrast experiment in experiment, experimental result is as shown in Fig. 3 (a) ~ Fig. 3 (d) He Fig. 4 (a) ~ Fig. 4 (f).

When Fig. 4 (a) ~ Fig. 4 (f) is for different weights during SNR=2dB, uses 32,64 and 96 pulses, obtain the image of 256 doppler values.As can be seen from the experiment, the result of the present invention shown in Fig. 4 (a) ~ Fig. 4 (f) has good experiment effect, and background is level and smooth, because make use of the geometry priori of image, and the clutter simultaneously also in Background suppression.And traditional range Doppler (RD) algorithm shown in Fig. 3 (a) ~ Fig. 3 (d) be imaged on 64 pulses time edge contour very fuzzy, this shows, experimental result of the present invention has significant raising than traditional range Doppler (RD) algorithm.

The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the prerequisite not departing from the technology of the present invention principle; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (1)

1., based on the Laplacian collaborative compression radar imaging method of remarkable figure, it is characterized in that, comprise the steps:

Step 1, adopts range Doppler algorithm to obtain the ISAR image of low resolution;

Step 2, structure sparse dictionary:

A) hypothesis does not get over parasang migration, and introduce noise, l parasang comprises K orientation to different scattering centers, then the echoed signal s of l parasang _l(t) be:

$s_l\left(t\right)=\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{B}_k\&CenterDot;\mathrm{rect}\left(\frac{t}{T_a}\right)\exp (-j2\mathrm{\&pi;}{f}_k\&CenterDot;t)+n_l$

Wherein f _kfor Doppler frequency, B _kfor the reflection amplitude of a kth scattering point, K is the number of scattering point, n _lit is the additive noise of l parasang;

B) definition time sequence t:[1:N] ^tΔ t, wherein N=T _a/ Δ t is umber of pulse, Δ t=1/f _rfor the time interval, f _rfor the pulse repetition rate of radar, the number of definition Doppler is Q, then corresponding DOPPLER RESOLUTION is Δ f _d=f _r/ Q, discrete Doppler sequence is f _d: [1:Q] ^tΔ f _d-(f _r/ 2), the Q of setting should be greater than umber of pulse N, constructs sparse dictionary thus as follows:

Wherein, then the matrix form of the echoed signal of l range unit is:

s _l＝Ψθ _l+n _l

Wherein vectorial θ _lthe l that (l=1,2...L) is image array arranges, θ _lthe complex amplitude at corresponding K the strong scattering center of nonzero element in (l=1,2...L);

Step 3, to preliminary ISAR imaging results the PCT conversion done based on following formula is significantly schemed I:

$P=\mathrm{sign}\left(\mathrm{CT}\left(\hat{S}\right)\right);F=\left|{\mathrm{CT}}^{-1}\left(P\right)\right|;I=G*F^2$

Wherein CT, CT ^-1be respectively profile transformation and inverse transformation, G is a dimensional Gaussian low-pass filter, and * is convolution operation, and sign () is sign function:

$\mathrm{sign}\left(t\right)=\left\{\begin{array}{cc}+1,& t\&GreaterEqual;0\\ -1,& t<0\end{array}\right.;$

Step 4, constructs the Laplacian matrix of remarkable figure:

L＝D-G

Similar matrix G is based on remarkable figure, is constructed and produce by gaussian kernel function, and its i-th row jth column element computing method are as follows:

Wherein I _ifor the orientation of i-th parasang in the remarkable figure that preliminary imaging low Resolution Radar ISAR image obtains is to coefficient vector, for k the neighbour nearest apart from i-th range unit,

D is the degree matrix of figure, represents the connection weights of each node and other nodes, and on diagonal line, the value of each element is

The form of the Laplacian Matrix of the remarkable figure after normalization is

Step 5, for obtaining high-resolution ISAR image, being namely used alternatingly Θ and T of base tracing algorithm and Analytic Method optimum, namely separating optimization problem:

Wherein W _l=diag{w _il, calculated by the weight matrix of remarkable figure:

1) initialization λ, initialization

2) following algorithm is repeated, until Θ does not change:

2.1) base tracing algorithm is utilized to solve following formula to upgrade Θ:

$\left\{\begin{array}{cc}\underset{{\mathrm{\&theta;}}_l}{\min }& \underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\left({\left|\right|W_l{\mathrm{\&theta;}}_l\left|\right|}_1\right)\\ s.t& \underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\left(\right|\left|{\left[\begin{array}{c}{\mathrm{\&lambda;s}}_l\\ {\mathrm{rt}}_l\end{array}\right]-\left[\begin{array}{c}\mathrm{\&lambda;\&Psi;}\\ \mathrm{rI}\end{array}\right]{\mathrm{\&theta;}}_l\left|\right|}_2^2\right)\&le;\mathrm{\&epsiv;}\end{array}\right.;$

2.2) following formula is adopted to upgrade T:

2.3) following formula is adopted to upgrade λ:

$\mathrm{\&lambda;}=\mathrm{\&lambda;}+\mathrm{\&mu;}\stackrel{L}{\mathrm{\&Sigma;}}{\left|\right|s_l-\mathrm{\&Psi;}{\mathrm{\&theta;}}_l\left|\right|}_2^2.$