CN104991241A - Target signal extraction and super-resolution enhancement processing method in strong clutter condition - Google Patents

Abstract

The invention discloses a target signal extraction and super-resolution enhancement processing method in a strong clutter condition. The main idea is that after an imaging result of point targets in a ground scene is obtained, target rectangular region slicing data including the point targets and a ground clutter is selected; a strong scattering point corresponding to a scattering point two-dimension position corresponding to the Fourier primary function correlation maximum value in the target rectangular region slicing data is extracted; weak scattering points nearby the strong scattering point are extracted using a gradient descent method, so as to obtain a weak scattering point area nearby the strong scattering point corresponding to the scattering point two-dimension position which meets conditions of the gradient descent method; subtraction of the weak scattering point area nearby the strong scattering point corresponding to the scattering point two-dimension position which meets conditions of the gradient descent method from the target rectangular region slicing data is performed, so as to obtain a complete SAR (Synthetic Aperture Radar) image; and the complete SAR image is subjected to the super-resolution enhancement processing using a regularization method, so as to obtain a final super-resolution SAR image.

Description

Under strong clutter background, the extraction of echo signal and super-resolution strengthen disposal route

Technical field

The invention belongs to Radar Signal Processing Technology field, in particular to extraction and the super-resolution enhancing disposal route of echo signal under a kind of strong clutter background, the echo signal be applicable under strong clutter background is extracted, or the regularization super-resolution SAR imaging processing of sparse signal.

Background technology

Synthetic-aperture radar (SAR) is a kind of high-resolution microwave imaging radar, compared with traditional optical imaging, microwave imaging radar does not limit by weather condition, can round-the-clock, round-the-clock echo signal or this echo signal place scene to be observed, utilize SAR image to obtain great attention to the research that battlefield echo signal carries out automatically identifying at present.Under normal circumstances, the resolution of SAR image is higher, automatically identifies that the accuracy of echo signal is higher.

But, when synthetic-aperture radar (SAR) is descended in working order, strong land clutter in this synthetic-aperture radar (SAR) antenna beam range of exposures, meeting and echo signal enter to synthetic-aperture radar (SAR) receiver by this synthetic-aperture radar (SAR) antenna in the lump, form the SAR image that signal to noise ratio is lower.In end processing sequences, if first accurately do not extracted echo signal, but directly the SAR image lower to signal to noise ratio carries out super-resolution enhancing process, and will introduce a large amount of false target signal, the accuracy that echo signal is identified automatically reduces.

Therefore, when SAR image signal to noise ratio is lower, carry out accurately extracting being necessary to echo signal.And extract for echo signal, traditional method utilizes two-dimension constant false alarm to process, and this two-dimension constant false alarm obviously can suppress weaker clutter, and accurately can extract echo signal; But during for the complex environment that land clutter is very strong or land clutter fluctuation is very large, adopt this two-dimension constant false alarm can produce a large amount of false-alarm, make accurately to extract echo signal; Although adopt slide window processing to the CFAR thresholding in this two-dimension constant false alarm, also can reduce false-alarm, this operation can cause damage to the weak scattering point of echo signal, cannot ensure the integrality of echo signal.

Summary of the invention

For above the deficiencies in the prior art, the extraction and the super-resolution that the object of the invention is to propose echo signal under a kind of strong clutter background strengthen disposal route, finally obtain super-resolution SAR image.

Realization approach of the present invention is: according to the correlativity of echo signal and Fourier's base height, and the regularization method improved can strengthen the advantage of echo signal resolution, by structure self-adapting clutter thresholding, echo signal is extracted, finally resolution is carried out to the echo signal after extraction and strengthen process, finally obtain super-resolution SAR image.

For reaching above-mentioned technical purpose, the present invention adopts following technical scheme to be achieved.

Under strong clutter background, the extraction of echo signal and super-resolution strengthen a disposal route, it is characterized in that, comprise the following steps:

Step 1, carries out motion compensation successively to the SAR radar echo signal of ground scene and conventional line frequency modulation becomes mark imaging processing, obtains the imaging results of p point target and then obtain the imaging results Img that a width comprises P point target; Wherein, p ∈ 1,2 ..., P}, P represent the total number of point target in ground scene;

Step 2, comprises in the imaging results Img of P point target at a width, extract the target rectangle region slice of data D comprising P point target and land clutter _a, obtain this target rectangle region slice of data D respectively _atwo-dimensional frequency discrete expression S (n, m) and this target rectangle region slice of data D _atime-domain expression G (k, h); Wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene;

Step 3, by target rectangle region slice of data D _afour clutter rectangular areas, corner of the same size and Fourier basis functions W _kh(n, m) carries out correlativity process respectively, calculates self-adapting clutter thresholding ε; Wherein, n ∈ [1, N], m ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene;

Step 4, by target rectangle region slice of data D _awith Fourier basis functions W _kh(n, m) carries out correlativity process, extracts and obtains target rectangle region slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^q; Wherein, k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations;

Step 5, adopts gradient descent method to target rectangle region slice of data D _awith Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^qcontiguous weak scattering point region is extracted, and is met strong scattering point G (k, h) of gradient descent method condition ^qcontiguous weak scattering point region wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], n represents target area slice of data D _adistance samples unit number, M represents target area slice of data D _aazimuth sample cells number, D _arepresent the target area slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations;

Step 6, by target rectangle region slice of data D _adeduct the strong scattering point meeting gradient descent method condition contiguous weak scattering point region obtain residue target rectangle region slice of data and to this residue target rectangle region slice of data step 4 and step 5 is utilized to carry out iterative operation, until residue target rectangle region slice of data in all scattering points and Fourier basis functions W _kh(n, m) degree of correlation maximal value is lower than self-adapting clutter thresholding ε, and iteration stopping, obtains the weak scattering region that P strong scattering point and this P strong scattering point are close to separately, and then is combined into the complete S AR image G that a width comprises P point target, and

Wherein, represent the scattering point two-dimensional position meeting gradient descent method condition corresponding all scattering points, n represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and P also represents target rectangle region slice of data D _ain the total number of strong scattering point, q represents iterations;

Step 7, utilizes regularization method to carry out super-resolution to the complete S AR image G that a width comprises P point target and strengthens process, obtain the super-resolution SAR image that a width is final wherein, P represents the total number of point target in ground scene.

Beneficial effect of the present invention: the present invention utilizes the strong correlation of echo signal and Fourier basis functions, the strong scattering point of echo signal is extracted, afterwards in order to protect weak scattering point, the weak scattering point utilizing gradient descent method contiguous to strong scattering point extracts further, and utilize iterative operation, extract from strong clutter and obtain complete echo signal; In addition, the present invention utilizes the regularization method of improvement to carry out resolution enhancing process to target, can strengthen target resolvability, suppressed sidelobes and noise, raising SAR image contrast, the super-resolution SAR imaging of realize target signal.

Accompanying drawing explanation

Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.

Fig. 1 is extraction and the super-resolution enhancing process flow schematic diagram of echo signal under a kind of strong clutter background of the present invention;

Fig. 2 is the target rectangle region slice of data D comprising earthwave rectangular area, four corners of the present invention _aschematic diagram;

Fig. 3 is that the present invention uses gradient descent method to extract the contiguous weak scattering point schematic diagram of strong scattering point G (k, h) corresponding to scattering point two-dimensional position, wherein, the strong scattering point G (k that scattering point two-dimensional position is corresponding, the first weak scattering point in h) contiguous weak scattering point region is G (k, h-1), the strong scattering point G (k that scattering point two-dimensional position is corresponding, the second weak scattering point in h) contiguous weak scattering point region is G (k-1, h-1), the strong scattering point G (k that scattering point two-dimensional position is corresponding, the 3rd weak scattering point in h) contiguous weak scattering point region is G (k-1, h), the strong scattering point G (k that scattering point two-dimensional position is corresponding, the 4th weak scattering point in h) contiguous weak scattering point region is G (k-1, h+1), the strong scattering point G (k that scattering point two-dimensional position is corresponding, the 5th weak scattering point in h) contiguous weak scattering point region is G (k, h+1), the strong scattering point G (k that scattering point two-dimensional position is corresponding, the 6th weak scattering point in h) contiguous weak scattering point region is G (k+1, h+1), the strong scattering point G (k that scattering point two-dimensional position is corresponding, the 7th weak scattering point in h) contiguous weak scattering point region is G (k+1, h), the strong scattering point G (k that scattering point two-dimensional position is corresponding, the 8th weak scattering point in h) contiguous weak scattering point region is G (k+1, h-1),

Fig. 4 is the actual measurement scene schematic diagram of triplane target of the present invention;

Fig. 5 (a) is the corner reflector image schematic diagram under strong clutter background, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit,

Fig. 5 (b) is that combining target strong scattering point extracts criterion and target area and extracts criterion and carry out corner reflector to the corner reflector image under the strong clutter background shown in Fig. 5 (a) and extract the result schematic diagram obtained, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit;

Fig. 6 (a) extracts the result schematic diagram of to carry out 8 times of interpolation after corner reflector and obtaining, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit,

Fig. 6 (b) is that the regularization method of the improvement utilizing the present invention to propose carries out resolution to Fig. 6 (a) and strengthens and process the result schematic diagram that obtains, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit;

Fig. 7 (a) be to 1 bugle reflector target in Fig. 6 (b) and 2 bugle reflector target to carry out respectively before and after superresolution processing along orientation to diagrammatic cross-section, wherein, transverse axis represent orientation to, unit is azimuth sample cells, the longitudinal axis represents normalized amplitude, unit is dB

Fig. 7 (b) be to 3 bugle reflector target in Fig. 6 (b) and 2 bugle reflector target carry out respectively before and after superresolution processing along distance to diagrammatic cross-section, wherein, transverse axis represent distance to, unit is distance samples unit, the longitudinal axis represents normalized amplitude, and unit is dB;

Fig. 8 (a) is the motor vehicle target image schematic diagram under strong clutter background, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit,

Fig. 8 (b) extracts criterion in conjunction with strong scattering point to carry out to Fig. 8 (a) result schematic diagram that motor vehicle Objective extraction obtains, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit;

Fig. 9 (a) carries out the result schematic diagram that 12 times of interpolation obtain after extractor motor-car target, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit,

Fig. 9 (b) is that the regularization method of the improvement utilizing the present invention to propose carries out super-resolution to Fig. 9 (a) and strengthens and process the result schematic diagram that obtains, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit.

Embodiment

With reference to Fig. 1, be that extraction and the super-resolution of echo signal under a kind of strong clutter background of the present invention strengthens process flow schematic diagram, under this kind strong clutter background, the extraction of echo signal and super-resolution strengthen disposal route, comprise the following steps:

Step 1, carries out motion compensation successively to the SAR radar echo signal in ground scene and conventional line frequency modulation becomes mark (Chirp Scaling, CS) imaging processing, obtains the imaging results of p point target and then obtain the imaging results Img that a width comprises P point target; Wherein, represent apart from the fast time, t _mrepresent the orientation slow time, p ∈ 1,2 ..., P}, P represent the total number of point target in ground scene.

Particularly, motion compensation and conventional line frequency modulation change mark (Chirp Scaling, CS) imaging processing are carried out successively to the SAR radar echo signal in ground scene, obtains the imaging results of p point target the imaging results of this p point target expression formula is:

$s_P(\hat{t},t_m)={\mathrm{\σ}}_p\sin c\[{\mathrm{\Δf}}_r(\hat{t}-2R(t_m)/c)\]\sin c\[{\mathrm{\Δf}}_a(t_m-X_n/V)\]$

Wherein, Δ f _rrepresent the signal band width after distance pulse pressure, represent apart from the fast time, t _mrepresent the orientation slow time, Δ f _arepresent Doppler frequency band width, σ _prepresent the scattering center amplitude of p point target, c represents the light velocity, and λ represents radar center wavelength, R (t _m) represent the oblique distance course of SAR radar to p point target place scene center,

And r _brepresent the shortest oblique distance of p point target place scene center, t _mrepresent the orientation slow time, sinc [] represents impact respective function, X _nrepresent p point target along the x-axis coordinate on course-and-bearing in rectangular coordinate system, V represents carrier of radar speed, p ∈ 1,2 ..., P}, P represent the total number of point target in ground scene.

Step 2, comprises in the imaging results Img of P point target at a width, extract the target rectangle region slice of data D comprising P point target and land clutter _a, obtain this target rectangle region slice of data D respectively _atwo-dimensional frequency discrete expression S (n, m) and this target rectangle region slice of data D _atime-domain expression G (k, h); Wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene.

Particularly, comprise at a width in the imaging results Img of P point target, extract the target rectangle region slice of data D comprising P point target and land clutter _a, obtain this target rectangle region slice of data D respectively _atwo-dimensional frequency discrete expression S (n, m) and this target rectangle region slice of data D _atime-domain expression G (k, h), this target rectangle region slice of data D _atwo-dimensional frequency discrete expression S (n, m) and this target rectangle region slice of data D _atime-domain expression G (k, h) can be expressed as respectively:

S(n,m)＝E(n,m)+C(n,m)

G(k,h)＝IFFT2{E(n,m)+C(n,m)}

Wherein, E (n, m) represents target rectangle region slice of data D _aall scattering points of a middle P point target discrete form, represent target rectangle region slice of data D _aall scattering points of a middle P point target,

And d _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, represent the imaging results of p point target, C (n, m) represents target rectangle region slice of data D _ain noise signal, and b _nmrepresent target rectangle region slice of data D _ain clutter amplitude, represent target rectangle region slice of data D _ain clutter phase place, N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, IFFT2{} represents two-dimentional inverse Fourier transform, represent apart from the fast time, t _mrepresent the orientation slow time.

Step 3, by target rectangle region slice of data D _afour land clutter rectangular areas, corner of the same size and Fourier basis functions W _kh(n, m) carries out correlativity process respectively, calculates self-adapting clutter thresholding ε; Wherein, n ∈ [1, N], m ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene.

Particularly, with reference to Fig. 2, be the target area slice of data D comprising land clutter rectangular area, four corners of the same size _aschematic diagram.Usually, land clutter rectangular area, these four corners of the same size is 5%D _a~ 15%D _athe land clutter that land clutter rectangular area, corner calculates time too little is too little, point target or scattering point (comprising strong scattering point and weak scattering point) may be introduced when land clutter rectangular area, corner is too large, calculated by mass data, land clutter rectangular area, four that choose in the present invention corners of the same size size is 10%D _a.

By target area slice of data D _aclutter rectangular area, four corners and Fourier basis functions W _kh(n, m) carries out correlativity process respectively, calculates self-adapting clutter thresholding ε.

Self-adapting clutter thresholding ε and Fourier basis functions W _khthe expression formula of (n, m) is respectively:

ε＝max{ε _i|ε _i＝max[Coh(D _i)]}

$W_{kh}(n,m)=\exp \{-j2\mathrm{\π}(\frac{nk}{N}+\frac{mh}{M})\},$

Wherein, Coh (D _i) represent target rectangle region slice of data D _ain i-th corner clutter rectangular area D in clutter rectangular area, four corners of the same size _icorrelation matrix, i ∈ 1,2,3,4}, and $Coh\left(D_i\right)={\left(\sqrt{E_{D_i}}\sqrt{NM}\right)}^{-1}\left|\underset{m=1}{\overset{M}{\Σ}}\underset{n=1}{\overset{N}{\Σ}}(D_i\·{W_{kh}}^{*}(n,m)\right)|,E_{D_i}$ Represent i-th corner clutter rectangular area D _ienergy summation, max [] represent ask for function maxima, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, W _kh(n, m) represents Fourier basis functions, represent Fourier basis functions W _khthe conjugation of (n, m).

Step 4, by target rectangle region slice of data D _awith Fourier basis functions W _kh(n, m) carries out correlativity process, calculates target rectangle region slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^q, and then extract target rectangle region slice of data D _ain with Fourier basis functions W _khthe strong scattering point that (n, m) degree of correlation maximal value is corresponding wherein, k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations.

The concrete sub-step of step 4 is:

4.1 by target rectangle region slice of data D _awith Fourier basis functions W _khafter (n, m) carries out correlativity process, obtain related coefficient Coh (k, h); Wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene.

Particularly, utilize point target signal and Fourier's base to have the correlativity of height, land clutter then shows this characteristic compared with weak dependence, by target rectangle region slice of data D _awith Fourier basis functions W _khafter (n, m) carries out correlativity process, obtain related coefficient Coh (k, h), its expression formula is:

$Coh(k,h)={\left(\sqrt{E_s}\sqrt{NM}\right)}^{-1}|\underset{m=1}{\overset{M}{\Σ}}\underset{n=1}{\overset{N}{\Σ}}\left(S\right(n,m)\·{W_{kh}}^{*}(n,m))|$

Wherein, E _srepresent target rectangle region slice of data D _aenergy, and s (n, m) represents target rectangle region slice of data D _atwo-dimensional frequency discrete expression, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, W _kh(n, m) represents Fourier basis functions, and k ∈ [1, N], h ∈ [1, M], D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene;

4.2 calculate target rectangle region slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^q; Wherein, k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations;

4.3 calculate target rectangle region slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^q, and then extract target rectangle region slice of data D _ain with Fourier basis functions W _khthe strong scattering point that (n, m) degree of correlation maximal value is corresponding wherein, k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations.

Particularly, the target rectangle region slice of data D of extraction _ain with Fourier basis functions W _khthe strong scattering point that (n, m) degree of correlation maximal value is corresponding extraction criterion be:

$G_{ext}^q(k,h)=G^q(k,h)$

s.t.max(Coh(k,h) ^q)≥ε

Wherein, G (k, h) ^qrepresent target area slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point, ε represents self-adapting clutter thresholding, Coh (k, h) ^qrepresent related coefficient, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations.

Step 5, adopts gradient descent method to target rectangle region slice of data D _awith Fourier basis functions W _khthe strong scattering point that (n, m) degree of correlation maximal value is corresponding contiguous weak scattering point region is extracted, and is met the strong scattering point of gradient descent method condition contiguous weak scattering point region wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], n represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations.

Particularly, with reference to Fig. 3, for the present invention uses gradient descent method to extract target rectangle region slice of data D _ain with Fourier basis functions W _khthe strong scattering point that (n, m) degree of correlation maximal value is corresponding contiguous weak scattering point schematic diagram, adopts gradient descent method to target rectangle region slice of data D _ain with Fourier basis functions W _khthe strong scattering point that (n, m) degree of correlation maximal value is corresponding contiguous weak scattering point region is extracted, and is met the strong scattering point of gradient descent method condition contiguous weak scattering point region its expression formula is:

$G_{ext}^q(\tilde{k},\tilde{h})=G_{ext}^q(k,h)+{\mathrm{\ΣG}}_{ext}^q(k+i,h+j)$

$\begin{array}{cc}\mathrm{st}.& \mathrm{grad}\left(G_{\mathrm{ext}}^q(k,h)\right)>\mathrm{grad}\left(G_{\mathrm{ext}}^q(k+i,h+j)\right);(-1\≤i\≤1,-1\≤j\≤1)\end{array}$

Wherein, represent the q time iterative extraction strong scattering point meeting gradient descent method condition out, grad () represents gradient algorithm, represent the q time iterative extraction strong scattering point meeting gradient descent method condition out one of them weak scattering point in the weak scattering point of surrounding eight vicinities ,-1≤i≤1 ,-1≤j≤1, k ∈ [1, N], h ∈ [1, M], n represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations.

Step 6, by target rectangle region slice of data D _adeduct the strong scattering point meeting gradient descent method condition contiguous weak scattering point region obtain residue target rectangle region slice of data and to this residue target rectangle region slice of data step 4 and step 5 is utilized to carry out iterative operation, until residue target rectangle region slice of data in all scattering points and Fourier basis functions W _kh(n, m) degree of correlation maximal value is lower than self-adapting clutter thresholding ε, and iteration stopping, obtains the weak scattering region that P strong scattering point and this P strong scattering point are close to separately, and then is combined into the complete S AR image G that a width comprises P point target, and

Wherein, represent the scattering point two-dimensional position meeting gradient descent method condition corresponding all scattering points, n represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and P also represents target rectangle region slice of data D _ain the total number of strong scattering point, q represents iterations.

Step 7, utilizes regularization method to carry out super-resolution to the complete S AR image G that a width comprises P point target and strengthens process, obtain the super-resolution SAR image that a width is final wherein, P represents the total number of point target in ground scene.

The concrete sub-step of step 7 is:

7.1 use regularization method process one width to comprise the complete S AR image G of P point target, obtain the High Resolution SAR Images that a width is to be reconstructed utilize the High Resolution SAR Images that this width is to be reconstructed amplitude and this width High Resolution SAR Images to be reconstructed gradient distribution openness, the High Resolution SAR Images that this width of vectorization is to be reconstructed be optimized function its expression formula is:

$\min J\left(\stackrel{\‾}{G}\right)=\left|\right|G-\mathrm{\Φ}\stackrel{\‾}{G}|{|}_2^2+{\mathrm{\λ}}_1|\left|\stackrel{\‾}{G}\right|{|}_k^k+{\mathrm{\λ}}_2\left|\right|D\left|\stackrel{\‾}{G}\right||{|}_k^k$

Wherein, Φ is expressed as picture operator, λ ₁, λ ₂all represent regularization parameter, represent that the High Resolution SAR Images that a width is to be reconstructed, G represent that a width comprises the complete S AR image of P point target, represent 2-norm, represent k-norm.

Particularly, the size that a width comprises the complete S AR image G of P point target is M × N, then the size of the High Resolution SAR Images G that a width is to be reconstructed is (M × N) × 1, majorized function section 1 in expression formula is data fidelity item, and this data fidelity item characterizes and minimizes actual observation amount and High Resolution SAR Images to be reconstructed between square error; Section 2 represents the sparse prior of target, and suitably the choosing of sparse prior of this target contributes to suppressing pseudo-target, reduces the secondary lobe of final super-resolution imaging result, can protect and strengthen the distinguishing of target scattering point; Section 3 represents the sparse prior of object edge, the sparse prior of this object edge is a slickness penalty term, this suitably choose the strong scattering gradient (as image border) that can retain final super-resolution imaging result, thus keep target shape.

7.2 skip optimisation functions section 3 in expression formula, utilizes regularization method, calculates optimization function and then obtain the final super-resolution SAR image of a width

Particularly, the gradient descent method used for the present invention carries out extraction process to the point target under strong clutter background, can the edge weak scattering dot information of retention point target, thus can holding point target shape.

Utilize the regularization method improved, i.e. skip optimisation function section 3 in expression formula, obtains optimization function image obtain the super-resolution SAR image that a width is final

Particularly, this optimization function image expression formula is:

$\mathrm{argmin}\hat{J}\left(\stackrel{\‾}{G}\right)=\mathrm{argmin}\left\{\right||G-\mathrm{\Φ}\stackrel{\‾}{G}|{|}_2^2+{\mathrm{\λ}}_1\left|\right|\stackrel{\‾}{G}\left|{|}_k^k\right\}$

Wherein, Φ is expressed as picture operator, and G represents that a width comprises the complete S AR image of P point target, λ ₁represent regularization parameter, represent 2-norm, represent k-norm.

By verifying validity of the present invention further to the emulation experiment of following measured data.

(1) point target measured data

In actual measurement scene, place three pieces of length of sides is 15 centimetre of three corner reflector, the actual measurement scene schematic diagram of triplane target as shown in Figure 4, and corner reflector distance is to apart 0.94 meter, and orientation is to apart 0.72 meter, and SAR radar basic parameter as shown in Table 1.

Table one

Actual measurement 1, accurately extract the triplane target in strong clutter background by method of the present invention, result is as shown in Fig. 5 (a) He Fig. 5 (b); Wherein, Fig. 5 (a) is the corner reflector image schematic diagram under strong clutter background, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit; Fig. 5 (b) is that combining target strong scattering point extracts criterion and target area and extracts criterion and carry out corner reflector to the corner reflector image under the strong clutter background shown in Fig. 5 (a) and extract the result schematic diagram obtained, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit.

As can be seen from the two width figure of Fig. 5 (a) and Fig. 5 (b), triplane target can successfully extract by using method of the present invention from land clutter.

Actual measurement 2, carry out resolution by method of the present invention to corner reflector after extraction and strengthen process, result is as shown in Fig. 6 (a) He Fig. 6 (b); Wherein, Fig. 6 (a) extracts the result schematic diagram of to carry out 8 times of interpolation after corner reflector and obtaining, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit; Fig. 6 (b) is that the regularization method of the improvement utilizing the present invention to propose carries out resolution to Fig. 6 (a) and strengthens and process the result schematic diagram that obtains, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit.

As apparent from Fig. 6 (a) and Fig. 6 (b) can, the resolution of using method of the present invention to corner reflector has enhancing effect, reaches super-resolution and strengthens object.

Actual measurement 3, carries out resolution comparative analysis to the corner reflector that resolution strengthens before and after process respectively by using method of the present invention, and result is as Fig. 7 (a) and Fig. 7 (b); Fig. 7 (a) be to 1 bugle reflector target in Fig. 6 (b) and 2 bugle reflector target to carry out respectively before and after superresolution processing along orientation to diagrammatic cross-section, wherein, straight line represents the diagrammatic cross-section along azimuth resolution before 1 bugle reflector target and 2 bugle reflector target superresolution processing, dotted line represents the diagrammatic cross-section along azimuth resolution after to 1 bugle reflector target and 2 bugle reflector target superresolution processing, transverse axis represent orientation to, unit is azimuth sample cells, the longitudinal axis represents normalized amplitude, and unit is dB; Fig. 7 (b) be to 3 bugle reflector target in Fig. 6 (b) and 2 bugle reflector target carry out respectively before and after superresolution processing along distance to diagrammatic cross-section, wherein, straight line represents the sectional view along range resolution before 3 bugle reflector target and 2 bugle reflector target superresolution processing, dotted line represents the sectional view along range resolution after to 3 bugle reflector target and 2 bugle reflector target superresolution processing, transverse axis represent distance to, unit is distance samples unit, the longitudinal axis represents normalized amplitude, and unit is dB.

As apparent from Fig. 7 (a) and Fig. 7 (b) can, triplane target strengthens after process through super-resolution, and the resolution of triplane target reduces greatly, and the measured data of this triplane target demonstrates validity of the present invention.

(2) Area Objects measured data

In actual measurement scene, place a frame motor vehicle target, distance is 5.5 meters to length, and orientation is 3 meters to length, and radar basic parameter as shown in Table 2.

Table two

Actual measurement 1, accurately extract the motor vehicle target under strong clutter background by using method of the present invention, result is as shown in Fig. 8 (a) He Fig. 8 (b); Fig. 8 (a) is the motor vehicle target image schematic diagram under strong clutter background, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit; Fig. 8 (b) extracts criterion in conjunction with strong scattering point to carry out to Fig. 8 (a) result schematic diagram that motor vehicle Objective extraction obtains, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit.

As can be seen from the two width figure of Fig. 8 (a) and Fig. 8 (b), motor vehicle target can successfully extract by using method of the present invention from strong Clutter Background.

Actual measurement 2, carry out super-resolution by method of the present invention to motor vehicle target after extraction and strengthen process, result is as Fig. 9 (a) and Fig. 9 (b); Fig. 9 (a) carries out the result schematic diagram that 12 times of interpolation obtain after extractor motor-car target, wherein, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit; Fig. 9 (b) is that the regularization method of the improvement utilizing the present invention to propose carries out super-resolution to Fig. 9 (a) and strengthens and process the result schematic diagram that obtains, transverse axis represent orientation to, unit is sampling unit, the longitudinal axis represent distance to, unit is sampling unit.

Can obviously find out from the circle 1 Fig. 9 (a) and Fig. 9 (b) and circle 2 respectively, the resolution of using method of the present invention to motor vehicle target has enhancing effect, wherein for circle 2, before motor vehicle target being carried out to super-resolution enhancing process, low to three point target resolution of motor vehicle target, these three aobvious point focusing of point target spy together, show as a bright spot; After super-resolution strengthens process, these three point targets show as three bright spots.Therefore, super-resolution strengthens the resolution that process greatly improves point target in image, can see more scattered information, effectively strengthens the contrast of image and the positioning precision of scattering center.

In sum, Simulation experiments validate correctness of the present invention, validity and reliability.

Obviously, those skilled in the art can carry out various change and modification to the present invention and not depart from the spirit and scope of the present invention; Like this, if these amendments of the present invention and modification belong within the scope of the claims in the present invention and equivalent technologies thereof, then the present invention is also intended to comprise these change and modification.

Claims (9)

1. under strong clutter background, the extraction of echo signal and super-resolution strengthen a disposal route, it is characterized in that, comprise the following steps:

Step 1, carries out motion compensation successively to the SAR radar echo signal of ground scene and conventional line frequency modulation becomes mark imaging processing, obtains the imaging results of p point target and then obtain the imaging results Img that a width comprises P point target; Wherein, p ∈ 1,2 ..., P}, P represent the total number of point target in ground scene;

Step 2, comprises in the imaging results Img of P point target at a width, extract the target rectangle region slice of data D comprising P point target and land clutter _a, obtain this target rectangle region slice of data D respectively _atwo-dimensional frequency discrete expression S (n, m) and this target rectangle region slice of data D _atime-domain expression G (k, h); Wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene;

Step 3, by target rectangle region slice of data D _afour clutter rectangular areas, corner of the same size and Fourier basis functions W _kh(n, m) carries out correlativity process respectively, calculates self-adapting clutter thresholding ε; Wherein, n ∈ [1, N], m ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene;

Step 4, by target rectangle region slice of data D _awith Fourier basis functions W _kh(n, m) carries out correlativity process, extracts and obtains target rectangle region slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^q; Wherein, k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations;

Step 5, adopts gradient descent method to target rectangle region slice of data D _awith Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^qcontiguous weak scattering point region is extracted, and is met strong scattering point G (k, h) of gradient descent method condition ^qcontiguous weak scattering point region wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], n represents target area slice of data D _adistance samples unit number, M represents target area slice of data D _aazimuth sample cells number, D _arepresent the target area slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations;

Step 6, by target rectangle region slice of data D _adeduct the strong scattering point meeting gradient descent method condition contiguous weak scattering point region obtain residue target rectangle region slice of data and to this residue target rectangle region slice of data step 4 and step 5 is utilized to carry out iterative operation, until residue target rectangle region slice of data in all scattering points and Fourier basis functions W _kh(n, m) degree of correlation maximal value is lower than self-adapting clutter thresholding ε, and iteration stopping, obtains the weak scattering region that P strong scattering point and this P strong scattering point are close to separately, and then is combined into the complete S AR image G that a width comprises P point target, and

Wherein, represent the scattering point two-dimensional position meeting gradient descent method condition corresponding all scattering points, n represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and P also represents target rectangle region slice of data D _ain the total number of strong scattering point, q represents iterations;

Step 7, utilizes regularization method to carry out super-resolution to the complete S AR image G that a width comprises P point target and strengthens process, obtain the super-resolution SAR image that a width is final wherein, P represents the total number of point target in ground scene.

2. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in step 1, and the imaging results of described p point target the imaging results of this p point target expression formula is:

$s_P(\hat{t},t_m)={\mathrm{\σ}}_P\sin c\[{\mathrm{\Δf}}_r(\hat{t}-2R(t_m)/c)\]\sin c\[{\mathrm{\Δf}}_a(t_m-X_n/V)\]$

Wherein, △ f _rrepresent the signal band width after distance pulse pressure, represent apart from the fast time, t _mrepresent the orientation slow time, △ f _arepresent Doppler frequency band width, σ _prepresent the scattering center amplitude of p point target, c represents the light velocity, and λ represents radar center wavelength, R (t _m) represent the oblique distance course of SAR radar to p point target place scene center,

And r _brepresent the shortest oblique distance of p point target place scene center, t _mrepresent the orientation slow time, sinc [] represents impact respective function, X _nrepresent p point target along the x-axis coordinate on course-and-bearing in rectangular coordinate system, V represents carrier of radar speed, p ∈ 1,2 ..., P}, P represent the total number of point target in ground scene.

3. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in step 2, and described this target rectangle region slice of data D _atwo-dimensional frequency discrete expression S (n, m) and this target rectangle region slice of data D _atime-domain expression G (k, h) be respectively:

S(n,m)＝E(n,m)+C(n,m)

G(k,h)＝IFFT2{E(n,m)+C(n,m)}

Wherein, E (n, m) represents target rectangle region slice of data D _aall scattering points of middle point target discrete form, represent target rectangle region slice of data D _ain all scattering points of point target, and p represents the total number of point target in ground scene, p ∈ 1,2 ..., P}, represent the imaging results of p point target, C (n, m) represents this target rectangle region slice of data D _ain noise signal, and b _nmrepresent this target rectangle region slice of data D _ain clutter amplitude, represent this target rectangle region slice of data D _ain clutter phase place, N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and IFFT2{} represents two-dimentional inverse Fourier transform, represent apart from the fast time, t _mrepresent the orientation slow time.

4. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in step 3, and described Fourier basis functions W _kh(n, m), its expression formula is:

$W_{th}(n,m)=\exp \{-j2\mathrm{\π}(\frac{nk}{N}+\frac{mh}{M})\}$

Wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene.

5. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in step 3, clutter rectangular area, described four corners of the same size, land clutter rectangular area, these four corners of the same size size is 10%D _a; Wherein, D _arepresent the target rectangle region slice of data comprising P point target and land clutter.

6. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in step 3, described self-adapting clutter thresholding ε, its expression formula is:

ε＝max{ε _i|ε _i＝max[Coh(D _i)]}

Wherein, Coh (D _i) represent target rectangle region slice of data D _ain i-th corner clutter rectangular area D in clutter rectangular area, four corners of the same size _icorrelation matrix, i ∈ 1,2,3,4}, and e _direpresent i-th corner clutter rectangular area D _ienergy summation, max [] represent ask for function maxima, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, W _kh(n, m) represents Fourier basis functions, W _kh ^*(n, m) represents Fourier basis functions W _khthe conjugation of (n, m).

7. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in step 4, and described target rectangle region slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^q, obtain target area slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^qsub-step be:

7.1 by target rectangle region slice of data D _awith Fourier basis functions W _khafter (n, m) carries out correlativity process, obtain related coefficient Coh (k, h); Wherein, n ∈ [1, N], m ∈ [1, M], k ∈ [1, N], h ∈ [1, M], N represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene;

7.2 calculate target rectangle slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^q; Wherein, k ∈ [1, N], h ∈ [1, M], N represents target rectangle slice of data D _adistance samples unit number, M represents target rectangle slice of data D _aazimuth sample cells number, D _arepresent the target rectangle slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations;

7.3 obtain target rectangle slice of data D _ain with Fourier basis functions W _khthe scattering point two-dimensional position (k, h) that (n, m) degree of correlation maximal value is corresponding ^qcorresponding strong scattering point G (k, h) ^q, extract this strong scattering point G (k, h) ^qstrong scattering point wherein, k ∈ [1, N], h ∈ [1, M], N represents target rectangle slice of data D _adistance samples unit number, M represents target rectangle slice of data D _aazimuth sample cells number, D _arepresent the target rectangle slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations.

8. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in steps of 5, and the described strong scattering point meeting gradient descent method condition contiguous weak scattering point region its expression formula is:

$G_{ext}^q(\tilde{k},\tilde{k})=G_{ext}^q(k,h)+{\mathrm{\ΣG}}_{ext}^q(k+i,h+j)$

$\begin{array}{cc}st.& grad\left(G_{\mathrm{\α}t}^q\right(k,h))>grad\left(G_{ext}^q\right(k+i,h+j));(-1\≤i\≤1,-1\≤j\≤1)\end{array}$

Wherein, represent the q time iterative extraction strong scattering point meeting gradient descent method condition out, grad () represents gradient algorithm, represent the q time iterative extraction strong scattering point meeting gradient descent method condition out one of them weak scattering point in the weak scattering point of surrounding eight vicinities ,-1≤i≤1 ,-1≤j≤1, k ∈ [1, N], h ∈ [1, M], n represents target rectangle region slice of data D _adistance samples unit number, M represents target rectangle region slice of data D _aazimuth sample cells number, D _arepresent the target rectangle region slice of data comprising P point target and land clutter, P represents the total number of point target in ground scene, and q represents iterations.

9. under a kind of strong clutter background as claimed in claim 1, the extraction of echo signal and super-resolution strengthen disposal route, it is characterized in that, in step 7, and the super resolution image that a described width is final obtain this width final super resolution image sub-step be:

9.1 use regularization method process one width to comprise the complete S AR image G of P point target, obtain the High Resolution SAR Images that a width is to be reconstructed utilize the High Resolution SAR Images that this width is to be reconstructed amplitude and this width High Resolution SAR Images to be reconstructed gradient distribution openness, the High Resolution SAR Images that this width of vectorization is to be reconstructed be optimized function its expression formula is:

$\min J\left(\stackrel{\‾}{G}\right)={||G-\mathrm{\Φ}\stackrel{\‾}{G}||}_2^2+k{||\stackrel{\‾}{G}||}_k^k+{\mathrm{\λ}}_2{||D\left|\stackrel{\‾}{G}\right|||}_k^k$

Wherein, Φ is expressed as picture operator, λ ₁, λ ₂all represent regularization parameter, represent that the High Resolution SAR Images that a width is to be reconstructed, G represent that a width comprises the complete S AR image of P point target, represent 2-norm, represent k-norm;

9.2 skip optimisation functions section 3 in expression formula, utilizes regularization method, calculates optimization function and then obtain the final super-resolution SAR image of a width