Summary of the invention
The object of the invention is in order to suppress the azimuth ambiguity of spaceborne band pattern SAR image, propose a kind of satellite-borne synthetic aperture radar band pattern azimuth ambiguity inhibition method.The SAR image orientation that the method can be removed the regions such as land and sea junction is effectively fuzzy, improves picture quality, and then strengthens the interpretability of image.
A kind of satellite-borne synthetic aperture radar band pattern azimuth ambiguity inhibition method of the present invention, comprises following step:
Step 1: determine position the classification of fuzzy region in SAR image
First, according to SAR operational factor, determine the side-play amount of azimuth ambiguity; Secondly, in conjunction with antenna radiation pattern, determine azimuth ambiguity present position in image; Last according to the size of blurred block, be divided into isolated fuzzyly and region is fuzzy, process respectively.
Step 2: adopt compression sensing method to remove isolated fuzzy
Two dimensional image is changed into a dimensional vector, utilize wavelet transformation and Gauss's sharpening matrix to obtain the rarefaction representation of image, re-use orthogonal matching pursuit (Orthogonal Matching Pursuit, OMP) method is rebuild image, finally isolated blurred block in original image is substituted with rebuilding image, remove isolated fuzzy.
Step 3: adopt the image mending method removal region based on masterplate fuzzy
Fuzzy for region, first calculate on fuzzy region border filling relative importance value a little, and determine the highest relative importance value region.Secondly, at source region, find immediate patch piece and fill fuzzy region.By that analogy, the filling that progressively extends internally along blurred block border, until complete the processing of All Ranges blurred block.
The invention has the advantages that:
(1) specific aim.Than traditional methods such as bandpass filtering, the method that the present invention proposes is directly judged azimuth ambiguity position, directly acts on targetedly fuzzy region, has reduced the impact on normal region.
(2) validity.Traditional methods such as bandpass filtering can reduce signal to noise ratio (S/N ratio) (signal-to-noise ratio, SNR), even can strengthen speckle noise.In addition, classic method is also difficult to remove completely fuzzy, still has part blurred signal residual on image.There are not the problems referred to above in the method that the present invention proposes.
Embodiment
Below in conjunction with drawings and Examples, the present invention is described in further detail.
The present invention is a kind of satellite-borne synthetic aperture radar band pattern azimuth ambiguity inhibition method, and flow process as shown in Figure 1, comprises following step:
Step 1: determine position the classification of fuzzy region in SAR image
Be specially:
(1) by haplopia complex pattern (Single-Look-Complex, SLC), obtain local average image, establishing former haplopia complex pattern is S, and the local average image obtaining is S
μ:
Wherein, S and S
μbe the complex matrix of m * n, m be orientation to always counting, n is for distance is to always counting, a be orientation to variable, r be apart to variable.L is the rectangle local average window centered by impact point, and the size of window is m
a* n
r, m
afor counting to window in orientation, n
rfor distance, to window, count, can choose m
a=40, n
r=15
(2) obtain fuzzy orientation to distance to skew
Wherein, Δ a be blurred signal orientation to skew, Δ r be blurred signal distance to skew, f
ddoppler centroid, f
rdoppler frequency rate, V
gsatellite ground speed, f
pbe pulse repetition rate, λ is SAR operation wavelength.
(3) obtain fuzzy ratio
Wherein, G (f
d) be antenna radiation pattern, f
dinstantaneous Doppler frequency, L
abe orientation to antenna length, R
refbe center constantly with reference to oblique distance, ξ is fuzzy ratio.
(4) obtain fuzzy judgment matrix B
1(a, r):
Wherein, Δ a be orientation to side-play amount, Δ r be distance to side-play amount, " 1 " representative exist fuzzy, " 0 " representative be not fuzzy.
(5) obtain fuzzy judgment matrix B
2(a, r):
Wherein, " 1 " representative exists fuzzy, and " 0 " representative is not fuzzy.
(6) obtain fuzzy judgment matrix B (a, r):
B(a,r)=B
1(a,r)∪B
2(a,r)
B (a, r) is B
1(a, r) and B
2the union of (a, r)." 1 " representative exists fuzzy, and " 0 " representative is not fuzzy.Fuzzy judgment matrix B (a, r) and former haplopia complex pattern S, local average image S
μon location of pixels, be one to one.So far, all lituras in image have been judged.
(7) according to fuzzy judgment matrix, carry out blurred picture classification.If have, surpass 49 fuzzy elements adjacent (noting: an element is adjacent with its 8 elements around), judge that these fuzzy elements belong to region fuzzy, and these fuzzy elements are all set to 2.At this moment " 0 " indicates without fuzzy, and " 1 " represents isolated fuzzy, and " 2 " represent that region is fuzzy.
Isolated fuzzy employing compressed sensing (Compressed Sensing, CS) method is removed (step 2), and image mending (the Exemplar-Based Image Inpainting) method of the fuzzy employing in region based on model removed (step 3).
Step 2: use compression sensing method to remove isolated fuzzy
Be specially:
(1) former haplopia complex pattern S is N=m * n mono-dimensional vector S' by row generate length:
S'=S(a,1)∪S(a,2)∪…∪S(a,n)
Wherein, a is that orientation is to variable.
(2) to S', sampling obtains the dimensional vector y that length is M
s:
y
s=ΦS'
Wherein, Φ is sampling matrix, only contains " 0 " and " 1 ".Every row only has 1 " 1 ", altogether M " 1 ".Obviously the column number of " 1 " has determined the position of sampled point in original image.If column number is y, sample coordinate position is
noting, do not sample isolated ambiguous location, is the element position of " 1 " in fuzzy judgment matrix B (a, r).
(3) build sparse property
Order:
S'=Γ
-1Ψ
TS
w
:
y
s=ΦΓ
-1Ψ
TS
w=ΘS
w
Θ=ΦΓ
-1Ψ
T
Wherein, Γ is Gauss's sharpening matrix (diagonal matrix, diagonal entry meets Gaussian distribution), and Ψ is wavelet transformation base.Θ can be obtained fom the above equation, Θ is called sensing matrix, S
wfor treating restructuring matrix.
(4) solve l
1norm minimum problem:
Also, meeting y
s=Θ S
wprerequisite under, reconstruct l
1the S of Norm minimum
w;
(5) adopt orthogonal matching pursuit algorithm to be reconstructed, concrete grammar is as follows:
(a) known vector of samples y
s, sensing matrix Θ.Making t is the circulation moment, r
t, Λ
t, Β
t, S
wtbe respectively t residual error constantly, indexed set, rebuilds atom set, treats restructuring matrix.Initialization r
0=y
s,
order circulation is t=1 constantly;
(b) find out residual error r
t-1respectively be listed as θ with sensing matrix Θ
jinner product the maximum, record the corresponding footmark λ of maximum column
t, that is:
λ
t=j
max=argmax
j=1…N|<r
t-1,θ
j>|
(c) upgrade indexed set Λ
t=Λ
t-1∪ { λ
t, the reconstruction atom set in the sensing matrix that record finds
(d) by least square method, obtain immediate S
wt,
(e) upgrade residual error
(f) if || rt||
2< δ (wherein δ is for stopping thresholding, generally, can choose δ=0.1 * || y
s||
2), stop iteration, output end product
if do not meet, return to step (b), continue iterative computation;
(g) by S'=Γ
-1Ψ
ts
wsolve S';
(6) the isolated fuzzy pixel in former haplopia complex pattern S is substituted with corresponding element in S'.So far, obtain removing isolated fuzzy image S
1.
Step 3: adopt the image mending method based on masterplate, removal region is fuzzy
Be specially:
(1) by S
1middle region vague image vegetarian refreshments (correspondence position is that 2 element is determined in fuzzy matrix) is whole 0, obtains X piece white space.Fig. 2 has provided the schematic diagram of a typical white space.
(2) establish the confidence level that in C (p) token image, p is ordered.Initialization C (p), order
and
point on the δ Ω of border, makes confidence level
(Ψ as shown in Figure 2,
pfor next step rectangular area centered by p point that is about to fill up, | Ψ |
pfor region Ψ
parea).Make weighted direction
(as shown in Figure 2,
for linear trend vector, be Ψ
pthe interior image gradient maximal value direction of ∩ (I-Ω), n
pfor the normal vector of p Dian Chu border δ Ω, α is quantizing factor, typical gray-scale map α=255).Make relative importance value
see intuitively, confidence level C (p) is in rectangular area to be filled up, and non-NULL part confidence level sum, divided by rectangular area.Weighted direction D (p) is the projection of the maximum direction of gradient on normal, divided by quantizing factor.
(3) white space is carried out to image mending, method is as follows:
(a) making t is the circulation moment, Ω
tfor t white space constantly, δ Ω
tfor t white space border constantly.Initialization t=1.
(b) determine t white space border δ Ω constantly
t;
(c) for border δ Ω
ton institute a little, calculate relative importance value P (p);
(d) for border δ Ω
ton institute a little, find the maximum point of relative importance value P (p)
that is:
Central point
after drawing, region to be filled up
determine immediately.
(e) in source region I-Ω, find and
immediate patch piece
for the central point of patch piece, that is:
Operational symbol wherein
represent
middle non-NULL pixel value and
qthe quadratic sum of the difference of respective pixel, for:
(f) for
will
copy to
that is,, by the blank in region to be filled up, with patch piece, directly fill.
(g) upgrade confidence level C (p),
also, upgrade the confidence level of being filled up region.Carry out t=t+1.
(h) if now white space filled complete,
stop iteration.If do not meet, return to step (b), continue to calculate.
(4) to all white space executable operations (3), until X piece white space all fills up complete.
(5) so far, all isolated fuzzy and region is fuzzy all removes, obtain finally removing the image S2 that azimuth ambiguity disturbs.Embodiment
For validity of the present invention is described, use this method to process actual measurement TerraSAR-X image.Required processing parameter is as shown in table 1.
Table 1 embodiment parameter
Fig. 3 has provided result.
Fig. 3 (a) is original image, can significantly find that the strong scattering target of picture centre forms azimuth ambiguity in upper and lower two parts, and wherein top is fuzzy due on sea thereby more obvious.In addition, the upper left white box of image is amplified and processed, can find wherein to exist speckle noise.
Fig. 3 (b) is the result after step 1 is processed, and what wherein white was decorated with grid is the definite fuzzy region of this method, can find that the azimuth ambiguity piece on image has all been distinguished and marked out, and classification situation is very desirable.
Fig. 3 (c) is traditional band-pass filtering method result, and removal is unclean can to find azimuth ambiguity, on the sea of image top, still can find that there is blurred signal remnants.In addition, from top, little figure can find, it is stronger that speckle noise becomes.
Fig. 3 (d) is result of the present invention, does not find that obvious azimuth ambiguity is remaining, and meanwhile, intensity of speckle noise does not change.
Therefore, the satellite-borne synthetic aperture radar band pattern azimuth ambiguity inhibition method that the present invention proposes can suppress azimuth ambiguity effectively, improves resolution and the recognition capability of target.