Subpixellic Methods for Sidelobes Suppression and Strong Targets Extraction in Single Look Complex SAR Images
Summary (3 min read)
Introduction
- Hence, a typical SAR image like the one shown in Fig. 1 (a) contains rather homogeneous areas with fluctuations due to speckle phenomenon (the larger the average intensity in the area, the larger are these fluctuations) and the signature of man-made structures in the form of intensities that are several orders of magnitude larger.
- Visually, these fluctuations appear grainy, but, more importantly, these correlations impact statistical methods for speckle reduction.
- Nonlinear processing methods were introduced in [5] under the name Spatially Variant Apodization (SVA) to reach this goal.
- By retaining at each pixel the smallest amplitude among all amplitudes obtained by applying the family of apodization functions, sidelobes are kept minimum while preventing the widening of the main lobe.
II. PSEUDO-RAW IMAGE AND PSEUDO-RAW SPECTRUM
- The SLC SAR images are, for reasons due to their acquisition process, band-limited and well-sampled signals.
- Ω̂ of size m× n, showing that the pixel spacing was adjusted during the sampling process to (over) satisfy the Shannon-Nyquist criterion.
- Besides, and as described in [29], the non-zero part of the Fourier spectrum is in fact apodized, which means that for any (α, β) ∈.
- The authors refer the reader to [3], [28] for more details about the computation of the pseudo-raw image u0 from u (in particular in the case when the frequency attenuating function γ is unknown).
- The reason why the spatial agencies introduce an apodization is to attenuate the sidelobes of the strong point target responses (highly present in urban areas) which are visible due to the SAR impulse response.
A. The cardinal sine impulse response model
- In stripmap mode, the SAR imaging system (acquisition + SAR synthesis) exhibits an approximate separability in range and azimuth, with rectangular spectra in both dimensions.
- Irregular resampling scheme proposed in [28], also known as Algorithm 1.
- In (11d), the value U0(k−tx, `− ty) can be efficiently computed by evaluating the Shannon interpolation of the mono-dimensional signal ` 7→ uxtx (k, `) at the subpixellic position `−ty (this operation simply involves the inner product between the mono-dimensional discrete signal and a cardinal sine function).
- In Fig. 5 (c), the authors observe an interesting sidelobes removal but the spreading effect of strong targets is still present and this image also suffers from a negative bias of the gray levels (the image is darker).
- The proposed irregular resampling procedure leads to a multi-look image with preserved spatial resolution and much well-localized targets.
C. Advantages and drawbacks
- The proposed irregular resampling strategy provides a simple and efficient alternative to the traditional use of apodization or SVA.
- More precisely, the complexities in time and memory are O(NTK|ω|) and O(NT |ω|) respectively, after the precomputation of NT FFTs in O(NT · |ω| log |ω|) during the initialization step.
- Unfortunately, irregularly resampling the two images independently introduces some decoherences between the images, and their interferometric phase is not well preserved by this procedure as the authors shall see in the next section.
- In summary, irregularly resampling pseudo-raw images in order to minimize the sidelobes effects of their neighboring bright targets leads to images that are at the same time of high resolution, statisti- cally accurate, and well-suited for visual interpretation.
IV. SPECKLE PLUS TARGETS DECOMPOSITION: A REVISITED CLEAN APPROACH
- One containing the bright targets (described by their sub-pixellic positions and complex amplitudes) and another one containing the image that would have been observed without the bright targets.the authors.
- The pixel with the brightest amplitude in the image is assumed to contain a target.
- Under that viewpoint, the CLEAN algorithm corresponds to the mere matching pursuit [37] while the RELAX algorithm is closer to the orthogonal matching pursuit [38].
- Is is important to note that both the CLEAN and RELAX approaches share the same weakness, they do not rely on a precise target detection criterion to decide whether a pixel of the image contains a strong target or not and the whole process does not rely on a satisfactory stopping criterion.
- Numerous approaches for target detection have been proposed in the literature (see e.g., [39], [40]).
A. A contrario detection of bright targets centers
- The a contrario methodology is a mathematical framework dedicated to the design of detectors providing a rigorous control of the number of false detections, that is, the average number of detections allowed in pure noise data.
- Thus, the quantity Ru0x (k0, `0) should efficiently measure the ratio between the target amplitude and the local reflectivity.
- The actual control of the average number of false detections (made in pure random data following H0) predicted by Proposition 1 is tested in Fig. 10.
- In practice, the authors observe that it can be slightly higher, which is due to the imperfect approximation of the distribution of the random variables {Ru0(k, `)}(k,`)∈ω by a Rayleigh function (in particular for small values of K).
B. Speckle plus targets decomposition
- One can remark that (19) also corresponds to the maximum likelihood estimator of A0 if the authors assume in (18) that w0 is a pure and stationary speckle (whatever its constant reflectivity).
- Usually, the practical implementation of the a contrario detectors consists in extracting the ε-meaningful structures in decreasing NFA order.
- Thus, in presence of several thousands of targets, the total execution time may exceed a day which can be problematic in many situations.
- In general situations, this strategy avoids the recomputation of Ru0 over all the domain ω and offers a nice reduction of the execution time (in practice, by a factor 15).
C. Applications of the decomposition
- First, it can be used to suppress the target sidelobes effect in the pseudo-raw image at arbitrary resolution.
- The image Dω′(C ) defined in (22) is a linear combination of discrete Diracs centered at different positions of the grid ω′, and does not present any sidelobes.
- In Fig. 15, the authors compare the coherence maps computed from apodized pairs of images (see Fig. 15 (a)), from pseudo-raw images (see Fig. 15 (b)), from irregular resamplings of those pseudo-raw images (see Fig. 15 (c)), or from the images obtained by applying the Rω operator to the pseudo-raw images (see Fig. 15 (d)).
- On the other hand, over-coherent values can also be found in Fig. 15 (b) along the sidelobes of the strong targets (see the cross shape on the left side of Fig. 15 (b)).
V. CONCLUSION AND PERSPECTIVES
- The authors addressed in two different ways the issue of sidelobes suppression with no loss of resolution and statistics preservation for pseudo-raw images.
- The first proposed approach consists in resampling the pseudo-raw image over an irregular grid that efficiently cancels the sidelobes in the vicinity of the strong targets while preserving the speckle statistics (in particular the valuable spatial uncorrelation of the speckle in the pseudo-raw images) in fully developed speckle areas.
- The derived revisited CLEAN algorithm exhibits a higher computational complexity than the resampling algorithm (its complexity is proportional to the size of the image multiplied by the number of targets found in the image).
- The decomposition of the image that it provides is particularly suited to numerous SAR applications.
- On the contrary, the authors can decide to only focus on the speckle-dominated component to perform tasks such as denoising, segmentation, classification, which may reveal more efficient in the absence of targets.
Did you find this useful? Give us your feedback
Citations
226 citations
72 citations
Cites methods from "Subpixellic Methods for Sidelobes S..."
...In this paper, the noisy TerraSAR-X images are decorrelated using the method proposed in [40] and the Sentinel-1 images are decorrelated by resampling because of its special acquisition model (the beam both steering in range direction and steering from backward to forward in azimuth direction)....
[...]
57 citations
Cites background from "Subpixellic Methods for Sidelobes S..."
...Whitening the spectrum [18], [25] or down-sampling the image...
[...]
...man’s fully developed speckle model generally assume an absence of spatial correlations [17], which is not the case in actual SAR images synthetized by space agencies [18], [24]....
[...]
33 citations
17 citations
Cites methods from "Subpixellic Methods for Sidelobes S..."
...For instance, a tailored procedure, decomposing single-polarization SAR imagery in speckle dominated areas and point targets, was proposed in [33]....
[...]
References
9,380 citations
"Subpixellic Methods for Sidelobes S..." refers methods in this paper
...Under that viewpoint, the CLEAN algorithm corresponds to the mere matching pursuit [37] while the RELAX algorithm is closer to the orthogonal matching pursuit [38]....
[...]
6,712 citations
"Subpixellic Methods for Sidelobes S..." refers background in this paper
...It follows that Uc0 is a band-limited signal which can be reconstructed exactly (neglecting sensor noise), according to the Shannon-Whittaker Sampling Theorem [30], [31], provided an infinite number of its samples {Uc0 (k · δr, ` · δaz)}(k,`)∈Z2 are observed at regularly spaced locations, with steps δr ≤ 1 along the range direction and δaz ≤ 1 along the azimuth direction....
[...]
...Sampling Theorem [30], [31], provided an infinite number of its samples {U 0 (k · δr, ` · δaz)}(k,`)∈Z2 are observed at regularly spaced locations, with steps δr ≤ 1 along the range direction and δaz ≤ 1 along the azimuth direction....
[...]
5,415 citations
"Subpixellic Methods for Sidelobes S..." refers methods in this paper
...Rather than locally selecting the best suited apodization, spectral super-resolution techniques (Capon [8] or APES [9]) are applied to improve the localization of point targets....
[...]
4,607 citations
"Subpixellic Methods for Sidelobes S..." refers methods in this paper
...Under that viewpoint, the CLEAN algorithm corresponds to the mere matching pursuit [37] while the RELAX algorithm is closer to the orthogonal matching pursuit [38]....
[...]
3,963 citations
"Subpixellic Methods for Sidelobes S..." refers background in this paper
...In particular, many works are devoted to identifying weak targets or moving targets [11]–[15], under foliage targets [16], using polarimetric data [17]–[19] or interferometric data [20], [21]....
[...]
Related Papers (5)
Frequently Asked Questions (8)
Q2. What future works have the authors mentioned in the paper "Sub-pixellic methods for sidelobes suppression and strong targets extraction in single look complex sar images" ?
It would be also interesting to investigate the possibility to design SAR image processing models that would rely on the two components jointly.
Q3. What is the main idea behind the revisited CLEAN procedure?
Their revisited CLEAN procedure relies on an efficient sub-pixellic target detection criterion based on the so-called a contrario methodology [43], which leads to a well justified stopping criterion and an accurate control of the false alarms.
Q4. What is the function that is used to determine the tx of the u0 field?
Initialization: precompute the horizontal and vertical translations of u0, that is, compute for all t ∈ T, uxt = U0(ω − (t, 0)) and uyt = U0(ω − (0, t)).
Q5. What is the advantage of the a contrario framework?
Compared to the classical statistical decision theory, the a contrario framework presents the advantage to get rid of the design of a H1 hypothesis, making the a contrario algorithm less sensitive to the modeling choice for the structures that the authors want to detect.
Q6. what is the recombination of C into discrete Diracs on the grid?
C the recombination of C into discrete Diracs on the grid ω′, which is defined by∀(x′, y′) ∈ ω′, Dω′(C )(x′, y′) = ∑(x,y,A)∈CAδπω′ (x,y)(x ′, y′) , (22)where πω′(x, y) = argmin(x′,y′)∈ω′ ‖(x − x′, y − y′)‖ denotes a projection of (x, y) over ω′, and δπω′ (x,y)(x ′, y′) is defined by∀(x′, y′) ∈ ω′ , δπω′ (x,y)(x ′, y′) = { 1 if (x′, y′) = πω′(x, y) 0 otherwise,so that δπω′ (x,y) simply represents a discrete Dirac centered at the position πω′(x, y) ∈ ω′.
Q7. Why is the amplitude of the target smeared in the vicinity of its?
due to the sampling, the total amplitude of the target can be smeared in the vicinity of its center (this is the sidelobe effect) which makes difficult the estimation of the reflectivity of its surrounding area.
Q8. How can the continuous signal Uc0 be modeled?
Under this model, the continuous signal Uc0 : R2 → C, before sampling, can be modeled as the convolution between the continuous latent scene and the impulse response.