betaFeedback

WorksEngineering

Spatial phase coherence and its applications to masking and adaptive neighborhood sizing in multitemporal InSAR

Abstract

We examine spatial phase coherence (SPC), a measure of the correlation between the wrapped phase time series of two pixels in a stack of single-look synthetic aperture radar acquisitions. A pixel and its adjacent neighbors share speckle through the point spread function, so that decorrelation alone does not lower their SPC below a floor. We find that masking pixels whose mean SPC to their eight neighbors is well below that floor, as in parts of layover and shadow areas, is an inexpensive complement to geometry-derived layover and shadow masks. For distributed scatterer pixels, averaging SPC over a fixed set of neighbors that share no speckle yields a proxy for the local coherence, the phase coherence indicator (PCI), that is available before the selection of statistically homogeneous pixels (SHPs). In phase estimation, averaging more pixels reduces noise but may also increase the bias in heterogeneous areas. To relax this trade-off, we propose to enhance SHP selection algorithms by scaling the search window size according to the PCI and truncating the set of selected pixels where a smaller nested set agrees clearly better with the phases at the center. In simulations of a TerraSAR-X and a Sentinel-1 stack with 44 and 79 acquisitions, we apply this adaptive neighborhood sizing (ANS) to established amplitude- and covariance-based SHP selection algorithms with thresholds that adapt to the PCI. Without ANS, the largest tested search window size has 13 to 29 % more phase linking error than the best of the tested sizes. With ANS, the largest window has the lowest error of the tested sizes, 11 % lower than that of the same selection at its best tested size without ANS. On a real TerraSAR-X stack, the mask based on SPC flags pixels in building shadows that a terrain model misses. On real Sentinel-1 stacks, ANS preserves dense fringes on a slowly deforming slope and in part on a fast landslide.

main.pdf

Introduction

In multitemporal interferometric synthetic aperture radar (InSAR), phase estimation and phase linking over distributed scatterers rely on averaging observations from selected neighboring pixels ((Ho Tong Minh and Tebaldini, 2023)). This raises three questions for each pixel to be estimated, namely whether an average over its neighbors can represent it at all, which neighbors are statistically homogeneous with it, and how large the set of averaged pixels should be. We call the pixel to be estimated the center pixel, or center, and the set of averaged pixels the ensemble. The first question arises in particular for pixels in layover and shadow areas, which are typically masked. In layover, scatterers at several heights are projected onto the same pixel in the synthetic aperture radar (SAR) image, so that subtracting a topographic phase may introduce systematic phase inconsistencies. In shadow, the phase is often dominated by noise that is less correlated between neighboring pixels than speckle. Geometric layover and shadow maps derived from the acquisition geometry and a digital elevation model (DEM) (Kropatsch and Strobl, 1990) can identify these areas, but they depend on the accuracy of the DEM. Layover and shadow caused by objects that are not included in the model, such as buildings and trees, are another limitation. Layover has also been detected from the data itself. Tupin et al. (1996) use the amplitude, Zhu et al. (2017) the local fringe frequency of an interferogram over high-rise buildings, and Rossi and Eineder (2015) the number of times each SAR pixel occurs in the geocoding of an elevation model generated from the interferometric data, without external data. Tupin et al. (1996) also detect areas without topographic phase from the filtered coherence and the confidence of the local fringe frequency estimates of an interferogram. These areas include shadow, decorrelated areas and surfaces without backscatter.

Established processing frameworks such as SqueeSAR [3] identify statistically homogeneous pixels (SHPs) by tests on the temporal amplitude distributions [14] (Even and Schulz 2018). Later tests also use the interferometric phases [24] (Dong et al. 2018; Song et al. 2017). Other approaches collect pixels of similar intensity by region growing (Vasile et al., 2006) or weight pixels by the similarity of their surrounding patches (Deledalle et al., 2011, 2015). SHP tests that compare the temporal mean amplitudes or intensities of two pixels lose their significance level when the speckle is temporally correlated. On a coherent target the speckle changes little between acquisitions, so that the means of two pixels of the same surface keep the difference of their speckle. A threshold set for independent samples or for one assumed coherence therefore holds only at that coherence.

Averaging reduces noise, but its extent determines how much spatial detail is retained. Strongly decorrelated targets need larger ensembles to reach a given precision (cf. Rocca, 2007, Eqs. (5.22) to (5.25)). Where the phase varies rapidly in space, as in areas with strong spatial gradients of line-of-sight deformation, a smaller ensemble may suffice if the coherence is high enough, whereas a large one blurs fringes and lowers the estimated coherence (Zebker and Chen, 2005). The averaging neighborhood has been adapted to this in shape and size. For single interferograms, directional windows follow the local fringe orientation (Lee et al., 1998), and the intersection of confidence intervals (ICI) rule chooses, for each pixel, the largest window whose local polynomial estimate stays consistent with those of smaller windows [8] (Katkovnik et al. 2002). For stacks, Pepe et al. [15] adapt a fixed multilook window to the phase by removing the pixels whose phases deviate from the multilooked phases of that window and comparing the deviations of the remaining pixels with those of the pixel that deviates least. In other SHP-based processing, however, the coherence matrix is usually estimated from all pixels that the SHP selection accepts (e.g., [6, 24]), also where the phase varies within the search window. Chi et al. (2022, Section 3.2 and Eq. (13)) instead shrink the search window of an SHP selection at each pixel until a coefficient of variation of the phase of the time-averaged differential interferogram falls below a threshold, which requires unwrapped interferograms.

In this paper, we examine a simple quantity, the squared magnitude of the mean phasor of the relative phase of two pixels across acquisitions, which we refer to as spatial phase coherence (SPC), and its use for all three questions. The quantity itself is known under several names (Section 2.2). It is used in directional statistics (Mardia and Jupp, 2000) and in InSAR for persistent scatterers (Ferretti et al., 2000), and in related forms for phase stability (Hooper et al., 2004, 2007), point coherence estimation (Vecchioli et al., 2023) and SHP selection [13, 15, 17]. Less attention has so far been paid to the behavior of the statistic on single-look data with spatially correlated speckle, and to the uses of its properties in the processing chain, from masking to the size of the ensemble that an SHP selection returns. Our contributions are as follows.

  1. A statistical model of SPC on single-look data, with (a) its expectation under the circular complex Gaussian (CCG) model, (b) its floor under speckle that adjacent pixels share through the system point

spread function, predicted from the processing parameters, and (c) an identity that links pairwise SPC to the mean squared phase-only coherence of an ensemble.

  1. Two maps, the neighborhood SPC (NSPC) and the phase coherence indicator (PCI), computed from fixed sets of neighbors at the cost of only one inner product over the acquisitions per neighbor, and available directly from the single-look complex (SLC) stack. A threshold on NSPC, the average SPC of a pixel with its eight direct neighbors, flags pixels where this average lies below the floor that shared speckle sets, and complements geometric layover and shadow maps. The PCI, an average of SPC over neighbors that share no speckle with the pixel, rises with its coherence and steers the SHP selection and the sizing of the ensemble, as described below.

  1. Thresholds for SHP tests that compare temporal mean amplitudes or intensities. They depend on the PCI and keep the false rejection rate of the order of the significance level at every coherence, also where the speckle of single-look data is temporally correlated.

  1. Adaptive neighborhood sizing (ANS) in two parts: a search window scaled with the PCI (S), and a truncation step (T) that reduces a selected ensemble to a smaller nested one only where that one agrees clearly better with the phases of the center and of the pixels that share its speckle. The agreement is measured with a generalization of SPC, and ANS can be combined with any SHP selection.

The two maps, the thresholds and ANS are evaluated in simulations with known phase and demonstrated on TerraSAR-X data over Visp, Switzerland, and Sentinel-1 data over Gilgit-Baltistan, Pakistan. ANS is combined with three kinds of SHP selection: (a) the amplitude-based fast SHP selection (FaSHPS) of [6], which tests the temporal amplitudes of two pixels, (b) the intensity- and phase-based covariance matrix patch (CMP) selection of [24] and (c), as their reference, all pixels of the search window (the full window). In § 2.1, we develop the theory, and in § 2.2, we define the mask, the PCI, the SHP selections, the two parts of ANS and the estimator of the coherence matrix. In § 2.3 and § 3, we present the simulations and the real data, including the runtime. In § 3.2, we discuss the applicability of the two maps and of ANS, the variants not adopted, and limitations.

Spatial phase coherence

Definition

Consider a stack of N>1N > 1 coregistered SLC images in radar coordinates, and let xp(n)x_p(n) denote the complex value of pixel pp in acquisition nn. A pixel is identified with its integer image coordinates (azimuth, range), so that q−pq-p denotes the spatial lag from pixel pp to pixel qq. A model phase, such as a simulated topographic phase, may have been subtracted beforehand (§ 2.1.6). For nonzero observations, define

up(n)=xp(n)∣xp(n)∣=eiϕp(n),up=(up(1),…,up(N))T.(1)u_p(n) = \frac{x_p(n)}{\lvert x_p(n)\rvert} = e^{i\phi_p(n)}, \qquad\mathbf{u}_p = \left(u_p(1),\ldots,u_p(N)\right)^{\mathsf{T}}. \tag*{(1)}

The coherence of two acquisitions s,ts,t is the normalized covariance of their complex values (Even and Schulz, 2018, Eq. (2)). It is estimated by the sample coherence over a neighborhood W\mathcal{W} (Even and Schulz, 2018, Eq. (8)), whose phase-only counterpart is

γ^st∞(W)=∣1∣W∣∑q∈Wuq(s)uq(t)‾∣.(2)\widehat{\gamma}_{st}^{\infty}(\mathcal{W}) = \left\lvert\frac{1}{\lvert\mathcal{W}\rvert} \sum_{q\in\mathcal{W}} u_q(s)\overline{u_q(t)} \right\rvert. \tag*{(2)}

Exchanging the roles of pixels and acquisitions in Eq. (2), i.e., fixing two pixels p,qp,q and averaging over the acquisitions (Figure 1), gives the SPC of the two pixels, which we define as the square of the resulting magnitude,

Figure 1. Illustration of phase-only coherence and SPC on a stack of images. (a) The phase-only coherence (Eq. (2)) of two acquisitions ss, tt is built from the sum of the products uq(s)uq(t)u_q(s)u_q(t) over the pixels qq of a neighborhood W\mathcal{W}, the same in both images. (b) SPC (Eq. (3)) of two pixels pp, qq is built from the sum of the products up(n)uq(n)u_p(n)u_q(n) over the NN acquisitions of the stack.

Spq=∣1N∑n=1Nup(n)‾uq(n)∣2=∣upHuq∣2N2.(3)S_{pq} = \left\lvert\frac{1}{N}\sum_{n=1}^{N}\overline{u_p(n)}u_q(n) \right\rvert^2 = \frac{\left\lvert\mathbf{u}_p^{\mathsf{H}}\mathbf{u}_q \right\rvert^2}{N^2}. \tag*{(3)}

It satisfies 0≤Spq≤10 \le S_{pq} \le1, with Spq=1S_{pq}=1 precisely when the two phase vectors differ by a constant offset. SPC is invariant to constant phase rotations of either pixel and to any acquisition-dependent rotation common to both. Referencing all phases to one acquisition is a rotation of the first kind and leaves SPC unchanged, provided that the reference acquisition, whose phase is then zero at every pixel, is kept as one of the NN terms of the sum. Like Eq. (2), SPC uses the phases only, as do the estimates of the coherence matrix in this paper (§ 2.2.8). Exchanging the roles in the sample coherence instead would give the correlation coefficient of the two complex time series. Bright samples and acquisitions would dominate it, and its expectation, the floor of § 2.1.5 and the thresholds derived from it would depend on amplitude changes over the acquisitions, e.g., from soil moisture or vegetation. SPC does not change under any positive scaling of the samples, and the amplitudes enter the processing only through the SHP tests (§ 2.2.4 and § 2.2.5).

If the relative phases ϕq(n)−ϕp(n)\phi_q(n)-\phi_p(n) are independent across the acquisitions and uniformly distributed, E[Spq]=1/N\mathbb{E}[S_{pq}]=1/N. Removing this bias and rescaling gives the corrected SPC and its clipped version

Qpq=NSpq−1N−1,Qpq+=max⁡(Qpq,0),(4)Q_{pq}=\frac{NS_{pq}-1}{N-1}, \qquad Q_{pq}^{+}=\max(Q_{pq},0), \tag*{(4)}

respectively. Where the distinction does not matter, we also call QpqQ_{pq}, Qpq+Q_{pq}^{+} and their square roots SPC.

Relation to existing measures

The quantities in Eqs. (3) and (4) are known in InSAR and in other scientific fields under different names. Section S1.1 of the supplement gives the details of the relations below. In directional statistics, testing SpqS_{pq} against a threshold is the Rayleigh test of uniformity, applied here to the relative phases ((Mardia and Jupp, 2000, Section 6.3.1, Eqs. (6.3.3) and (6.3.4)). In InSAR, with the phase model of the multi-image coherence set to zero, Spq\sqrt{S_{pq}} has the form of the multi-image phase coherence of a pair of neighboring persistent scatterers pp and qq ((Ferretti et al., 2000, Eq. (2.7))), summed over the acquisitions instead of single-reference interferograms. Vecchioli et al. (2023) estimate the coherence of each point of a stack from such coherences of pairs of nearby points, computed after a motion model is fitted to each pair. They assume that the phase noise of two points is independent, which may require leaving out the nearest pixels of oversampled images. The expected coherence of a pair is then the product of the coherences of its two points, and the logarithms of these products form an overdetermined linear system for the coherence of every point. The phase-stability measure of Hooper et al. (2007, Eq. (20)) applies the same construction, the magnitude of a mean phasor, to the phases of one pixel relative to the filtered phase of the surrounding pixels, after a fitted look angle error is removed. For SHP selection, the zero-lag phase cross-correlation of Narayan et al. [13] has magnitude NISpqN_I\sqrt{S_{pq}} when SpqS_{pq} is computed over the NIN_I single-reference interferograms that it uses, instead of over the acquisitions. The magnitude of the centered complex correlation of Shi et al. [17]Eq. (1)) differs from $\sqrt{S_{pq}}$ in general and coincides with it when both temporal phasor means vanish. The circular variance used for adaptive multilooking by {citet}Pepe et al. (2021, Eqs. (9) and (10)) is one minus the magnitude of a mean phasor, taken over the interferograms and computed between the interferometric phases of a pixel and the multilooked phases of the window rather than those of a second pixel. The SPC formula is therefore not a contribution of this paper. The contributions concern its behavior on single-look SAR data and its use for masking, for an indicator of local coherence that steers the SHP selection and for the size of an ensemble. Its statistics have also been examined in neuroscience: QpqQ_{pq} is the pairwise phase consistency of Vinck et al. [21]Eqs. (14) and (15), an unbiased estimator of squared phase synchrony between neuronal signals recorded over repeated trials, and $\sqrt{S_{pq}}$ is the phase-locking value of {citet}Lachaux et al. (1999, p. 195). Acquisitions take the place of trials here. The unbiasedness of the pairwise phase consistency assumes independent, identically distributed trials. Acquisitions are neither, and $\mathbb{E}[Q_{pq}]$ is then the mean expected cosine agreement ({xref}Eq. (5)`).

Agreement across interferograms and coherence of an ensemble

Phase linking uses the coherence of an ensemble, whereas SPC compares pixels pairwise. On average the two agree: an ensemble is coherent over the interferograms exactly when its pixels agree in phase pairwise over the acquisitions, and the coherence of an ensemble is maximal in every interferogram exactly when the SPC of each pair of its pixels is maximal. Write Δϕpst=ϕp(s)−ϕp(t)\Delta\phi_{p}^{st}=\phi_p(s)-\phi_p(t), vpst=eiΔϕpstv_p^{st}=e^{i\Delta\phi_p^{st}} and M=(N2)M=\binom{N}{2}. Expanding Eq. (3) gives

Qpq=1M∑s<tcos⁡(Δϕpst−Δϕqst).(5)Q_{pq}=\frac{1}{M}\sum_{s<t}\cos\left(\Delta\phi_p^{st}-\Delta\phi_q^{st}\right). \tag*{(5)}

Corrected SPC is thus the mean cosine of the difference between the interferometric phases of the two pixels over all MM interferograms (§ 13.1.1 of the supplement), computed in O(N)O(N) operations without forming the interferograms or a coherence matrix. The identity with the coherence of an ensemble holds for any set E\mathcal{E} of L=∣E∣≥2L=|\mathcal{E}|\ge2 pixels,

1M∑s<tLγ~st(E)2−1L−1=Qˉ(E)=1L(L−1)∑q,q′∈Eq≠q′Qqq′.(6)\frac{1}{M}\sum_{s<t}\frac{L\tilde{\gamma}_{st}(\mathcal{E})^2-1}{L-1}=\bar{Q}(\mathcal{E})=\frac{1}{L(L-1)}\sum_{\substack{q,q'\in\mathcal{E}\\q\ne q'}}Q_{qq'}. \tag*{(6)}

The left side corrects the squared coherence magnitude of each interferogram for its bias 1/L1/L at zero coherence for LL independent looks (cf. Touzi et al. (1999, Eq. (5)), as Eq. (4) corrects SpqS_{pq}. The mean corrected SPC over all pairs of an ensemble is thus the mean corrected squared phase-only coherence of that ensemble (§ 13.1.1 of the supplement). Unlike this mean over all pairs, which includes pairs that do not contain the center, the maps of § 2.2.1 and § 2.2.2 are built from the pairs of each pixel with fixed neighbors and describe that pixel and its immediate surroundings. The two maps, NSPC and the PCI, need eight and 24 inner products of length NN per pixel, against O(LN2)O(LN^2) operations for the coherence of an ensemble of LL pixels.

Expected value under the CCG model

The expected value of SPC shows how much of a low SPC comes from decorrelation and how much from different true phase histories. It follows from the expectation of each cosine in Eq. (5), which depends on the distribution of the single-look interferometric phase Δϕ\Delta\phi of an interferogram with coherence γ\gamma. Let gg denote the mean cosine of its phase error Δϕ−arg⁡γ\Delta\phi-\arg\gamma (cf. Just and Bamler, [7]Eq. (15); {citet}Lee et al., 1994, Eq. (21)), in the closed form of {citet}López-Martínez and Fàbregas (2002, Eq. (8))`, with 2F1_2F_1 the Gauss hypergeometric function,

g(∣γ∣)=E[cos⁡(Δϕ−arg⁡γ)]=π4∣γ∣ 2F1(12,12;2;∣γ∣2).(7)g(|\gamma|)=\mathbb{E}[\cos(\Delta\phi-\arg\gamma)]=\frac{\pi}{4}|\gamma|\,{}_2F_1\left(\frac{1}{2},\frac{1}{2};2;|\gamma|^2\right). \tag*{(7)}

For example, g(0.2)=0.16g(0.2) = 0.16, g(0.5)=0.41g(0.5) = 0.41, g(0.8)=0.70g(0.8) = 0.70. Suppose pixels pp and qq are independent and each follows a zero-mean CCG distribution with true coherence matrix entries γstp\gamma_{st}^{p}, γstq\gamma_{st}^{q} (not sample estimates) and true phase histories θp\theta_p, θq\theta_q, with arg⁡γstp=Δθpst=θp(s)−θp(t)\operatorname{arg}\gamma_{st}^{p} = \Delta\theta_{p}^{st} = \theta_p(s) - \theta_p(t). Since pp and qq are independent and the phase error of a CCG pixel is symmetric about zero (Section S1.2 of the supplement), taking expectations in Eq. (5) gives

E[Qpq]=1M∑s<tg(∣γstp∣)g(∣γstq∣)cos⁡(Δθpst−Δθqst).(8)\mathbb{E}[Q_{pq}] = \frac{1}{M}\sum_{s<t} g\left(\lvert\gamma_{st}^{p}\rvert\right)g\left(\lvert\gamma_{st}^{q}\rvert\right)\cos\left(\Delta\theta_{p}^{st}-\Delta\theta_{q}^{st}\right). \tag*{(8)}

Decorrelation lowers E[Qpq]\mathbb{E}[Q_{pq}] through g2g^2, and differences in phase history through the cosine. Even two pixels with the same true phase history and ∣γst∣=0.5\lvert\gamma_{st}\rvert= 0.5 for all acquisition pairs give E[Qpq]≈0.17\mathbb{E}[Q_{pq}] \approx0.17, so that where the coherence is low, a single value of SPC says little about whether two pixels share a true phase history.

When the relative phasors zn=up(n)‾uq(n)z_n = \overline{u_p(n)}u_q(n) are independent across acquisitions with means ξn\xi_n, the expectation (Section S1.2 of the supplement) is

E[Qpq]=∣∑nξn∣2−∑n∣ξn∣2N(N−1),(9)\mathbb{E}[Q_{pq}] = \frac{\left\lvert\sum_n \xi_n\right\rvert^2-\sum_n\lvert\xi_n\rvert^2}{N(N-1)}, \tag*{(9)}

which reduces to ∣ξ∣2\lvert\xi\rvert^2 when all ξn\xi_n equal a common mean ξ\xi (cf. [21 Eqs. (13) and (19]). We use Eq. (9) for the null hypothesis below.

SPC under correlated speckle

References

  1. [1]Bradley Efron and Charles Stein. The jackknife estimate of variance. The Annals of Statistics, 9(3): 586–596, 1981. doi: 10.1214/aos/1176345462.DOI
  2. [2]Bradley Efron and Robert J. Tibshirani. An Introduction to the Bootstrap, volume 57 of Monographs on Statistics and Applied Probability. Chapman & Hall, New York, 1993. ISBN 0-412-04231-2.
  3. [3]Alessandro Ferretti, Alfio Fumagalli, Fabrizio Novali, Claudio Prati, Fabio Rocca, and Alessio Rucci. A new algorithm for processing interferometric data-stacks: SqueeSAR. IEEE Transactions on Geoscience and Remote Sensing, 49(9):3460–3470, 2011. doi: 10.1109/TGRS.2011.2124465.DOI
  4. [4]Ramon F. Hanssen. Radar Interferometry: Data Interpretation and Error Analysis, volume 2 of Remote Sensing and Digital Image Processing. Kluwer Academic Publishers, Dordrecht, 2001. ISBN 0-7923-6945-9. doi: 10.1007/0-306-47633-9.DOI
  5. [5]Sajid Hussain, Bin Pan, Zeeshan Afzal, Muhammad Ali, Xianlong Zhang, Xianjian Shi, and Muhammad Ali. Landslide detection and inventory updating using the time-series InSAR approach along the Karakoram Highway, northern Pakistan. Scientific Reports, 13:7485, 2023. doi: 10.1038/s41598-023-34030-0.DOI
  6. [6]Mi Jiang, Xiaoli Ding, Ramon F. Hanssen, Rakesh Malhotra, and Ling Chang. Fast statistically homogeneous pixel selection for covariance matrix estimation for multitemporal InSAR. IEEE Transactions on Geoscience and Remote Sensing, 53(3):1213–1224, 2015. doi: 10.1109/TGRS.2014.2336237.DOI
  7. [7]Dieter Just and Richard Bamler. Phase statistics of interferograms with applications to synthetic aperture radar. Applied Optics, 33(20):4361–4368, 1994. doi: 10.1364/AO.33.004361.DOI
  8. [8]Vladimir Katkovnik, Jaakko Astola, and Karen Egiazarian. Phase local approximation (PhaseLa) technique for phase unwrap from noisy data. IEEE Transactions on Image Processing, 17(6):833–846, 2008. doi: 10.1109/TIP.2008.916046.DOI
  9. [9]Nicolaas H. Kuiper. Tests concerning random points on a circle. Proceedings of the Koninklijke Nederlandse Akademie van Wetenschappen, Series A, 63:38–47, 1960.DOI
  10. [10]Jean-Philippe Lachaux, Eugenio Rodriguez, Jacques Martinerie, and Francisco J. Varela. Measuring phase synchrony in brain signals. Human Brain Mapping, 8(4):194–208, 1999. doi: 10.1002/(SICI)1097-0193(1999)8:4<194::AID-HBM4>3.0.CO;2-C.doi.org/10.1002/(sici)1097-0193(1999)8:4<194::aid-hbm4>3.0.co;2-c
  11. [11]Jong-Sen Lee, Karl W. Hoppel, Stephen A. Mango, and Allen R. Miller. Intensity and phase statistics of multilook polarimetric and interferometric SAR imagery. IEEE Transactions on Geoscience and Remote Sensing, 32(5):1017–1028, 1994. doi: 10.1109/36.312890.DOI
  12. [12]Carlos López-Martínez and Xavier Fàbregas. Modeling and reduction of SAR interferometric phase noise in the wavelet domain. IEEE Transactions on Geoscience and Remote Sensing, 40(12):2553–2566, 2002. doi: 10.1109/TGRS.2002.806997.
  13. [13]Avadh Bihari Narayan, Ashutosh Tiwari, Ramji Dwivedi, and Onkar Dikshit. A novel measure for categorization and optimal phase history retrieval of distributed scatterers for InSAR applications. IEEE Transactions on Geoscience and Remote Sensing, 56(10):5843–5849, 2018. doi: 10.1109/TGRS.2018.2826842.DOI
  14. [14]Alessandro Parizzi and Ramon Brçic. Adaptive InSAR stack multilooking exploiting amplitude statistics: A comparison between different techniques and practical results. IEEE Geoscience and Remote Sensing Letters, 8(3):441–445, 2011. doi: 10.1109/LGRS.2010.2083631.DOI
  15. [15]Antonio Pepe, Pietro Mastro, and Cathleen E. Jones. Adaptive multilooking of multitemporal differential SAR interferometric data stack using directional statistics. IEEE Transactions on Geoscience and Remote Sensing, 59(8):6706–6721, 2021. doi: 10.1109/TGRS.2020.3030003.DOI
  16. [16]F. W. Scholz and M. A. Stephens. K-sample Anderson–Darling tests. Journal of the American Statistical Association, 82(399):918–924, 1987. doi: 10.1080/01621459.1987.10478517.DOI
  17. [17]Guoqiang Shi, Peifeng Ma, Hui Lin, Bo Huang, Bowen Zhang, and Yuzhou Liu. Potential of using phase correlation in distributed scatterer InSAR applied to built scenarios. Remote Sensing, 12(4):686, 2020. doi: 10.3390/rs12040686.DOI
  18. [18]Xiaojun Su, Yi Zhang, Xingmin Meng, Dongxia Yue, Jinhui Ma, Fuyun Guo, Ziqiang Zhou, Mohib Ur Rehman, Zainab Khalid, Guan Chen, Runqiang Zeng, and Fumeng Zhao. Landslide mapping and analysis along the China-Pakistan Karakoram Highway based on SBAS-InSAR detection in 2017. Journal of Mountain Science, 18(10):2540–2564, 2021. doi: 10.1007/s11629-021-6686-6.DOI
  19. [19]Emmanuel Trouvé, Jean-Marie Nicolas, and Henri Maître. Improving phase unwrapping techniques by the use of local frequency estimates. IEEE Transactions on Geoscience and Remote Sensing, 36(6):1963–1972, 1998. doi: 10.1109/36.729368.DOI
  20. [20]Mohib Ur Rehman, Yi Zhang, Xingmin Meng, Xiaojun Su, Filippo Catani, Gohar Rehman, Dongxia Yue, Zainab Khalid, Sajjad Ahmad, and Ijaz Ahmad. Analysis of landslide movements using interferometric synthetic aperture radar: a case study in Hunza-Nagar valley, Pakistan. Remote Sensing, 12(12):2054, 2020. doi: 10.3390/rs12122054.DOI
  21. [21]Martin Vinck, Marijn van Wingerden, Thilo Womelsdorf, Pascal Fries, and Cyriel M. A. Pennartz. The pairwise phase consistency: A bias-free measure of rhythmic neuronal synchronization. NeuroImage, 51(1):112–122, 2010. doi: 10.1016/j.neuroimage.2010.01.073.DOI
  22. [22]Urs Wegmüller, Charles Werner, Tazio Strozzi, Andreas Wiesmann, Othmar Frey, and Maurizio Santoro. Sentinel-1 support in the GAMMA software. Procedia Computer Science, 100:1305–1312, 2016. doi: 10.1016/j.procs.2016.09.246.
  23. [23]Howard A. Zebker and John Villasenor. Decorrelation in interferometric radar echoes. IEEE Transactions on Geoscience and Remote Sensing, 30(5):950–959, 1992. doi: 10.1109/36.175330.DOI
  24. [24]Changjun Zhao, Zhen Li, Bangsen Tian, Ping Zhang, Wenhao Wu, Shuo Gao, Yuechi Yu, and Yunyun Dong. A statistically homogeneous pixel selection approach for adaptive estimation of multitemporal InSAR covariance matrix. International Journal of Applied Earth Observation and Geoinformation, 110:102792, 2022. doi: 10.1016/j.jag.2022.102792.DOI

Paper details

vlnc:W_qvcwqunf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Claude (Anthropic)

View statement
Looking for related papers…
Contents