IEEE Geoscience and Remote Sensing Magazine - March 2020 - 51

PIXEL SELECTION
The purpose of pixel selection is to select pixels with stable phase information and good coherence. Deformation
information of the observed scene is obtained based on
the selected pixel-point phase. The quality of the differential interferometry phase directly affects the accuracy
of the deformation measurement. When deformation
analysis is performed on the point where phase stability
is poor and coherence low, large measurement errors exist. Therefore, it is usually necessary to select some highquality pixels derived from the permanent scatterer (PS).
The most commonly used selection algorithms include
those based on both coherence information [59] and
amplitude information [60]. A reasonable PS point is selected by one or more algorithms to set an appropriate
threshold, where the threshold is typically adjusted based
on the phase quality and density between the selected
pixel points [61]. The following section introduces classical point-selection algorithms and the three-threshold
selection method in detail.
AMPLITUDE-DISPERSION INDEX
Based on the statistical characteristics of the amplitude
dispersion and phase-standard deviation, in the case of
high signal-to-noise ratio, the stability of the target point is
directly represented by the amplitude dispersion. The formula for calculating the amplitude-dispersion index [60],
[62], [63] is given as
d
D A = mAA , (21)

where D A is the amplitude-dispersion index, d A is the
standard deviation, and m A is the point-amplitude mean.
The smaller the amplitude-dispersion index, the more
stable the amplitude information. A reasonable amplitude-dispersion index threshold TD (usually 0.25) is selected, and the pixel that satisfies the condition D A 1 TD
is selected as the PS point. The amplitude-dispersion index method is more suitable for selecting point scatterers
and uses only the amplitude information of the time-series images. It has high efficiency and a small calculation
amount and is suitable for application in the real-time
monitoring system.
INTERFEROMETRIC COHERENCE
The interferometric phase quality of each pixel is estimated by
calculating its coherence. The coherence formula [64], [65] is

c = ; c ; $ exp" jz , =

E " S 1 $ S 2) ,
, (22)
E " ; S 1 ;2 , $ ; " ; S 2 ;2 ,

where S 1 and S 2 are the complex values corresponding to
the same pixel of the acquisitions that form an interferogram, ; $ ; indicates the absolute value operator, and E " $ ,
refers to the expectation value.
The points satisfying c 2 Tc are considered the PS
points. Tc should be a reasonable coherence threshold. The
MARCH 2020

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

coherence value is estimated by taking into consideration
the moving window, as its size directly affects the result of
the coherence value. The larger the selected window, the
more reliable the coherence value; however, because the
coherence calculation reduces the resolution, some isolated points will be missed. The coherence coefficient value
ranges between zero and one. The higher the threshold, the
higher the reliability of the PS points; however, the high coherence is likely to cause a missed selection (usually the coherence value threshold is set between 0.6 and 0.8) [66]. In
addition, the coherence method can effectively remove pixels with serious coherence, i.e., vegetation and water [67].
THREE-THRESHOLD METHOD
A single threshold for PS point selection emphasizes only
one feature of the PS points. There may be erroneous or
missing selections. We have designed a three-threshold
method combining the amplitude-dispersion index, interferometric coherence, and phase-error combination for PS
point selection [10]. The phase-error term is referred to as
original interferometry phase subtraction, with the filtered interferometry phase represented by

z noise = z - Filter (Tz). (23)

A reasonable phase-error threshold Tz is set. The pixel
with z noise 1 Tz is used as the PS point. The three-threshold
combination method comprehensively considers amplitude, coherence, and phase information, respectively. This
method has been successfully applied to landslide data
analysis in Mao County, Sichuan Province, China [10].
PHASE UNWRAPPING
According to its working mode, GB-DInSAR is divided into
continuous operation and discontinuous operation modes.
In the case of fast displacement, the continuous operation
mode is normally used to continuously capture the images,
thereby avoiding the DIn phase exceeding one phase period
(2 2r).
For the discontinuous mode of operation, the image pair
has a long time interval. When the observed area deforms
significantly, the DIn phase may exceed one phase period
(2 2r). Consequently, the DIn phase z will be entangled.
At the edge of the fringe, the interferometric phase value
has a step change, which induces one or n integer periods.
Phase unwrapping uses a mathematical method to obtain
the number of integer cycles in the interferometric phase
to recover the continuously varying phase. There are already several 2D spatial phase unwrapping methods used
in spaceborne InSAR, among which the branch cutting [68]
and network cost-flow methods [69] are the common.
ATMOSPHERIC PHASE COMPENSATION
The atmosphere can cause electromagnetic wave refraction, which induces a change of the propagation path and
direction. Changes in atmospheric conditions (such as
51

IEEE Geoscience and Remote Sensing Magazine - March 2020

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2020