inbul 2004
d LIDAR,
M.
Support of
blications,
EMs from
g Laser
e Sensing,
)
(mitto, K.,
‘ross, P.A.
EM of the
rchives of
693.
rces — the
candLaser
of Forest,
r 112 2003
d.
| and and
/2, Three
Mf Canada
ruction by
criterion.
te Sensing
f Filtering
rammetry,
Sciences,
G, (2004).
D Features
1 Archives
)
anning of
candLaser
of Forest,
r 112 2003
2
| automatic
\R DSMs.
70(4):427-
THE FILTERING AND PHASE UNWRAPPING OF INTERFEROGRAM
Xu Qing, Jin Guowang, Zhu Caiying, Wang Zhengde, He Yu , Yang Peizhang
Institute of Surveying and Mapping, 450052, Zhengzhou, China
Xg64(037]1.net , jgw77(tsohu.com, caivine.zhu(@inta.com.en
Commission VI, WG VI/4
KEY WORDS: Remote Sensing, SAR, Image, DEM/DTM, Data mining,
ABSTRACT:
INSAR(Interferometric Synthetic Aperture Radar) has been studied and used widely in the field of generating high accuracy
DEM(Digital Elevation Model). The processing of INSAR includes many key technologies, such as image matching, phase
unwrapping, baseline estimation, etc. The phase unwrapping is difficult, but it can be made easy and simple by an excellent
interferogram filtering, as is often ignored by many people. So, in this paper, the interferogram filtering and phase unwrapping are both
studied. In the aspect of filtering, some filtering methods such as mean filtering, median filtering, adaptive filtering, multi-look
filtering and the vector filtering are analyzed and compared. The filtering results show that the vector filtering is more excellent than
others. Then, it is presented that the interferogram filtering is a critical technology for the phase unwrapping, we should filter the
interferogram noises before phase unwrapping, and a new scheme for phase unwrapping is put into practice in the paper. Firstly, the
interferogram is filtered by vector filtering method. Secondly, The filtered images generated from different filtering methods are
unwrapped successfully by selecting a compatible gate value of coherence or of fake coherence with region grow method. Both the
vector filtering for interferogram and the region grow phase unwrapping method can be applied in the successful phase unwrapping.
They will promote a wider use of INSAR in generating DEMs.
1. INTRODUCTION
INSAR is one of the best techniques to derive DEM(Digital
Elevation Model). It has some prominent advantages, such as
fast, high precision, and working without the limitation of time
and climates. But the phase unwrapping, which is one of the key
processing steps of INSAR, cannot be practically used. So, this
paper plays emphases on the studying of phase unwrapping. We
all know that there are a lot of methods for phase unwrapping,
but none of them can be used for all practical data. Most of them
can get successful unwrapping results for simulated data, but
unsuccessful results for practical data. They can't be practically
used in the processing of INSAR due to the neglect of
interferogram's filtering in former processing. The great
influence of noises on phase unwrapping being considered, this
paper presents that the filtering of interferogram is a key
technology of phase unwrapping. Several common filtering
methods, such as non Adaptive filtering, Adaptive filtering and
Multilooking filtering, are analyzed and compared, and a new
vector filtering method is presented, whose filtering results are
better than others. Then the region grow phase unwrapping
method based on the coherence image or on the fake coherence
image is used for unwrapping the interferograms filtered by
vector filtering.
The results show that the vector filtering and the region
grow phase unwrapping based on the coherence image or on the
fake one by selecting a compatible gate value are successful for
phase unwrapping and can make INSAR be widely used.
2. SOME COMMON PHASE UNWRAPPING METHODS
The phases of interferogram must be unwrapped for
deriving DEMs, but phase unwrapping is known as the most
difficult technique of INSAR. There are three main classes of
phase unwrapping approaches: Branch-cut methods, LS (Least
Square) methods and other methods.
a) Branch-cut methods: These algorithms attempt to isolate
phases of error prior to integration. The basic idea is to unwrap
the phase by choosing only paths of integration that lead to
self-consistent solutions (R. M. Goldstein, 1988).
b) LS methods: These algorithms of phase unwrapping
minimize the difference between the gradients of the solution
and unwrapped phase in an LS sense. They include FFT/DCT
Algorithm, PCG (Preconditioned conjugate gradient) Algorithm,
etc;
c) Other methods: There are a lot of other methods for phase
unwrapping, such as line detection Algorithm, knowledge
intervention algorithm, etc.
Al of the above phase unwrapping methods can resolve
some problems caused by noises and sub-sample in some extent.
but each of them has its flaw. Usually, they can unwrap the
simulated phases successfully while cannot unwrap the practical
phases, or they are available for some specific interferogram
while not for others.
The phase unwrapping methods are always based on the
hypothesis as follows: The signal of SAR can make full samples
to the ground terrain, and so the difference of phases between a
pixel and the adjacent one is in the range of half cycle, that is the
absolute value of phase difference is less then 7 . If there are
not noises, phase unwrapping is very easy. In Figure 1(a), the
ideal interferometric phases is shown. The phases change from
0 to 27 , then rapidly to 0, and then 0 to 2 77 again. The
periodicity of the phases is clear. In this condition, we can
calculate the difference in vertical and horizontal directions,
then we can unwrap the phase through integral along the two
directions. However, there are a lot of noises in the
inteferograms of INSAR, which can be caused by the system
temperature, overlays and inaccurate matching, etc. The phases
with the effect of noises are shown as Figurel(b), where, we can
see that the periodicity is not so clear that the phase unwrapping
is getting much difficult.
