IEEE Geoscience and Remote Sensing Magazine - December 2021 - 156

data. In many studies, the training data set is built by corrupting
optical images through multiplicative noise. This is
far from realistic for despeckling applied to real SAR data.
Therefore, despeckling in an unsupervised manner would
be highly desirable and worth attention.
InSAR
InSAR is one of the most important SAR techniques, and
it is widely used in reconstructing the topography of the
Earth's surface, i.e., DEM generation [65], [154], [155], and
detecting topographical displacements, e.g., monitoring
volcanic eruptions [156]-[158], earthquakes [159], [160],
land subsidence [161], and urban areas by using time-series
methods [162]-[164]. The principle of InSAR is to first measure
the interferometric phase between signals received by
two antennas located at different positions and then extract
topographic information from the obtained interferogram
by unwrapping and converting the absolute phase to height.
However, an actual interferogram often suffers from a large
number of singular points, which originate from the interference
distortion and noise in radar measurements. These
points result in unwrapping errors and, consequently, lowquality
DEMs.
To tackle this problem, Ichikawa and Hirose [165] applied
a complex-valued neural network (CV-NN) in the
spectral domain to restore singular points. With the help of
the complex Markov random field filter [166], they aimed
at learning ideal relationships between the spectrum of
neighboring pixels and that of the center pixels via a onehidden-layer
CV-NN. Notably, the center pixels of each
training sample are supposed to be ideal points, which indicates
that singular points are not fed to the network during
the training procedure. Similarly, Oyama and Hirose [167]
restored singular points with a CV-NN in the spectrum
domain. Related to topography extraction, Costante et al.
[169] proposed a full CNN encoder-decoder architecture
for estimating DEMs from single-pass image acquisitions.
They demonstrated that this model was capable of extracting
high-level features from input radar images using an
encoder section and then reconstructing full-resolution
DEMs via a decoder section. Moreover, the network can potentially
solve the layover phenomenon in one single-look
SAR image that has contextual features.
In addition to reconstructing DEMs, Schwegmann et al.
[170] presented a CNN-based technique to detect subsidence
deformations from interferograms. They employed
a nine-layer network to extract salient information from
interferograms and displacement maps for discriminating
deformation targets from deformation-like targets.
Furthermore, Anantrasirichai et al. [10], [171], [172] used
a pretrained CNN to automatically detect volcanic ground
deformation through InSAR images. They divided each
image into patches and relabeled it with binary labels, i.e.,
" background " and " volcano, " and finally fed it to the network
to predict volcano deformation. In [173], they further
improved their method to be able to detect slow-moving
156
volcanoes using a time series of interferograms. In another
study related to automatic volcanic deformation detection, Valade
et al. [168] designed and trained a CNN from scratch to
learn a decorrelation mask from input wrapped interferograms;
the CNN then was used to detect volcanic ground
deformation. A flowchart of this approach appears in Figure
8. The training in both [168] and [173] was based on
simulated data.
Another geophysics-motivated example of using deep
learning on InSAR data, which was actually proposed earlier
than the previously mentioned CNN-based studies,
can be found in [174]-[176], where the authors used simple
feed-forward shallow neural networks for seismic event
characterization and automatic seismic source parameter
inversion by exploiting the power of neural networks in
solving nonlinear problems. Recently, deep learning has
been utilized for tomographic processing, as well. An unfolded
deep network that involves vector-approximate
message-passing algorithms was proposed in [177]. Experiments
with simulated and real data were performed, showing
the spectral estimation gains and achieving competitive
performance. In [178], a real-valued deep neural network
was applied for multiple-input, multiple-output SAR 3D
imaging. It displayed a better superresolution power compared
with other compressive sensing-based methods.
In summary, it can be concluded that the use of deep
learning methods in InSAR is still at a very early stage. Although
deep learning has been incorporated in different
applications combined with InSAR, the full potential of
interferograms has not been fully exploited, except in the
pioneering work of Hirose [179]. Many applications treat
interferograms and deformation maps obtained from interferograms
as images similar to RGB and gray-scale ones,
and therefore the complex nature of interferograms has remained
unnoticed. Apart from this issue, such as the SAR
despeckling problem related to deep learning, the lack of
ground truth data for detection and image restoration problems
provides motivation to focus on developing semisupervised
and unsupervised algorithms that combine deep
learning and InSAR. Otherwise, a training database consisting
of interferograms for different scenarios and different
phase contributions could be beneficial for supervised
learning applications. Simulation-based interferogram generation
for the latter was recently proposed in [180].
SAR-OPTICAL DATA FUSION
The fusion of SAR and optical images can provide complementary
information about targets. However, considering
the two different sensing modalities, the prior identification
and the coregistration of corresponding images are
challenging [181] but compulsory. For the purpose of identifying
and matching SAR and optical images, many current
methods resort to deep learning, given its powerful
capabilities of extracting effective features from complex
images. In [58], the authors proposed a CNN for identifying
corresponding image patches of VHR optical and SAR
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE DECEMBER 2021
IEEE Geoscience and Remote Sensing Magazine - December 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2021
