IEEE - Aerospace and Electronic Systems - May 2022 - Tutorial XV - 33

Oveis et al.
Figure 16.
Proposed GAN architecture of [206]. (a) Generator. (b) Discriminator. (c) Residual block.
small amount of labeled samples and then extending
the labeled set iteratively. They used a cotraining
[209] method to perform this task by utilizing a few
labeled samples to predict the labels of the unlabeled
samples at first. Afterward, samples with high confidence
were chosen and added to the previous labeled
set for the next training iteration. Shi et al. [210]
employed DCGAN for SAR image enhancement and
used the result for the SAR-ATR task. Their proposed
architecture is depicted in Figure 16 as an example of
DCGAN. Zhang et al. [211] used DCGAN for transferring
knowledge from unlabeled SAR images. They
trained a DCGAN with unlabeled samples to learn
generic features of SAR images and reused the learned
parameters for the SAR-ATR task.
CONCLUSION
CNN is undoubtedly the most successful DL technique in
computer vision and it has attracted much attention in both
civilian and military applications. In the same way, SAR
plays a significant role in RS by offering high-resolution
images with all-day and all-weather capabilities. In this survey,
we have systematically reviewed the publications that
applied CNN to different subareas ofSARdata analysis such
as ATR, scene classification with an emphasis on LULC,
object detection with an emphasis on ships, segmentation,
change detection, denoising, and regression for parameters
MAY 2022
estimation. Moreover, we discussed different topics related
to CNN such complex architectures, transfer learning, and
data augmentation. Despite the fact that CNN showed
its promising role in tackling different SAR problems,
we believe that this field is young and in the following
years, more attention should be paid to address the
remaining challenges. Some future research directions
are listed below.
1) Taking the discrepancy between natural and SAR
images into account, there is still potential to study
the feasibility of transfer learning from well-known
CNN architectures in the computer vision domain
to SAR domain.
2) CV-CNNs have been developed to exploit the phase
information embedded in SAR images. There are
only a few studies that applied CV-CNNs to SAR
images and it is expected that in the upcoming years
more scholars will explore the capabilities of CVCNNs.
3)
The deeper the CNN is, the more complicated features
can be extracted. However, more parameters
have to be trained and more training samples are
required to prevent overfitting. Due to the lack
of large-scale annotated SAR images, effective
data augmentation techniques, exploitation of
unlabeled SAR images, and other possible solutions
should be studied in more detail. Moreover,
IEEE A&E SYSTEMS MAGAZINE
33
IEEE - Aerospace and Electronic Systems - May 2022 - Tutorial XV

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - May 2022 - Tutorial XV
