IEEE - Aerospace and Electronic Systems - May 2022 - Tutorial XV - 18

Tutorial:
DOI. No. 10.1109/MAES.2021.3117369
A Survey on the Applications of Convolutional Neural
Networks for Synthetic Aperture Radar: Recent
Advances
Amir Hosein Oveis , Elisa Giusti , and Selenia Ghio , National
Inter-University Consortium for Telecommunications, 56124 Pisa, Italy
Marco Martorella
, National Inter-University Consortium for
Telecommunications, 56124 Pisa, Italy, and also University of Pisa, 56124
Pisa, Italy
In recent years, convolutional neural networks (CNNs)
have drawn considerable attention for the analysis of
synthetic aperture radar (SAR) data. In this study, major
subareas of SAR data analysis that have been tackled by
CNNs are systematically reviewed, such as automatic
target recognition, land use and land cover classification,
segmentation, change detection, object detection, and
image denoising. Special emphasis has been given to
practical techniques such as data augmentation and
transfer learning. Complex-valued CNNs, which have
been introduced to exploit phase information embedded
in SAR complex images, have also been extensively
reviewed. To conclude this review paper, open challenges
and future research directions are highlighted.
INTRODUCTION
In our data-intensive world, machine learning (ML) techniques
are becoming increasingly important. Features are
extracted to reduce the dimension of input data to a more
manageable scale by holding only informative and nonredundant
parts. Feature extraction and feature classification are
Authors' current address: Amir Hosein Oveis, Elisa
Giusti, Selenia Ghio, and Marco Martorella, Radar
and Surveillance Systems, National Inter-University
Consortium for Telecommunications, 56127, Pisa, Italy
(e-mail: aoveis@cnit.it, elisa.giusti@cnit.it, selenia.
ghio@cnit.it); Marco Martorella, Department of Information
Engineering, University of Pisa, 56127, Pisa,
Italy (e-mail: marco.martorella@iet.unipi.it).
Manuscript received September 30, 2020, revised
September 15, 2021; accepted September 15, 2021, and
ready for publication September 30, 2021.
Review handled by Dale Blair.
0885-8985/21/$26.00 ß 2021 IEEE
18
two steps of most ML problems. In the conventional
ML algorithms, handcrafted features are extracted.
However, the emphasis is currently on the automatic
learning of features. Deep learning (DL) is undoubtedly
the most popular trend in the ML domain. During
the past decade, it has achieved impressive performance
in the field of computer vision and image processing,
with applications such as image classification,
object detection, super-resolution restoration, etc. [1].
Based on the availability of ground truth or labels for
the training set, learning procedures can be accomplished
in supervised, unsupervised, or semi-supervised
manners. In addition, reinforcement learning is another
type of learning that is often discussed under the scope
of semi-supervised and sometimes under unsupervised
learning approaches [2]. Although supervised DL
brings outstanding results, in case of insufficient
labeled training samples [for example, in the remote
sensing (RS) domain], other DL approaches can be
employed to extract features from unlabeled images.
Convolutional neural network (CNN) [3], known also
as ConvNet, is one of the most popular DL approaches
in computer vision and RS fields. A CNN, inspired by
the human's visual perception mechanism, consists of
different convolutional and pooling layers. Typically,
features are extracted at first, and then they are fed to
a multilayer perceptron (MLP) for classification or
regression purposes. Other types of deep neural networks
such as autoencoders (AEs) [4], deep belief networks
(DBNs) [5], [6], and recurrent neural networks
(RNN) [7] will be briefly introduced later in the section
" Deep Neural Networks in Computer Vision and
Remote Sensing. "
In this article, applications ofCNNs to SARwill be thoroughly
analyzed. Moreover, some other DL approaches that
incorporate convolution operation will be mentioned. We
IEEE A&E SYSTEMS MAGAZINE
MAY 2022
https://orcid.org/0000-0001-5278-6509 https://orcid.org/0000-0002-9894-0839 https://orcid.org/0000-0002-6057-8459 https://orcid.org/0000-0002-8985-5069
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
