IEEE Geoscience and Remote Sensing Magazine - December 2022 - 5

" Data " are, of course, the essential product generated
from remote sensing. Before moving on to our articles
on various techniques for working with data, it is good
to pause and reflect upon some of the ethical issues associated
with the use of Earth observation (EO) data,
particularly when emerging methods of artificial intelligence
(AI) are involved. With this in mind, I call your
attention to the review by Kochupillai et al. [A5], resulting
from 18 months of discussions with researchers working
in these fields. In this review, the major ethical issues
related to AI for EO (AI4EO) are studied under six broad
principles: privacy, honesty, integrity, fairness, responsibility,
and sustainability.
From this theoretical background, the authors provide
a practical road map for use in identifying ethical issues
in research. They also provide some specific examples
from different research fields to illustrate how science
questions can be redesigned to capture the ethical dimension.
In particular, they show how the restructuring
of research questions can better support the United
Nations Sustainable Development Goals with example
applications in the classification of slum areas and monitoring
illegal small-scale mining. They also provide some
considerations for future work. These include looking beyond
dominant Western concepts of ethics in terms of
consequences and the duty to incorporate multicultural
approaches based on, for example, intention and action.
Ethical considerations should be incorporated into all
stages of a project ( " ethics by design " ), from proposal
development through to the integration of EO data into
policy-making, to ensure that research and applications
are better aligned with " on-the-ground " realities of human
values and human rights.
Obtaining data is only one part of a typical remote
sensing application; doing anything useful with the data
will most often require some amount of computation.
With the enormous growth in the volume of Earth observation
data being produced, remote sensing researchers
and practitioners are increasingly being required to
employ various high-performance computing (HPC)
tools. The so-called " big data " problem arises in many
fields, but the volume of data from remote sensing, combined
with the complexity of realistic geophysical models,
makes this a significant challenge. Zhang et al. [A6]
provide a review of the state of the art in various highperformance
and high-throughput computing methods
that are applicable to remote sensing data processing.
Specific approaches covered in this review include supercomputers,
parallel programming, cloud computing,
grid computing, graphics processing units (GPUs), and
field-programmable gate arrays. An extensive bibliography
of 248 references is provided, and a large sampling
of different applications is categorized. The authors enumerate
several types of challenges in applying HPC: versatility,
sharing, security, heterogeneity, and timeliness.
They conclude by presenting their viewpoints on the fuDECEMBER
2022 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
ture development of the HPC field as applied to remote
sensing data.
On the subject of data, remote sensing is but one source
of data that can be used for observing, understanding, and
making decisions about our environment. The interpretation
of remote sensing data often requires some
context
that
can be obtained
only through reference
to ground-level observations.
Social media, in all
forms, is another enormous
source of data, which could
be used to provide this context
in some applications.
This massive trove of global
multiperspective data is
constantly being generated
by almost 4 billion users
but has variable and often
unknown quality. Although promising, integrating social
media data with remote sensing is not trivial due to
the significant differences in the characteristics of both
data types.
Zhu et al. [A7] review some of the key aspects of social
OUR THIRD AND FINAL
FEATURE IN THE " RADAR "
THEME IS A REVIEW OF
METHODS FOR ESTIMATING
SUBSURFACE PROPAGATION
VELOCITY IN GROUNDPENETRATING
RADAR.
media data relevant to its use for this purpose, such as data
availability; preparation; management of " analysis-ready "
datasets; geoinformation extraction (from text or images);
and methods of the fusion of social media and remote sensing
data. Example applications include detailed information
about buildings that can be used for the supervised
classification of satellite images.
Referring to the earlier article by Kochupillai et al. [A5]
in this issue, the use of social media data does raise ethical
issues in terms of privacy, bias, and national security. While
mindful of these concerns, the authors encourage the remote
sensing community to look at using social media in
the spirit of " open science, open data. "
Li et al. [A8] review the SV248S dataset collected from
the Jilin-1 commercial remote sensing satellite and the use
of video images for tracking objects. They present 10 challenges
and compare a total of 28 deep learning-based algorithms.
Two evaluation methods, enhanced intersection
over union and enhanced union score, are applied to evaluate
the accuracy of the different algorithms for detecting
and tracking different objects.
Our last feature is a review of self-supervised learning
(SSL), a technique that has shown big success in the computer
vision community. SSL has great, but mostly unrealized,
potential for remote sensing. Wang et al. [A9] provide an
introduction to SSL and a literature review of the applications
to remote sensing. They summarize the three main
categories of SSL techniques and important differences between
their application to natural images (common in
computer vision) to the broader types of remote sensing
data. Results are compared to various remote sensing modalities,
multispectral, hyperspectral, and SAR. Performance
5
IEEE Geoscience and Remote Sensing Magazine - December 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2022
