IEEE Geoscience and Remote Sensing Magazine - June 2016 - 5

from the Guest editor

from the Guest editors
By Gustau Camps-Valls, José BiouCas-Dias,
anD melBa CrawforD

A Special Issue on Advances in
Machine Learning for Remote Sensing
and Geosciences
"All models are wrong, but some are useful."
- George E.P. Box (1987)

achine learning has become a standard paradigm
for the analysis of remote sensing and geoscience
data at both local and global scales. In the upcoming
years, with the advent of new satellite constellations,
machine learning will have a fundamental role in processing large and heterogeneous data sources. Machine
learning will move from mere statistical data processing to actual learning, understanding, and knowledge
extraction. The ambitious goal is to provide responses
to the challenging scientific questions about the earth
system. This special issue aims at providing an updated,
refreshing view of current developments in the field.
For this special issue, we have collected five articles that present snapshots of the recent advances in
machine-learning methodologies for remote sensing
and geosciences. The articles were identified through a
two-step submission/review process including a selection within white article submissions, followed by a
traditional blind peer review of full article submissions.
The result is a comprehensive set of contributions covering the main challenges in this arena, i.e, classification,
change detection, feature extraction for domain adaptation, parameter retrieval, and model inversion.
The first article, "Monitoring Land-Cover Changes,"
deals with the highly relevant problem of land-cover
change detection using machine-learning techniques.
The authors present an insightful overview of the posed
challenges from the perspective of machine learning
and discuss some of the recent advances to address
them. Multitask, multi-instance, and multiview learning paradigms are reviewed.
Digital Object Identifier 10.1109/MGRS.2016.2548646
Date of publication: 13 June 2016

june 2016

ieee Geoscience and remote sensing magazine

The second article, "Deep Learning for Remote Sensing Data," reviews the emerging field of deep learning
in remote sensing data classification. This methodology is now considered to be the most powerful in many
related fields dealing with structured data, such as images in computer vision or speech in natural language
processing. The deep-learning models learn hierarchical (bottom-up) spatial-spectral feature representations
with high discrimination power. The authors review the
role and architecture of deep learning in remote sensing data analysis from the traditional topics of image
preprocessing, pixel-based classification, and target
recognition to the recent challenging tasks of high-level
semantic feature extraction and scene understanding.
Some experiments support the excellent performance
of the deep-learning algorithms for remote sensing big
data analysis and open the doors for tackling further
challenging problems.
The third article is titled "Domain Adaptation for
the Classification of Remote Sensing Data." Due to accessibility and cost, the labeled data for the classification of remotely sensed data are typically collected
for a single image acquired by a given sensor, and it is
often within a limited area if a scene covers an extended region. Changing conditions during multiple time
periods and extended areas, as well as intersensor
variability, frequently result in shifts in class-specific
signatures, thereby affecting classification results. Approaches for addressing this issue are often referred to
as transfer-learning strategies. In this article, the authors provide an overview of machine-learning-based
approaches for dealing with this problem via domain
adaptation, which seeks to adapt models trained for
one task to a new, but related, task. Four categories
of approaches that have been shown to be effective
for domain adaptation (i.e., the selection of invariant features, representation matching, classifier adaptation, and selective sampling of labeled data) are
5

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2016