IEEE Geoscience and Remote Sensing Magazine - June 2019 - 143

Thus, ad hoc techniques are required to identify these minor spectral variations in the spectral-temporal domain.
In this article, we analyze the CD problem in HS images
and present a comprehensive overview of the existing
methods, including the most widely used and most recently published ones.
CATEGORIES OF CHANGE-DETECTION
APPROACHES FOR HYPERSPECTRAL IMAGES
CD approaches can be grouped in several ways by considering different perspectives. In the literature, there are some
excellent review articles that provide categorization of CD
techniques [13], [19]-[21]. However, they focus mainly on
MS images, and very few works (e.g., [22], [23]) provide a
review of CD techniques in HS images.
In general, we can divide CD approaches in HS images
by considering different possible variables (see Figure 2).
1) Application purpose: Four classes can be defined:
anomalous, binar y, multiclass, and time-series CD
methods.
2) Availability of ground reference data: Three classes can be
defined: supervised, semisupervised, and unsupervised
CD methods.
3) Automation degree: Three classes can be defined: fully automatic, semiautomatic, and manual CD methods.
In this article, we use the first categorization because it
is the most interesting from methodological and application perspectives. We performed an analysis on CD in HS
images articles published in IEEE journals and conference

Anomalous CD
Binary CD
Application
Purpose

Multiclass CD
Time-Series CD

CD
Techniques

Availability
of Ground
Reference Data

Supervised CD

Fully Automatic CD
Semiautomatic CD
Manual CD

FIGURE 2. The categories of techniques for CD in HS images.
JUNE 2019

CHANGE-DETECTION PROBLEM
IN HYPERSPECTRAL IMAGES
CD is a systematic process that requires a series of comprehensive processing steps, which include [20], [25] 1) understanding the CD problem, 2) selecting suitable remote
sensing data, 3) accurate image preprocessing, 4) selecting
suitable features, 5) designing the CD algorithm, and 6) evaluating the CD performance. In each step, significant effort
should be aimed at defining successful procedures to obtain
high-quality data input through the CD technique and then
a high-accuracy CD output. If we consider the CD problem
in a pair of bitemporal HS images, we can define the block

Time-Series CD
13 (11.93%)

Semisupervised CD
Unsupervised CD

Automation
Degree

papers during the previous five years of our study (i.e.,
2012-2017), as shown in Figure 3. This was accomplished
by searching articles according to the keywords in the first
category (i.e., anomalous, binary, multiclass, and time-series
CD) jointly with the keyword "hyperspectral." Then, a
manual filtering was carried
out by analyzing the title
and context of each article.
Relatively few articles on CD
RELATIVELY FEW ARTICLES
in HS images have been pubON CD IN HS IMAGES HAVE
lished compared with other
BEEN PUBLISHED
hot topics in the HS field, such
COMPARED WITH OTHER
as classification and unmixHOT TOPICS IN THE HS
ing. Among them, in the past
FIELD, SUCH AS
five years, the topic of anomaCLASSIFICATION
lous CD (ACD) had the most
AND UNMIXING.
published articles (25 out of
109). Multiclass CD and timeseries CD were the topics of
the second and third groups,
with 16 and 13 articles, respectively. Binary CD was addressed in only seven articles. In addition, 48 articles focused on specific CD applications (e.g., cryosphere CD
[24]) or could not be assigned to any of the categories. Figure 3 shows a pie chart representing the distribution of the
articles according to our categorization.

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

Multiclass CD
16 (14.68%)

109
Works

Others
48 (44.04%)

Binary CD
7 (6.42%)
Anomalous CD
25 (22.93%)
FIGURE 3. The statistics on CD as the subject of HS image articles

published in IEEE journal and conference papers during the years
2012-2017, according to the category of CD application purpose.
143

IEEE Geoscience and Remote Sensing Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2019