IEEE Geoscience and Remote Sensing Magazine - June 2019 - 145

of the number of change types K, and 3) separation of the
samples in X C into the classes ~ k (k =1, 2, f, K ) representing different kinds of change [11], [22].
In [35], the authors analyzed the concept of change in
bitemporal HS images. A qualitative illustration example
is shown in Figure 6. In particular, conceptually, SCVs can
be separated into major changes and subtle changes. Major
changes are related to land-cover transitions having spectral signatures that significantly differ from each other [Figure 6(a)]. Subtle changes show similar behavior to a given
major change for most of the analyzed wavelengths but differ significantly in only some components of the spectrum
[Figure 6(b)]. Subtle changes can be seen as subclasses of
the corresponding major change. As an example, a major
change representing a major land-cover transition (e.g., from
noncultivated to cultivated) could be divided into subtle
changes corresponding to different variations, such as water
content or different growth rates of the crops. Although major changes may be discriminated by using MS images, this
is not the case for subtle changes because they would be lost
(i.e., grouped with the most similar major change) due to
the coarse spectral resolution of such images.

I1

CVA

I2

ρ

Ωc

B

SCV

CHALLENGES OF CHANGE DETECTION
IN HYPERSPECTRAL IMAGES
Although the use of HS images has key advantages in CD
applications versus MS images, it also results in significant
challenges that must be properly addressed to exploit the

ID
B

Accordingly, HS images can be used to detect a wide variety of changes, ranging from major to subtle. A qualitative illustration of the major- and subtle-change concepts
is shown in Figure 6. X C ={C 1, C 2, f} is the set of major
changes. Each major change may include subtle changes (i.e., C 1 ={C 1, 1, C 1, 2, f} a nd C 2 ={C 2, 1, C 2, 2, f}). By
iterating the process, it is possible to state that each subtle
change can be further split
DEPENDING ON THE
until it is not possible to deAPPLICATION, ONE MAY
tect spectral inhomogeneity.
BE MORE INTERESTED IN
Different changes can have
DETECTING THE RELEVANT
significantly different relevance
BUT SUBTLE LAND-COVER
from a semantic point of
TRANSITIONS.
view; depending on the application, one may be more
interested in detecting the relevant but subtle land-cover transitions. The definition of
what it is considered "change" is not straightforward and
depends significantly on the application. Therefore, this
aspect should be taken into account when developing CD
methods for HS images-for example, by organizing the
detected changes in such a way that the relevance of the
different changes is highlighted [35].

0

C1
C2
ωn

h (ρ)

(a)
I1
I2

Is

(a)
SCV

I1
B

Wavelength

ωn

B

Spectral
Vector
Change Vector
Analysis
Stacking

I2

2B

0

Wavelength

C1
C1,1
C1,2
C2
C2,1
C2,2
ωn

B
(b)
FIGURE 5. An illustration of the CD problem in HS images with

(a) a comparison operator (i.e., CVA) and (b) an image-stacking
operation.
JUNE 2019

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

(b)
FIGURE 6. A qualitative illustration showing examples of (a) major
changes and (b) subtle changes (solid line) within the given major
changes (dotted line) defined in the SCV domain.

145
IEEE Geoscience and Remote Sensing Magazine - June 2019

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