IEEE Geoscience and Remote Sensing Magazine - June 2019 - 154

pixel (which represents 30 m), it results in a spectral change
at the subpixel level.
CHANGE DETECTION IN IMAGE TIME SERIES
CD in the time series of HS images requires the availability of a long time sequence of HS images. The user-interest
changes may be either abrupt
changes that occurred on the
land surface or subtle changWITH THE NUMEROUS HS
es in a given time period. The
MISSIONS THAT WILL BE
latter are valuable to reveal
LAUNCHED IN THE NEAR
land-cover evolution trends
by investigating change patFUTURE, THE NUMBER OF
terns from the time perspecMULTITEMPORAL HS DATA
tive. From current HS satelSETS WILL SIGNIFICANTLY
lites, it is still difficult to
INCREASE.
have a high temporal resolution (e.g., EO-1 Hyperion
had a revisit time between 16
and 30 days). Thus, it is difficult to construct long time
series for change analysis. For future HS satellite missions
(see Table 1), which will have a decreasing revisit-time
resolution, the use of a denser time series of HS images
will provide opportunities to detect subtle changes in the
land surface. Until now, in the context of HS image time
series CD, there are few works in the literature due to the
limitations of not having a long time series of HS images
in real applications.
In [73], the authors propose a method for modeling the
temporal variation of the reflectance response as a function
of time period and wavelength. It is designed based on a

new 3D model of the spectral signature that concurrently
considers the spectral, spatial, and temporal information. A
library of known endmembers was used to perform a classification, which depends mainly on 3D surface reconstruction quality and similarity measures. Experiments prove
that this method can better model the change-distinctive
patterns associated with seasonal variations.
In [72], an unsupervised CD method is proposed for detecting small changes in a multitemporal HS image sequence.
The feature space was built using block processing and locally linear embedding, whereas the CD map was generated
by clustering the change and no-change binary classes.
A framework for dynamically modeling and efficiently
unmixing a time series of HS images is proposed in [74].
Based on the linear mixing process at each time, the spectral signatures and fractional abundances of the pure materials in the scene are viewed as latent variables that follow a dynamic structure. An efficient spectral unmixing
algorithm is developed to estimate the latent variables by
performing alternating minimizations.
In [75], the authors apply canonical polyadic tensor
decomposition techniques to the blind analysis on HS
big data that can be either time series or multiangular
acquisitions. The method can be interpreted as multilinear blind spectral unmixing, where the big HS tensor
is decomposed into three factors, such as spectral signatures, fractional spatial abundances, and temporal/
angular changes. Both quantitative and qualitative evidence of the validity of this methodology were obtained
using the MODIS HS data set during the 2012 snow season in the French Alps.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)
1

FIGURE 18. The subpixel-level multiclass CD results, i.e., class abundances, obtained using
the MSU approach on the Benton County Hyperion data set [38]. (a)-(g) The seven change
classes. (h) The no-change class.

154

CONCLUSIONS
The role of multitemporal HS images
is becoming increasingly important
in remote sensing applications. They
provide finer and deeper insights to
discover, describe, and discriminate
the land-cover changes on Earth's
surface. Multitemporal HS images
are an important data source for CD
applications to detect both abrupt
and subtle changes using bitemporal
images and long time series. With
the numerous HS missions that will
be launched in the near future, the
number of multitemporal HS data
sets will significantly increase. This
will require the design of advanced
techniques to take full advantage
of HS images in the multitemporal framework.
In this article, the problem of
CD in HS images was analyzed in
terms of basic concept definition,
category description, and discussion

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

JUNE 2019

IEEE Geoscience and Remote Sensing Magazine - June 2019

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