IEEE Geoscience and Remote Sensing Magazine - June 2021 - 71

◗ SVs: Spectral signatures inevitably suffer from the various
SVs caused by illumination and topography changes,
the noise effects from external conditions and internal
equipment, atmospheric interference, and the
complex mixing of materials during the imaging process.
Figure 10 depicts a visual example that specifies
the SVs (e.g., trees) in a real HS scene. The considerable
uncertainties brought about by these factors have a significant
negative impact on the accurate estimation of
abundances and endmembers in SU.
◗ Nonlinearity: The complex interactions (e.g., the intimate
and multilayered mixing [94]) among multiple materials,
also known as nonlinearity, inevitably occur during
the process of HS imaging. Nonlinearity in SU is a longstanding,
pending challenge. Most of the existing nonlinear
unmixing models attempt to consider only certain
special cases [125], e.g., bilinear mixing, intimate
mixtures, and so on. Consequently, there is still the lack
of a general and powerful model that can robustly address
the various nonlinearities in SU.
◗ Model explainability: The nonnegativity and sum-to-one
constraint considered in LMM are the basic priors for
spectral signals in HS images. However, only the two constraints
fail to model the complex unmixing process in an
explainable fashion. To further enhance the explainability,
new spectral mixture models should be developed beyond
the classic LMM by fully excavating the intrinsic attribute
knowledge that lies in the
HS image.
DATA FUSION AND
ENHANCEMENT
The high-spectral resolution
of HS images enables the identification
and discrimination
of materials, while high-spatial
resolution can provide a
possibility of the derivation
of surface parameters [126].
However, due to equipment
limitation, there is usually
a tradeoff between the spatial
and spectral resolutions,
and the HS images obtained by the spaceborne imaging
spectrometers are usually done so with a moderate GSD
[126]. To enhance the spatial resolution, one popular way
is to fuse the HS images with high-spatial MS images to
generate new high-spatial-spectral HS (HrHS) images. In
THE COMPLEX
INTERACTIONS (E.G., THE
INTIMATE AND
MULTILAYERED MIXING)
AMONG MULTIPLE
MATERIALS, ALSO KNOWN
AS NONLINEARITY,
INEVITABLY OCCUR DURING
THE PROCESS OF HS
IMAGING.
Trees
0.7
0.6
0.5
0.4
0.3
0.2
0.1
020406080 100 120
Spectral Bands
(a)
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
020406080 100 120
Spectral Bands
(b)
0.3
0.2
0.1
-0.1
-0.2
-0.3
-0.4
020406080 100 120
Spectral Bands
(c)
FIGURE 10. A visual example clarifying the SVs in a real HS scene (Pavia city center). (a) An image patch is cropped from the scene to show spectral
bundles involving the spectral variations of trees. (b) The pure spectral signature (i.e., the endmember) of trees acquired from the laboratory.
(c) The differences between (a) and (b), which are seen as SVs.
JUNE 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
71
Reflectance
SV
Reflectance

IEEE Geoscience and Remote Sensing Magazine - June 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2021