IEEE Geoscience and Remote Sensing Magazine - June 2021 - 53

and engineers who already have some background knowledge
in HS RS, ML, and optimization. This can further help
them launch new investigations based on the aforementioned
topics and interpretable AI techniques for their focused
fields.
OVERVIEW
BACKGROUND AND SIGNIFICANCE OF HS RS
Imaging spectroscopy, which was first conceptualized by
Goetz1 et al. in the 1980s [1], is the first HS imaging technique
to truly achieve integration of the 1D spectrum and
2D imaging for Earth RS. Imaging spectroscopy is a typical
passive RS technique, assembling spectroscopy and digital
photography into a unified system. Figure 1 shows the dataacquisition
process of two different imaging patterns: active
and passive RS [2]. The resulting HS image collects hundreds
of 2D images finely sampled from the approximately
contiguous wavelength across the whole electromagnetic
spectrum [3] (see Figure 2). This enables the recognition
and identification of materials, particularly those that have
extremely similar spectral signatures in visual cues [e.g.,
red, green, blue (RGB)] [4] at a more accurate and finer level.
As a result, HS RS has been significantly advanced and
widely applied in many challenging tasks of Earth observation
[5], such as fine-grained land cover classification, mineral
mapping, water-quality assessment, precision farming,
urban planning and monitoring, disaster management and
prediction, and concealed-target detection. More specifically,
characterized by the distinctive 3D signaling structure,
the advantages of HS images over conventional 1D or 2D
signal products can be summarized as follows:
◗ Compared to the common 1D signal system, the HS 2D
spatial pattern provides more structured information,
enabling the discrimination of underlying objects of interest
at a more semantically meaningful level.
◗ Beyond natural 2D images, the rich spectral information
in HS images is capable of detecting materials through
tiny discrepancies in the spectral domain because the
Satellite (or Sensor)
Active RS
Main Abbreviations Used in This Article
AI
Artificial intelligence
ANN
CML
DL
DR
GSD
HS
KNN
LDA
LMM
MA
ML
MML
MS
NMF
RS
SAR
SOTA
SSL
SU
TV
1D
2D
3D
Artificial neural network
Cross-modality learning
Deep learning
Dimensionality reduction
Ground-sampling distance
Hyperspectral
k-nearest neighbor
Linear-discriminant analysis
Linear mixing model
Manifold alignment
Machine learning
Multimodality learning
Multispectral
Nonnegative matrix factorization
Remote sensing
Synthetic aperture radar
State of the art
Shared subspace learning
Spectral unmixing
Total variation
One dimensional
Two dimensional
Three dimensional
HS imaging instruments exploit the sensors to collect
hundreds or thousands of wavelength channels with an
approximately continuous spectral sampling at a subtle
interval (e.g., 10 nm).
Furthermore, the significance of HS images compared
to other RS imaging techniques with a lower spectral resolution,
e.g., multispectral (MS) imaging or RGB imaging,
mainly lies in the following three aspects:
1) HS images are capable of finely discriminating among
different classes that belong to the same category, such as
bitumen and asphalt, stressed grass and synthetic grass,
Solar (Lighting Source)
Passive RS
Satellite (or Sensor)
ForestsRocks
Grass
Built-Up Area
ForestsRocks
Grass
Built-Up Area
FIGURE 1. An illustration that clarifies the data-acquisition process of two different imaging patterns, i.e., active and passive RS.
JUNE 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
53
IEEE Geoscience and Remote Sensing Magazine - June 2021

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