IEEE - Aerospace and Electronic Systems - September 2021 - 19

Soldi et al.
exploited fruitfully for target classification beyond simple
size-based categorization toward more refined vessel-type
identification [59]. More recently, the increasing amount
of free and open optical data, e.g., S-2, Landsat 8, along
with the advances in hardware (e.g., GPU computing), has
boosted the application of machine learning paradigms,
e.g., deep learning, convolutional neural networks, and
support-vector machines, to both detection and classification
purposes for MS (see further details in the companion
paper [8]). To summarize, a high level of spatial detail
along with the easier interpretability and the absence of
speckle noise make optical satellite imagery a very attractive
solution for MS purposes. However, compared to
SAR, MSP imagery is affected by weather conditions,
e.g., clouds, and ocean waves.
HSP OPTICAL SENSORS
Figure 12.
(a) Detected vessels in a high-resolution MSP image (RGB)
acquired by WorldView-2 satellite (2 m resolution). (b) Same
detected vessels in the corresponding high-resolution PAN image
(0.5 m resolution). (c) Pansharpened (PAN + RGB) image showing
the two detected vessels. Scene extent: 1.2 (width) by 0.8 km
(height).
illustrated in [59]. However, most common techniques
include threshold-based methods [50],[54], where candidate
targets are selected by comparing image pixels with a
properly evaluated (possibly local) threshold, and methods
based on shape and texture features [52],[53],[55], where
potential targets are extracted by taking advantage of the
different geometrical and spectral characteristics of ships
versus the surrounding sea. The richer spectral information
provided by optical sensors compared to SAR can be
SEPTEMBER 2021
As opposed to MSP instruments, HSP remote sensing is a
relatively new technology able to detect and measure target
radiance up to hundreds of bands over the VIS/IR portion
of the EM spectrum. HSP systems offer a very high
spectral resolution (less than 1 nm), which allows nearlaboratory
quality measurements of radiation from far distances
and a " continuous " mapping of the spectral signature
of the sensed object compared with the discrete
spectral sampling provided by MSP sensors. This peculiar
feature enables the quantitative analysis and identification
of any substance, such as minerals, chlorophyll, chemical
agents, and gases, by comparing its measured spectral profile
with prepopulated libraries containing spectral signatures
of different targets. Accordingly, HSP data are
typically exploited in applications such as natural hazard
forecasting, atmosphere analysis, precision farming, water
quality control, urban mapping, and snow monitoring
[60]. The ability to reveal molecular absorption and transmissivity
bands with fine spectral details, unattainable by
current MSP sensors, makes optical HSP sensors also
referred to as imaging spectrometers, in analogy with
chemical spectroscopy [43].
HSP systems offer a unique tool in the framework of
MS, thanks to the high spectral resolution that adds a further
dimension to space-based measurements. Indeed, HSP
imagery extends surveillance and monitoring capabilities
by providing key support for detecting difficult targets, i.e.,
objects that are smaller, more fleeting, and/or are immersed
in more complex clutter, than those typically detectable
with other remote sensing technologies. For instance, in
maritime environments, typical low-observable targets
detectable using HSP data include chlorophyll and sediment
load in physical oceanographic applications, submarines,
and sea mines in military applications and schools of
cetaceans in marine biology applications. In all these scenarios,
the detection problem refers to low-contrast targets
IEEE A&E SYSTEMS MAGAZINE
19
IEEE - Aerospace and Electronic Systems - September 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - September 2021
