IEEE Geoscience and Remote Sensing Magazine - March 2022 - 263

NONRECTILINEAR TRAJECTORIES
A critical point in airborne GPR applications is related to
the accurate knowledge of the platform's position during the
survey. Indeed, in airborne or UAV-borne GPR imaging, the
height of the radar's antenna and the flight's velocity are not
constant; in addition, the trajectory, followed by the platform,
may deviate significantly from a rectilinear one. As a
result, positioning errors, which are comparable to or higher
than the operative wavelength, arise, and the radargrams
are severely distorted [129]. To avoid this detrimental effect
on the radar-imaging's performance, compensation of the
deviations with respect to the ideal flight track is a key issue,
requiring accurate knowledge of the UAV's platform position
and velocity. However, the quality of such information
depends on the accuracy of both the embarked navigation
sensors and the deployed ground-based tracking devices.
The most popular navigation sensors are IMUs with gyroscopes
and accelerometers, magnetometers, and Global
Navigation Satellite System (GNSS) receivers (GPS and multiconstellation
receivers). The data provided by these devices
are typically fused, and the final positioning accuracy is
driven by the GNSS receiver, which may reach centimeterlevel
accuracy. Unfortunately, onboard GPS devices may
have performance issues in complex scenarios, such as narrow
valleys or near mountains. In these conditions, various
solutions can be exploited to correct the motion errors of the
platform, such as range finders [130], edge-detection algorithms
[80], ground-based tracking devices as robotic total
stations [131], and multiple GNSS receivers that enable more
sophisticated trajectory reconstruction techniques [86].
An enhanced strategy for reconstructing the UAV's position
is based on CDGPS, a recognized technique that
enhances positioning accuracy. CDGPS exploits the differences
between the measurements collected by two relatively
close receivers (one onboard and one ground based)
to filter out the common errors affecting them (i.e., satellite
clock errors and tropospheric and ionospheric errors).
CDGPS data can be generated both in real time or in postprocessing.
An RTK CDGPS solution was proposed in [130].
This solution exploits a radio link between a ground receiver
and the onboard receiver to perform a real-time control
of the flight trajectory with an accuracy of 0.5 cm in the
east-north plane and of 1 cm along the vertical direction,
thus avoiding artifacts in the image-reconstruction process.
A postprocessing kinematic (PPK) CDGPS solution was applied
in [86]. The PPK strategy is simpler and less expensive
as it does not require any communication link between the
GPS receivers. Although it does not allow for real-time control
of the airborne trajectory, the PPK technique achieves
very accurate offline trajectory estimation, and this information
can be used in radar data processing to obtain accurate
imaging results.
UAV CHALLENGES
A further key point concerns contactless GPRs mounted
on-board a mini- and micro-UAV. This technology offers
MARCH 2022 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
several advantages compared to the classic airborne platforms,
such as ease of use, flexibility, limited cost, and
high survey accuracy. Mini- and micro-UAVs are extremely
compact platforms, but they support only a limited payload
mass. This entails that a necessary series of tradeoffs have
to be sought, concerning the type of components and sensors
that can be acquired. For instance, low-weight batteries
with limited capacity, compact GPR systems with small and
light antennas, and a small number of navigation systems
can typically be installed on these platforms. These constraints
unavoidably imply a loss in radar penetration capability,
reduced autonomy, and a less accurate estimation of
the flight's trajectory. Nevertheless, it must be stressed that
UAV technology is worth being investigated for its improvements
in the coming years.
LARGE-SCALE DATA PROCESSING
Contactless GPR imaging applications require the processing
of large-scale (in terms of probing wavelengths) data.
In this framework, the inversion of the linear integral
equation (3) based on the TSVD regularization scheme is
computationally demanding and may become impractical.
Accordingly, the SZ approach can be conveniently exploited
to reduce the computational effort (see [62] and the
references therein).
The SZ approach consists of focusing the data taken
from partially overlapping measurement domains and
combining the corresponding images in such a way as to
get an overall focused image. As schematically represented
in Figure 3, the main steps of the SZ procedure are detailed
as follows.
◗ The measurement line C and the surveyed domain D
are divided into K partially overlapping subdomains i
and iKD ,i
C
= 12 g ,, ,
procedure in (4).
◗ The tomographic image of the overall surveyed domain
D is obtained by combining the K reconstructions i
|u
achieved for each subdomain D .i
Γ1
Γ2
Γ3
ΓK
Γ
.
◗ For each subdomain ,Di
tion i|u
the tomographic reconstrucis
obtained by exploiting the TSVD inversion
x
z
D1
D2
D3
DK
FIGURE 3. The SZ imaging approach. The measurement line C
and the surveyed domain D are divided into K partially overlapping
subdomains.
263
IEEE Geoscience and Remote Sensing Magazine - March 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2022
