IEEE - Aerospace and Electronic Systems - July 2021 - 27

Credit: Image licensed by Ingrahm Publishing
literature toward the application of AI for clutter suppression
and classification while presenting and interpreting
their results. In the section " Target Classification, " we
give an overview of the research approaches in the field of
AI-based target classification highlighting the key ideas.
In " Target Tracking Support Through AI, " we analyze
each part of target tracking and point out potential for
applications in practice. Finally, we briefly summarize our
investigations.
AI-BASED CLUTTER IDENTIFICATION IN AIR TRAFFIC
CONTROL
In the area of air traffic control, we identified two major
research streams toward the application of AI for clutter
identification. While some researchers tried to distinguish
between targets and several different clutter classes
[12], [30], [48], [54], others focused on the reliable differentiation
between the two classes clutter and targets
[31], [51].
More precisely, the research group around Haykin et al.
AI IN CLUTTER IDENTIFICATION
In general, radar systems particularly aim to distinguish
between two types of radar returns: On the one hand, signal
echoes arising from the target types of interest (e.g.,
aircraft or cars) and, on the other hand, so-called clutter
coming from the reflection of undesired objects or phenomena
(e.g., birds or certain weather conditions) [48],
[30]. Identification of clutter is important for improving
target detection by reliably suppressing clutter [32], [49],
[50] as well as for discovering potential obstacles or
unsafe weather conditions [30], [48]. Conventionally,
clutter suppression is achieved by applying algorithms as
the moving target indicator (MTI) and the moving target
detector (MTD) [20], [48], [51] or using a constant false
alarm rate (CFAR) [20], [26], [52]. While the MTI aims
at removing stationary clutter signals (e.g., ground clutter),
the MTD is responsible for suppressing clutter arising
from moving objects [48]. In contrast, a CFAR
system is responsible for minimizing undesired signal
echoes that result from clutter or noise by setting a detection
threshold [53]. Moreover, numerous AI-based
approaches have been developed over the years in order
to identify clutter (cf. Figure 1). Hence, our focus lies on
the two most prominent radar application areas for clutter
identification with AI, which are air traffic control [12],
[30], [31], [48], [51], [54] and marine environments [20],
[26], [32], [49], [50], [52], [55], [56]. In the following,
we present for each application area the most representative
literature developing AI-based approaches for clutter
identification.
JULY 2021
[12], [30], [48], [54] started an extensive research program
on clutter classification aiming at the differentiation
between multiple major classes of radar returns including
target, weather, and birds. In a later phase of their program
from 1989-1991, their research in particular included the
investigation oftraditional AI-based approaches.
The data they used for their investigations has been
recorded with the help of two L-band air traffic control
area surveillance radars being part of the Terminal
Radar and Control Systems (TRACS) located on the
Canadian Forces' Bases at Moose Jaw in Saskatchewan
and Trenton in Ontario. This two-channel radar is a derivate
of the Westinghouse ASRS-3 whose both channels
can be used at the same time, each one on a different frequency
and polarization. However, in their investigations
only one horizontally polarized channel has been
used. As their research originally aimed at identifying
clutter in terms of birds, the majority of the data has
been recorded mainly at night time in 1985 during the
bird-migration season.
To classify the radar returns, features were extracted
and selected based on single scans. On this account, they
used feature sets which are mainly comprised of the
reflection coefficients that have been calculated based on
time series of radar measurements using the multisegment
Burg algorithm and an order-5 prediction error filter.
The filter order determines the highest reflection
coefficient order, which reflects the number of preceding
or following measurements considered for prediction.
The reflection coefficients of different orders are each
estimated based on the total prediction error consisting
IEEE A&E SYSTEMS MAGAZINE
27
IEEE - Aerospace and Electronic Systems - July 2021

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