signals

Article
Classification and Discrimination of Birds and Small Drones
Using Radar Micro-Doppler Spectrogram Images †
Ram M. Narayanan 1, * , Bryan Tsang 1 and Ramesh Bharadwaj 2

1 Department of Electrical Engineering, The Pennsylvania State University, University Park, PA 16802, USA;
[email protected]
2 Center for High Assurance Computer Systems, Code 5546, U.S. Naval Research Laboratory,
Washington, DC 20375, USA; [email protected]
* Correspondence: [email protected]; Tel.: +1-814-863-2602
† This work was presented partially at Tsang, B.; Narayanan, R.M.; Bharadwaj, R. Experimental analysis of
micro-Doppler characteristics of drones and birds for classification purposes. In Proceedings of the SPIE
Conference on Radar Sensor Technology XXVI, Orlando, FL, USA, 3–7 April 2022; doi: 10.1117/12.2622408.

Abstract: This paper investigates the use of micro-Doppler spectrogram signatures of flying targets,
such as drones and birds, to aid in their remote classification. Using a custom-designed 10-GHz
continuous wave (CW) radar system, measurements from different scenarios on a variety of targets
were recorded to create datasets for image classification. Time/velocity spectrograms generated for
micro-Doppler analysis of multiple drones and birds were used for target identification and movement
classification using TensorFlow. Using support vector machines (SVMs), the results showed an
accuracy of about 90% for drone size classification, about 96% for drone vs. bird classification, and
about 85% for individual drone and bird distinction between five classes. Different characteristics of
target detection were explored, including the landscape and behavior of the target.

Keywords: drone detection; UAV detection; bird detection; micro-Doppler; spectrogram; continuous-
wave radar; target classification; time-frequency analysis

Citation: Narayanan, R.M.; Tsang, B.;

Bharadwaj, R. Classification and
1. Introduction
Discrimination of Birds and Small
Drones Using Radar Micro-Doppler With the advent of commercial Unmanned Aerial Vehicles (UAVs), drone detection has
Spectrogram Images. Signals 2023, 4, become more of a necessity in recent years. The remote operative capabilities of UAVs pose
337–358. https://doi.org/10.3390/ threats to security, privacy, and other potential legal issues. Additionally, the number of
signals4020018 applications for UAVs has also increased, spanning recreational, commercial, and military
uses. Nowadays, state-of-the-art drones even possess wireless networking capabilities.
Academic Editors: Chi-Yi Tsai and
Most modern drones are equipped with video cameras and can even deliver small packages.
Francesco de Paulis
The relatively cheap costs of UAVs paired with remote operability allows anyone to become
Received: 31 January 2023 a pilot with low risk to their health. If used incorrectly, many hazards may be overlooked
Revised: 26 March 2023 and the degree of risk involved increases.
Accepted: 11 May 2023 Although guidelines have been created by the FAA for remote pilots, the ease of access
Published: 18 May 2023 to these systems can prove to be dangerous. People with malicious intentions pay no heed
to the warnings and rules created by the FAA. For example, in 2018, hundreds of flights
were canceled at Gatwick Airport due to a drone sighting close to the runway, causing flight
delays of two days [1]. Methods for detecting drones in the past include acoustic, optical,
Copyright: © 2023 by the authors.
and radar methods. RCS-based radar methods, such as high-resolution range profiles
Licensee MDPI, Basel, Switzerland.
and synthetic aperture radar, can be difficult when considering the close RCS values of
This article is an open access article
birds and drones. To avoid this, we investigated the benefits of using micro-Doppler
distributed under the terms and
conditions of the Creative Commons
signatures from targets using a CW radar system. In addition to detecting the flying target,
Attribution (CC BY) license (https://
it is possible to discern the type of target (bird or drone) based on their micro-motions
creativecommons.org/licenses/by/ such as propeller blade spinning or bird wing flapping. Using the short-time Fourier
4.0/). transform (STFT), we generated time-frequency plots called spectrograms to observe the
RCS - Radar Cross Section

Signals 2023, 4, 337–358. https://doi.org/10.3390/signals4020018 https://www.mdpi.com/journal/signals

Signals 2023, 4 338

individual micro-Doppler characteristics. The classification efforts involved the overall

Doppler bandwidth spread, base velocity, and cycle frequency.
Fortunately, the Federal Aviation Administration (FAA) created the Part 107 regu-
lations for commercial small UAVs in June 2016. These regulations require UAV pilots
to register their UAVs if they weigh over 250 g [2]. In addition, the pilot must carry a
small UAV operator license, which can be obtained after passing the Part 107 exam. This
exam covers a wide range of information regarding UAV safety. Some topics include UAV
limitations, flight operations, airspace classification and operating requirements, flight
restrictions, weather awareness, emergency procedures, radio communication procedures,
and more.
State-of-the-art techniques for detecting drones include acoustic [3,4], optical camera [5,6],
lidar [7,8], and radar [9–11]. Despite the wide range of sensing practices, there are several
trade-offs associated with each method. In Table 1, the advantages and disadvantages of
sensing drones with acoustics, cameras, lidar, and radar sensors are listed.

Table 1. Sensing methods advantages and disadvantages.

Sensor Type Advantages Disadvantages

No LOS
Ineffective in noisy conditions
Acoustic Low visibility
Ineffective in windy conditions
Low cost
Very sensitive operability
Easy operability
Camera High cost
High accuracy
Ineffective in darkness
Expensive
Ineffective in foggy and rainy
Lidar High accuracy
conditions
Very narrow FOV
Low-cost
Targets have small visibility
Radar Weather resistant
Large FOV
Robust

Some of the advantages of acoustics include no line-of-sight (LOS) requirements,

effective performance in low-visibility conditions, and relatively low implementation
costs. However, in noisy conditions, such as windy gusts, microphones will not be able to
distinguish or detect drones. Using cameras for detecting targets is advantageous due to
their high accuracy and easy operability. Although it is easy to detect flying targets, it is
difficult to classify the subject further. There is also a high cost associated with using and
maintaining cameras. Lidar sensing is an accurate method for detecting targets, but the
cons greatly outweigh the pros. They are expensive, have a narrow LOS, and are heavily
weather-dependent. Finally, radar sensing provides an effective method for detection due
to the low cost of implementation and robustness of operability. The biggest weakness of
using radar is that such targets tend to have a low radar cross section (RCS) or visibility,
which can hinder their ability to detect them.
RCS-based radar methods such as high-resolution range profiles and synthetic aper-
ture radar (SAR) can be difficult when considering the similar RCS values of birds and
drones [12–15]. To avoid this, the benefits of leveraging the micro-Doppler signatures from
targets were investigated using a continuous-wave (CW) radar system. Through processing
algorithms, it was possible to discern the type of target (bird or drone) based on their
micro-motions, such as spinning of propeller blades or flapping of bird wings.
For automatic target recognition (ATR) purposes, many recent studies have utilized
machine learning to classify images of micro-Doppler signatures between drones and birds.
Unfortunately, the lack of real-world data makes improving classification efforts difficult.
The remainder of this paper is structured as follows: Section 2 discusses the concepts
and theory related to the micro-Doppler features of birds and drones. Section 3 describes
Signals 2023, 4 339

our simulation setup and results in order to provide an understanding of the experimental
data. Section 4 discusses the design and operational characteristics of the constructed
10-GHz radar systems. Section 5 outlines the experimental setup and the results obtained.
In Section 6, the classification setup and processing tactics are described. In Section 7, the
conclusions are presented with possible directions for future research. In this paper, please
note that “UAV” and “drone” are used interchangeably.

2. General Characteristics of Micro-Doppler Signatures from Birds and Drones

As is well known, the reflected signal from a target that is in motion relative to a radar
contains a frequency shift relative to the transmit frequency due to the well-known Doppler
effect [16]. Quite simply, if f is the transmit frequency of a monostatic radar (i.e., transmitter
and receiver collocated, as is the usual case), for a target moving at a speed of v at an angle
of θ towards the radar, the Doppler frequency shift f d is given by f d = 2v cos θ/λ, where
λ is the transmit wavelength. The radial velocity vr is given by vr = v cos θ. For directly
approaching targets, θ = 0, the Doppler shift is positive, whereas for directly receding
targets, θ = π, the Doppler shift is negative. For a stationary target (v = 0) or for a target
moving at right angles to the LOS to the radar (θ = π/2), the Doppler shift is zero.
In many cases, a structural component attached to a target may exhibit micromotions
in addition to the main motion of the object (including zero bulk motion). Micromotions
can be caused by small movements around the rest position, such as vibration, oscillation,
rotation, swinging, flapping, flutter, etc. [17]. The source of micromotion may be a revolving
surface, such as the rotor blades of a drone, or a vibrating surface, such as the swinging
arms and legs of a walking human, or the flapping wings of a bird, for example.
These micromotions induce frequency modulations in the main Doppler frequency
shift caused by the bulk motion of the target. Micromotions consisting of pure periodic
vibration or rotation generate Doppler frequency shifts around the central Doppler-shifted
carrier frequency and appear as sidebands. The modulation contains harmonic frequencies
determined by the carrier frequency, the vibration or rotation rate, and the angle between
the direction of vibration and the direction of the incident wave. The time-varying frequency
modulations provide unique signatures of the object for further classification, recognition,
and identification.
Micro-Doppler signatures of birds and drone targets are characterized by sinusoidal
modulation owing to the periodic motion of the rotating blades of drones and UAVs and the
flapping wings of birds [18]. The modulation frequency of drones and UAVs is higher by
one order of magnitude compared to that of birds because these targets contain high-speed
rotation blades. Since the number of blades on a drone is usually an even number, when a
blade approaches the radar, another blade recedes from the radar; thus, its micro-Doppler
signature spectrogram is symmetric. In contrast, both wings flap up and down for birds at
the same time; thus, their spectrograms are asymmetric.
The main source of micro-Doppler signatures from birds arises from the oscillatory
motion of their flapping wings. The frequency of the Doppler fluctuations is primarily
related to the wing flapping (or wingbeat) frequency and its harmonics [19,20]. Periodic
fluctuations in signal intensity are attributed to changes in body shape due to flapping [19].
In addition, periodic intensity modulations can be caused by the appearance and disap-
pearance of the “corner reflector effect” caused by the wings as they flap to and fro [20].
Therefore, the unique micro-Doppler signatures of birds are caused by several mechanisms.
The micro-Doppler spread of a bird is quite narrow, and its characteristic signature takes
the form of periodic flashes corresponding to the wingbeat frequency [9]. The micro-
Doppler signatures of birds depend on their size, wingspan, wing flapping frequency, radar
frequency, and viewing angle.
In the case of drones and UAVs, the spinning of the rotor blades is the primary cause
of their micro-Doppler signatures [21,22]. The rotation is deterministic and controlled, and
coupled with the fact that these targets have a high RCS, their micro-Doppler signatures are
clean, symmetric, and recognizable compared to those of birds. Assuming pure periodic
 In the case of drones and UAVs, the spinning of the rotor blades is the primary cause
of their micro-Doppler signatures [21,22]. The rotation is deterministic and controlled, and
coupled with the fact that these targets have a high RCS, their micro-Doppler signatures
Signals 2023, 4 340
are clean, symmetric, and recognizable compared to those of birds. Assuming pure peri-
odic rotation, the micromotion dynamics generate sideband Doppler frequency shifts
about the scattered signal. The modulation contains spectral components that depend on
rotation,
the the or
vibration micromotion dynamics
rotation rate, generate
the vibration sideband
amplitude, andDoppler
the anglefrequency
between theshifts about
direction
thevibration
of scatteredandsignal. The modulation
the direction contains
of the incident spectral
wave [23]. components that depend
A small asymmetry may be ondue
the
vibration
to or rotation
the asymmetry of rate, the vibration
the propellers, amplitude,
as the and
propeller theisangle
that between
moving towardsthe the
direction
radar of
is
vibration
curved in and the direction
the opposite wayof the respect
with incidenttowave [23]. A small
the propeller asymmetry
moving away from maythe
beradar
due to[24].
the
asymmetry
The of the propellers,
micro-Doppler signaturesasofthe propeller
drones that on
depend is moving towards
their size, number theof
radar is curved
propellers or
in the opposite way with respect to the propeller moving
rotors, rotational frequency, radar frequency, and viewing angle. away from the radar [24]. The
micro-Doppler
The uniquesignatures
signaturesofofdrones depend
birds and on can
drones theirbesize, number
exploited notofonly
propellers or rotors,
to perform bird
rotational
vs. frequency, radar
drone classification but frequency, and viewing
also to classify angle. sizes and drones of different
birds of different
The unique
structures signatures
and types [25–28].of birds and drones can be exploited not only to perform bird
vs. drone classification but also to classify birds of different sizes and drones of different
structures
3. Simulated andMicro-Doppler
types [25–28]. Signatures
3.1. TheoreticalMicro-Doppler
3. Simulated Considerations Signatures
The equation
3.1. Theoretical for Doppler frequency f d for a rotating target is given by
Considerations
The equation for Doppler frequency f for a rotating target is given by
f  2 vd  (1)
d r

f d = 2vand
where vr is the radial velocity of the target, r /λ  is the radar wavelength. The corre-
(1)

where vr isthefor
sponding a 10-GHz
radial velocitycarrier
of thefrequency
target, andisλ3iscm.
theA rotor’s
radar rotationalThe
wavelength. velocity  in
correspond-
revolutions per second
ing λ for a 10-GHz is related
carrier to the
frequency is 3maximum expected of
v r rotational
cm. A rotor’s the blade
velocity v intip through
revolutions
per second is related to the maximum vr expected of the blade tip through
  vr 2 l (2)
v = vr /2πl (2)
where l is the blade length. In the following simulations,  is modeled at 56 revolu-
tions
whereper
l issecond, which
the blade corresponds
length. to a vr simulations,
In the following value of 35.2vm/s for an latvalue
is modeled of 10 cm.
56 revolutions
This maps towhich
per second, an expected maximum
corresponds to a vDoppler
r value ofshift
35.2around
m/s for2.35
an lkHz occurring
value of 10 cm.atThis
the blade
maps
tips.
to an expected maximum Doppler shift around 2.35 kHz occurring at the blade tips. Itor
It is assumed that the blade is along the LOS of the radar. Most UAVs have two is
three
assumedblades
thatassociated with
the blade is each
along therotor.
LOS Figure 1 shows
of the radar. how
Most the radial
UAVs component
have two of the
or three blades
rotating
associated blade
withchanges overFigure
each rotor. time. 1 shows how the radial component of the rotating blade
changes over time.

Figure 1. Radar signal interaction with a single quadcopter rotor.

Figure 1. Radar signal interaction with a single quadcopter rotor.
The radar system transmits a continuous signal at a single frequency. These signals
The radar
bounce off system transmits
the rotating propellers,aand
continuous
the radarsignal at athese
receives single frequency.
reflected These
signals. Thesignals
radial
bounce off the rotating propellers, and the radar receives these reflected signals.
velocity vr of the spinning blade tips does not remain constant, but the periodicity The radial
can
velocity of thethe
v r with
be modeled spinning
use of ablade tips doesplot
velocity-time not and
remain constant, but
a spectrogram. Inthe periodicity
addition, can
a simple
Fast Fourier Transform (FFT) of the raw data provides a power spectral density plot that
denotes the peaks of frequencies, corresponding to the radial velocities observed within
the recording. An FFT takes a window size of samples and computes the power spectral
density of that instance. The spectrogram is known as the Short-Time Fourier Transform
(STFT), which computes multiple FFTs and displays them as frequency or velocity over
 be modeled with the use of a velocity-time plot and a spectrogram. In addition, a simple
Fast Fourier Transform (FFT) of the raw data provides a power spectral density plot that
denotes the peaks of frequencies, corresponding to the radial velocities observed within
Signals 2023, 4 341
the recording. An FFT takes a window size of samples and computes the power spectral
density of that instance. The spectrogram is known as the Short-Time Fourier Transform
(STFT), which computes multiple FFTs and displays them as frequency or velocity over
time.
time. In Figure
In Figure 2, the
2, the power
power spectral
spectral density
density and and the spectrogram
the spectrogram of a single
of a single propeller
propeller are
are
shown.shown.

(a) (b)
Figure 2. Plots
Figure related
2. Plots to to
related a single propeller
a single drone.
propeller (a)(a)
drone. power spectral
power density;
spectral (b)(b)
density; spectrogram.
spectrogram.

AsAsthe
theblade
blademoves
moveshorizontally
horizontally along the
the radar’s
radar’sLOS,
LOS,vr vbecomes
r
becomes 0. When
0. When the
the blade
begins
blade to move
begins directly
to move to or away
directly to or from
awaythe radar,
from the the maximum
radar, vr valuesvarevalues
the maximum detected.
are A
r
target moving towards the radar results in a positive (+) Doppler shift, whereas a target
detected. A target moving towards the radar results in a positive (+) Doppler shift,
moving away results in a negative (−) Doppler shift. The maximum peaks in the power
whereas a target moving away results in a negative (−) Doppler shift. The maximum peaks
spectrum are around 2.5 kHz. In the resulting spectrogram, the maximum velocity is
in the power spectrum are around 2.5 kHz. In the resulting spectrogram, the maximum
confirmed to be roughly 37.5 m/s which is per Equation (1).
velocity is confirmed to be roughly 37.5 m/s which is per Equation (1).
Additional Doppler frequencies can be caused by the main body movement of the
Additional Doppler frequencies can be caused by the main body movement of the
drone. For a moving drone, the maximum Doppler shift is given by
drone. For a moving drone, the maximum Doppler shift is given by

fd fd = f d_body
fd_body + f d_micro-Doppler
 fd_micro-Doppler (3) (3)
where f d_body is the Doppler shift induced by the main body movement of the drone and
where fd_body is the Doppler shift induced by the main body movement of the drone and
f d_micro-Doppler is the micro-Doppler shift due to the rotating blades. Furthermore, because
fdthe is the measures
radar system
_micro-Doppler
micro-Doppler
radialshift due the
velocity, to the rotating
look blades.
angle from the Furthermore, because
radar to the target plays
a vital role in the displayed Doppler frequency. Equation (3) can be rewritten
the radar system measures radial velocity, the look angle from the radar to the target plays as follows
whenrole
a vital considering the lookDoppler
in the displayed angle dependence
frequency. Equation (3) can be rewritten as follows
when considering the look angle dependence
f d (θ ) = f d cos θ (4)
f d ( )  f d cos  (4)
where θ is the angle between the drone velocity vector and the radar LOS.
where The  ismicro-Doppler
the angle between the drone velocity
characteristics of flying vector
objectsandvary
the radar LOS. to target. For
from target
The micro-Doppler
instance, a quadcoptercharacteristics
signature willofbeflying objects
different fromvary from plane
a fixed target or
to target. For in-In
a helicopter.
stance, a quadcopter
addition, signaturesystems
many surveillance will beconfuse
different fromwith
drones a fixed
birdsplane
due or a helicopter.
to their In ad-
similar sizes and
dition,
speeds.many surveillance
Conveniently, systems
the confuse drones
micro-Doppler with birds
phenomenon due to
is a great their similarbetween
discriminator sizes andthe
speeds. Conveniently,
two classes. In birds, the
the micro-Doppler of phenomenon
the flappingispattern
a greatproduces
discriminator between
a periodic shape
thebut
two classes.
is not In birds,
as strong the micro-Doppler
and quick as a drone. Theofrotationthe flapping
speed pattern produces
for a drone’s blades a can
periodic
exceed
shape
6000butRPM,is not
butas
thestrong
wing and
flapsquick as aare
of a bird drone.
much The rotation
slower. speed the
Through for implementation
a drone’s bladesof
can exceed 6000 the
spectrograms, RPM, but the wingsignatures
micro-Doppler flaps of a bird are much
of different slower.
flying Through
targets can be the imple-
classified.
mentationThe of spectrograms,
mechanics theinmicro-Doppler
of drones flight are unique. signatures of different
By generating thrust flying
through targets can
the rotation
beofclassified.
the four motors, each with two blades attached, the UAVs can support their weight and
ascend. Different combinations
The mechanics of rotation
of drones in flight speeds
are unique. Byresult in different
generating thrustUAV flight
through themotions.
rota-
In of
tion thisthestudy, the weight
four motors, eachand
withsize
twoofbladesthe target are discriminating
attached, the UAVs can supportfactors their
for which
weight the
micro-Doppler signature belongs to which target. For heavier UAVs, more thrust is required
to lift the drone, resulting in a higher detected micro-Doppler bandwidth.
Signals 2023, 4 342

3.2. Simulation Setup and Results

Before the data were collected experimentally, simulations were created to replicate the
micro-Doppler detection scenario. These simulations were created in the MATLAB (R2021a)
environment using the Radar Toolbox. The Radar Toolbox expediates and supports the
simulation of radar systems through design, deployment, and field data analysis. First
and foremost, constants were set for the initial simulation parameters. The simulation
parameters are listed in Table 2, denoting the setup for drone behavior throughout the
detection process.

Table 2. Simulation setup parameters.

Parameter Value
Environment Free space
Range 10 m
Dwell time 0.5 s
Sampling rate 12 ksps
Number of rotors on drone 4
Blades per rotor 2
Average blade rotation rate (v) 56 rps
Blade length 0.1 m
Radar cross section 0.3 m2

Using the phased toolkit, a rectangular waveform was used to simulate the CW
behavior of the 10-GHz signal. The sampling rate and the Pulse Repetition Frequency (PRF)
were at the same value of 44.1 kHz, the default sampling rate of a laptop’s audio card. By
setting the sampling rate and the PRF to the same value with the pulse width equal to the
inverse of the PRF, a 100% duty cycle was simulated.
Two antenna elements were used: one for transmission and one for reception. These
antenna elements were spaced at half-wavelength (i.e., λ/2) distances from each other. The
transmit power used was 70.79 mW with a gain of 22 dB. The receiver preamplifier was set
to a gain of 19 dB and a noise figure of 1.7 dB. The environment was free space, where the
propagation speed was 3 × 108 m/s.
The drone body was positioned at (10, 0, 10), where the z-axis is the height. The target
velocity was 0 for all axes, meaning the drone was hovering in place. The drone behavior
was simulated by consisting of the main body RCS of 0.3 m2 with four accompanying
rotors, each evenly positioned 15 cm away from the body.
A for loop was created to simulate each ‘pulse’, sample. With each sample, the drone
and each blade position were updated. The sample was transmitted to the target and
reflected on the radar. The value was stored in an array until each sample was transmitted
and received, summing up to 0.5 s.
The first plot generated was a time plot shown in Figure 3. This represents the raw
values collected by the radar system.
Subsequently, the Fast Fourier Transform (FFT) operation was performed on the time
signal. Since the FFT is a cyclical process, the frequency axis was split into negative and
positive frequencies. The Doppler frequency f d axis was converted to velocity v using
v = λ f d /2. The frequency response is shown in Figure 4. The maximum detectable velocity
was around ±30.24 m/s but the strongest return was at ±23.52 m/s.
Finally, spectrograms were generated with a window size of 128 segments with
120 overlapping segments. The time axis, originally in seconds, was converted to millisec-
onds. The frequency axis was converted to velocity. The spectrogram in Figure 5 accurately
displays the micro-Doppler signature of the hovering quadcopter. Periodic maximum
velocity values were observed, confirming the cyclical nature of the rotating blades.
Signals 2023,
Signals 2023,4,4FOR PEER REVIEW 343

Figure 3. Simulated quadcopter time signal.

Subsequently, the Fast Fourier Transform (FFT) operation was performed on the tim
signal. Since the FFT is a cyclical process, the frequency axis was split into negative an
positive frequencies. The Doppler frequency f d axis was converted to velocity v usin
v   f d 2 . The frequency response is shown in Figure 4. The maximum detectable veloc
Figure 3. Simulated quadcopter time signal.
ity was around ±30.24 m/s but the strongest return was at ±23.52 m/s.
Subsequently, the Fast Fourier Transform (FFT) operation was performed on the tim
signal. Since the FFT is a cyclical process, the frequency axis was split into negative an
positive frequencies. The Doppler frequency f d axis was converted to velocity v usin
v   f d 2 . The frequency response is shown in Figure 4. The maximum detectable velo
ity was
Figure
Figure around
3.3.Simulated
Simulated±30.24 m/stime
quadcopter
quadcopter but
timethesignal.
signal.strongest return was at ±23.52 m/s.

Subsequently, the Fast Fourier Transform (FFT) operation was performed on the tim
signal. Since the FFT is a cyclical process, the frequency axis was split into negative and
positive frequencies. The Doppler frequency f d axis was converted to velocity v using
v   f d 2 . The frequency response is shown in Figure 4. The maximum detectable veloc
ity was around ±30.24 m/s but the strongest return was at ±23.52 m/s.
Figure 4. Simulated quadcopter frequency response, converted to velocity response.

Finally, spectrograms were generated with a window size of 128 segments with 12
overlapping segments. The time axis, originally in seconds, was converted to milliseconds
The frequency axis was converted to velocity. The spectrogram in Figure 5 accurately dis
plays the micro-Doppler signature of the hovering quadcopter. Periodic maximum veloc
Figure
ity
Figure 4. 4.
values Simulated
Simulated quadcopter
were observed,
quadcopter frequency
confirming
frequency response,
the cyclical
response, converted
nature
converted oftothe
to velocity velocity response.
rotating
response. blades.

Finally, spectrograms were generated with a window size of 128 segments with 12
overlapping segments. The time axis, originally in seconds, was converted to millisecond
The frequency axis was converted to velocity. The spectrogram in Figure 5 accurately di
plays the micro-Doppler signature of the hovering quadcopter. Periodic maximum velo
ity values were observed, confirming the cyclical nature of the rotating blades.
Figure 4. Simulated quadcopter frequency response, converted to velocity response.

Finally, spectrograms were generated with a window size of 128 segments with 12
overlapping segments. The time axis, originally in seconds, was converted to milliseconds
The frequency axis was converted to velocity. The spectrogram in Figure 5 accurately dis
plays the micro-Doppler signature of the hovering quadcopter. Periodic maximum veloc
ity values were observed, confirming the cyclical nature of the rotating blades.
Figure5.5.Simulated
Figure Simulated quadcopter
quadcopter spectrogram.
spectrogram.

Figure 6 shows the simulated spectrograms for the different window sizes. By adjust-
ing the window size N of the STFT, a tradeoff between the time resolution and Doppler
resolution was observed. When the window size was set to 64 samples, the time resolution
was very fine, emphasizing the micro-Doppler periodicity of each blade tip. However,
the velocity readings spanned a range of values, resulting in ambiguity regarding the
Figure
true 5. Simulated
Doppler velocity.quadcopter spectrogram.
On the other hand, with a window size of 512 segments, the time
resolution became ambiguous, but the velocity measurements were finer and more precise.
For the images generated for the datasets in this work, a window size of 128 segments
was utilized.

Figure 5. Simulated quadcopter spectrogram.

 ments
pler was utilized.
resolution was observed. When the window size was set to 64 samples, the time res-
olution was very fine, emphasizing the micro-Doppler periodicity of each blade tip. How-
ever, the velocity readings spanned a range of values, resulting in ambiguity regarding
the true Doppler velocity. On the other hand, with a window size of 512 segments, the
Signals 2023, 4
time resolution became ambiguous, but the velocity measurements were finer and more 344
precise. For the images generated for the datasets in this work, a window size of 128 seg-
ments was utilized.

(a) (b)

(a) (b)

(c) (d)
(c) (d)
Figure 6. Spectrograms with different window sizes. (a) N  64 ; (b) N  128 ; (c) N 
N N64=; (b) N N128 (c) N(c)N256
N 6.6.
Figure
Figure Spectrograms
512 with
Spectrograms
. different
with window
different sizes.
window (a) (a)
sizes. 64; (b) = ; 128; = ; 256;
(d)
N  512 .
(d) N = 512.

In In Figure
In Figure
Figure 7,
7, the7,integration
the the integration
integration time
time is time isto
is lowered
lowered lowered
to to the
emphasize
emphasize emphasize
the periodicitythe
periodicity periodicity
within
within the
the w
blade
blade
blade rotations.
rotations.
rotations. Peaks
Peaks Peaks
are are detectable
aredetectable
detectable around
around±40 ±around
m/s. ±40 m/s.
40m/s.

Figure 7. Simulated quadcopter results with shortened integration time.

Figure
Figure 7. Simulated
7. Simulated quadcopter
quadcopter results
results with with shortened
shortened integration
integration time. time.

In Figure 8, simulations are generated to replicate the ascending, hovering, and

descending motions of a quadcopter. By increasing the acceleration of the blades through
time, the simulation mimics the vertical motions of a quadcopter.
 Signals 2023, 4, FOR PEER REVIEW 9
Signals 2023, 4, FOR PEER REVIEW 9

In Figure 8, simulations are generated to replicate the ascending, hovering, and de-
scending motions
In Figure of a quadcopter.
8, simulations By increasing
are generated the the
to replicate acceleration of hovering,
ascending, the bladesand
through
de-
Signals 2023, 4 345
time, the motions
scending simulation
of amimics the vertical
quadcopter. motions of
By increasing thea acceleration
quadcopter.of the blades through
time, the simulation mimics the vertical motions of a quadcopter.

(a) (b)
(a) (b)

(c)
(c)
Figure 8. Different rotation accelerations. (a) Ascending; (b) Hovering; (c) Descending.
Figure 8.
Figure 8. Different
Different rotation
rotation accelerations.
accelerations. (a)
(a)Ascending;
Ascending;(b)
(b)Hovering;
Hovering;(c)
(c)Descending.
Descending.
4. Radar SystemDesign
4. Designand andConstruction
ConstructionDetails
Details
4. Radar System Design and Construction Details
AnX-Band
An X-Band10-GHz
10-GHzCW CWradar
radarsystem
systemwaswasdesigned
designedin-house
in-house toperform
performDoppler
Doppler
10-GHz CW radar system was designed in-house totoperform Doppler
measurements.
measurements. By transmitting a constant frequency signal, the returned signal is received
measurements. By transmitting aa constant
constantfrequency
frequencysignal,
signal,the
thereturned
returnedsignal
signalisisreceived
received
andprocessed.
and processed.Since
Sincethisthisisisaacontinuous-wave
continuous-wavesystem,
system,there
thereare
areno
noranging
rangingcapabilities
capabilities
and processed. Since this is continuous-wave system, there are no ranging capabilities
available. Figure
available. Figure999shows
shows
shows the block diagram of of
thethe
10-GHz radar system, which waswas con-
available. Figure thethe block
block diagram
diagram of the 10-GHz
10-GHz radarradar system,
system, whichwhich
was con-
structedand
structed andand
constructed used
used fordata
used
for data
for datacollection.
collection.
collection.

Figure
Figure 9.9.Block
Figure9. diagram
Blockdiagram
Block of
diagramof X-Band
ofX-Band 10-GHz
X-Band10-GHz radar
10-GHzradar system.
radarsystem.
system.

The 10-GHz signal is generated by powering the voltage-controlled oscillator (VCO)

with a 12-V power supply. This signal is sent to the splitter, where half of the power is
sent to the transmit chain and the rest is sent to the In-Phase/Quadrature (I/Q) mixer. In
this step, it is necessary to compare and mix the transmitted and received signals to obtain
the Doppler frequencies. To avoid saturation within the power amplifier (PA), a 12-dB
Signals 2023, 4 346

attenuator was used to reduce the output power before amplification. A +18 dBm power
output is transmitted through a 10-dB gain horn antenna.
After the signal is reflected from a target, it is received through an identical horn
antenna and fed through two low-noise amplifiers (LNA2 and LNA1). Next, the signal
passes through a bandpass filter (BPF) to filter any undesired reflections or clutter. This
filtered signal is mixed with the local oscillator signal to produce the Doppler frequency in
two channels. The inphase channel maintains a pure Doppler signal, while the quadrature
channel implements a 90◦ phase shift. Both signals are passed through two low-pass filters
(LPF1 and LPF2) before going through an auxiliary cable to be processed within a laptop
environment through MATLAB.
Table 3 lists the overall parameters of the radar system, while Table 4 lists the relevant
component specifications.

Table 3. Radar system parameters.

Parameter Value
Frequency 10 GHz
Waveform CW
Transmit power +18 dBm
Antenna gain 10 dBi
Receiver noise figure 1.8 dB
Sampling rate 12 ksps
Sampling interval 500 ms

Table 4. Component specifications.

Operating Noise Power

Component Part # Gain VDC Current Supplier
Frequency Figure Output
VCO V2555C 9.95–10.66 GHz – – +12 dBm 12 V 150 mA GMI
Power Splitter ZX10-2-126-S+ 7.4–12.6 GHz – – – – – Mini-Circuits
Attenuator FW-12+ DC–12 GHz – – – – – Mini-Circuits
Power Amplifier ZX60-183-S+ 6–18 GHz 23 dB – +18.5 dBm 5V 260 mA Mini-Circuits
Horn Antenna WR90 8.2–12.4 GHz 10 dBi – – – – A-InfoMW
Low Noise APTMP5-
8–12 GHz 40 dB 3 dB +25 dBm 1 15 V 450 mA AmpliTech
Amplifier 1 08001200-3025-D6
Low Noise
ZX60-05113LN+ 5–11 GHz 19 dB 1.8 dB +12.2 dBm 1 5V 42 mA Mini-Circuits
Amplifier 2
Band Pass Filter ZVBP-10R5G-S+ DC–40 GHz – – – – – Mini-Circuits
Marki
Frequency Mixer IQB0618LK 6–18 GHz – – – – –
Microwave
Low Pass Filter 1 VLF-3400+ DC–3.4 GHz – – – – – Mini-Circuits
Low Pass Filter 2 EF502 DC–100 kHz – – – – – Thorlabs
1 Power output at 1-dB gain compression.

A photograph of the breadboard radar system is shown in Figure 10. The radar system
was mounted on a cart, as shown in Figure 11. The 10-GHz radar system setup consists of
three shelves. The top shelf contains the laptop and the horn antennas. The middle shelf is
home to the DC power supply and the RF components of the 10-GHz system. The bottom
shelf carries the portable power station that supports the DC power Supply.
The horn antennas were configured on a wooden mount pointed at 30◦ above the
horizon, as shown in Figure 12. There is also a webcam placed above the horn antennas for
ground truth when recording data. By using a 30◦ look angle, the micro-Doppler signatures
of the drones were more pronounced, which will be discussed in greater depth in Section 5.
The polarization used was horizontal-transmit horizontal-receive (HH), which is expected
to provide the best UAV detection results at far ranges because the blade faces are wider
horizontally than vertically when viewed afar at low grazing angles. Similar conclusions
were drawn using forward-scattering data under different polarizations [29].
 in Section 5. The polarization used was horizontal-transmit horizontal-receive (HH
which is expected
signatures to provide
of the drones thepronounced,
were more best UAV detection results
which will at far ranges
be discussed because
in greater depththe blad
faces
in are wider
Section 5. Thehorizontally
polarization than
used vertically when viewed afar
was horizontal-transmit at low grazing
horizontal-receive angles. Sim
(HH),
which is expected to provide the best UAV detection results at far ranges because
ilar conclusions were drawn using forward-scattering data under different polarization the blade
faces
[29].are wider horizontally than vertically when viewed afar at low grazing angles. Sim-
Signals 2023, 4 347
ilar conclusions were drawn using forward-scattering data under different polarizations
[29].

Figure 10.
10. Photograph
Photograph of
of X-Band 10-GHz
10-GHz breadboard
breadboard radar
radar system.
Figure
Figure 10. PhotographX-Band
of X-Band 10-GHz breadboardsystem.
radar system.

Figure 11. Photograph of field portable cart-mounted radar system.

Figure11.11.
Figure Photograph
Photograph of field
of field portable
portable cart-mounted
cart-mounted radar system.
radar system.
Signals 2023, 4, FOR PEER REVIEW 1
Signals 2023, 4, FOR PEER REVIEW 1

Signals 2023,
Signals 4 FOR PEER REVIEW
2023, 4, 348
12

Figure
Figure 12.
12. 30°
30° LOS
LOS antenna
antenna setup.
setup.
Figure
Figure12. 30◦LOS
12.30° LOSantenna
antennasetup.
setup.
5.
5. Experimental
Experimental Micro-Doppler
Micro-Doppler Signatures
Signatures
5.5.5.1.
ExperimentalMicro-Doppler
Experimental
Data
Micro-DopplerSignatures
Collection Process
Signatures
5.1. Data Collection
5.1. Data Collection ProcessProcess
5.1. Data Collection
Outdoor Process
data were collected by
Outdoor
Outdoor data
data werewere collected
collected by placing
placing
by placing
the
the drone
the drone drone 3–15 m
3–15 m3–15
away
awaymfrom
away from the
thefrom
radarthe
radar
radar system
system system
and Outdoor data
performing were collected
different by
flight placing
motions, the drone
as shown 3–15 m
in away
Figure from
13. the radar
There weresystem
a few difficul
andandperforming
performing different
different flightflight motions,
motions, as shown
as shown in Figurein13.
Figure
There13. There
were a fewwere a few difficu
difficulties
andtiesperforming
when different flight motions, as shown in Figure 13. There weredrones.
a few difficul-
when trying trying
to obtainto obtain
the the micro-Doppler
micro-Doppler signatures ofsignatures
the drones.of the
The RCS of theThe RCS of th
blades
ties when trying to obtain the micro-Doppler signatures of the drones.
ties when trying to obtain the micro-Doppler signatures of the drones. The RCS of theRCS of th The
blades
was was
too low;
blades was too low;
therefore, therefore,
methods formethods
increasingfor increasing
the the
detectability detectability
thewere conducted. were
were conducted.
blades was tootoo
low;low; therefore,
therefore, methodsmethods for increasing
for increasing detectability
the detectability were conducted. conducted.

Figure 13. Outdoor drone data collection setup.

Figure 13. Outdoor drone data collection setup.
Figure
Figure
The13.13. Outdoor
Outdoor drone
drone
micro-Doppler datadata collection
collection
signature wassetup.setup.by adding reflective tape onto the blade
enhanced
tips of the quadcopter rotors, as shown in Figure 14. This significantly improved its de-
The
The
The micro-Doppler
micro-Doppler
micro-Doppler signature
signature was was
signature was enhanced
enhanced by
by adding
by adding
enhanced reflective
reflective
adding tape
tape onto
reflective theonto
tape bladethe
onto the blad
blad
tectability.
tips
tips of the quadcopter rotors, as shown in Figure 14. This significantly improved its
tips ofof the
the quadcopter
quadcopter rotors,
rotors, as as shown
shown in in Figure
Figure 14. 14. This
This significantly
significantly improved
improved its its de
de
tectability.
detectability.
tectability.

Figure 14. Akaso UAV with reflective tape on the blade tip.

5.2. UAV Characteristics

Figure
Figure
The
Figure 14.
14. Akaso
Akaso UAV
UAV
description
14. Akaso of
UAV with
with
each
with reflective
reflective
UAV istape tape
on theon
provided
reflective tape oninthe
blade blade
tip.
Table
the blade5. tip.
tip.

5.2.
5.2. UAV
UAV Characteristics
Characteristics
The
The description of
description of each
each UAV
UAV is
is provided
provided in
in Table
Table 5.
5.
Signals 2023,
Signals 4 FOR PEER REVIEW
2023, 4, 349
13
Signals 2023, 4, FOR PEER REVIEW 13

5.2. UAV
Table Characteristics
5. UAV characteristics.
TableThe
5. UAV characteristics.
description of each UAV is provided in Table 5.
Akaso BlackHawk
Dimension Metabird
Akaso
Quadcopter BlackHawk
Quadcopter
Table 5.Dimension
UAV characteristics. Metabird
Weight 10 gm (0.022 lb) Quadcopter
750 gm (1.65 lb) Quadcopter
1775 gm (3.87 lb)
Weight
Body length
Dimension 10
16 gmcm (0.022
(6.3
Metabird in)lb) 750 gm
Akaso
– (1.65 lb) 1775BlackHawk
gm–(3.87 lb)
Quadcopter Quadcopter
Body length
Wingspan 16
33 cm (6.3
(13 in)in) –
38.9 cm (15.3 in) 77.7 cm –(30.6 in)
Weight 10 gm (0.022 lb) 750 gm (1.65 lb) 1775 gm (3.87 lb)
Wingspan
End-to-end span 33 cm–(13 in) 38.9
19.5 cmcm (15.3 in) 77.7 (30.6 in)
Body length 16 cm (6.3 in) – (7.7 in) 26.0 cm–(10.2
End-to-end
Wingspan span
Blade axis to blade axis 33 cm –
(13 in) 19.5
16.3
38.9 cm
cm (7.7
(6.4
(15.3 in)
in) 26.0
22.0 cm (10.2
77.7 cm
cm (8.7in)
(30.6 in)
in)
End-to-end
BladeBlade
axis to span axis
blade –– 19.5 cm
16.3 cm (7.7
(6.4in)in) 26.0 cm
22.0 cm (10.2
(8.7 in)
in)
length 7.0 cm (2.8 in) 9.5 cm (3.7 in)
Blade axis to blade axis – 16.3 cm (6.4 in) 22.0 cm (8.7 in)
Blade length
Height
Blade length –
– 7.0
10.5 cm
cm (2.8
(4.1
7.0 cm (2.8 in) in)
in) 9.5
9.5cm
12.0 cm (3.7
(4.7in)in)
cm (3.7 in)
Height
Height –– 10.5 cm(4.1
10.5 cm (4.1in)in) 12.0
12.0 cm (4.7in)in)
cm (4.7
Figure 15 shows the smallest UAV, a Metabird, and its spectrogram. The Metabird
mimicsFigure 15 shows
a flying
Figure 15 shows
bird andtheissmallest
the smallest
controlled UAV,
UAV,by aa smartphone.
Metabird, and
Metabird, and
It is itsequipped
its spectrogram.
spectrogram. The Metabird
with flexible
The Metabird
wings
mimics
mimics a flying bird and is controlled by a smartphone. It is equipped with flexible wings
made ofa flying
carbon bird and
fiber, is
liquid controlled
crystal by
polymers,a smartphone.
and an It is equipped
indestructible with
foam flexible
body. It flies
wingsby
made of
flapping carbon
its wingsfiber, liquid
without crystal
the aid polymers,
of a and
propeller. an indestructible
The micro-Doppler
made of carbon fiber, liquid crystal polymers, and an indestructible foam body. It flies foam body.
signature It flies
of by
the
flapping
Metabird
by flappingits wings
consists without
of periodic
its wings the
without flapsaid
the aid of
of of a propeller.
consistent The
velocities.
a propeller. micro-Doppler signature
The body hassignature
The micro-Doppler an undulating of the
of the
Metabird
nature as itconsists
flies. TheofDoppler
periodicbandwidth consistent
changes velocities.
flaps of consistent over time due Thetobody
the rotational behavior
has an undulating
nature
of as it flies.
the UAV. The
The Doppler
The MetabirdDopplerdoesbandwidth
not always
bandwidth changes over time due to the
travel perpendicular the rotational
rotational behavior
radar.
of the UAV. The The Metabird
Metabird does
does not
not always
always travel
travel perpendicular
perpendicular to to the
the radar.
radar.

Figure
Figure 15.
15. Metabird
Metabird (small
(small UAV)
UAV) and
and its
its associated
associated spectrogram.
spectrogram.
Figure 15. Metabird (small UAV) and its associated spectrogram.
Figure
Figure 16
16 shows
showsthethemedium-sized
medium-sizedAKASO AKASO drone
droneand
anditsits
associated spectrogram.
associated spectrogram.In
this Figure 16 shows
spectrogram, the
periodic medium-sized
shapes are AKASO
present that drone
are and
similarits
to associated
the spectrogram.
simulations.
In this spectrogram, periodic shapes are present that are similar to the simulations. This This In
spec-
this spectrogram,
trogram shows
spectrogram periodic
the the
shows behaviorshapes are present
of aofdrone
behavior slowly
a drone that
slowly are similar
descending
descending to the
ininthe simulations.
thevertical This
verticaldirection. spec-
direction. This
This
trogram
was shows
observed the behavior
through
through the of a drone
the maximum
maximum slowly
velocity
velocity of descending
ofthe
theblade
bladetips,inwhich
tips, the vertical
which direction.
wasaround
was around This
2626m/s.
m/s.
was observed through the maximum velocity of the blade tips, which was around 26 m/s.

Figure 16. Akaso drone (medium UAV) and its associated spectrogram.
Figure 16. Akaso drone (medium UAV) and its associated spectrogram.
Figure 16. Akaso drone (medium UAV) and its associated spectrogram.
 Figure 17 shows the large BlackHawk drone and its associated spectrogram. Th
Signals
Signals 2023,
2023, 4,
4 FOR PEER REVIEW
maximum blade tip velocity was much higher compared to the AKASO drone. 350 This
14 is du
to the overall weight of the UAV being higher than the AKASO. We observed that th
negative velocities showed more pronounced micro-Doppler peaks compared to the pos
itiveFigure
side. This
Figure 17 was because
17 shows
shows the largethe
the large positive peaks
BlackHawk
BlackHawk drone wereits
and
drone and itshidden behind
associated
associated the arms of
spectrogram.
spectrogram. the drone
The
The
concealing
maximum the
blade sharp
tip micro-Doppler
velocity was much pattern.
higher compared to the AKASO drone.
maximum blade tip velocity was much higher compared to the AKASO drone. This is This is due
to
duethetooverall weight
the overall of theofUAV
weight beingbeing
the UAV higher thanthan
higher the AKASO. We observed
the AKASO. We observedthat that
the
negative velocities
the negative showed
velocities moremore
showed pronounced micro-Doppler
pronounced peakspeaks
micro-Doppler compared to thetopos-
compared the
itive side.side.
positive This This
was because the positive
was because peaks peaks
the positive were hidden behindbehind
were hidden the arms
theofarms
the drone,
of the
concealing the sharp
drone, concealing themicro-Doppler pattern.pattern.
sharp micro-Doppler

Figure 17. BlackHawk drone (large UAV) and its associated spectrogram.

Figure 17. BlackHawk

5.3. Bird drone (large UAV) and its associated spectrogram.
Characteristics
Figure 17. BlackHawk drone (large UAV) and its associated spectrogram.
In Characteristics
5.3. Bird this section, the setup for collecting data on birds is explained. Bird flights wer
5.3. Bird
collectedCharacteristics
In thiswith a 10-GHz
section, radar
the setup at Shaver’s
for collecting Creek,
data a bird
on birds sanctuaryBird
is explained. nearflights
Penn were
State Univer
sity.InMost
collected this
withofathe
section, birds
10-GHz in this
the radar
setup for
at aviary
collecting
Shaver’s are disabled,
data
Creek, on birds
a bird butisthis
sanctuary did Penn
explained.
near notBird
significantly
flights
State wereaffect thei
University.
collected
Most with
flightofpatterns.
the a 10-GHz
birds in this radar
Through aviarytheatare
Shaver’s
disabled,
help Creek,
of the but a bird
this
trainers did sanctuary
at not near
significantly
Shaver’s Penn
Creek, State
affect
birds Univer-
their
flewflight
from perch
patterns.
sity. MostThrough
to perch, ofas
thethe the
birds help
in
radar thisofaviary
the trainers
system at Shaver’s
are disabled,
recorded thebut Creek,
this did
induced birds
not flew from
significantly
Doppler perch
shifts. toenclosure
affect
The perch,
their wa
as
wooden with wire meshing in between columns. Since the radar system onlywith
the
flight radar
patterns.system
Throughrecorded
the the
help induced
of the Doppler
trainers at shifts.
Shaver’s The enclosure
Creek, birdswas
flew wooden
from perch
tracks radia
wire
to meshing
perch, as the in between columns. Sincethetheinduced
radar system only tracks radial movement,
movement, theradar system
enclosure recorded
did not play a role in Doppler shifts.
the spectrogram The enclosure
quality. Thewas
reflection
the enclosure did not play a role in the spectrogram quality.
wooden with wire meshing in between columns. Since the radar system only tracks The reflections from the wall
radial
from the
behind the bird
wall behind
did not affect
the bird did not affect its micro-Doppler signature. However, thi
movement, the enclosure did itsnotmicro-Doppler
play a role in signature. However,
the spectrogram this was
quality. Thenot verified
reflections
was
by
from
not flight
a bird verified
the wall
by the
outside
behind
aofbird
the flight
bird enclosure.outside of the enclosure.
did not affect its micro-Doppler signature. However, this
was not Figure
Figure 18
18 shows
verified shows the
the bird
by a bird bird enclosure
flightenclosure
outside of and and
theradar radar
system.
enclosure. system. Four different
Four different birds
birds were were used
used
inthis
in this
Figureexperiment:
experiment:
18 shows a bald
a bald
the eagle,
bird eagle,
golden
enclosure golden
eagle,
and eagle, broad-winged
broad-winged
radar system. hawk,
Four hawk, andwere
and American
different birds American
used kestre
kestrel.
These
These areareshown
shown in Figure
in Figure 19 (obtained
19 (obtained from https://www.allaboutbirds.org/guide/,
from https://www.allaboutbirds.org/guide/,
in this experiment: a bald eagle, golden eagle, broad-winged hawk, and American kestrel. ac
accessed
cessed
These areonon 11 March
11 March
shown 2022).
2022).
in Figure 19 (obtained from https://www.allaboutbirds.org/guide/, ac-
cessed on 11 March 2022).

Figure 18. Bird data collection setup.

Figure18.
Figure 18.Bird
Bird data
data collection
collection setup.
setup.
 Signals
Signals 4, 4FOR PEER REVIEW
2023,
2023, 15351

(a) (b)

(c) (d)
Figure
Figure19.19.
Bird types.
Bird (a) (a)
types. Small BirdBird
Small 1: American kestrel;
1: American (b) Small
kestrel; Bird Bird
(b) Small 2: Broad-winged hawk;hawk;
2: Broad-winged (c)
Big Bird 1: Golden eagle; (d) Big Bird 2: Bald eagle.
(c) Big Bird 1: Golden eagle; (d) Big Bird 2: Bald eagle.

The
Thephysical
physicalcharacteristics ofof
characteristics each bird
each are
bird areprovided
providedininTable
Table6.6.

Table 6. 6.
Table Bird characteristics.
Bird characteristics.
American
American Broad-Winged
Broad-Winged Golden Eagle
Bird
BirdFeature
Feature Golden Eagle Bald Eagle
Bald Eagle
Kestrel
Kestrel Hawk
Hawk
68–163 gmgm
68–163 (0.15– 340–454
340–454gmgm 2.95–6.12
2.95–6.12kg kg 3.00–6.30 kg (6.6–
3.00–6.30 kg
Weight
Weight
0.36 lb)lb)
(0.15–0.36 (0.75–1
(0.75–1lb)
lb) (6.5–13.5
(6.5–13.5lb)
lb) 13.9 lb) lb)
(6.6–13.9
21.0–31.8 cm
21.0–31.8 cm 34.3–44.5 cm
34.3–44.5 cm 69.9–83.8cm
69.9–83.8 cm 71.1–96.5
71.1–96.5 cm
cm (28–
Length
Length (8.25–12.5 in) (13.5–17.5 in) (27.5–33 in) (28–38 in)
(8.25–12.5 in) (13.5–17.5 in) (27.5–33 in) 38 in)
50.8–61.0 cm 81.3–101.6 cm 1.83–2.13 m 1.83–2.29 m
Wingspan 50.8–61.0 cm 81.3–101.6 cm 1.83–2.13 m (6–7 1.83–2.29 mft)(6–
Wingspan (20–24 in) (32–40 in) (6–7 ft) (6–7.5
(20–24 in) (32–40 in) ft) 7.5 ft)

Figure 20 shows the bald eagle spectrogram. Within this spectrogram, a large velocity
Figure 20 shows the bald eagle spectrogram. Within this spectrogram, a large velocity
bandwidth is initially observable at around 300 ms. This represents the downstroke of the
bandwidth is initially observable at around 300 ms. This represents the downstroke of the
bird, quickly followed by an upward flap. The back half of the spectrogram shows the
bird, quickly followed by an upward flap. The back half of the spectrogram shows the
soaring of the bird, as it mechanically controls the rest of the flight.
soaring of the bird, as it mechanically controls the rest of the flight.
Figure 21 shows the golden eagle spectrogram. The golden eagle spectrogram is quite
Figure 21 shows the golden eagle spectrogram. The golden eagle spectrogram is quite
similar to the bald eagle, with strong flaps present at 200 and 500 ms. Unlike the UAVs, the
similar to the bald eagle, with strong flaps present at 200 and 500 ms. Unlike the UAVs,
flapping is not consistently periodic because flapping requires large amounts of energy. If
the flapping is not consistently periodic because flapping requires large amounts of en-
the birds had a larger environment to fly in, the micro-Doppler signatures may look quite
ergy. If the birds had a larger environment to fly in, the micro-Doppler signatures may
different because soaring would be more viable.
look quite different because soaring would be more viable.
Figure 22 shows the spectrogram for the broad-winged hawk. The broad-winged hawk
Figure
is the 22 smallest
second shows the spectrogram
bird. forrequire
Smaller birds the broad-winged hawk. The
a higher periodicity broad-winged
in flaps compared to
hawk is the second smallest bird. Smaller birds require a higher periodicity
eagles, but their flaps do not need to be as strong because they are supporting in flaps com-
a smaller
pared to eagles,
payload. The but their flaps dosignature
micro-Doppler not needistoquite
be ascluttered
strong because they
because theare supporting
body a
of the bird
smaller payload. The micro-Doppler signature is quite cluttered because the body
undulates. This makes the peaks of the wing flaps less pronounced. The maximum velocity of the
bird undulates.
achieved This
by the makes the peaks
broad-winged hawk of thearound
was wing flaps less pronounced. The maximum
5 m/s.
velocity achieved by the broad-winged hawk was around 5 m/s.
 Signals 2023, 4 352

Signals
Signals 2023,
2023, 4,
4, FOR
FOR PEER
PEER REVIEW
REVIEW Figure 23 is the spectrogram of an American kestrel. Similar to the broad-winged 11
Signals 2023, 4, FOR PEER REVIEW hawk, the total Doppler bandwidth was quite small, but the flapping periodicity was higher 16
than the big birds.

Figure
Figure 20.
Figure20.
20.Spectrogram
Spectrogram ofof
Spectrogram ofthe
the thebald
bald eagle.
eagle.
bald eagle.
Figure 20. Spectrogram of the bald eagle.

Figure 21. Spectrogram of the golden eagle.

Figure21.21.
Figure
Figure 21. Spectrogram
Spectrogram
Spectrogram of golden
of
of the the golden
the golden eagle.
eagle.eagle.

Figure 22. Spectrogram of the broad-winded hawk.

Figure
Figure
Figure 22.22.
22.
Figure Spectrogram
Spectrogram
Spectrogram of the
the broad-winded
of broad-winded
the broad-winded
23 is theof spectrogram hawk.
hawk.
of hawk.
an American kestrel. Similar to the broad-winged
hawk, the total Doppler bandwidth was quite small, but the flapping periodicity was
Figure
higher 23 big
than the is the spectrogram of an American kestrel. Similar to the broad-winge
birds.
hawk, the total Doppler bandwidth was quite small, but the flapping periodicity wa
higher than the big birds.
Signals
Signals2023,
2023,4,4 FOR PEER REVIEW 353 1

Signals 2023, 4, FOR PEER REVIEW 17

Figure
Figure
Figure 23.23.
23. Spectrogram
Spectrogram
Spectrogram ofAmerican
ofofthe
the the American
American kestrel.
kestrel.
kestrel.

While
While
While Figures
Figures
Figures 20–23
20–23 depict
depict
20–23 thespectrograms
the
depict spectrograms
the spectrograms ofofeach
eachof bird
bird
each flying
flying
bird almost
almost horizontally
flyinghorizontally
almost horizontall
from
from one end of its enclosure to the other, Figure 24 shows a
from one end of its enclosure to the other, Figure 24 shows a comparison abetween
one end of its enclosure to the other, Figure 24 shows a comparison
comparison between
between abald
bald a bal
eagleflying
eagle flyingdown
downvs. vs. flyingup.up. Both flight types (flying down and flying up) were
eagle flying down flyingvs. flying Both
up. flight types
Both flight (flying down
types (flying and
downflying and up)flying
were col- were co
up)
collected
lected independently
independently andand separately,
separately, as as
thethe bird
bird flew
flew from
from oneone endendofofitsitsenclosure
enclosuretotothe
the
lected
other.The
independently
Themicro-Doppler
and
micro-Dopplerbandwidth
separately,
bandwidthisislarger
as the
largerwhen
bird
whenflying
flew
flyingup,
from
up,which
one
whichmakes
end sense
makessense
of its because
enclosure to th
other. because
other.
the wingsThe micro-Doppler
need bandwidth is larger
thrustwhen flying up, which makes sense becaus
the wings need totoflap
flapharder
hardertoto generatemore
generate more thrust (similar
(similar totothe
the quadcopter).
quadcopter). Through
Through
thedata
bird
bird wings
data need tothe
collection,
collection, flap harderchallenge
thebiggest
biggest to generate
challenge was
was more
getting
getting thrust (similar
thebirds
the birdstotofly.to the
fly. OnOnquadcopter).
hotdays,
hot the Throug
days,the
bird
birds data
did notcollection,
want to the biggest
cooperate. Thischallenge
resulted was
in large getting
birds did not want to cooperate. This resulted in large recordings (5–10 min) for only aadays, th
the
recordings birds
(5–10 to fly.
min) On
for hot
only
birdsofof
couple
couple did not want
recorded
recorded to cooperate. This resulted in large recordings (5–10 min) for only
flights.
flights.
couple of recorded flights.

(a) (b)
Figure
Figure24.
24.Bald
Baldeagle
eaglespectrograms.
spectrograms.(a)
(a)Flying
Flyingdown;
down;(b)
(b)Flying
Flyingup.
up.
(a) (b)
6.6.Target
TargetClassification
Classification
Figure 24. Bald eagle spectrograms. (a) Flying down; (b) Flying up.
Forclassification,
For classification, the
the main
main features
featuresof ofinterest
interestarearethe
thetotal Doppler
total Doppler bandwidth
bandwidth and the
and
micro-Doppler
the6.micro-Doppler periodicity. With drones, the micro-Doppler bandwidth is much
periodicity. With drones, the micro-Doppler bandwidth is much larger larger than
the Target
than birds Classification
due
the birds to
duethetohigh RPM RPM
the high of theofspinning blades.
the spinning In terms
blades. of periodicity,
In terms the drones
of periodicity, the
have aFor classification,
constant the
micro-Doppler main features
pattern becauseof interest
it is are
necessary the
drones have a constant micro-Doppler pattern because it is necessary to stay into total
stay in Doppler
flight. bandwidth
However,
flight. an
birds
However, are not constantly
the micro-Doppler
birds flapping their
are notperiodicity.
constantly Withwings.
flapping drones, the micro-Doppler bandwidth is much large
their wings.
than the birds due to the high RPM of the spinning blades. In terms of periodicity, th
6.1. Spectrogram
drones have Generation
a constantandmicro-Doppler
Curation pattern because it is necessary to stay in fligh
However, birds are anot
For classification, constantly
large flapping
dataset was their
essential forwings.
providing confident results. To
achieve this, many flights of each drone and bird were recorded. For the UAVs, the flights
6.1.snipped
were Spectrogram Generation
into equal and The
segments. Curation
radar-reflected data from the targets (UAVs and
birds) were processed to generate spectrograms. Only the spectrograms that contained the
For classification, a large dataset was essential for providing confident results. T
micro-Doppler signatures were retained for further analysis.
achieve this, many flights of each drone and bird were recorded. For the UAVs, the flight
were snipped into equal segments. The radar-reflected data from the targets (UAVs an
birds) were processed to generate spectrograms. Only the spectrograms that contained th
micro-Doppler signatures were retained for further analysis.
Signals 2023, 4 354

6.1. Spectrogram Generation and Curation

For classification, a large dataset was essential for providing confident results. To
Signals 2023, 4, FOR PEER REVIEW achieve this, many flights of each drone and bird were recorded. For the UAVs, the flights
18
were snipped into equal segments. The radar-reflected data from the targets (UAVs and
birds) were processed to generate spectrograms. Only the spectrograms that contained the
micro-Doppler
A MATLABsignatures
script waswere retained
used to readfor further
every WAVanalysis.
file within a folder and generate
spectrograms. The titles and axis labels were removed file
A MATLAB script was used to read every WAV for within a folder andsaved
each spectrogram, generate
as
spectrograms. The titles and axis labels were removed for each spectrogram, saved as PNG
PNG files, and stored in the database.
files, and stored in the database.
Before the images were classified within the TensorFlow environment, each image
Before the images were classified within the TensorFlow environment, each image
was set to 128 × 128 with a batch size of 32 and converted to grayscale.
was set to 128 × 128 with a batch size of 32 and converted to grayscale.
6.2.
6.2.Support
SupportVector
VectorMachine
Machine
For
For classificationefforts,
classification a Support
efforts, Vector
a Support Machine
Vector Machine(SVM) was utilized
(SVM) [30]. The
was utilized [30].SVM
The
algorithm
SVM algorithm produces relatively high accuracy with low computational power. is
produces relatively high accuracy with low computational power. SVM SVMtyp-is
ically usedused
typically in binary classification
in binary classificationproblems
problemsby by
separating
separatingtwotwoclasses
classeswith
withananoptimal
optimal
hyperplane
hyperplane or decision boundary. This implies maximizing the distance betweendata
or decision boundary. This implies maximizing the distance between data
points
pointsfrom
fromthetheseparate
separateclasses.
classes.By Bymaximizing
maximizingthe thedistance
distancebetween
betweendata
datapoints,
points,images
images
can
canbebeclassified
classifiedwith
withhigher
higherconfidence.
confidence.
For
For classification efforts,the
classification efforts, themain
mainfeatures
featuresofofconcern
concernarearethe
themicro-Doppler
micro-Dopplerband-band-
width
widthandandthethemicro-Doppler
micro-Dopplerperiodicity.
periodicity. The
Themicro-Doppler
micro-Dopplerbandwidth
bandwidthdescribes
describesthe the
maximum
maximumvelocity
velocity achieved
achieved by by the
themicro-motions,
micro-motions,whereas
whereasthethe periodicity
periodicity provides
provides in-
insight
sight into how
into how oftenoften the motion
the motion occurs.occurs.

6.3.UAV
6.3. UAVClassification
Classification
Thefirst
The firstclassification
classificationeffort
effort was
was conducted
conducted to to distinguish
distinguish between
between thethe three
three types
types of
of UAVs:
UAVs: small small (Metabird),
(Metabird), medium
medium or mid
or mid (Akaso),
(Akaso), and(BlackHawk).
and big big (BlackHawk). The dataset
The dataset was
was into
split split80%
intofor
80% for training
training and for
and 20% 20%testing.
for testing. For testing,
For testing, 86 UAV 86 UAV
imagesimages
werewere
used.used.
10-
Fold cross-validation was utilized to test the performance of multiple classifiers. EightEight
10-Fold cross-validation was utilized to test the performance of multiple classifiers. dif-
different
ferent classifiers
classifiers were
were tested,
tested, which
which were:Logistic
were: LogisticRegression,
Regression,K-Nearest
K-NearestNeighbor
Neighbor
(KNN),Support
(KNN), SupportVectorVectorMachine
Machine(SVM),
(SVM),Linear
LinearSVM,SVM,Multi-layer
Multi-layerPerceptron
PerceptronClassifier,
Classifier,
Random Forest
Random Forest Classifier,
Classifier,Gaussian
Gaussian Naïve
NaïveBayes Classifier,
Bayes and Adaboost
Classifier, Classifier.
and Adaboost Among
Classifier.
the above eight candidates, the top three performers were Logistic Regression,
Among the above eight candidates, the top three performers were Logistic Regression, SVM, and
Random Forest. Overall, each class was distinguished with a high confidence
SVM, and Random Forest. Overall, each class was distinguished with a high confidence above 90%.
In Figure
above 90%.25,In the confusion
Figure 25, thematrices
confusion display
matricesthedisplay
results of
thethe classifiers.
results of the Out of the three
classifiers. Out
top performers, the SVM classifier performed the best.
of the three top performers, the SVM classifier performed the best.

(a) (b)

Figure 25. Cont.

Signals 2023, 4, FOR PEER REVIEW 19
Signals 4
2023,2023,
Signals 4, FOR PEER REVIEW 355
19

(c)
(c)
Figure 25. 25.
Figure Confusion
Confusionmatrix results
matrix ofofclassifier
results comparison for UAV classification. (a)Logistic
LogisticRe-
Re-
Figure 25. Confusion matrix results of classifier
classifiercomparison
comparisonforfor
UAV classification.
UAV (a)
classification. (a) Logistic
gression; (b) SVM; (c) Random Forest.
gression; (b) SVM; (c) Random Forest.
Regression; (b) SVM; (c) Random Forest.

6.4.6.4.
6.4. Binary
Binary
Binary (UAV(UAV
(UAV vs.vs.
vs. Bird)
Bird)
Bird) Classification
Classification
Classification
Subsequently,
Subsequently,
Subsequently, birdbird
bird micro-Dopplerimages
micro-Doppler
micro-Doppler imageswere
images wereincluded
were included in
included inthe
theclassification
the classificationefforts.
classification efforts.
efforts.
Initially,
Initially, a
Initially, aa binary binary classification
binary classification
classification task task
task for for distinguishing
for distinguishing
distinguishing UAVs UAVs
UAVs andand birds
and birds was
birds was conducted.
was conducted.
conducted.
TheThe
The sklearn.svm
sklearn.svm
sklearn.svm toolkit
toolkit
toolkit was
was
was utilizedwith
utilized
utilized withthe
with the linear
thelinear SVC
linearSVC classifier
SVCclassifier
classifier with
with
withaamaximum
amaximum
maximum itera-
itera-
iter-
tiontion of
ationofof1000. 1000.
1000.The The linear
Thelinear
linearSVC SVC
SVCusesuses
usesthe the l2 penalty
thel2l2penalty
penaltyand and a squared-hinge
andaa squared-hinge
squared-hinge loss loss function
functionwith
loss function with
with
0.0001
0.0001 tolerance.
tolerance.
0.0001 tolerance.
With With a 2/3 training and 1/3testing
testing split,152 152 drone images
images and 86 bird images were
With aa 2/3 2/3training
trainingand and1/3
1/3 testingsplit,split, 152drone
drone imagesand and86 86bird
bird images
images were
were
used for testing. The confusion matrix in Figure 26 shows high accuracy in classifying
used
used for
for testing.
testing. The The confusion
confusion matrix
matrix in in Figure
Figure 2626 shows
shows high
high accuracy
accuracy in in classifying
classifying
between drones and birds with 96% accuracy. The precision for drones is 98% and for
between
between dronesdrones and and birds
birds with
with 96%
96% accuracy.
accuracy. The The precision
precision forfor drones
drones isis 98%
98% and
and for
for
birds is 92%.
birds
birds is
is 92%.
92%.

Figure 26. Drones vs. birds confusion matrix.

Figure 26.
Figure Drones vs.
26. Drones vs. birds
birdsconfusion
confusion matrix.
matrix.
6.5. 5-Class Classification
6.5. 5-Class Classification
For Classification
multiclass classification, the SVM classifier utilizes a one-versus-rest scheme. This
6.5. 5-Class
For multiclass
strategy classification,
splits a five-class datasetthe SVM
into classifier
multiple utilizes a one-versus-rest
binary-classification problems.scheme. This
In this clas-
For multiclass classification, the SVM classifier utilizes a one-versus-rest scheme. This
strategy splits
sification a five-class
problem, the threedataset intoUAVs
separate multiple binary-classification
are considered problems.
in addition to In this
the two catego-
strategy splits a five-class dataset into multiple binary-classification problems. In this clas-
ries of birds,problem,
classification namely, big theand small
three bird categories.
separate UAVs areThe big birds consist
considered of theto
in addition bald
theand
two
sification problem, the three separate UAVs are considered in addition to the two catego-
golden eagles,
categories while
of birds, the small
namely, birdssmall
big and pertain to categories.
bird the broad-winged
The bighawk
birdsand American
consist of thekes-
bald
riestrel.
of birds,
Figure namely,
27 shows big and
thethe small bird
confusion categories.
matrix for the The big birds consist of the bald and
and golden eagles, while small birds pertain to 5the
classes.
broad-winged hawk and American
golden eagles, while the small birds pertain to the broad-winged hawk and American kes-
kestrel. Figure 27 shows the confusion matrix for the 5 classes.
trel. Figure 27 shows the confusion matrix for the 5 classes.
It is noted that the quadcopters and small birds are classified with strong confidence,
while the Metabird and the big birds have a harder time. For the big birds, the confusion is
across the board, but is mostly mistaken for a small bird. The inaccuracy within the small
UAV class, the Metabird, may be due to the smaller number of images available for training
and testing. Most of the mispredictions of the Metabird were from the other drones.
Signals 2023,44, FOR PEER REVIEW
Signals2023, 356
20

Figure 27.
Figure 27. 5-class
5-class confusion
confusion matrix.
matrix. In the class description labels, “B_”,
“B_,” “M_”,
“M_,” and “S_” refer to big,
medium, and small, respectively.
medium, and small, respectively.

7. Conclusions
It is noted that the quadcopters and small birds are classified with strong confidence,
while Inthe
thisMetabird
paper, we and the bigmicro-Doppler
analyzed birds have a harder time. Forby
data collected thea big birds, the confusion
custom-designed radar
is across
system the board,
utilized but is mostly
in different scenarios mistaken
with a for a small
variety bird. The
of targets, suchinaccuracy
as UAVs within
and birds.the
Asmall UAVwas
dataset class, the Metabird,
created based onmay the be due to the smaller
micro-Doppler images number of images
obtained, available for
and classification
results
trainingwere andachieved with high
testing. Most of theconfidence. Different
mispredictions of thecharacteristics
Metabird were of target
from detection
the other
were
drones.explored, including the landscape and the behavior of the target. Using a custom-built
10-GHz CW radar, over 700 images were collected and used for target distinction. For
classification
7. Conclusions between UAV sizes, an accuracy of 90% was observed, whereas for the binary
classification of two
In this paper, weclasses,
analyzed namely birds and data
micro-Doppler UAVs, an accuracy
collected of 96% was obtained.
by a custom-designed radar
More specific classification among the five classes showed
system utilized in different scenarios with a variety of targets, such as UAVs an accuracy of 85%.and birds. A
Some
dataset was limitations of ouron
created based work
the include the fact that
micro-Doppler imagesthe obtained,
bird data were obtained under
and classification re-
non-natural conditions because the birds were caged and could
sults were achieved with high confidence. Different characteristics of target detection not be observed inwere
full
flight.
explored,In addition,
includingthe thedata were limited
landscape and thetobehavior
the observation angles
of the target. that awere
Using available10-
custom-built at
the test site. Furthermore, drone data were acquired under limited
GHz CW radar, over 700 images were collected and used for target distinction. For classi- conditions of flying
towards and away from the radar, but not performing maneuvers. Future work should
fication between UAV sizes, an accuracy of 90% was observed, whereas for the binary
address some of the limitations stated above.
classification of two classes, namely birds and UAVs, an accuracy of 96% was obtained.
Future work in this field would also include methods for automating a method for
More specific classification among the five classes showed an accuracy of 85%.
detecting targets with a low-powered SDR, then notifying and turning on the 10-GHz radar
Some limitations of our work include the fact that the bird data were obtained under
to collect the micro-Doppler signature. This would allow for more efficient automatic target
non-natural conditions because the birds were caged and could not be observed in full
recognition.
flight. In addition, the data were limited to the observation angles that were available at
Another avenue for future work would involve the distinction of UAV motion as the
the test site. Furthermore, drone data were acquired under limited conditions of flying
target rolls and pitches in different directions. Due to the cluttered motions in the varying
towards and away from the radar, but not performing maneuvers. Future work should
velocities, it would be of interest to design an algorithm for separating these rotor velocities
address some of the limitations stated above.
and characterizing the overall target motion. Detecting micro-Doppler signatures in urban
Future areas
or cluttered workwould
in thisalso fieldbewould also include methods for automating a method for
beneficial.
detecting targets with a low-powered
Using micro-Doppler images, classification SDR, thenefforts
notifying and turning
regarding on the 10-GHz
UAV motions and targetra-
dar to collect the micro-Doppler
recognition can be confidently characterized. signature. This would allow for more efficient automatic
target recognition.
Author Another avenueConceptualization,
Contributions: for future work would R.M.N.involve
and R.B.;the distinction B.T.;
methodology, of UAV motion
software, asvali-
B.T.; the
target rolls
dation, R.M.N.,andB.T.
pitches in different
and R.B.; directions.
formal analysis, DueB.T.
R.M.N., to the
andcluttered motions in
R.B.; investigation, theresources,
B.T.; varying
velocities,
R.M.N. and it would
R.B.; data be of interest
curation, to design an algorithm
B.T.; writing—original for separating
draft preparation, these
R.M.N.; rotor veloci-
writing—review
ties editing,
and and characterizing
R.M.N., B.T. and theR.B.;overall target motion.
visualization, B.T. andDetecting micro-Doppler
R.B.; supervision, signatures
R.M.N.; project admin- in
istration,
urban orR.M.N.
clutteredandareas
R.B.; would
fundingalso acquisition, R.M.N. All authors have read and agreed to the
be beneficial.
published
Using version of the manuscript.
micro-Doppler images, classification efforts regarding UAV motions and tar-
get recognition
Funding: can bewas
This research confidently
funded bycharacterized.
U.S. Office of Naval Research, grant number N00014-20-1-
2064. We appreciate the assistance provided by the following individuals at Shaver’s Creek: Abigail
Author Contributions:
Flanders, Conceptualization,
Joseph Whitehead, R.M.N.
Shannon Davidow, Jonand R.B.; methodology,
Kauffman, Jack Meyer,B.T.;
and software, B.T.; vali-
Joshua Potter.
dation, R.M.N., B.T. and R.B.; formal analysis, R.M.N., B.T. and R.B.; investigation, B.T.; resources,
R.M.N. and R.B.; data curation, B.T.; writing—original draft preparation, R.M.N.; writing—review
Signals 2023, 4 357

Data Availability Statement: The data are not publicly available due to sponsor restrictions.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or
in the decision to publish the results.

References
1. O’Malley, J. The no drone zone. Eng. Technol. 2019, 14, 34–38. [CrossRef]
2. Federal Aviation Administration. How to Register Your Drone. Available online: https://www.faa.gov/uas/getting_started/
register_drone (accessed on 20 March 2023).
3. Pham, T.; Srour, N. TTCP AG-6: Acoustic detection and tracking of UAVs. In Proceedings of the SPIE Conference on Unat-
tended/Unmanned Ground, Ocean, and Air Sensor Technologies and Applications VI, Orlando, FL, USA, 12–16 April 2004;
pp. 24–30. [CrossRef]
4. Bernardini, A.; Mangiatordi, F.; Pallotti, E.; Capodiferro, L. Drone detection by acoustic signature identification. In Proceedings of
the 2017 IS&T International Symposium on Electronic Imaging (EI 2017), Burlingame, CA, USA, 29 January–2 February 2017;
pp. 60–64. [CrossRef]
5. Liu, H.; Qu, F.; Liu, Y.; Zhao, W.; Chen, Y. A drone detection with aircraft classification based on a camera array. IOP Conf. Ser.
Mater. Sci. Eng. 2018, 322, 052005. [CrossRef]
6. Omkar, S.N.; Asogekar, N.; Rathi, S. Detection, tracking and classification of rogue drones using computer vision. Int. J. Eng.
Appl. Sci. Technol. 2022, 7, 11–19. [CrossRef]
7. Hammer, M.; Hebel, M.; Borgmann, B.; Laurenzis, M.; Arens, M. Potential of lidar sensors for the detection of UAVs. In
Proceedings of the SPIE Conference on Laser Radar Technology and Applications XXIII, Orlando, FL, USA, 15–19 April 2018;
pp. 1063605-1–1063605-7. [CrossRef]
8. Salhi, M.; Boudriga, N. Multi-array spherical lidar system for drone detection. In Proceedings of the 22nd International Conference
on Transparent Optical Networks (ICTON), Bari, Italy, 19–23 July 2020; pp. 1–5. [CrossRef]
9. de Wit, J.J.M.; Harmanny, R.I.A.; Prémel-Cabic, G. Micro-Doppler analysis of small UAVs. In Proceedings of the 9th European
Radar Conference, Amsterdam, The Netherlands, 31 October–2 November 2012; pp. 210–213.
10. Rahman, S.; Robertson, D.A. Radar micro-Doppler signatures of drones and birds at K-band and W-band. Sci. Rep. 2018, 8, 17396.
[CrossRef] [PubMed]
11. Zhang, Y.D.; Xiang, X.; Li, Y.; Chen, G. Enhanced micro-Doppler feature analysis for drone detection. In Proceedings of the 2021
IEEE Radar Conference (RadarConf21), Online, 7–14 May 2021; pp. 1–4. [CrossRef]
12. Molchanov, P.; Egiazarian, K.; Astola, J.; Harmanny, R.I.A.; de Wit, J.J.M. Classification of small UAVs and birds by micro-Doppler
signatures. In Proceedings of the 10th European Radar Conference, Nuremberg, Germany, 9–11 October 2013; pp. 172–175.
13. Harmanny, R.I.A.; de Wit, J.J.M.; Cabic, G.P. Radar micro-Doppler feature extraction using the spectrogram and the cepstrogram.
In Proceedings of the 11th European Radar Conference, Rome, Italy, 8–10 October 2014; pp. 165–168. [CrossRef]
14. Markow, J.; Balleri, A. Examination of drone micro-Doppler and JEM/HERM signatures. In Proceedings of the 2020 IEEE Radar
Conference (RadarConf20), Online, 21–25 September 2020; pp. 1–6. [CrossRef]
15. Gérard, J.; Tomasik, J.; Morisseau, C.; Rimmel, A.; Vieillard, G. Micro-Doppler signal representation for drone classification
by deep learning. In Proceedings of the 28th European Signal Processing Conference (EUSIPCO), Online, 18–21 January 2021;
pp. 1561–1565. [CrossRef]
16. Doppler, C. Über das farbige Licht der Doppelsterne und einiger anderer Gestirne des Himmels. Abh. Königlich Böhmischen Ges.
Wiss. 1843, 5, 465–482.
17. Chen, V.C. The Micro-Doppler Effect in Radar, 2nd ed.; Artech House: Norwood, MA, USA, 2019; p. 19.
18. Wu, Q.; Zhao, J.; Zhang, Y.; Huang, Y. Radar micro-Doppler signatures model simulation and feature extraction of three typical
LSS targets. In Proceedings of the 6th International Conference on Information Science and Control Engineering (ICISCE),
Shanghai, China, 20–22 December 2019; pp. 1103–1108. [CrossRef]
19. Zaugg, S.; Saporta, G.; Van Loon, E.; Schmaljohann, H.; Liechti, F. Automatic identification of bird targets with radar via patterns
produced by wing flapping. J. R. Soc. Interface 2008, 5, 1041–1053. [CrossRef]
20. Gong, J.; Li, D.; Yan, J.; Hu, H.; Kong, D. Measurement and analysis of radar signals modulated by flapping wings of birds. IEEE
Access 2022, 10, 85137–85145. [CrossRef]
21. Fang, X.; Xiao, G. Rotor blades micro-Doppler feature analysis and extraction of small unmanned rotorcraft. IEEE Sens. J. 2020,
21, 3592–3601. [CrossRef]
22. Kolev, N.; Sivkov, Y.; Sirakov, E. K band radar drone signatures. Inf. Secur. Int. J. 2020, 47, 349–354. [CrossRef]
23. Passafiume, M.; Rojhani, N.; Collodi, G.; Cidronali, A. Modeling small UAV micro-Doppler signature using millimeter-wave
FMCW radar. Electronics 2021, 10, 747. [CrossRef]
24. Petrović, P.Z.; Savić, S.V.; Ilić, M.M. Electromagnetic modelling of micro-Doppler Signatures of commercial airborne drones. In
Proceedings of the 29th Telecommunications Forum (TELFOR), Online, 23–24 November 2021; pp. 1–4. [CrossRef]
25. Torvik, B.; Gusland, D.; Olsen, K.E. Signatures of small drones and birds as emerging targets. In Micro-Doppler Radar and Its
Applications; Fioranelli, F., Griffiths, H., Ritchie, M., Balleri, A., Eds.; SciTech Publishing: Stevenage, UK, 2020; pp. 257–290.
Signals 2023, 4 358

26. Chen, X.; Chen, W.; Rao, Y.; Huang, Y.; Guan, J.; Dong, Y. Progress and prospects of radar target detection and recognition
technology for flying birds and unmanned aerial vehicles. J. Radars 2020, 9, 803–827. [CrossRef]
27. Liu, J.; Xu, Q.Y.; Chen, W.S. Classification of bird and drone targets based on motion characteristics and random forest model
using surveillance radar data. IEEE Access 2021, 9, 160135–160144. [CrossRef]
28. Palamà, R.; Fioranelli, F.; Ritchie, M.; Inggs, M.; Lewis, S.; Griffiths, H. Measurements and discrimination of drones and birds
with a multi-frequency multistatic radar system. IET Radar Sonar Navig. 2021, 15, 841–852. [CrossRef]
29. Musa, S.A.; Abubakar, S.A.; Girei, S.H.; Habush, A.H.H. A preliminary investigation of copter drone detection by using forward
scattering radar. J. Aerosp. Eng. Mech. 2022, 6, 588–595. [CrossRef]
30. Cortes, C.; Vapnik, V. Support-vector networks. Mach. Learn. 1995, 20, 273–297. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.