See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/364324778

APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY

PURPOSES

Conference Paper · October 2022

CITATIONS READS

3 2,184

2 authors:

Aleksandar Petrovski Marko Radovanović

University Goce Delcev" Stip Military academy, Skopje, North Macedonia Vojna akademija Beograd
30 PUBLICATIONS 65 CITATIONS 57 PUBLICATIONS 210 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Marko Radovanović on 28 June 2023.

The user has requested enhancement of the downloaded file.

 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR
MILITARY PURPOSES
ALEKSANDAR PETROVSKI
University “Goce Delchev” - Shtip, Military academy “General Mihailo Apostolski”, Skopje,
[email protected]

MARKO RADOVANOVIĆ
University of defence, Military academy, Belgrade, [email protected]

ANER BEHLIC
University “Goce Delchev” - Shtip, Military academy “General Mihailo Apostolski”, Skopje,
[email protected]

Abstract: One does not have to look far to see how Artificial Intelligence (AI) – the ability of machines to perform tasks
that typically require human intelligence – is transforming the international security environment in which all involved
security forces operates. Due to its cross-cutting nature, AI will pose a broad set of international security challenges,
affecting both traditional military capabilities and the realm of hybrid threats, and will likewise provide new
opportunities to respond to them. AI will have an impact on all of core tasks of every collective defense, crisis
management, and cooperative security systems. With new opportunities, risks, and threats to prosperity and security at
stake, the promise and peril associated with this foundational technology are too vast for any single actor to manage
alone. As a result, cooperation is inherently needed to equally mitigate international security risks, as well as to
capitalize on the technology’s potential to transform enterprise functions, mission support, and operations.
The aim of the paper is to present drones that learn on their own, i.e., that possess artificial intelligence, and which can
be used for military purposes. The paper presents the possibility of autonomous use of drones with artificial intelligence
in combat and non - combat operations of the army. Drones supported by GIS, C5IRS (command, control, computers,
communications, cyber-defense (C5), intelligence, surveillance, and reconnaissance (ISR) and AI will give significantly
advantage on the ground, because they can operate alone and will learn in line with the ground situation. With the
fusion of human, information, and physical elements increasingly determining decisive advantage in the battlespace,
interoperability becomes even more essential.
Keywords: Drone, UAV, Artificial Intelligence, GIS, C5IRS, Security, YOLOv5.

1. INTRODUCTION drones in a counter-terrorist operation. [5] Petrovski and

Radovanović analyze the use of drones in cooperation
Modern combat operations require the use of the most with the C5IRS system for the needs of the army.[6] Ilić
sophisticated combat means in order to efficiently and Tomašević analyze the impact of the Nagorno-
perform the assigned tasks. The use of unmanned aerial Krabah conflict on the perception of combat drones.[7]
vehicles is an indispensable segment of modern combat Radovanovic et al. analyzes the possibility of
operations. Due to its versatile use and different implementing drones in mortar units in order to increase
capabilities, it provides a wide range of capabilities to the efficiency of fire support units by applying a fire
units equipped with this type of combat equipment. By management system in cooperation with the C4IRS
applying unmanned aerial vehicles integrated into the system.[8] Bares performs interoperability modeling for
C5IRS system, it is possible to have a real-time image the C4IRS system in a collective security system.[9]
from the battlefield, which gives the decision maker the Petrovski et al. analyzes the application of GIS in
possibility of timely and efficient command of forces in cooperation with the C5IRS system in geography for the
the operation. Modern types of unmanned aerial vehicles needs of the military.[10] Radovanovic et al. analyzes the
are used in modern combat operations, from commercial selection of UAVs for the needs of military and police
to armed unmanned aerial vehicles. Both foreign and tactical units using the fuzzy AHP - VIKOR model of
domestic authors have dealt with this topic. Milić et al. multicriteria decision making.[11] Žnidaršič et al. shows
analyze the possibility of using drones in operations in several types of drones and anti-drone means for
urban environments.[1] Radovanovic et al. (2020) shows implementation in the units of the Serbian Army.[12]
the possibility of using civilian drones in the protection Petrovski and Toshevski present the application of GIS in
and monitoring of the land security zone. [2] Adamski geo-reconnaissance and C5IRS for military purposes. [13]
analyzes the effectiveness of UCAVs used in modern
armed conflicts. [4] Jović analyzes the combat use of

92
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

2. DEFINITION AND CLASSIFICATION Petrovski and Radovanović defined the terms drone and
OF UNMANNED AIRCRAFT UAV and their classification (figure 1). [6] The term
UAV has a broad meaning, it means with a motor that is
So far, there is no generally accepted definition of remotely controlled by the operator or it is a means that
unmanned aerial vehicles as well as their classification, so has a certain level of autonomy (control is done using
the European Association of Unmanned Vehicles Systems communication software, and often uses artificial
- EUROUVS has defined the classification of unmanned intelligence and different types of sensors), which they
aerial vehicles in relation to purpose flight altitude, flight can be used once or repeatedly and can carry deadly or
duration, speed, Maximum takeoff weight - MTOW, non-lethal cargo, transmit data in real time, using as a
aircraft dimensions, signal range, etc. [14] WiFi station and ect. It is a synthesis of the means and
devices necessary to manage it. They differ in purpose,
According to the model of control and management of
construction characteristics (shape, dimensions, weight,
unmanned aerial vehicles, they are divided into
payload, maximum flight altitude, maximum range, flight
autonomous systems, self-control systems, radar or radio
time, speed, etc.) of the environment in which they are
beam control systems, telecommand control systems and
used, the energy source with which they are driven.
combined systems (autonomous, non-autonomous). In
Depending on the purpose, they can be used in different
relation to flight altitude, take-off weight and maximum
environments such as land, water, air and space, and a
range, drones are divided into four categories. [14][15]
wide range of possibilities has created a condition for
- Category 1 (weight up to 1 kg, flight altitude up to 50 application in defense and security (for the needs of the
m, AGL - Above Ground Level, range up to 150 m); army and police - original purpose). agriculture,
- Category 2 (mass greater than 1 kg to 5 kg, flight construction, traffic, trade, communication, science,
altitude up to 150 m AGL, range up to 500 m); medicine, research, architecture, video and photography,
geology, forestry, mining, oceanography, environmental
- Category 3 (mass greater than 5 kg to 20 kg, flight
management, sports, mapping, etc. The term drone is
altitude up to 300 m AGL, range up to 2500 m);
more general than the term unmanned aerial vehicle,
- Category 4 (mass greater than 20 kg, flight altitude because all unmanned aerial vehicles can be called
greater than 300 m AGL, range greater than 2500 m) drones, while a drone does not necessarily have to be an
unmanned aerial vehicle. [6]

Figure 1: Classification of drones

93
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

Based on the classification of unmanned aerial vehicles, it needs to be trained using a supervised learning process. I
is concluded that the characteristics of unmanned aerial think that this will be much harder to achieve than you
vehicles usually depend on their purpose. It is necessary can imagine as combat drone technology is in my opinion
to analyze the tactical and technical characteristics of capable of being the most lethal and effective tools ever
unmanned aerial vehicles in order to consider their created for intermediate & close combat scenarios.
possibility of use in modern combat operations.
The Army’s office for countering small drones sees
Military use of unmanned aerial vehicles can be classified artificial intelligence and machine learning as key
into three groups: naval, land and air use, while for technologies for defeating enemy systems, service
civilian purposes it can be used in various areas of human officials said Oct. 15. And drone swarms could be used
activity, such as. in geodesy (photogrammetry), for simultaneous, multi-directional attacks in ways that
agriculture, industrial production, civil protection, disaster could overwhelm human defenders. The aerospace and
management, critical infrastructure surveillance, defence industry initially made drones for
environmental protection, police surveillance, protection counterinsurgency and defence, which have proven
and rescue of people, intelligence and security services, helpful in these environments. Today, many Military
journalism, commercial activities and leisure. Warfare industries develop drone technology integrated
into more military programs worldwide. They provide
3. APPLICATION OF DRONES WITH many benefits and advantages that make them extremely
ARTIFICIAL INTELLIGENCE FOR useful for different roles. Military drones have come to
revolutionize warfare. They are roving on land, streaking
MILITARY PURPOSES through the skies, and diving under the seas. Since their
Drones have improved military capabilities around the creation, more than fifty years ago, drones have
world in many ways. It will also continue to change constantly evolved to the present, becoming one of the
military warfare through the following: The Use of main artificial intelligences (AI) weapons, integrated into
Military Drones: The Impact on Land Forces and Legal military forces throughout the world. As a result, more
Implications Command and Control –Drones can relay military forces are looking to use drones to increase their
crucial information on enemy movements, locations, and combat and surveillance capacity.
positions of strategic targets. This information allows Military advantages are in autonomous weapons systems
commanders to be more efficient and make better act as a force multiplier. That is, fewer warfighters are
decisions when in the field. needed for a given mission, and the efficacy of each
Drone technology itself is a relatively new area of warfighter is greater. Next, advocates credit autonomous
military technology, but military engineers took to weapons systems with expanding the battlefield, allowing
combining drones with artificial intelligence rather combat to reach into areas that were previously
quickly to create a product that in some cases might be inaccessible. Finally, autonomous weapons systems can
comparable to the performance of human reconnaissance reduce casualties by removing human warfighters from
teams. Shield.AI, AeroVironment, and Lockheed Martin dangerous missions.
all showcase how military defense contractors are
combining current computer vision technology and image 4. YOLOV5 METHOD FOR ENCHANCED
recognition with drones to solve military problems MILITARY DRONE SURVEILLANCE
without the need of endangering human life. Shield AI’s
WITH AI SOFTWARE CAPABILITIES
drone can purportedly navigate unknown lands without
the need for GPS tracking. Such a UAV could give This article covers real-world model incorporate in either
military troops the capability to collect data that could drones or the AI software behind them: In this article,
allow them to move faster and check if they are being we’ll look at several military drones and UAVs with AI
pursued during tactical reconnaissance, tracking, combat capabilities. There are a variety of use cases for AI when
assessment, and cartographic missions. Drones could it comes to drone technology. The military seems to
allow operators to make decisions without being commonly apply AI for allowing its drones to fly on their
concerned that they might be ambushed from the rear, for own, which requires machine vision.
example.
Implementing artificial intelligence for drones is a
combination of mechanical devices, navigational
instruments, and machine vision. The AI behind the drone

94
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

Figure 2: Model of Geo-reconnaissance and commanding (GRC) information system in UAV supported with C4IRS
systems and AI (Petrovski A. and Mihajlo T. defined this model - figure 2)

4.1. YOLOv5 Method for enhanced military from various internet and local sources. Furthermore,
drone surveillance upon that dataset a YOLOv5 model is trained and
evaluated. A Surveillance Camera is set up on a military
The proposed framework aims to detect objects of critical surveillance drone, which is connected to the System
military importance based on the images received from Model. The camera forwards real-time video through the
the video stream of military surveillance drones. The system model which processes the data in real time.
actions will be taken depending on the obtained result. Moreover, the processed image is projected on the Real-
Figure 3 summarizes the suggested methodology of this Time Monitoring Interface which shows the result of
paper. image processing, including alarms for detection of vital
military objects. False positives and vital military objects
which weren’t detected by the system but self-identified
by the human operator, are captured and further
forwarded to the unified Vital Objects Database, where
they are re-evaluated and used for further training of the
system.
This paper proposes a real-time vital Military Objects
detection system based on Convolution Neural Network
using the YOLOv5 (You Only Look Once) framework.
The main contributions of this approach include:
1) The proposed system can be implemented easily on
already existing surveillance drones
2) It detects vital military objects with high accuracy
from large distances.
3) The software used for development is free and open
source.
4) A unified large dataset for vital military objects
including weapons, vehicles, tanks, buildings,
Figure 3: Overview of the Methodology bunkers and artillery pieces, with various scenarios
and conditions.
Firstly, a unified Vital Objects Database is created by 5) It aids military commanders and personnel in
using verified images of objects that are critical for decision making, tracking, and identification of vital
military surveillance such as weapons, bunkers, military military objects.
buildings, vehicles, tanks, and artillery pieces, collected

95
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

4.2. YOLOv5 Approach for Vital Military military surveillance task, detection speed and accuracy
Objects Detection are imperative, and compact model size also determines
its inference efficiency on resource-poor edge devices.
YOLO is based on Convolution Neural Network (CNN) Secondly, the Yolov5 applied path aggregation network
and represents a very fast end-to-end object detection (PANet) [19] as its neck to boost information flow.
algorithm. [16] Continuous improvements have made it PANet adopts a new feature pyramid network (FPN)
achieve top performances on two official object detection structure with enhanced bottom-up path, which improves
datasets: Pascal VOC (visual object classes) [17] and the propagation of low-level features. At the same time,
Microsoft COCO (common objects in context). [18] adaptive feature pooling, which links feature grid and all
The network architecture of Yolov5 is shown in Figure 4. feature levels, is used to make useful information in each
There are three reasons why Yolov5 is chosen as the main feature level propagate directly to following subnetwork.
method for training the system model. Firstly, Yolov5 PANet improves the utilization of accurate localization
incorporated cross stage partial network (CSPNet) [18] signals in lower layers, which can obviously enhance the
[19]into Darknet, creating CSPDarknet as its backbone. location accuracy of the object. Thirdly, the head of
CSPNet solves the problems of repeated gradient Yolov5, namely the Yolo layer, generates 3 different sizes
information in large-scale backbones, and integrates the (18 × 18, 36 × 36, 72 × 72) of feature maps to achieve
gradient changes into the feature map, thereby decreasing multi-scale prediction [20], enabling the model to handle
the parameters and FLOPS (floating-point operations per small, medium, and big objects.
second) of model, which not only ensures the inference
speed and accuracy, but also reduces the model size. In

Figure 4: The network architecture of Yolov5. It consists of three parts: (1) Backbone: CSPDarknet, (2) Neck: PANet,
and (3) Head: Yolo Layer. The data are first input to CSPDarknet for feature extraction, and then fed to PANet for
feature fusion. Finally, Yolo Layer outputs detection results (class, score, location, size) [22]

96
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

In the YOLO method, the input images are only seen once time, forests, desert, and urban, as seen in Figure 6.
through the neural network, and it predicts the detected However, even then the dataset is severally limited,
object in the image. It works by dividing the input image therefore, we included several videos with different
into different grids based on predefined grid size and then Military Targets in different conditions such as size,
predicts the probability of the desired object in each grid. shape and color. In order to successfully adopt and train
It predicts all the classes and the object bounding that are the dataset in YOLOv5 the dataset itself was assigned
in the image in one run of the Algorithm.[23] labels and the annotating bounding boxes. The value of
the annotating box coordinate in each image is then
normalized between 0-1. This process was carried out
using Roboflow [23], which makes it possible to annotate
and make data labels in the desired format. Furthermore,
the dataset is carefully split into 3 different categories for
training, validation and testing. Additionally, by carefully
assigning the images to each category, we avoided a
biased model by using data with a similar number of
images for each class which forms a balanced dataset.
First and foremost, in the model training process we fine-
tune the YOLOv5m model which is already pre-trained
on the COCO data set. Furthermore, Transfer learning
was used to train on the unified dataset. Moreover,
various augmentation processes such as HSV, color
spacing, mosaic, image scaling were applied. The
hyperparameters that are fine-tuned are used here, such as
SGD optimizer, 0.01 learning rate, 0.0005 weight decay
and then finally 600 epochs on batch size 32.
The model was built using Google Collaboratory
resources to process the dataset.
Figure 5: Overview of military vehicles that can be
recognize with drone pre-input images in data bases.
The proposed YOLOv5 based (pre-trained) solution is
extremely lightweight and fast to recognize the critical
objects of military importance. Good accuracy and the fps
are two important things to consider when working on
real-time surveillance and data gathering.

4.3. Model Training Process & Dataset

To train, validate and evaluate the model a customized
dataset was built. The finalized dataset contains around
10000 images of military trucks, tanks, that were
collected from publicly available datasets, Google images,
and local sources as presented in Table 1.
Table 1. Distribution of images in the dataset
Dataset type Size & Type Source
Military Trucks, Military Vehicles –
6772 images
Tanks, APC Mendeley Data[23]
Military Tanks –
Military Tanks 1078 images
Kaggle[23]
Moving and Stationary
Military Artillery,
Target Acquisition and
Tanks, MRL 1000 images
Recognition (MSTAR)
Systems, Trucks
Dataset[23]
Stationary
Military
Equipment, 2000 images
Local Sources
Artillery, & 20 videos
Command posts &
fire stations Figure 6: Training Process on the Unified Dataset,
annotating Military Targets by class (Tanks, APC’s,
The images cover a verity of scenarios Military Targets in Artillery, MRL Systems)
various scenarios and terrain including day and night

97
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

5. RESULTS AND DISCUSSION from the unified dataset, we achieved on average 0.922
[email protected]. Furthermore, From the displayed confusion
To evaluate the performance of the proposed system, the matrix in Figure 7 it can be seen that the models can label
trained model was tested in different environments maximum of the data quite correctly Also from Figure 9 it
including forests, mountains, open fields, etc. The results can be seen that the model can confirm whether there is a
are performed with a confidence threshold of 0.4, Military Target on the image quite correctly.
Initially, when performed on the handpicked test data

Figure 7: Confusion Matrix of the model

The confidence value of the Military Targets the model
This obtained value and prediction results shown on
recognizes is quite high also. Moreover, we can see the
performance graphs in Figure 8 prove that there is a pretty
performance of the suggested model applied to validation
good potential to use the YOLOv5 algorithm to detect
data through performance graphs in Figure 8.
Military Targets in real-time during drone surveillence.

Figure 8: Result Graphs obtained from model training and validation

98
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

However, it is difficult to say that the proposed model

The model is lightweight and fast enough to deploy. It can
does not have any shortcomings. The model may cause
be incorporated with in-built capturing tools, in this case,
errors and consider, non-military vehicle, and other
Surveillance Cameras mounted on drones, and predict
military like objects, as real targets. These issues mainly
recognition in almost real time with minimum delays.
increase as the distance of the target increases, which the
camera then captures in an unclear image, which confuses
the model. However, longer training times, more concrete
and larger datasets would highly likely yield a good result
in this process and eliminate most of the false positives.
Further research should be focused on the integration of
the modern C6ISR system (combat command) to the
combat systems of the army and the advantages that the
mentioned system provides to units during the execution
of a modern combat operation in a different environment.

ACKNOWLEDGEMENTS
Figure 9a: Examples of predicted output's This scientific work in same parts was written as a result
of unifying the results of research in two scientific
research projects funded by the Ministry of Defense of the
Republic of Serbia, under numbers VA-DH/1/21-23
„Uticaj savremenog okruženja na izvođenje borbenih
dejstava u urbanim sredinama” (Influence of
contemporary environment on the conduct of combat
activities in urban spaces). i VA-DH/1/22-24 „Model
upravljanja razvojem sposobnosti sistema odbrane”
(Management model for defense system capability
development).

References
Figure 9b: Examples of predicted output's [1] MILIĆ, A., RANĐELOVIĆ, A., RADOVANOVIĆ,
M.: Use of drones in operations in the urban
6. CONCLUSION environment, 5th International Scientific conference
Safety and crisis management – Theory and practise
The use of unmanned aerial vehicles equipped with AI Safety for the future – SecMan 2019, Belgrade:
and modern technical devices (high resolution cameras, Regional Association for Security, (2019) 124-130.
infrared and thermal cameras, microphones, various types
[2] RADOVANOVIĆ, M., MILIĆ, A., RANĐELOVIĆ,
of sensors, guided and non-guided missiles and other
A.: Mogućnost upotrebe dronova u zaštiti Kopnene
accessories) integrated into the C5IRS system
zone bezbednosti, 15. Međunarodna konferencija
significantly increases the efficiency of units engaged in
rizik i bezbednosni inženjering, Visoka tehnička
combat operations. The use of this sophisticated
technology enables timely and accurate information about [3] škola strukovnih studija u Novom Sadu i Fakultet
the event in real time, as well as the destruction of the tehničkih nauka, Novi Sad, (2020) 303 - 311.
target without risk to humanity, while transmitting the [4] ADAMSKI, M.: Effectiveness analysis of UCAV used
situation to the battlefield command center. Unmanned in modern military conflicts, Aviation, 24 (2) (2020).
aerial vehicles can still be used to deliver medical [5] JOVIĆ, Ž.: Combat engagement of drones in USA
supplies and other necessary combat and non-combat counterterrorist operations, Bezbednost, 58(3)
equipment to units in the area of operation. (2016) 171-190.
The use of unmanned aerial vehicles in combat [6] PETROVSKI, A., RADOVANOVIĆ, M.:
operations, regardless of limitations, increases the Application of detection reconnaissance technologies
efficiency and effectiveness of the units engaged in the use by drones in collaboration with C4IRS for
operation, and also increases the protection and reduces military interested, Contemporary Macedonian
the risk to the engaged personnel. Based on the tactical Defence, 21 (40) (2021) 117-126.
and technical characteristics of unmanned aerial vehicles, [7] ILIĆ, D., TOMAŠEVIĆ, V.: The impact of the
it is possible to realize various tasks. Nagorno-Karabakh conflict in 2020 on the
Today’s environment also requires the collective perception of combat drones, Serbian Journal of
capabilities and efforts of multiple organizations. Engineering Management, 6(1) (2021) 9-21.
Unfortunately, traditional C4ISR systems weren’t built [8] RADOVANOVIĆ, M., SAMOPJAN, M.,
with interoperability in mind—in the past, they were PETROVSKI, A.: Possibility of Implementation of
designed to meet mission-specific requirements and to Drons in Mortar Units in Order to Increase the
solve a particular set of problems facing an organization. Efficiency of Fire Support Units, 24. Međunarodna

99
 APPLICATION OF DRONES WITH ARTIFICIAL INTELLIGENCE FOR MILITARY PURPOSES OTEH 2022

DQM konferencija Upravljanje kvalitetom i [17] EVERINGHAM, M., ESLAMI, S.A., VAN GOOL,
pouzdanošću ICDQM -2021, Prijevor, (2021) 307– L., WILLIAMS, C.K., WINN, J., ZISSERMAN, A.:
315. The pascal visual object classes challenge: A
[9] BARES, M.: Interoperability Modeling of the C4ISR retrospective. International Journal of Computer
Systems, RTO SCI Symposium on “System Concepts Vision, 111 (1) (2015) 98–136.
for Integrated Air Defense of Multinational Mobile [18] LIN, T.Y., MAIRE, M., BELONGIE, S., HAYS, J.,
Crisis Reaction Forces”, held in Valencia, Spain, PERONA, P., RAMANAN, D., DOLLÁR, P.,
published in RTO MP-063. (2001) 16-1 – 16-16. ZITNICK, C.L.: Microsoft coco: Common objects in
[10] PETROVSKI,A., TANESKI,N., BOGATINOV,D.: context, In Proceedings of the 13th European
Geography in Geospatial Intelligence - C4IRS and Conference on Computer Cision (ECCV 2014),
Cyber Security, 5th International Scientific Zurich, Switzerland, (2014) 740–755.
conference Safety and crisis management – Theory [19] WANG, C.Y., MARK LIAO, H.Y., WU, Y.H.,
and practise Safety for the future – SecMan 2019, CHEN, P.Y., HSIEH, J.W., YEH, I.H.: CSPNet: A
Belgrade: Regional Association for Security, (2019) new backbone that can enhance learning capability
64-72. of cnn, In Proceedings of the IEEE Conference on
[11] RADOVANOVIĆM., PETROVSKI,A., Computer Vision and Pattern Recognition (CVPR
ŽINDRAŠIČ, V., RANĐELOVIĆ, A.: Application 2020), Washington, DC, USA, (2020) 390–391.
of the fuzzy AHP -VIKOR hybrid model in the [20] WANG, K., LIEW, J.H., ZOU, Y., ZHOU, D.,
selection of an unmanned aircraft for the needs of FENG, J.: Panet: Few-shot image semantic
tactical units of the armed forces, Scientific segmentation with prototype alignment, In
Technical Review, 71(2) (2021) 26-35. Proceedings of the IEEE International Conference on
[12] ŽNIDARŠIČ,V., RADOVANOVIĆ,M., Computer Vision (ICCV 2019), Seoul, Korea, (2019)
STEVANOVIĆ,D.: Modeling the organisational 9197–9206.
implementation of a drone and counter-drone [21] REDMON, J., FARHADI, A.: Yolov3: An
operator into the Serbian Armed Forces rifle incremental improvement, arXiv 2018,
section, Vojno delo, 72 (3) (2020) 84 -109. arXiv:1804.0276
[13] PETROVSKI,A., TOSHEVSKI,M.: Gis in Army: [22] XU, R., LIN, H., LU, K., CAO, L., LIU, Y.: A Forest
Application of GIS in Geo-Reconnaissance And C4IS Fire Detection System Based on Ensemble Learning,
in Army Purposes, 2nd International Scientific Forests, 12 (2) (2021) 217: 1-17.
Conference GEOBALCANICA, Skopje, (2016) [23] REDMON,J., FARHADI,A.: Yolo9000: better,
153-160 faster, stronger, in Proceedings of the IEEE
[14] AGBEYANGI,A., ODIETE,J., conference on computer vision and pattern
OLORUNLOMERUE,A.: Review on UAVs used for recognition, (2017) 7263–7271.
AerialSurveillance, Journal of Multidisciplinary [24] https://data.mendeley.com/datasets/njdjkbxdpn/1
Engineering Science and Technology, 3 (10) (2016)
[25] https://www.kaggle.com/datasets/antoreepjana/milita
5713-5719.
ry-tanks-dataset-images
[15] ARJOMANDI, A., AGOSTINO, S., MAMMONE,
[26] https://www.sdms.afrl.af.mil/index.php?collection=m
M.: Classification of Unmanned AerialVehicle,
star
Report for Mechanical Engineering class, University
of Adelaide, Australia, 2006. [27] www.roboflow.com
[16] REDMON, J., DIVVALA, S., GIRSHICK, R.,
FARHADI, A.: You only look once: Unified, real-
time object detection, in Proceedings of the IEEE
conference on computer vision and pattern
recognition, (2016) 779-788

100

View publication stats