=Paper=
{{Paper
|id=Vol-3654/paper5
|storemode=property
|title=A YOLO-based Method for Object Contour Detection and Recognition in Video Sequences
|pdfUrl=https://ceur-ws.org/Vol-3654/paper5.pdf
|volume=Vol-3654
|authors=Mariia Nazarkevych,Maryna Kostiak,Nazar Oleksiv,Victoria Vysotska,Andrii-Taras Shvahuliak
|dblpUrl=https://dblp.org/rec/conf/cpits/NazarkevychKOVS24
}}
==A YOLO-based Method for Object Contour Detection and Recognition in Video Sequences==
https://ceur-ws.org/Vol-3654/paper5.pdf
A YOLO-based Method for Object Contour Detection
and Recognition in Video Sequences
Mariia Nazarkevych1, Maryna Kostiak1, Nazar Oleksiv1, Victoria Vysotska1,
and Andrii-Taras Shvahuliak2
1 Lviv Polytechnic National University, 12 Stepan Bandera str., Lviv, 79013, Ukraine
2 Lviv Ivan Franko National University, 1 Universytetska str., Lviv, 79000, Ukraine
Abstract
A method for recognizing the contours of objects in a video data stream is proposed. The
data will be uploaded using the video camera. Objects will be recognized in real-time. We
will use YOLO—a method of identification and recognition of objects in real-time.
Recognized objects will be recorded in a video sequence showing the contours of the
objects. The approach proposed in the project reasonably synthesizes methods of
artificial intelligence, theories of computer vision on the one hand, and pattern
recognition on the other; it makes it possible to obtain control influences and
mathematical functions for decision-making at every moment with the possibility of
analyzing the influence of external factors and forecasting the flow of processes and refers
to the fundamental problems of mathematical modeling of real processes. The installation
of the neural network is shown in detail. The characteristics of the neural network and its
capabilities are shown. Approaches to computer vision for object extraction are shown.
Well-known methods are methods of expanding areas, methods based on clustering,
contour selection, and methods using a histogram. The work envisages building a system
for rapid identification of combat vehicles based on the latest image filtering methods
developed using deep learning methods. The time spent on machine identification will be
10–20% shorter, thanks to the developed new information technology for detecting
objects in conditions of rapidly changing information.
Keywords 1
Artificial intelligence, tracking, selection of objects, image recognition, YOLO, segmentation.
1. Introduction preparation of graphic images—selection of
objects, segmentation, and selection of
Video surveillance is a common means of contours.
solving problems related to security and event Tracking is determining the location of a
monitoring [1–3]. One of the main tasks arising moving object [7] or several objects over time
from video surveillance is detection [4], using a video camera (Fig. 1). The algorithm
tracking [5] and identification [6] of moving analyzes video frames and outputs the position
objects. Video cameras are near us and record of moving objects relative to the frame.
data about us. Therefore, there is a need to
recognize data and objects. To recognize the Contrast Blur Selection
data, you need to go through the stage of pre- enhancement reduction of contours
Filtration
processing, that is, improving the visual Video stream preprocessing
quality—increasing the contrast, distinguishing
the boundaries, removing blurring, and Figure 1: Scheme of preprocessing of a video
filtering. Then there is an operation for the sequence with object capture
CPITS-2024: Cybersecurity Providing in Information and Telecommunication Systems, February 28, 2024, Kyiv, Ukraine
EMAIL : mariia.a.nazarkevych@lpnu.ua (M. Nazarkevych); maryna.y.kostiak@lpnu.ua (M. Kostiak); nazar.oleksiv.mnsa.2020@lpnu.ua
(N. Oleksiv); victoria.a.vysotska@lpnu.ua (V. Vysotska); andrii-taras.shvahuliak@lnu.edu.ua (A.-T. Shvahuliak)
ORCID: 0000-0002-6528-9867 (M. Nazarkevych); 0000-0002-6667-7693 (M. Kostiak); 0000-0001-7821-3522 (N. Oleksiv); 0000-0001-
6417-3689 (V. Vysotska); 0009-0002-0319-1909 (A.-T. Shvahuliak)
©️ 2024 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
49
The main tracking problem is matching the Grayscale methods perform segmentation—
positions of the target object in successive dividing a digital image into several sets of
frames, especially if the object is moving fast pixels [11]. Image segmentation is commonly
compared to the frame rate. Thus, tracking used to highlight objects and boundaries. More
systems usually use a movement model [8], precisely, image segmentation is the process of
how the image of the target object can change assigning such labels to each pixel of an image
during various movements (Fig. 2). so that pixels with the same labels share visual
characteristics.
Block-based methods do not process
Selection of
objects
Segmentati
on
Selection
of contours
individual pixels [12], but groups of pixels
combined into blocks. If the block contains a
Department of objects boundary, then in such areas the boundary of
Figure 2: Scheme of separation of objects the object is determined inaccurately [13, 14].
The disadvantage of methods [15] based on
Examples of such simple movement patterns are the energy function is the low speed of
flat object tracking—affine transformation or operation.
object image homography [9].
The target can be a rigid three-dimensional
2.2. Methods of Expanding Regions
object, the motion model determines the
appearance depending on its position in space
and orientation. The methods of this group are based on the use
For video compression, keyframes are of local features of the image [16]. The idea of
divided into macroblocks. A motion model is a the region expansion method is to analyze first
burst of keyframes where each macroblock is the starting point, then its neighboring points
transformed using a motion vector. according to the criterion of homogeneity of
The image of a deformable object can be the analyzed points into one or another group.
covered with a grid and the movement of the In more effective variants of the method, the
object is determined by the position of the starting point is not individual pixels, but the
vertices of this grid. division of the image into several small areas.
When an object is to be searched and Each region is then checked for homogeneity,
matched against a given one, a new set of key and if the result of the test is negative, the
points is extracted into the test image the two corresponding area is divided into smaller
sets are matched and a similarity score is sections.
calculated. Threshold segmentation and segmentation
[17] according to the homogeneity criterion
based on average brightness (Fig. 4) often do
2. Review of Literature not give the desired results.
2.1. Object Selection
Methods of expanding regions
Before selecting an object from a video stream.
there are pixel-by-pixel, block-by-block, and By the According to
methods based on energy functional criterion of the texture- Threshold
minimization [10] (Fig. 3). homogeneity based segmen-
based on tation
Pixel-by-pixel methods of object selection brightness
homogeneity
criterion
process all points of the image. These methods
are highly accurate, but they are sensitive to Figure 4: Methods of expanding regions in a
noise. video stream
Selecting an object from a video stream
Such segmentation usually results in a large
number of small regions. The most effective
MInimiza-
results are given by the segmentation based on
Pixel Post-block tion of the
energy
the homogeneity criterion based on the texture
functional [18].
Figure 3: Methods of selecting an object from
a video stream
50
2.3. Selection of Contours The difference is most often based on color,
brightness, texture, and pixel location, or a
In the video, heterogeneous objects are often balanced combination of these factors.
observed, so you have to face the task of finding
perimeters, curvature, form factors, specific 2.5. Methods Using a Histogram
surface area of objects, etc. All these tasks are
related to the analysis of contour elements of Histogram-based methods [23] are very
objects. efficient when compared to other image
Methods for highlighting contours in an segmentation methods because they require
image can be divided into three main classes: only a one-pixel pass.
1. High-frequency filtering methods [19]. A histogram is calculated over all pixels in
2. Methods of spatial differentiation [20]. the image and its minima and maxima are used
3. Methods of functional approximation to find clusters in the image. Color or
[21] (Fig. 5). brightness can be used when comparing.
Common to all these methods is the An improvement on this is to recursively
development of the boundary as a region of a apply it to the clusters in the image to divide
sharp drop in the image brightness function them into smaller clusters. The process is
(𝑖,), which is distinguished by the introduced repeated gradually with smaller and smaller
mathematical contour model. clusters until the moment when new clusters
stop appearing altogether.
Selection of contours
Approaches based on the use of histograms
can also be quickly adapted for multiple frames
Methods of
high-frequency
Methods of Methods of
functional
while retaining their single-pass speed
spatial
filtering differentiation approximation. advantage.
Figure 5: Methods of highlighting contours in 2.6. YOLO—Object Detection
a video stream
By the tasks, contour selection algorithms are You-Only-Look-Once (YOLO) [24] is an
subject to requirements: the selected contours independent video object detection system
must be thin, without gaps, and closed. The that can operate in real-time at very high frame
process of selecting contours is complicated rates—the common limit is 45 frames per
due to the need to apply algorithms for second, with a claimed useful frame rate of up
thinning and eliminating gaps. However, the to 155 frames per second. YOLO was originally
contours are not closed and unsuitable for released in 2015 with Facebook research.
analysis procedures. YOLO consists of two main parts: a class
detector and a framework detector. The class
2.4. Methods Based on Clustering detector determines which objects are present
in the image. The frame detector determines
The 𝐾-means method is an iterative method the location of objects in the image. The class
used to divide an image into 𝐾 clusters. The detector works by using a regression neural
basic algorithm is given below [22]: network that learns to predict the value of a
Step 1. Choose 𝐾 cluster centers, randomly variable. It learns to predict the probability
or based on some heuristics. that a certain object is present in the image.
Step 2. Place each image pixel in a cluster The YOLO class detector is a regression
whose center is closest to that pixel. neural network with 24 deep layers. The input
Step 3. Recalculate the cluster centers by layer of the network receives a 448×448 pixel
averaging all the pixels in the cluster. image. The output layer of the network
Step 4. Repeat steps 2 and 3 until contains 84 values. Each value corresponds to
convergence (for example, when the pixels the probability that a certain object is present
remain in the same cluster). in the image.
The distance is usually taken as the sum of
squares or absolute values of the differences
between the pixel and the center of the cluster.
51
2.7. Features of Video Tracking 3. Problem Statement
Digital IP cameras are increasingly used in Let’s install object tracking in the video stream
modern video surveillance systems. and examine the speed of object detection.
Connecting an IP camera to an already existing To do this, we need to write the following
local network can guarantee minimal command in the terminal:
installation costs. pip install ultralytics
Let’s consider the characteristics that must And then import it into the code:
be taken into account when choosing computer from ultralytics import YOLO
technologies for a digital video surveillance Now everything is ready to create a neural
system. network model:
The first characteristic is the number of model = YOLO(“yolov8m.pt”)
physical ports to which other devices can be As mentioned earlier, YOLOv8 is a group of
connected. Will this parameter determine the neural network models. These models were
maximum number of IP cameras that can be built and trained using PyTorch and exported
connected? For a home video surveillance as .pt files.
system, a switch that has 4 ports is often used. The first time you run this code, it will
Equipment with 8–16–24 ports is used for download the yolov8m.pt file from the
professional systems [25]. Ultralytics server to the current folder. It will
The second characteristic is bandwidth. At then construct a model object. You can now
the same time, the bandwidth of each port is train this model, detect objects, and export
taken into account. The most common values them for use. There are convenient methods
are 10/100 Mbps and 1 Gbps. It should be for all these tasks:
taken into account that often the total train({dataset descriptor file path})—used
bandwidth of the switch can be lower than the to train the model on the image dataset.
total value of all ports. When choosing the predict({image})—Used to predict the
bandwidth of a candle, you need to determine specified image, for example, to detect the
what data transfer rate your network can bounding boxes of all objects that the model can
handle. find in the image.
The third characteristic is the speed of data export({format})—used to export the model
transmission, which will limit the possibility of from the default PyTorch format to the specified
receiving and transmitting information. format.
The fourth feature of PoE is the function All YOLOv8 object detection models are
that allows you to power other devices through already pre-trained on the COCO dataset,
the same cable that transmits data. This is very which is a huge collection of images of 80
important for the organization of video different types.
surveillance, as it allows you to get rid of The prediction method accepts many
unnecessary wires, and also simplifies the different types of input data, including a path
process of installation and organization of the to a single image, an array of image paths, an
power supply of connected devices. Image object from Python’s well-known PIL
The fifth characteristic is management library, and others [26].
protocols. Yes, PoE switches are divided into After running the input data through the
managed and unmanaged. Managed switches model, it returns an array of results for each
are devices that support several protocols input image. Since we only provided one
(functions) of network management and data image, it returns an array with one element,
transmission. which you can extract like this:
To build simple and small IP surveillance
systems, physically isolated from networks in
which other critical data is transmitted The result contains detected objects (Fig. 6) and
(telemetry data, banking and financial data, convenient properties for working with them.
video conferences, etc.), it is possible to The most important is the boxes array with
dispense with the use of unmanaged PoE information about the detected bounding boxes
switches. on the image (Fig. 7).
52
You can analyze each box either in a loop or
manually. Let’s take the first object:
The box object contains bounding box
properties, including:
xyxy—coordinates of the box in the form of
an array [x1,y1,x2,y2]
cls—object type identifier
conf—confidence level of the model
regarding this object. If it’s very low, like<0.5,
you can just ignore the field.
Figure 6: Video capture of an object and Let’s display information about the object:
“person” recognition
For the first object, you will receive
information:
Since YOLOv8 contains PyTorch models, the
outputs from PyTorch models are encoded as
an array of PyTorch Tensor objects, so you
Figure 7: Capturing the object on video and need to extract the first element from each of
recognizing the “chair” these arrays:
You can determine how many objects are
found by running the len function:
After launch, “2” was received, which means
that two boxes were detected: one for a mobile
phone, and the other for a person (Fig. 8).
Now we see the data as Tensor objects. To extract
the actual values from a Tensor, you need to use
the .tolist() method for tensors with an array
inside, and the .item() method for tensors with
scalar values.
Let’s load the data into the corresponding
variables:
Figure 8: Video capture of two objects and
recognition of “person” and “cell phone”
53
5. Implementation
The video_detection function takes a video
path as input and performs object detection
Now we see the actual data. Coordinates can be using the YOLO model.
rounded, and probability can also be rounded The YOLO model from Ultralytics is loaded
to two decimal places. from the specified checkpoint file (yolov8n.pt).
All objects that a neural network can detect The frames of the detected objects on each
have digital identifiers. In the case of the pre- frame are determined, and the processed
trained YOLOv8 model, there are 80 feature frames are returned.
types with IDs from 0 to 79. The COCO feature Class names corresponding to detected
classes are public. Additionally, the YOLOv8 objects are defined in the classNames list.
result object contains a convenient names Video capture and processing:
property to retrieve these classes. OpenCV is used to capture video frames
from the specified path.
4. Data to Proposed Model On each frame, detected objects are drawn
along with class labels and confidence levels.
The web application consists of 2 main files: The processed frames are returned for
Flaskapp.py is a file responsible for the streaming in the user’s browser.
project itself, its appearance, and its structure. The general course of work:
YOLO_Video.py is a file that is responsible • The user uploads a video file through
for the YOLO algorithm, namely for the the interface.
implementation of object recognition in the • The file is saved and its path is stored
video stream. in the Flask session.
Implementation of the Flaskapp.py file: • The object detector is called from the
Configuring the Flask application: received video path.
A web application is created using the Flask • The processed video frames are
class. transmitted for real-time viewing
Configuration parameters such as the secret through the user’s browser.
key and the file download folder are set. This project uses Flask for the web
Defines UploadFileForm class using Flask- application and integrates YOLO for real-time
WTF to handle file uploads. video processing. The YOLO_Video.py file
Video processing functions: isolates functionality related to YOLO, making
Generate_frames and generate_frames_web it modular and reusable.
functions are defined to generate frames based When entering the web application, we are
on output from YOLO detection. greeted by the “title” page:
These functions use the video_detection There are two buttons on this page:
function from the YOLO_Video.py file to perform The first Video button sends us to a page
object detection on video frames. where we can upload a video, press the Submit
Routes are defined for the home page (/ and button, and receive the processed video [27].
/home), the webcam page (/webcam), and the The second LiveWebcam button sends us to
video download page (/FrontPage). a page where the webcam is connected
The /video and /webapp routes are automatically and displayed on the screen in a
responsible for broadcasting video frames with processed format.
object detection results. In the images Figs. 9–16 we can see that the
The webcam and front routes render HTML YOLOv8 algorithm is running on the webcam.
templates for webcam pages and video uploads. Our model is based on pre-trained OSFA and
The UploadFileForm class is used to handle is built on top of PyTorch. The training image size
uploads of video files. was up to 256×128. A batch size of 64 randomly
The application runs on the development selected data was then fed to the network.
server if the script is executed directly. During the test, the test images are also resized
to 256×128. Our model is trained on 100 epochs.
The values of α1, α2 and the learning rate are the
54
same as those set by OSFA. α1, α2 and learning
rate are set to 1, 0.0007, 3.5×10−5, respectively.
In SAM, the number of horizontal parts is 4. All
experiments are performed with a hardware
environment of 11th Gen Intel(R) Core(TM) i7-
11800H at 2.30 GHz and NVIDIA GeForce RTX
3060.
Figure 12: Human and tank recognition
results in the Yolo video sequence
Figure 9: Results of recognition and
identification of two people and a truck in the
Yolo video sequence
Figure 13: Recognition results of two people, a
tank in the Yolo video sequence
Figure 10: Results of recognition and
identification of a person and a Yolo video
sequence Figure 14: The results of object recognition
and identification of the Yolo video sequence
Figure 11: Results of object recognition and Figure 15: Results of recognition and
identification of the Yolo video sequence identification of three objects of the Yolo video
sequence
55
Network, in: Workshop on Cybersecurity
Providing in Information and
Telecommunication Systems, vol. 3550
(2023) 240–245.
[2] H. Hulak, et al., Dynamic Model of
Guarantee Capacity and Cyber Security
Management in the Critical Automated
System, in: 2nd International Conference
on Conflict Management in Global
Figure 16: Results of recognition and Information Networks, vol. 3530 (2023)
identification of three objects of the Yolo video 102–111.
sequence [3] V. Grechaninov, et al., Decentralized
Access Demarcation System
The data were taken from [28] to form the Construction in Situational Center
dataset. The protected technology was Network, in: Workshop on Cybersecurity
developed in [29], from where the protected Providing in Information and
communication channels are taken. The Telecommunication Systems II, vol.
structural diagram of the model was taken in 3188, no. 2 (2022) 197–206.
[30], and the use of methods in [31]. Also, the [4] J. Liu, et al., Deep Industrial Image
approaches used for image preprocessing Anomaly Detection: A Survey, Mach.
were taken in [32] and [33]. Data processing Intell. Res. 21(1) (2024) 104–135. doi:
was formed thanks to [34]. 10.1007/s11633-023-1459-z.
[5] E. Kruger-Marais, Subtitling for
6. Conclusions Language Acquisition: Eye Tracking as
Predictor of Attention Allocation in
This study analyzed the YOLOv8 object Education, Int. J. Lang. Stud. 18(2)
recognition algorithm and its differences from (2024) 129–150. doi: 10.5281/zenodo.
other machine learning algorithms. A web 10475319.
application for object recognition in a video [6] P. Li, et al., Efficient Long-Short
stream was also created, analyzed, and tested. A Temporal Attention Network for
clear overview of the development tools for a Unsupervised Video Object Segmen-
web application using YOLOv8 has been tation, Pattern Recognit. 146 (2024)
provided. The PyCharm programming 110078. doi: 10.1016/j.patcog.2023.
environment, the Flask framework, and its 110078.
advantages compared to other frameworks were [7] E. Kawamura, et al., Ground-Based
studied. Also explored is Ultralytics, which helps Vision Tracker for Advanced Air Mobility
in the development and testing of web and Urban Air Mobility, AIAA SciTech
applications for video streaming. 2024 Forum (2024). doi:
As a result of the work performed, a high level 10.2514/6.2024-2010.
of understanding of the YOLOv8 algorithm, its [8] Z. Yin, et al., Numerical Modeling and
features, and its capabilities in the field of object Experimental Investigation of a Two-
detection in the video stream was achieved. The Phase Sink Vortex and its Fluid-Solid
developed web application is not only a practical Vibration Characteristics, J. Zhejiang
application of thialgorithm but can also serve as University-SCIENCE A 25(1) (2024) 47–
a basis for further developments and 62. doi: 10.1631/jzus.a2200014.
improvements in this direction. [9] T. Wolf, D. Fridovich-Keil, B. Jones,
Mutual Information-Based Trajectory
Planning for Cislunar Space Object
References Tracking using Successive Convexifi-
cation, AIAA SCITECH 2024 Forum
[1] P. Anakhov, et al., Protecting Objects of (2024). doi: 10.2514/6.2024-0626.
Critical Information Infrastructure from [10] H. Zhang, et al., Quality Assessment for
Wartime Cyber Attacks by DIBR-synthesized Views based on
Decentralizing the Telecommunications Wavelet Transform and Gradient
56
Magnitude Similarity, IEEE Transactions (2024) 1-19. doi: 10.1007/s11042-023-
on Multimedia (2024) 1–14. doi: 17914-1.
10.1109/tmm.2024.3356029. [20] Y. Su, et al., Enhancing Concealed Object
[11] Q. Qin, Y. Chen, A Review of retinal Detection in Active Millimeter Wave
Vessel Segmentation for Fundus Image Images Using Wavelet Transform, Signal
Analysis, Eng. Appl. Artif. Intell. 128 Process. 216 (2024) 109303. doi:
(2024) 107454. doi: 10.1016/j.sigpro.2023.109303.
10.1016/j.engappai.2023.107454. [21] J. Bhandari, D. Russo, Global Optimality
[12] M. Li, et al., Ftpe-Bc: Fast Thumbnail- Guarantees for Policy Gradient Methods,
Preserving Image Encryption Using Oper. Res. (2024). doi: 10.1287/opre.
Block-Churning, SSRN 4698446 (2024). 2021.0014.
[13] M. Vladymyrenko, et al., Analysis of [22] K. Qian, H. Duan, Optical Counting
Implementation Results of the Platform of Shrimp Larvae Using Masked
Distributed Access Control System. in: K-Means and a Side Window Filter, Appl.
IEEE International Scientific-Practical Optics 63(6) (2024) A7-A15. doi:
Conference Problems of Infocom- 10.1364/ao.502868.
munications, Science and Technology [23] S. Lee, et al., Intensity Histogram-Based
(2019). doi: 10.1109/picst47496.2019. Reliable Image Analysis Method for
9061376. Bead-Based Fluorescence Immunoassay,
[14] V. Buriachok, V. Sokolov, P. Skladannyi, BioChip J. (2024) 1–9. doi:
Security Rating Metrics for Distributed 10.1007/s13206-023-00137-9.
Wireless Systems, in: Workshop of the [24] J. Redmon, et al., You Only Look Once:
8th International Conference on Unified, Real-Time Object Detection,
“Mathematics. Information Tech- IEEE Conference on Computer Vision
nologies. Education:” Modern Machine and Pattern Recognition (2016) 779–
Learning Technologies and Data Science, 788. doi: 10.1109/CVPR.2016.91.
vol. 2386 (2019) 222–233. [25] T. Zhang, et al., The Design and
[15] D. Clayton‐Chubb, et al., Metabolic Implementation of a Wireless Video
Dysfunction‐Associated Steatotic Liver Surveillance System, 21st Annual
Disease in Older Adults is Associated International Conference on Mobile
with Frailty and Social Disadvantage, Computing and Networking (2015) 426–
Liver Int. 44(1) (2024) 39-51. doi: 438. doi: 10.1145/2789168.2790123.
10.1111/liv. 15725. [26] M. Nazarkevych, et al., The Ateb-Gabor
[16] E. Mira, et al., Early Diagnosis of Oral Filter for Fingerprinting, Conference on
Cancer using Image Processing and Computer Science and Information
Artificial Intelligence, Fusion: Practice Technologies, AISC 1080 (2019) 247–
Appl. 14(1) (2024) 293–308. doi: 255. doi: 10.1007/978-3-030-33695-
10.54216/fpa.140122. 0_18.
[17] L. Qiao, et al., A Multi-Level Thresholding [27] V. Sokolov, P. Skladannyi, A. Platonenko,
Image Segmentation Method Using Video Channel Suppression Method of
Hybrid Arithmetic Optimization and Unmanned Aerial Vehicles, in: IEEE 41st
Harris Hawks Optimizer Algorithms, International Conf. on Electronics and
Expert Syst. Appl. 241 (2024) 122316. Nanotechnology (2022) 473–477. doi:
[18] Z. Wang, et al., Fibrous Whey Protein 10.1109/ELNANO54667.2022.9927105.
Mediated Homogeneous and Soft- [28] M. Nazarkevych, Data Protection Based
Textured Emulsion Gels for Elderly: on Encryption Using Ateb-Functions,
Enhancement of Bioaccessibility for XIth International Scientific and
Curcumin, Food Chemistry 437(1) Technical Conference Computer
(2024) 137850. doi: 10.1016/j. Sciences and Information Technologies
foodchem.2023.137850. (2016) 30–32. doi: 10.1109/STC-
[19] J. Gao, Y. Huang, FP-Net: Frequency- CSIT.2016.7589861.
Perception Network with Adversarial [29] M. Medykovskyy, et al., Methods of
Training for Image Manipulation Protection Document Formed from
Localization, Multimed. Tools Appl. Latent Element Located by Fractals, Xth
57
International Scientific and Technical
Conference “Computer Sciences and
Information Technologies” (CSIT)
(2015) 70–72. doi: 10.1109/STC-
CSIT.2015.7325434.
[30] V. Sheketa, et al., Formal Methods for
Solving Technological Problems in the
Infocommunications Routines of
Intelligent Decisions Making for Drilling
Control, IEEE International Scientific-
Practical Conference Problems of
Infocommunications, Science and
Technology (2019) 29–34. doi:
10.1109/PICST47496.2019.9061299.
[31] V. Sheketa, et al., Empirical Method of
Evaluating the Numerical Values of
Metrics in the Process of Medical
Software Quality Determination,
International Conference on Decision
Aid Sciences and Application (2020) 22–
26. doi: 10.1109/DASA51403.2020.
9317218.
[32] N. Boyko, N. Tkachuk, Processing of
Medical Different Types of Data Using
Hadoop and Java MapReduce, in: 3rd
International Conference on Informatics
& Data-Driven Medicine Vol. 2753
(2020) 405–414.
[33] N. Boyko, et al., Fractal Distribution of
Medical Data in Neural Network, in:
IDDM Vol. 2448 (2019) 307–318.
[34] I. Tsmots, et al., The Method and
Simulation Model of Element Base
Selection for Protection System
Synthesis and Data Transmission, Int. J.
Sensors Wireless Commun. Control
11(5) (2021) 518–530. doi:
10.2174/22103279109992010221946
30.
58