For authors
Search
Search Results
##search.searchResults.foundPlural##
-
MULTIMODAL DATA FEATURE EXTRACTION METHOD FOR NETWORK ATTACK CLASSIFICATION
A.V. Balyberdin6-162025-07-24Abstract ▼An intrusion detection system (IDS) is an important component of corporate data network (CDN) protection. IDS analyzes network traffic and detects network attacks. Depending on the detection methods, IDS can be classified into the following types of systems: signature-based analysis systems, anomaly detection systems (ADS), and hybrid systems combining the aforementioned approaches. Recently, anomaly detection systems (IDS) have been actively developing. For anomaly detection systems, network attacks are anomalous behavior of network traffic consisting of a set of features or event attributes. Modern IDS are based on machine and deep learning methods, and therefore the detection of network attacks and anomalies is formulated as a classification and clustering problem. To solve these problems, methods for optimizing the feature space of network traffic are required. The aim of the work is to develop a feature extraction method based on a multimodal approach to representing network traffic data for classifying network attacks. The paper considers the analysis of relevant studies on feature extraction methods from various fields. The objective of the study is to improve classification efficiency using a multimodal representation of network traffic features. The result of the work is a method for extracting data features based on two modalities: a spectral representation of network traffic features and an image feature matrix. The novelty of the presented method lies in the application of the windowed Fourier transform method for network traffic events, followed by the calculation of spectral features for discrete signals, as well as the transformation of data features into an image matrix and its expansion to optimize the feature space using a convolutional neural network (CNN). Evaluation of the multimodal method showed that this method increased the classification accuracy for unbalanced classes of network attacks
-
CONTROL OF A MOBILE ROBOT ON BASE OF NEURAL NETWORK FOR THE PATH PLANNING IN UNMAPPED OBSTRUCTED ENVIRONMENT
А. К. Farhood99-1142022-01-31Abstract ▼In this work, a neural network of deep learning of a special structure is used. The neural
network allows a mobile robot to move without encountering obstacles in an unknown environment.
The main problems that the efforts of researchers in the field of neural network traffic planners
are aimed at solving are improving the performance of neural networks, optimizing their
structure and automating learning processes. The main result of this article is a new iterative algorithm
for developing a training set. At the first iteration, the initial training set is developed and
the initial training of the neural network is performed. In the following iterations, the neural network
trained at the previous stage is used as a filter for the following training sets. The filter selects
trajectories with collisions caused by neural network errors. During the learning process, the
number of convolutional and fully connected layers increases iteratively. Thus, the proposed algorithm
makes it possible to develop both a training set and a neural network architecture. Training
results are compared for filtered and unfiltered sets. The high efficiency of filtering has been confirmed,
as a result of which the distribution of examples in the training sample changes. The algorithm
can be used to develop a planning block for a mobile ground control system. The article
provides an example of training a neural network in a Matlab modeling environment. In the example,
five iterations of training were carried out, during which an accuracy of more than 90% was
achieved. This accuracy was obtained using the collected statistics on the movement of the mobile
robot in a randomly generated environment. The density of filling the environment with obstacles
was up to 40%, which corresponds to urban conditions. The comparison of neural network planners
trained using the proposed iterative procedure and with conventional training is carried out.
The comparison showed that the use of an iterative procedure increases the accuracy of planning
up to 12-15%. At the same time, the initial volume of the resulting sample is reduced several times
due to the applied filtering. -
ANALYSIS OF ENCRYPTED NETWORK TRAFFIC BASED ON ENTROPY CALCULATION AND APPLICATION OF NEURAL NETWORK CLASSIFIERS
V.A. Bukovshin, P.A. Chub, D.A. Korochentsev, L.V. Cherkesova, N.V. Boldyrikhin, O.A. Safaryan2021-02-13Abstract ▼Network traffic analysis allows you to solve many problems, such as: determining the pattern
of data transmission over the network, collecting statistics on the use of web applications,
monitoring and further researching network load, identifying potential malicious software and
network attacks, etc. 40% of Internet traffic belongs to unknown applications. This suggests that
for the area of network traffic analysis, the task of classifying applications has acquired particular
importance. Improvements in software in the field of network technologies have contributed to the
discovery of serious vulnerabilities in the implementation of some network protocols, namely TCP
and HTTP. By using network traffic analyzers, an attacker gained access to the contents of data
packets transmitted over the network. However, with the increasing qualifications of the information
community in the field of computer security, as well as with the development of network
technology standards, the analysis of network traffic has become noticeably more complicated.
The increased use of mathematical methods for protecting information, such as symmetric and
asymmetric cryptographic protocols, has led to the fact that most approaches to the analysis of
network traffic have lost their meaning and are no longer used. Therefore, the search for new
solutions to the problem of classifying network traffic, taking into account the possibility of its
encryption, is relevant. The article is devoted to the description of a new mixed approach to the
analysis of network traffic, based on the combined use of information theory and machine learning
algorithms. It also provides a comparative analysis of the proposed method with existing approaches
based on both information theory and machine learning. The aim of the research is to
develop an algorithm based on an intelligent approach to the analysis of network traffic. The proposed
algorithm is based on calculating entropy and using neural network classifiers. Research
objectives include: theoretical substantiation of the proposed approach in the field of information
theory, as well as machine learning algorithms; carrying out a structural description of the implemented
algorithms for calculating entropy and classifying applications that generate encrypted
traffic; comparative analysis of the proposed algorithm with existing approaches to the analysis of
encrypted network traffic. The result of the research is a new algorithm that allows classifying
various types of encrypted traffic with a high degree of reliability. -
ASSESSMENT OF INFLUENCING FACTORS AND FORECASTING OF POWER CONSUMPTION IN THE REGIONAL POWER SYSTEM, TAKING INTO ACCOUNT ITS OPERATING MODE
N.K. Poluyanovich, М. N. Dubyago2022-05-26Abstract ▼The article is devoted to the research of the assessment of influencing factors and forecasting
of power consumption in the regional power system, taking into account its operating modes.
The analysis of existing methods of forecasting energy consumption is carried out. The choice of a
forecasting method using an artificial neural network is justified. An algorithm for creating a neural
network for short-term prediction of electrical load is considered. The relevance of the work is
due to the requirements of the current legislation for forecasting electricity consumption in order
to solve the problem of maintaining a balance of power between the generating side and the consumption
of electric energy. At the same time, one of the main tasks related to the generation of
electric energy and its consumption is the task of maintaining a balance of capacities. On the one
hand, with an increase in the planned load, interruptions in the supply of electricity may occur, on
the other hand, a decrease in electricity consumption will also lead to a decrease in the efficiency
of power plants, and ultimately to an increase in the cost of electricity both for the wholesale electricity
market and for the end user. The developed neural network model reduces the task of shortterm
forecasting of power consumption to the search for a matrix of free coefficients by training
on available statistical data (active and re-active power, ambient temperature, date and index of
the day). The received NS model of short-term forecasting of power consumption of a section of
the district 10 kV electric grid takes into account the factors: – time, - meteorological conditions,
– disconnections of individual power supply lines of cottages, – operating mode of electricity consumers.
Predictive estimates of the power consumption of the power system have been obtained
based on the data of the electricity consumed by the outdoor temperature, the type of day, etc. The
model for predicting the magnitude of the consumed active and reactive power is quite workable,
but at this stage still has a fairly high level of forecasting error. To improve the accuracy of forecasting,
it is necessary to increase the database that makes up the training sample, because at the
moment the available data cover a time period of only 3–4 months. The results of the analysis
showed that forecasting reactive power consumption causes the greatest difficulties. -
USING FAST PROTOTYPING FACILITIES FOR IMPLEMENTATION OF A CONVOLUTION NEURAL NETWORK ON A FPGA
V. V. Bakhchevnikov , V. A. Derkachev , A. N. Bakumenko2020-10-11Abstract ▼Research in the field of artificial intelligence is carried out with increasing interest every
year. The fields of application of artificial intelligence are quite extensive: automation, analysis of
a large amount of data, smart home technology, machine vision, etc. Artificial intelligence technologies
are based on the use of artificial neural networks, which are based on the principles of
the animal nervous system. In this case, the actual issue is the implementation of artificial neural
networks on various software and hardware platforms: programmable logic integrated circuits of
the FPGA type (Field Programmable Gate Array), on special purpose integrated circuits (Application-
Specific Integrated Circuit, ASIC), GPU, CPU etc. FPGA performs best in low-power mobile
systems. ASIC demonstrates the highest performance at a fairly high development cost.
The problem of rapid prototyping of projects based on the use of artificial neural networks for
FPGAs using conventional methods (using HDL languages, HDL encoders, graphic programming)
is that either such a project is complex and time-consuming to debug (HDL languages), or
the resulting code is not optimal (HDL encoders), or the duration of the project development and
the complexity of reconfiguring the neural network (graphical programming) are high. Therefore,
in the framework of this work, an effective method for designing fully connected and convolutional
neural networks for their implementation on FPGAs using the Xilinx System Generator for DSP
and Matlab / Simulink package is considered. Artificial neural networks generated in this way are
easily reconfigurable and allow solving the following problems: image recognition, optimal filtering
(for example, for problems of subsurface radar). -
QUANTUM DEEP LEARNING OF CONVOLUTIONAL NEURAL NETWORK USING VARIATIONAL QUANTUM CIRCUIT
S.М. Gushanskiy, V. Е. Buglov167-1772021-10-05Abstract ▼Quantum computing in general and quantum deep learning represent a promising field re-lated to the research of modern methods and algorithms of quantum computing used for the pur-pose of teaching and developing new architectures of artificial neural networks. Recently, there has been a trend that research conducted in the field of quantum deep learning is becoming in-creasingly widespread among specialists. This can be explained by the fact that it has been estab-lished that quantum circuits are capable of functioning like artificial neural networks, while demonstrating the best results in solving several tasks, including, for example, the actual task of classifying objects in an image or in a video stream. Thanks to the rapid development of quantum computing in the field of deep learning, optimal solutions have been found for such urgent prob-lems as the vanishing gradient problem, finding a local minimum, improving the efficiency of large-scale parametric machine learning algorithms, eliminating decoherence and quantum er-rors, etc. Within the framework of this work, the process of functioning of a quantum variational scheme is described, its main characteristics are established, and disadvantages are identified. The key features of quantum computing, on which the process of implementing quantum deep learning with the reinforcement of a convolutional neural network is based, are also analyzed. In addition, quantum deep learning of a convolutional neural network has been carried out using a variational quantum scheme, which leads to an increase in the performance of a convolutional neural network in solving the problem of image processing, namely its classification, using a quantum computing environment. The relevance of this article consists in the implementation of a quantum deep learning algorithm with the reinforcement of a convolutional neural network for image processing, as well as the great importance of the subject of this study for the future devel-opment of quantum computing devices that can be used in artificial intelligence systems, etc., which corresponds to the priority direction of the development of domestic science
-
THE METHOD OF ESTIMATION POSITIONS OF THE UAVS BY MEASURING THE DISTANCES BETWEEN ELEMENTS OF THE GROUP
V.A. Kostjukov, M.Y. Medvedev, V.K. Pshikhopov, E.Y. Kosenko2021-04-04Abstract ▼Currently, the active use of groups of robots has begun to solve a number of tasks for civil
and military purposes. In this regard, problems arise associated with group management, the organization
of reliable communication channels and ensuring the effective functioning of the group
with limited energy resources. When solving the problem of optimizing energy consumption, the
problem of increasing the efficiency of interaction of the elements of the group with stationary
charging stations arises. This problem can only be solved by considering an integrated system,
which includes robots and charging stations. Centralized management of such a system is justified
in the case of a small number of its elements. However, with an increase in the number of elements
in a group, the complexity of management increases, so a combination of centralized and decentralized
management methods becomes a higher priority solution. The complex of problems of
decentralized management of such a group includes the task of organizing the optimal interaction
of its elements in order to achieve the goal of its functioning. When organizing energy exchange
between robots and charging stations, solving this problem plays a key role in optimizing energy
consumption. In this article, the concept of the interaction of mobile and stationary objects is developed,
which implies the possibility of each agent choosing an appropriate companion for interaction.
This choice is made taking into account the current state of the system and the assessment
of the history of interaction results. The developed concept is detailed for a system that includes
UAVs and their recharging stations. An algorithm is proposed for the decentralized selection of
pairs of interacting elements "UAV - charging station" based on two indicators - the energy efficiency
of the charging process, and the time spent by the UAV to reach the target point. Both indicators
are taken into account when choosing the weights assigned to each charging station as its
degrees of efficiency. Also, these indicators are included in the optimized quality criterion. An
optimization procedure has been developed, the result of which is the number of the charging station
that is most suitable for a given mobile object for interaction. -
VIBRATION MONITORING OF INTERNAL COMBUSTION ENGINE
A.V. Logunov, A.L. Beresnev2022-01-31Abstract ▼The work is devoted to the problem of diagnostics of automotive internal combustion engines.
The problem of monitoring the state of internal combustion engine is now most relevant due
to the increase in the number of cars and the tightening of environmental requirements. In the
work the consequences of operation of faulty internal combustion engine are considered. The purpose
of the work is to justify the choice from existing diagnostic methods of such a method, which
can help to detect the fault most accurately and quickly. For this purpose, the work details modern
diagnostic tools, highlights the principles of work, advantages and disadvantages. With the advent
of modern technologies, the long-known method of estimating the state of internal combustion
engine by sound can become the most advanced, as the human factor is excluded, for signal processing
the computational technique of analysis of the audio spectrum in which is carried out with
the help of artificial neural networks is used. The use of artificial neural networks for sound spectrum
analysis has found application in speech recognition and for diagnosis of respiratory system
diseases. The article considers mechanisms that are capable of generating sound signals during
internal combustion engine operation, some of them are phased, i.e. they are tied to operating
cycles, some are not phased. The proposed diagnostic technique allows to distinguish "useful"
sounds from the total number of internal combustion engine noises, after comparative analysis to
point to the node the sound of which differs from the reference, serviceable one. Scientific novelty
consists in the fact that the diagnostic process becomes automated, all sounds captured by sensors
are processed in a computer or a special scanner, the display shows information about the condition of certain nodes, unlike traditional methods where the diagnosis is carried out visually or by
ear. This increases diagnostic accuracy and reduces overall labor intensity by avoiding partial or
complete engine disassembly -
ALGORITHM FOR TRAINING THE ARTIFICIAL NEURAL NETWORK OF FACTOR PREDICTING THE POWER CABLE LINES INSULATING MATERIALS LIFE
N.K. Poluyanovich, M. N. Dubyago2021-07-18Abstract ▼The article is devoted to the research of thermofluxtual processes in accordance with the
theory of thermal conductivity for solving the problems of factor prediction of the residual life of
insulating materials based on the non-destructive temperature method. The relevance of the task of
developing algoritma for predicting the temperature of SCL cores in real time based on the data of
the temperature monitoring system, taking into account the change in the current load of the line
and external heat removal conditions, is justified. The experimental method revealed the types of
artificial neural networks, their architecture and composition, which provide maximum prediction
accuracy with a minimum set of significant factors. A neural network has been developed to determine
the temperature regime of the current-carrying core of the power kawhite. The minimum
set of significant factors and the dimension of the input training vector is determined, which provides
the versatility of the neural network prediction method. A neural network for determining the
temperature mode of the current-carrying core is designed to diagnose and predict the electrical
insulation (EI) life of a power cable. The model allows assessing the current isolation state and
predicting the residual resource of the SCL. Comparative analysis of experimental and calculated
characteristics of learning algorithms of isostic neural is carried out. It has been found that the
proposed algorithm of artificial neural network can be used for prediction of current-carrying
core temperature mode, three hours in advance with accuracy up to 2.5% of actual value of core
temperature. The main field of application of the developed neural network for determining the
temperature mode of the current-carrying core is in di-agnostics and predicting the electrical
insulation (EI) life of the power cable. The development of an intelligent system for predicting the
temperature of the LCS core contributes to the planning of the operation modes of the electric
network in order to increase the reliability and energy efficiency of their interaction with the integrated
energy system. -
APPLICATION OF THE NEURAL NETWORK APPROACH TO DIAGNOSE THE INTERNAL COMBUSTION ENGINE OF VEHICLES
А. V. Logunov, А. L. Beresnev2022-04-21Abstract ▼The work is devoted to the problem of diagnosing the internal combustion engine of vehicles
this problem is now the most relevant due to the constant growth of the car fleet and the tightening
of requirements for safe operation. Timely and accurate control of the internal combustion engine
is able to prevent the failure of entire vehicle assemblies, as well as to avoid such serious consequences
as a traffic accident. With the advent of modern technologies the long-known method of
engine condition estimation by sound can become the most advanced, since the human factor is
excluded, for signal processing the computer technique is applied, the analysis of a sound spectrum
in which is carried out by means of artificial neural networks. The application of artificial
neural networks for analyzing the sound spectrum has found application in speech recognition and
for diagnosing diseases of the respiratory system. The article deals with the failure of one of the
main parts of internal combustion engine - the bearing. All possible types of bearing faults and the
reasons why they occur are presented. The nodes and mechanisms of the internal combustion engine
in which bearings are used are listed. The algorithm of the experimental part is described.
The experiment which includes transformation of the received sound signals into spectrograms
and extraction of features with the help of which the classification is carried out, is executed. The
executed experimental part has proved the possibility of diagnosing of the internal combustion
engine by means of artificial neural networks. Scientific novelty lies in the fact that the diagnostic
process becomes automated, all the sounds taken by sensors are processed in a computer or in the
future in a special scanner, the display shows information about the state of certain nodes, unlike
traditional methods where the diagnosis is carried out visually or by ear. Thus, the diagnostic
accuracy increases and the overall labor intensity decreases due to the exclusion of partial or
complete engine disassembly. -
CONTROL SYSTEM DESIGN AND AUTONOMY FOR TWO-WHEELED MOBILE ROBOT
А. А. Tkachenko, D.D. Devyatkin2022-04-21Abstract ▼Model Predictive Control is an advanced process control method that used while meeting a
set of constraints. From an engineering point of view, the MPC method of designing control systems
is attractive, because is relatively simple in design, including for solving complex production
problems. This method is similar to the classical synthesis of a control system based on a linearquadratic
controller (LQR). The key difference between MPC and LQR is that predictive control
solves the optimization problem within a sliding time horizon, while the linear quadratic method
used to solve the same problem over a fixed time window. The paper considers a method for constructing
two-wheeled mobile robot control system using Model Predictive Control. The process of
building a mathematical model of the mechanical system of the robot is given, as well as the linearization
of the resulting model is performed. The basic principles of constructing a control system
based on MPC for linear systems without external disturbances, as well as using an observer to
assess the state of the model under the influence of additive white Gaussian noises, are presented.
A variant of the synthesis of a control system with imposed restrictions on the input signal is considered.
Also presented is a method for determining the position of a two-wheeled robot in space
using a vision system, which is based on the use of a neural network. The architecture of the used
model is given, as well as a stereo camera, which used to build an image depth map. In addition to
the above, the work describes in detail the principle of the deep learning model – YOLOv3, which
based on several blocks of input data processing. A detailed description of the implementation of a
stereo camera in conjunction with an artificial neural network model using the Python programming
language and libraries for working with video data and a stereo camera is presented. -
VERIFICATION OF DYNAMIC BIOMETRIC PARAMETERS OF A PERSONALITY BASED ON A PROBABLE NEURAL NETWORK
Y.A. Bryuhomitsky2021-01-19Abstract ▼Biometric identity verification is used primarily for access to computer and mobile systems, as
well as for remote (voice) verification. In fact, the most widespread systems are biometric verification
systems based on a fixed passphrase, which are quite simple to implement, but very vulnerable to
attacks of reproduction of a compromised short text. To eliminate this drawback, it is proposed to
carry out identity verification using a text that is arbitrary in terms of volume, content and language
(text-independent biometric verification). This paper proposes a generalized approach to solve the
problem of identity verification by dynamic biometric parameters of different modality (keyboard
writing, handwriting, voice). The presentation of dynamic biometrics signals is carried out by converting
them into a sequences of information units, each of which contains the same number of counts
of biometric signal of corresponding modality. The solution to this problem is carried out by monitoring
the degree of concentration of closely located information units (clusters) at certain points of the
multidimensional feature space. Such control is implemented on a probabilistic neural network thatstatistically evaluates the probability density of the distribution of information units in the corresponding
clusters with the subsequent determination of the total probability density for the entire
class of objects. The advantages of the proposed approach are: generalization of substantially different
methods of text-independent identity verification by dynamic biometric parameters of different
modality; the ability to make a verification decision for a fixed time of receipt of biometric data, determined
by the size of the model used; the ability to set the verification accuracy by changing the
dimension of the layer of probabilistic network samples. The disadvantage of the proposed approach
is the need for software implementation of a large-scale neural network. However, this drawback is
quickly leveled with an increase in the productivity of computer technology. -
NEUROCOMPUTER CONTROL OF CABLE NETWORKS BANDWIDTH THROUGH ACCOUNTING AND CONTROL OF THEIR PARAMETERS
N.К. Poluyanovich, N. V. Azarov, М.N. Dubyago84-1032025-07-31Abstract ▼The article discusses a neurocomputer system for predicting the resource of a power cable
line (РCL) using neural network technologies. A hardware modular implementation of a
neurocomputer (NC) implemented on the basis of FPGA was selected. To solve the problem of
predicting thermal processes of РCL, it was decided to use a NeuroMatrix NM6404 digital
neurochip with a variable structure due to their high performance compared to power consumption,
a high degree of versatility. To predict the temperature conditions of the РCL, an artificial
neural network (INS) was developed to determine the current temperature regime for the currentcarrying
core of the РCL. The architecture of the INS for the implementation of the NC of the SCL
temperature prediction system has been selected, which allows for long-term prediction of РCL
temperatures in real time. The choice of the activation function of the INS neurons for the implementation
of the NC of the SCL temperature prediction system, which allows for a long-term forecast
of SCL temperatures without increasing the error with an increase in the forecast range. The
proposed neural network algorithm that predicts the characteristics of the electrical insulation of
the РCL, based on the sliding window method for predicting time series, was tested on a control
sample of experimental data not included in the sample for training the INS. Experimental studies
of the proposed adaptive forecasting method have been carried out, namely, an adaptive algorithm
has been developed and the prediction of thermal processes in the isolation of the SCL from the
load current has been performed. Analysis of the results showed that the longer the aging time, the
greater the temperature difference between the original and aged sample. When analyzing the
data obtained, it was determined that the maximum deviation of the data obtained from the INS
during the experiment from the data in the training sample was less than 3%, which is quite acceptable
for this study result. It is shown that the developed methods and algorithms are elements
of an integrated power grid management system, and the developed adaptive NC model makes it
possible to assess the current state of insulation and predict the remaining life of the РCL -
DEVELOPMENT OF A METHOD FOR PERSONAL IDENTIFICATION BASED ON THE PATTERN OF PALM VEINS
V.А. Chastikova, S.А. Zherlitsyn2022-11-01Abstract ▼The article describes the work on the creation of a neural network method for identifying
a person based on the mechanism of scanning and analyzing the pattern of palm veins as a biometric
parameter. As part of the study, the prerequisites, goals and reasons for which the deve lopment
of a reliable biometric identification system is an important and relevant area of activity
are described. A number of problems are formulated that are inherent in existing methods for
solving the problem: the graph method and the method based on calculating the distance expressed
in various interval metrics. The description of the principles of their work is given.
The tasks solved by personal identification systems are formulated: comparison of the subject of
identification with its identifier, which uniquely identifies this subject in the information system.
A mechanism for reading a pattern of veins from the palm of the hand, developed for analyzing
an image obtained with a digital camera sensitive to infrared radiation, is described. When the
palm is in the frame, illuminated by the light of the near infrared range, the image obtained
from the camera becomes noticeable pattern of veins, vessels and capillaries that lie under the
skin. Depending on the organization, the identification system may, based on the provided identifier,
determine the appropriate access subject or verify that the same identifier belongs to the
intended subject. Three methods for further analysis of biometric data and personal identification
are given: approaches based on categorical classification and binary classification, as well
as a combined approach, in which identification is first used by the first method, and then, by
the second, but already for a known access identifier defined on the first stage. The resulting
architecture of the neural network for the categorical classification of the vein pattern is pr esented,
a method for calculating the number of model parameters depending on the number of
registered subjects is described. The main conclusions and experimental measurements of the
accuracy of the system when implementing various methods are presented, as well as diagrams of
changes in the accuracy of models during training. The main advantages and disadvantages of the
above methods are revealed. -
CLASSIFICATION OF RADAR IMAGES OF MULTI-ROTOR UNMANNED AERIAL VEHICLES USING THE YOLO11 ALGORITHM
V.А. Derkachev171-1802025-07-24Abstract ▼This article discusses a classifier of radar images of unmanned aerial vehicles based on a neural network built on the YOLO algorithm version 11. Solving the problem of detecting and classifying unmanned aerial vehicles has become one of the priority tasks at present. The increase in the number of modifications of unmanned aerial vehicles greatly complicates the use of statistical classification methods, which requires the use of new approaches to solving the classification problem. The development of neural network methods, simultaneously with an increase in the performance of computers for training, on the one hand, and embedded solutions, on the other, allows for the classification of aircraft using radar images in real time. The use of the YOLO11 algorithm allows, in addition to determining the class of the target, to estimate the range to the observed object. The use of radar images is justified due to the fact that visual observation is not always possible due to difficult weather conditions and darkness. To train the neural network, it is proposed to use a set of radar images obtained using the author's model of data generation with an arbitrary configuration of unmanned aerial vehicles. The neural network of the Detection YOLO11s class (9.4 million parameters) was trained on a sample of radar images of two classes, a total of 8192. As a result of training, an accuracy of 0.99 was obtained for classification in 2 classes of objects (on test model data). Tests were conducted using natural data taken using the TI IWR1642 millimeter-range radar system, as a result of which error-free classification of objects on a small sample was achieved
-
ALGORITHM FOR TRAINING DATA PREPARATION OF CONVOLUTIONAL NEURAL NETWORKS FOR LETTER AND CHARACTER RECOGNITION
D.А. Bezuglov , М.S. Mishchenko , S.E. Mishchenko134-1442025-07-24Abstract ▼The accuracy of text image recognition remains limited in practice. This is due to the fact that the alphabet of symbols can include lowercase and uppercase letters with a similar font, as well as composite characters formed from several simpler characters. To solve this problem, the character recognition system is supplemented with semantic or structural analysis systems, which significantly complicates the information system for text recognition. Currently, convolutional neural networks are widely used for recognizing single characters, for which a database with images of recognized characters is used for training. The paper proposes an algorithm characterized in that the image of a single character for a training sample includes fragments of characters that can be located in a line in close proximity to the recognized character. This allows you to expand the set of images for training and additionally include information in the image about the placement of the symbol in the string, its relative size and whether this symbol is composite. The formation of images for the training sample simulates the process of segmentation of a symbol by brightness, which is usually used when selecting a symbol for further recognition.
At the same time, the size of the symbol is estimated, it is supplemented with images of neighboring symbols, and then the size of the area, the image that will be placed in the training sample, is estimated. The resulting image is scaled and cropped in such a way that images of a given size are received at the input of the neural network. In the work, to recognize the alphabet of symbols, including uppercase and lowercase characters of the Russian and English alphabets, numbers, symbols and punctuation marks, it is proposed to use a variety of convolutional neural networks, each of which is trained to recognize one character. The symbol is selected by comparing the responses of all neural networks and selecting the maximum response. The proposed algorithm for training data preparation is compared with a well-known algorithm based on the use of images of single characters. It is established that the proposed algorithm for preparing data for training provides an increase in the accuracy of recognizing the alphabet of 138 characters by more than two times. -
COMBINING SEGMENTATION, TRACKING, AND CLASSIFICATION MODELS TO SOLVE VIDEO ANALYTICS PROBLEMS
V.D. Matveev, А. Е. Arkhipov, I. S. Fomin2025-04-27Abstract ▼The task of detecting obstacles in front of a mobile robot has been successfully solved long ago using
laser and ultrasonic sensors. However, obstacles that are not detected by these types of sensors may endanger
the safety of the robot. To detect them in the work, it is proposed to use a technical vision system
(STZ), the information from which is processed by a semantic segmentation neural network, which returns
the mask of the obstacle on the frame and its class. The basis for such a network was the SAM universal segmentation
network, which requires further development to be applied to the semantic segmentation task.
The peculiarity of this network is its universal applicability, that is, the ability to select any objects in any
filming situation. At the same time, SAM does not predict the semantics of the object. In this paper, an additional
module is proposed that makes it possible to implement semantic segmentation by classifying the features
of the selected objects. The possibility of using such a module to solve the problem of supplementing the
network output with new information is substantiated. The classification result is then fed into the same filtering
algorithm as the masks to ensure consistency between the result of the universal network and the complementary
module. After integrating the module with the model, a new semantic segmentation model was
obtained, called RTC-SAM in the work. It was used to perform semantic segmentation of a publicly available
dataset with images of an open area. The 45% result obtained by the IoU metric exceeds the result of existing
methods by 13%. The images of the results of using the new network shown in the work make it possible to
verify its performance. It also describes the testing of the developed solution with a study of the performance
of the developed model on a PC and a mobile computer. The algorithm on the mobile computer shows insufficient
speed to enter real-time mode – more than 3.5 seconds to process one frame. In this regard, one of
the directions of further research in the field of improving system performance. -
RESEARCH OF AN INTELLIGENT ADAPTIVE CONTROL ALGORITHM BASED ON THE REINFORCEMENT LEARNING METHOD
А. N. Karapeev, Е.Y. Kosenko, М. Y. Medvedev, V. K. Pshikhopov2025-04-27Abstract ▼An algorithm for adaptive control of a DC motor based on the use of machine learning technology
with reinforcement is proposed and investigated. An overview and brief analysis of the state of affairs in
the field of intelligent motor control systems is given. A mathematical model of the DC motor is presented,
and a structural scheme for training an intellectual agent is presented. An intelligent adaptive motor speed
control system is proposed. The DC motor is represented as a black box with the limited input and output.
The control system is based on a zero-order Q-learning algorithm. It is assumed that the output of the
intelligent agent is a control applied to the motor input. The intelligent system uses a tabular approximation
of the value of each of the control action. In this article, we study the effect of the discreteness of the
representation of state, the set of control effects used, the applied rewards, and the parameters of the
learning algorithm on the control error. The sensitivity of the control system to the parameters of the motor
and an unmeasured moment is investigated. Based on the results of the study, a modified algorithm is
proposed, which assumes the measurement or evaluation of the current of the motor stator. The control
algorithm provides robustness to parameters and external disturbance. Additionally, the approximation of
the control value function using polynomials and using a neural network are investigated -
OBJECT IDENTIFICATION METHOD FOR INTEGRATION WITH ROBOTIC SYSTEMS
N.М. Chernyshov, I. К. Romanova-Bolshakova2025-04-27Abstract ▼The aim of the research is to develop a methodology for identifying and determining the location of objects
under conditions of low visibility and potential changes in their shape, with a focus on extracting parts
created using selective laser sintering (SLS) from a powder medium. The study examines two fundamentally
different approaches to forming control algorithms for a robotic manipulator. The first approach, trust-based, is
based on the assumption of minimal displacement of the object during manipulation. The manipulator moves
along a trajectory calculated from a preliminary three-dimensional model without correction until the moment
of capture. This method is characterized by high operational speed and minimal computational costs. However,
it carries risks such as object deformation due to environmental resistance, displacement of the part upon contact
with the tool, and the inability to capture the object if it deviates significantly from its nominal position.
The second approach, cautious, involves the gradual removal of powder layers to visualize the object and adjust
the trajectory before capture. This method includes several stages: removing the top layer of the medium to
partially expose the part, analyzing data to refine the object's position, and constructing an adaptive trajectory
considering possible displacement. Special attention in the article is given to data generation for training neural
networks, which are used for object identification under noisy conditions. Two methods of artificial modeling of
powder coatings are considered. The primitive method involves expanding the vertices of a three-dimensional
model along their normals with the addition of random noise. The improved method proposes differentiated
powder distribution considering local surface curvature. Subsequent experimental results showed that training a
neural network using real data has low efficiency. Recognition accuracy ranged from 60% to 75%, which is
attributed to the small sample size and the influence of external factors such as lighting and interference. At the
same time, the use of synthetic data, prepared according to the methodology presented in the study, increased
recognition accuracy to 92%. The practical significance of the work lies in the development of a methodology
for searching, detecting, and identifying a part immersed in powder, which can be used to automate postprocessing
processes in industries utilizing selective laser sintering. The developed solutions are adapted for
integration into robotic systems operating under conditions of limited visibility. The proposed methods can be
scaled to a wide range of tasks in additive manufacturing and robotics, making them promising for implementation
in industrial processes. -
COMPREHENSIVE APPROACH TO INTRUDER RECOGNITION BASED ON VIDEO IMAGERY
А. Е. Kolodenkova , М. О. Bochkarev105-1132026-07-07Abstract ▼Intruder recognition in uncontrolled environments is a critical function of biometric-based security control systems (SCS), ensuring protection at facilities with large crowds. The accuracy of such systems can be improved by combining multiple biometric traits extracted from video imagery. However, processing video data involves several challenges, including viewpoint variation, occlusion, and selecting an appropriate feature fusion level. To resolve these challenges, a complete approach to intruder recognition is proposed. The approach is based on video preprocessing, the combination of two convolutional neural networks (CNNs), Dempster–Shafer theory, and the random forest method. These techniques fuse behavioral biometric features at the score level to classify video sequences. Gait and gestural behavior are selected as biometric modalities, as they can be captured without direct subject interaction. For the classification task, three classes of video data are defined: class 1 – person is not intruder, class 2 – a potential intruder, class 3 – intruder. The study also presents a general architecture for the proposed approach, along with a detailed description of its processing stages. The effectiveness of the approach is measured through experiments performed on the KTH dataset, which comprises six types of simple human actions performed by different subjects under different background conditions. Experimental results show that the proposed approach improves intruder recognition accuracy in uncontrolled environments, achieving an 87 % classification rate.
-
IMAGE MATCHING SYSTEM WITH USING INTUITIONISTIC FUZZY SETS
К.I. Morev286-2982026-04-29Abstract ▼This paper presents a fully learnable system for solving the problem of matching two images. All the main elements of the system are trainable, i.e. their final form corresponds to the target dataset on which the training was carried out. The fact that the system is trainable, the methods used in training and the architecture of the system allow using the system to solve a large number of various computer vision problems. The system consists of a convolutional neural network that serves both to extract key points and their descriptors, as well as a trainable matcher of the extracted key points based on their description and mutual arrangement in the observed scene. The used convolutional neural network processes full-size images and calculates both the location of interest points with pixel accuracy and the descriptors associated with them in a single forward pass. Matching key points is a separate step and is performed after the forward pass of the neural network. In the process of training the model for calculating the positions of key points and their descriptors, a method for forming a training sample is used, called homographic adaptation - an approach that helps to increase the repeatability and accuracy of detecting key points. The process of training the feature point detection model consists of obtaining new weights in the process of additional training of the base detector, which represents the initialization weights of the model. The final feature point detection model, trained on the universal MS-COCO image set using homographic adaptation, repeatedly outperforms the original base detector in terms of the number, reliability and repeatability of feature points, and also outperforms any other traditional corner detector based on classical approaches
-
AN ALGORITHM FOR CONTROLLING AN AUTONOMOUS UNDERWATER VEHICLE WHEN SEARCHING FOR A DESIGNATED BOTTOM OBJECT WITH THE INTEGRATED USE OF VARIOUS BOTTOM MONITORING TOOLS
V.S. Bykova , А.I. Mashoshin2026-04-29Abstract ▼The search for designated bottom objects is one of the most difficult tasks solved by the AUV, due to a number of factors, the main of which are: the variety of search objects (sunken submarines, surface ships, airplanes, helicopters, mines, underwater pipelines and communication cables, various underwater infrastructure), the need for integrated use for search for various bottom monitoring tools that differ in their physical principles of operation, resolution, and search performance. The purpose of the work, the results of which are presented in the article, was to develop an algorithm for managing the AUV when searching for a designated bottom object that meets these requirements, and to verify it using a digital polygon and a digital twin of the AUV. The probability of correctly attributing the detected bottom object to the search object was chosen as a criterion for choosing a bottom monitoring tool in each specific case. As a result, the logic of searching for a designated bottom object is as follows. The search for bottom objects is carried out using a tool with maximum search performance. When a bottom object is detected, the probability of its belonging to the search object is determined. If it exceeds the set high threshold, a decision is made to locate the designated bottom object. If it is less than the specified low threshold, it is decided that an extraneous object has been detected. In other cases, a decision is made on the need to examine the object with a higher resolution. The technology of classification of bottom objects based on the training of an artificial neural network trained using synthesized training material is described. The results of checking the effectiveness of the developed algorithm using a digital polygon and a digital twin of AUV are presented. The simulation of the developed algorithm showed that the integrated use of bottom monitoring tools increases the likelihood of a successful solution to the problem and reduces the time needed to solve it.
-
ESTIMATION OF THE SPATIAL POSITION OF AN ON-BOARD CAMERA BY COMPARING AERIAL IMAGES AND SATELLITE IMAGE DATA
А.Y. Budko , Т.А. Gaida , Z.А. Ponimash2026-02-27Abstract ▼The article describes a method for estimating the spatial position of an onboard camera of an aircraft. This method involves comparing aerial photographs and georeferenced remote sensing (RSS) data by using a neural network detector to detect stable spatiotemporal reference points in both datasets. This method then solves the well-known Perspective-n-Point (PnP) problem for estimating rotation and translation matrices that minimize the reprojection error based on the correspondences between 3D world points and 2D points of their projections onto the onboard camera matrix. This approach can be used to solve the pressing problem of aircraft localization in the absence of global navigation satellite system signals. Road intersections are selected as stable spatiotemporal reference points that are clearly visible in RSS data and aerial photographs. Other local semantic image patterns characteristic of a particular area may serve as an alternative. Since direct comparison of remote sensing and airborne images is difficult due to significant differences in shooting conditions, the use of robust landmark detectors based on artificial neural network (ANN) algorithms is proposed. To train the robust detector, a mixed dataset was created using satellite and airborne imagery. The mixed dataset was labeled using a 3D Gaussian function normalized to unity with a apex at the intersection center, the graph of which is projected onto a 2D mask of the training set. The parameters of the Gaussian function are calculated based on the radius of the circle enclosing the intersection. Using a normalized 3D Gaussian function with a apex at the geometric center of the intersection projection allows the network to predict the probability of each image pixel belonging to the intersection, with a maximum at the intersection center, which increases positioning accuracy due to more precise georeferencing of the landmark point in the global 3D dataset. A U-Net-type artificial neural network was trained as an intersection detector. A differentiable analog of the Dice metric was used as a training quality metric. AdamW, coupled with a CosineAnnealingLR cosine learning rate planner, was used as an optimizer. The final section of the paper presents the results of comparing satellite data and airborne imagery using the proposed method.
-
ALGORITHM FOR SIGN LANGUAGE RECOGNITION USING CONVOLUTIONAL NEURAL NETWORK
V.Е. Bondareva , Т.S. Chernomorova , А.V. Krivtsun , Abdulkarem Abeer2026-02-27Abstract ▼This paper addresses the problem of automatic recognition of Russian Sign Language (RSL) using computer vision and deep learning methods. The relevance of the study is driven by a steady increase in the number of people with hearing impairments: according to the World Health Organization, there are currently about 70 million deaf and hard-of-hearing individuals worldwide, and this number is projected to reach 630 million by 2035. The development of effective gesture recognition algorithms is an important direction for creating contactless human–machine interaction systems aimed at improving accessibility of information technologies and enhancing the quality of life for people with hearing disabilities. The aim of the study is to develop and experimentally validate an algorithm for real-time recognition of Russian Sign Language alphabet gestures in a video stream using a convolutional neural network. A specialized dataset was created, consisting of 430 images of hand gestures corresponding to the letters of the RSL alphabet, captured from different angles and under varying lighting conditions. The model was implemented using TensorFlow and Keras libraries, while integration with the video stream was performed using OpenCV and a marker-based hand tracking system. As a result of training and testing, the proposed model achieved a recognition accuracy of 99% on the test dataset. A comparative analysis with classical machine learning methods demonstrated the superiority of the convolutional neural network in terms of classification accuracy and robustness to external noise. The obtained results confirm the effectiveness of the proposed approach and its applicability for real-time systems intended for communication, educational, and rehabilitation applications, as well as for the development of advanced human–machine interaction interfaces.
-
ANALYSIS OF TRADITIONAL AND NEURAL NETWORK-BASED CONTROL METHODS FOR ELECTRIC DRIVES IN ROBOTICS AND PERSPECTIVES OF HYBRID APPROACHES
А. I. Tataurov , V.Е. Vavilov287-2982025-12-30Abstract ▼The objective of this study is to conduct a comparative analysis of traditional and neural network-based control methods for electric drives in robotics, with an emphasis on identifying their strengths and weaknesses, determining their areas of application, and assessing the prospects for the development of hybrid approaches. Effective control of electric drives is critically important for modern robotic systems, which must demonstrate high performance, reliability, and versatility in various application domains. Specifically, key challenges include high-precision trajectory tracking, energy-efficient control, robust control under uncertainties and disturbances, constraint-aware control, as well as synchronized and coordinated control of multiple electric drives. In this regard, optimizing the control of electric drives to ensure motion accuracy, energy efficiency, and adaptation to changing conditions becomes a top priority. To achieve this goal, the study systematizes and analyzes the characteristics and applications of traditional electric drive control methods, such as PID controllers, Kalman filters, sliding mode control, and model predictive control. It also examines key neural network-based approaches to electric drive control, including feedforward neural networks, recurrent neural networks, radial basis functions, neuro-fuzzy systems, and reinforcement learning. A comparative analysis of these methods is conducted to identify their advantages and limitations based on key parameters such as trajectory tracking accuracy, robustness to disturbances and uncertainties, adaptability to changing operating conditions, and computational complexity. Additionally, the study investigates and assesses the prospects for hybrid electric drive control methods that combine the reliability and control quality of traditional methods in linear and structured environments with the flexibility and adaptability of neural network-based methods in complex and dynamic robotic systems. The study’s key findings indicate that traditional electric drive control methods, such as PID controllers and sliding mode control, remain effective and preferable in linear and well-defined systems due to their simplicity and reliability. At the same time, neural network-based approaches demonstrate significant advantages in controlling complex nonlinear systems, as well as in uncertain conditions requiring adaptation to changing environments. Special attention is given to hybrid control methods, which integrate the strengths of both traditional and neural network-based approaches. These methods are regarded as the most promising and advanced direction, enabling the development of intelligent and robust electric drive control systems capable of operating efficiently in complex and dynamic environments.
