РУС
ENG
№ 4(39)'2025
Aims and Scope
Founder and Publisher
Editorial board
Licensing terms
Privacy Statement
Plagiarism policy
Publication ethics
Archiving Policy
Subscription
Instructions for authors
Reviewing proccess
Copyright Notice
License agreement
Article processing charges
All issues
Search
Contacts
Section
ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
A. I. Paramonov, K. A. Pavluchenko
MULTI-ASPECT ANALYSIS OF EMOTIONAL CONTEXT FOR TEXT DOCUMENT CATEGORIZATION
UDC: 004.891.3:004.4
DOI: 10.24412/2413-7383-2025-4-39-4-14
Language: Russian
Abstract: Рассматривается проблема интеллектуального анализа текстов с целью выявления их эмоционального контекста. Представлены итоги обзора современных подходов к задаче выявления эмоций в текстах, обозначены ключевые ограничения и проблемы при обработке сложных лингвистических конструкций. Выдвигается гипотеза: для повышения качества категоризации цифровых текстов по эмоциональному фону необходимо учитывать семантические и прагматические особенности их представления (саркастические высказывания, эмодзи, контекстуальные зависимости и другие лингвистические маркеры). Предложен метод многоаспектного анализа текста для категоризации документов на основе выявления их эмоционального контекста. Представлена многомодульная архитектура метода, основным компонентом которой является классификатор, построенный на
Keywords: Natural Language Processing, Sentiment Analysis, Emotional Context, Document Categorization, Transformer-based Models, RoBERTa.
A.G. Gurin, A.V. Grigoriev
COMPARATIVE ANALYSIS OF METHODS FOR ENHANCING THE EFFICIENCY OF PRODUCTION SYSTEMS FOR BIG DATA PROCESSING
UDC: 004.825
DOI: 10.24412/2413-7383-2025-4-39-15-26
Language: Russian
Abstract: An analysis of logical inference in production systems for big data processing tasks has been conducted. Despite the effectiveness of classical production systems, the combination of two approaches remains understudied: the use of AND-OR trees for intuitive logic representation and the application of tensor representations for performance enhancement. The research demonstrates the complementarity of these approaches, justifying the feasibility of creating an adaptive meta-algorithm that dynamically selects the method based on the characteristics of the problem being solved. For educational systems, such a hybrid approach is particularly promising, as it allows for combining high data processing performance with complete interpretability of the decisions made.
Keywords: production systems, AND-OR-trees, tensor representations, ontologies, education system.
D. I. Dzeboev
SEARCH OF UNITS IN TENSORIAL ALGEBRA
UDC: 512.554.1
Language: Russian
Abstract: The paper considers the analytical problem of finding unit elements in tensorial algebra — a generalized algebraic system of bilinear vector multiplication defined by an algebra tensor. It is shown that the action of any tensorial number in the algebra can be represented as a linear operator, and that the search for the unit element reduces to solving a system of linear equations where the contraction of the algebra tensor with an unknown vector yields the identity matrix. General formulas for the left and right units are derived in terms of Moore –Penrose pseudoinverses. It is proven that if at least one right and one left unit exist, they coincide, forming a unique two -sided (commutative) unit that renders the algebra unital.
Keywords: tensorial algebra, algebra tensor, algebraic unit, right and left unit, Moore– Penrose pseudoinverse, linear operator, unital structure
V. O. Eliseev1, V. I. Bondarenko2, A. Y. Maksimova1
DEVELOPMENT OF AN INTELLIGENT ASSISTANT FOR AN EDUCATIONAL INSTITUTION
UDC: 004.912
DOI: 10.24412/2413-7383-2025-4-39-36-48
Language: Russian
Abstract: The article proposes the method for development intelligent assistants for educational institutions based on Large Language Models (LLMs). This enables users to receive informational responses about an institution within a min ute, as opposed to the hours or days typically required. The method was implemented to create an assistant for Donetsk State University (DonSU). The study involved selecting a method for integrating corporate data into an LLM, followed by the direct collection and preprocessing of this data. The final system is a standard Retrieval-Augmented Generation (RAG) pipeline, which utilizes a fine-tuned LLaMA-3.1-8b model as the generator and Giga-Embeddings-instruct as the retriever. To evaluate the performance of the generator model, we used token-level similarity metrics and assessed the semantic similarity between the assistant's responses and the target answers for a set of predefined scenarios. The source code for all implemented steps and the final version of the DonSU assistant are open-source what makes the proposed approach applicable for other educational institutions.
Keywords: large language model, natural language processing, question answering, Retrieval-Augmented Generation, LLAMA, fine-tuning.
O. A. Lyamtsev1,2, I. I. Maksimenko1,2
OVERVIEW AND CHALLENGES OF USING DEEP LEARNING FOR THREE-DIMENSIONAL ESTIMATION OF A PERSON'S POSTURE FROM A SINGLE IMAGE
UDC: 004.932.2
DOI: 10.24412/2413-7383-2025-4-39-49-59
Language: Russian
Abstract: The article considers the task of 3D human pose estimation from monocular image, provides a mathematical description, highlights the main approaches, problems in the task and their solutions, neural network architectures. Among neural network architectures, graph convolutional networks are the most popular, extracting local information from nearby joints and transformers, modeling global dependencies through the internal attention mechanism. Among the key problems in solving the task, we can distinguish: 1) depth uncertainty; 2) occlusion; 3) insufficient volume and diversity of training data.
Keywords: 3D human pose estimation, computer visi on, deep learning, neural networks, monocular image
Ye. S. Moroz, Y.S. Pikaliov
ANALYSIS OF SOTA-ARCHITECTURES FOR SEMANTIC AUGMENTATION
UDC: 004.932.2
DOI: 10.24412/2413-7383-2025-4-39-60-70
Language: Russian
Abstract: The paper considers semantic segmentation tasks in the context of applied computer vision and related problems such as object detection, instance segmentation, and panoptic segmentation. Lightweight and high-speed real-time architectures are analyzed in terms of the trade-off between accuracy and latency on the Cityscapes, CamVid, and UAVid datasets. Special attention is paid to data preparation issues, including augmentation, image annotation, preprocessing, and model interpretability.
Keywords: semantic segmentation, computer vision, object detection, instance segmentation, panoptic segmentation, data augmentation, cross-view geolocalization.
B.V. Pavlenko
AN APPROACH TO MULTIMODAL DATA FUSION IN TASKS OF REGRESSION AND CLASSIFICATION IN DRONE IMAGES
UDC: 004.932.2
DOI: 10.24412/2413-7383-2025-4-39-71-83
Language: Russian
Abstract: In this paper, a method of multimodal feature combination is proposed using images from an unmanned aerial vehicle (UAV) and encoded metadata. The applicability of the proposed approach to the tasks of UAV position regression and object classification in images with a minimal set of known metadata is considered. For regression, a mechanism for interpolating altitude values is introduced, which allows for a more accurate association between the visual representation of an image and its altitude, and provides greater stability to changes in altitude and camera rotation angles of the UAV. To regulate the contribution of textual modality, a method is used that ensures random ignoring of textual features and reduces overfitting on textual modality in addition to regularization by the feature alignment error and its coefficient of influence on the overall model error. The same approach is used for classification, with the comparison of several variants of textual templates. Experiments were conducted on the following datasets.
Keywords: multimodal neural networks, regression, classification, feature fusion, positioning, UAV.
Ustenko V.Y. Blizno M.V.
DATA AUGMENTATION METHODOLOGY FOR SEGMENTING KEY OBJECTS IN UAV AERIAL IMAGERY
UDC: 004.932.2
DOI: 10.24412/2413-7383-2025-4-39-84-97
Language: Russian
Abstract: In this work, the authors propose a method (StageAug) that uses a limited amount of machine-learning server resources while increasing robustness. Experiments are conducted on the synthetic SynDrone dataset (road, natural objects, buildings, obstacles, water, other objects) using the pretrained StripNet neural architecture designed for extracting high-level features. The key element is a staged augmentation pipeline: from mild photometric transformations to Copy -Paste of rare objects and 2×2/3×3 mosaic compositions. The experiments show that the proposed method stabilizes early epochs and achieves mIoU = 0.6264 and FWIoU = 0.9565. The results confirm the hypothesis that harmonizing augmentations and using the enhanced StripNet -FPN indeed help to reduce the synthetic -to-real gap and improve robustness.
Keywords: semantic segmentation, StripNet, FPN, SynDrone, UAV, aerial imagery, augmentation, mosaic.
Section
MATHEMATICAL MODELING, NUMERICAL METHODS AND SOFTWARE SYSTEMS
D. L. Ostapushchenko, A. E. Voronov, V. V. Chernykh, M. O. Petrenko
NUMERICAL SOLUTION OF THE AERODYNAMICS PROBLEM IN MODELING THE FLIGHT OF UNMANNED AERIAL VEHICLES WITH INTELLIGENT CONTROL SYSTEMS
UDC: 004.8+533.65.013.622
DOI: 10.24412/2413-7383-2025-4-39-99-111
Language: Russian
Abstract: Developing intelligent control systems for unmanned aerial vehicles, especially those using reinforcement learning methods, requires a physically adequate model of the controlled object. The work is devoted to the constructio n of a mathematical model of the process of air flow around an unmanned aerial vehicle, based on potential theory, theory and numerical methods for solving integral equations. This model is designed to determine the phase trajectory of an aircraft in real time at various stages of the creation of intelligent flight control systems.
Keywords: unmanned aerial vehicle, intelligent control systems, artificial intelligence, reinforcement learning, aerodynamics, integral equations, flight dynamics
V. S. Solod
ALGORITHM AND PROGRAM FOR OPTIMIZING THE ACCELERATED COOLING OPERATION MODE OF REINFORCEMENT BARS IN THE THERMAL UNIT OF THE "THERMEX" DESIGN
UDC: 621.771.07:004.42
DOI: 10.24412/2413-7383-2025-4-39-112-123
Language: Russian
Abstract: The objective of this study is to develop software tools for optimizing the thermal hardening process for reinforcing sections with minim al cooling water consumption while adhering to process constraints. This is the first time the optimization problem has been formulated taking into account the chemical composition of the steel and the constraints. The objective function is the yield strength, and constraints include permissible water pressures at the nozzles, the absence of water boiling in the section, and the presence of hydro transportation of the rolled product through the unit. A flowchart of the algorithm is shown, the program execution procedure is described, and examples of optimization modes are given.
Keywords: accelerated cooling, rebar section, mathematical model, algorithm, optimization program, yield strength.
Section
AUTOMATION AND CONTROL OF TECHNOLOGICAL PROCESSES AND PRODUCTION
Н. В. Водолазская, М.К. Вотинцев, Е. В. Копытова, А.И. Хиценко
плексных минеральных удобрений с созданием специального SCADA -интерфейса. Анализи-
UDC: 658.011.56
DOI: 10.24412/2413-7383-2025-4-39-125-136
Language: Russian
Abstract: An information management system is proposed for monitoring the production process of complex mineral fertilizers with the creation of a special SCADA interface. The requirements for it are analyzed, including the need to display dynamic process parameters, t he status of technological equipment and emergency signals. This interface provides a display of the current equipment status, component dosing, granulation, product drying in real time, and offers the operator tools for managing the technological process.
Keywords: SCADA, production automation, complex mineral fertilizers, information management system, technological control.
Д.М. Логунова, А.М. Володина, Е.В. Курнасов, С.К. Кириллов, Н.Д. Тонконоженко
производстве SCADA- и MRP-системами
UDC: 004.896:681.518.5+65.011.56
DOI: 10.24412/2413-7383-2025-4-39-137-148
Language: Russian
Abstract: The article presents an information and analytical system that provides automated quality control of wooden blanks in the manufacture of guitars using machine vision technologies and neural networks. The key components of the system focused on the detection of defective products and the automated selection of pairs of blanks with a matching wood pattern are considered. The results of the image analysis module, which uses neural networks with YOLO and EfficientNet architectures to dete ct product defects, are presented. It is shown that the proposed method of product quality control will significantly eliminate the subjective human factor in identifying defects and selecting guitar blanks. Algorithms and fragments of the software implementation of the system are presented, which can be integrated with SCADA and MRP systems operating in production.
Keywords: neural networks, information and analytical systems, production automation, woodworking industry, machine vision, product quality control.
Section
Методы и системы защиты информации, информационная безопасность
Ya. I. Rushechnikov, I. A. Tretiakov, V. V. Danilov
ADAPTIVE AUTOMATED SYSTEM FOR SUPPRESSION OF ELECTROMAGNETIC SIDE EFFECTS AND INTERFERENCE
UDC: 681.5:621.37:004.056
DOI: 10.24412/2413-7383-2025-4-39-150-161
Language: Russian
Abstract: This paper describes a comprehensive technique for searching for and subsequent suppression of detected electromagnetic side effects and interference signals within a single feedback suppression software and hardware package. The proposed technique will ma ke it possible to use the most accessible software-defined platforms for scanning radio placement, which will reduce the final cost of an adaptive automated system for suppressing such signals.
Keywords: ASRS, TEMPEST, radio monitoring, model signal, SDR, signal suppression, SI5351 generator, coincidence coefficient, correlation
Section
Роботы, мехатроника. Робототехнические системы
М. Л. Абдулвадудов, К.Л. Вахидова, С.М. Ногамирзаев
сетей на примере детектора объектов YOLO V5s. Предложенные подходы к распараллеливанию
UDC: 004.032.26
DOI: 10.24412/2413-7383-2025-4-39-163-172
Language: Russian
Abstract: The aim of this work is to study methods for improving the efficiency of neural network algorithms on heterogeneous RoboDeus systems-on-a-chip (SoCs), aimed at application in robotic systems. An analysis of the RoboDeus SoC architecture, including the Elcore50 DSP core, is presented, and optimization of convolutional neural networks is demonstrated using the YOLO V5s object detector as an example. The proposed approaches to parallelizing computations, efficient use of memory, and accumulator registers enable significant performance gains when executing computer vision tasks in real time. Experimental results show that the us e of multi -core processing based on RoboDeus provides a frame rate of up to 33.3 FPS for the task of detecting objects in 640×640 images. Prospects for integrating optimized neural network algorithms into autonomous robotic systems for navigation, environment recognition, and interaction with objects are discussed.
Keywords: heterogeneous computing systems, RoboDeus, YOLO V5s, convolutional neural networks, DSP core, instruction set architecture (ISA), system on a chip (SoC).
V. M. Zuev
ERROR IN PRAEDICTIONE SITUS PROXIMAE IMAGINIS PELLICULAE
UDC: 004.93
DOI: 10.24412/2413-7383-2025-4-39-173-181
Language: Russian
Abstract: This article examines the error in predicting the next frame based on a dynamic and kinematic model of camera movement, assuming the camera is rigidly mounted on board the vehicle. The resulting mathematical dependencies of the error on various conditions, unlike neural networks often used for this purpose, yield mathematical dependencies from which the necessary numerical characteristics can be easily derived.
Keywords: detection, forecasting, error Hic articulus errorem in praedictione proximae imaginis examinat, fretus exemplo dynamico et cinematico motus camerae, assumendo cameram rigide in vehiculo affixam esse. Dependentiae
Section
SYSTEM ANALYSIS, CONTROL AND INFORMATION PROCESSING, STATISTICS
1Institute of Mechanics named after R.R. Mavlyutov UFIC RAS
APPLICATION OF AN ASSOCIATIVE CONTROLLER IN THE CONTROL CIRCUIT OF A MECHATRONIC MODULE
UDC: 519.4
DOI: 10.24412/2413-7383-2025-4-39-183-193
Language: Russian
Abstract: The article discusses the urgent task of improving the efficiency of me chatronic module control systems in the context of the "Industry 4.0" concept. A comparative analysis of four different methods of using an associative controller based on the CMAC neural network (Cerebellar Model Articulation Controller) is carried out to improve the quality of the classic PID controller. The following architectures are considered: parallel operation of a PID controller and an associative controller to compensate for disturbances, dynamic adjustment of the coefficients of the PID controlle r using associative memory, a hybrid method combining both approaches, as well as the complete replacement of the PID controller with an associative one. The analysis is carried out using the example of a servo drive model in the Matlab/Simulink environmen t. The paper evaluates the effectiveness of the proposed methods in terms of performance and overshoot, and demonstrates the advantages of hybrid intelligent control systems.
Keywords: PID controller, associative memory, adjustment of the controller, CMAC.
2JSC M. A. Kartsev Research Institute of Computing Systems
METHOD FOR INFORMATION PROCESSING AND MODIFICATION OF THRESHOLD PARAMETERS OF RULES IN A MIVAR EXPERT SYSTEM USING LARGE LANGUAGE MODELS
UDC: 004.89 + 004.021 + 007.52
DOI: 10.24412/2413-7383-2025-4-39-194-205
Language: Russian
Abstract: The aim of the work is to automate the change of threshold parameters of rules in a multi-dimensional expert system using large language models. A method for dynamically changing the threshold parameters of mivar rules based on the DeepSeek large language model is proposed, which ensures a constant change in the mivar knowledge base through human-machine interaction. The method is aimed at a scenario of threshold parameter changes, in which historical data is analyzed to adjust these parameters. The scientific novelty lies in the creation of a comprehensive method of human - machine interaction that systematizes procedures for changing parameters using advanced capabilities of large language models, which ensures higher performance of multi-dimensional expert systems in real dynamic conditions.
Keywords: algorithm, expert system, mivar, mivar expert system, large language model, automatic updating of the knowledge base, artificial intelligence, parameter changes.
S. A. Izosimova, K. S. Ivashko
THE RESULTS OF THE PILOT OPERATION OF AN INTELLIGENT COMPUTER SYSTEM FOR THE DIAGNOSIS AND CORRECTION OF STRESS STATUS
UDC: 004.8: 159.9.072
DOI: 10.24412/2413-7383-2025-4-39-206-218
Language: Russian
Abstract: В настоящей статье исследованы результаты экспериментального этапа разработки компьютерной интеллектуальной системы диагностики и коррекции стрессовых состояний, направленного на валидацию программного инструментария. По результатам онлайн-тестирования, проведенного с использованием авторской программы для диагностирования психоэмоциональных состояний личности, представлен анализ выраженности тревожно-депрессивной симптоматики у лиц, подверженных стрессу военного времени.
Keywords: stress diagnostics, online testing, web technologies, software validation.
E. G. Kurgan, K. S. Petrova, A. S. Sobchishen
DEVELOPMENT OF AN ELECTRONIC DATABASE AS A TOOL FOR DIGITALIZATION OF THE CENTRAL SCIENTIFIC AND TECHNICAL LIBRARY
UDC: 004.932.2
DOI: 10.24412/2413-7383-2025-4-39-219-228
Language: Russian
Abstract: This article examines the digitalization of libraries. The effects of digitalization in the library sector are examined. The need for digitalization of the Central Scientific and Technical Library of the Federal State Budgetary Scientific Institution «ISTI» is su bstantiated. An electronic database «Guide to the Library's Depository Fund» has been developed, which has the property of scalability and an intuitive interface, providing convenient search for literary sources based on a thematic principle and centralized storage of metadata. Based on a comparative analysis, the PostgreSQL DBMS was selected. A description of the relational structure consisting of tables and a web interface for data management is provided. The developed system serves as the basis for integ rating the library into the unified information space of Russia.
Keywords: electronic database, digitalization, library, data accessibility.
K. A. Nikitenko, A. V. Zviagintseva
ASSESSMENT OF SEMANTIC SIMILARITY OF SENTENCES USING PROBABILISTIC MEAUSRES BASED ON EMBEDDINGS
UDC: 004.8+004.912
DOI: 10.24412/2413-7383-2025-4-39-229-240
Language: Russian
Abstract: The article addresses the problem of evaluating the semantic similar ity of sentences using probabilistic measures of word embeddings. An method is proposed based on comparing the probabilities of independent and d ependent events corresponding to the words within a sentence. The methods for constructing attention matrices a nd digitizing words are analyzed, and the choice of probabilistic embeddings as the core of the semantic representation is justified. A procedure is proposed for comparing probabilities derived from embeddings and attention weights, which enables formalization of semantic relations between words and sentences.
Keywords: semantic similarity, probabilistic model, attention matrix, embedding, probabilistic text analysis.
D.A. Filipishin, S.A. Zori
THE RESOLUTION METHOD AS A MEANS OF EXTENDING THE ONTOLOGY INFERENCE MACHINE
UDC: 681.3
DOI: 10.24412/2413-7383-2025-4-39-241-253
Language: Russian
Abstract: The article discusses the use of the Prolog language solver as a means of extending the inference machine (reasoner) to build domain ontologies. A new approach to solving production tasks within the framework of the ontology editor, considered as an integrated development environment (IDE), is proposed. The perspective of using a Prolog language solver based on the predicate language of the mathematical logic of Horn disjuncts, semantic connections and axiomatics of binary relations inherent in ontologies is analyzed. A solution to the existing shortcomings in both systems is proposed in order to create a new ontological system that includes the capabilities of both approaches.
Keywords: ontology, Prolog, IDE, reasoner, solver, products, Einstein's Riddle, the Holmes game.
I. V. Chernyadev, T. V. Yermolenko
IMAGE IMPROVEMENT USING DEEP LEARNING: ANALYSIS AND PROSPECTIVE DEVELOPMENTS
UDC: 004.932.4
DOI: 10.24412/2413-7383-2025-4-39-254-265
Language: Russian
Abstract: The article discusses the current problem of improving imag e quality using deep learning methods, with a focus on super-resolution, noise reduction, and artifact removal. It analyzes key architectures based on transformers and convolutional neural networks, their advantages and disadvantages, such as high resource consumption and narrow specialization. The article proposes improvements in the form of randomized data augmentation for synthesizing the training set and replacing the standard convolution with Fast Fourier Convolution to capture global context. The goal of the work is to improve the quality of image restoration without high computational costs.
Keywords: super-resolution, neural networks, computer vision, deep learning, image processing.
