Dmytro Uhryn
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Dmytro Uhryn has a total of 107 co-authors. Here are the top co-authors by number of shared publications:
- Yuriy A. Ushenko
Chernivtsi National University
43 shared publications - Victoria Vysotska
Lviv Polytechnic National University, Ukraine
28 shared publications - Zhengbing Hu21 shared publications
- Artem Kalancha
Yuriy Fedkovych Chernivtsi National University
10 shared publications - Lyubomyr Chyrun
Lviv Polytechnic National University
9 shared publications - Vasyl Lytvyn
Lviv Polytechnic National University, Lviv, Ukraine
9 shared publications
Latest Publications
- Open Access
- 15 July 2026
- Published by Elsevier in Next Materials
- Vol. 13
Ensemble machine learning models for predicting mechanical properties of epoxy-based nanocomposites
- 15 July 2026
- Published by Elsevier in Next Materials
- Vol. 13
Open AccessEnsemble machine learning models for predicting mechanical properties of epoxy-based nanocomposites
- Open Access
- 30 June 2026
- Published by Yuriy Fedkovych Chernivtsi National University in Security of Infocommunication Systems and Internet of Things
- Vol. 4 (1) , 01012
Comparative Analysis of Detection and Segmentation Models for Building Damage Assessment from UAV Imagery
This paper presents a systematic comparison of object detection models (Faster R-CNN, RetinaNet, YOLOv8) and semantic segmentation models (U-Net, DeepLabV3+, SegFormer) for automated building damage assessment from unmanned aerial vehicle (UAV) imagery. The comparison covers both two-stage (Faster R-CNN) and single-stage (RetinaNet, YOLOv8) approaches to detection, as well as convolutional (U-Net, DeepLabV3+) and transformer-based (SegFormer) segmentation architectures. The RescueNet dataset, a high-resolution post-disaster semantic segmentation benchmark, was adapted for detection by extracting bounding boxes from semantic masks via connected component analysis. Dataset adaptation was performed through separate pipelines for each annotation format (YOLO txt, torchvision JSON), and for segmentation, individual building crops were used, simulating the real-world cascaded pipeline scenario. At the segmentation stage, the task is formulated as binary building-footprint delineation within detected crops; per-building damage-level grading is performed by the subsequent classification stage of the cascade and is beyond the scope of this study. To ensure fair comparison, all models within each task were trained with identical hyperparameters, loss functions, and equivalent pretrained weights (COCO for detection, ImageNet for segmentation). Detection models were evaluated using standard COCO metrics (mAP@0.5, mAP@0.5:0.95) computed via pycocotools, while segmentation models were assessed using globally accumulated IoU, Dice, Precision, and Recall. Inference latency was additionally measured for all models. The results provide an evidence-based foundation for selecting architectures in a cascaded Detection-Segmentation-Classification pipeline for emergency response scenarios. The practical significance of the study lies in the unified evaluation framework that minimizes confounding factors such as different pretraining levels or optimizer configurations, enabling a substantially more direct comparison of architectural properties. This work is part of a broader project on developing an automated UAV-based building damage monitoring system.
- 30 June 2026
- Published by Yuriy Fedkovych Chernivtsi National University in Security of Infocommunication Systems and Internet of Things
- Vol. 4 (1) , 01012
Open AccessComparative Analysis of Detection and Segmentation Models for Building Damage Assessment from UAV Imagery
This paper presents a systematic comparison of object detection models (Faster R-CNN, RetinaNet, YOLOv8) and semantic segmentation models (U-Net, DeepLabV3+, SegFormer) for automated building damage assessment from unmanned aerial vehicle (UAV) imagery. The comparison covers both two-stage (Faster R-CNN) and single-stage (RetinaNet, YOLOv8) approaches to detection, as well as convolutional (U-Net, DeepLabV3+) and transformer-based (SegFormer) segmentation architectures. The RescueNet dataset, a high-resolution post-disaster semantic segmentation benchmark, was adapted for detection by extracting bounding boxes from semantic masks via connected component analysis. Dataset adaptation was performed through separate pipelines for each annotation format (YOLO txt, torchvision JSON), and for segmentation, individual building crops were used, simulating the real-world cascaded pipeline scenario. At the segmentation stage, the task is formulated as binary building-footprint delineation within detected crops; per-building damage-level grading is performed by the subsequent classification stage of the cascade and is beyond the scope of this study. To ensure fair comparison, all models within each task were trained with identical hyperparameters, loss functions, and equivalent pretrained weights (COCO for detection, ImageNet for segmentation). Detection models were evaluated using standard COCO metrics (mAP@0.5, mAP@0.5:0.95) computed via pycocotools, while segmentation models were assessed using globally accumulated IoU, Dice, Precision, and Recall. Inference latency was additionally measured for all models. The results provide an evidence-based foundation for selecting architectures in a cascaded Detection-Segmentation-Classification pipeline for emergency response scenarios. The practical significance of the study lies in the unified evaluation framework that minimizes confounding factors such as different pretraining levels or optimizer configurations, enabling a substantially more direct comparison of architectural properties. This work is part of a broader project on developing an automated UAV-based building damage monitoring system.
- Open Access
- 30 June 2026
- Published by Yuriy Fedkovych Chernivtsi National University in Security of Infocommunication Systems and Internet of Things
- Vol. 4 (1) , 01007
Evaluating Information Sources on Social Media: A Systematic Review and Conceptual Model
This article presents a systematic review and conceptual framework for analyzing information sources in social networks, with a focus on Telegram as the dominant news platform in Ukraine. The relevance of the study is determined by the dramatic shift in media consumption patterns following Russia's full-scale invasion in February 2022: Telegram's audience tripled from 20% to 60% within months, and by 2025 it remains the primary news source for 52% of the adult population, surpassing television (25%), YouTube (32%), and other platforms. Despite this dominance, only 29% of users trust Telegram channels, creating a paradox where mass consumption coexists with low trust and minimal verification ‒ only 6% of Ukrainians consult fact-checking organizations. The article examines the mechanisms of information manipulation, including the five-level cascading narratives model (from "Russian World" to alternative information) and the five-level amplification pyramid that describes the transformation of deliberate disinformation into unwitting misinformation through state media, expert legitimization, portal localization, social media dissemination, and organic user spread. The 4D tactics framework (Dismiss, Distort, Distract, Dismay) systematizes the core techniques of these operations. A comparative analysis of ten methods for information source analysis spanning 2008-2025 reveals a clear evolution from isolated statistical approaches through graph-based methods to modern hybrid systems integrating natural language processing, graph neural networks, and attention mechanisms. The highest performance was demonstrated by the Lightning system (AUC 86.83%) and OnlineAgglomerative (93.2% accuracy). However, none of the reviewed methods simultaneously provides comprehensive semantic, temporal, and network analysis for source evaluation. To address this gap, the article proposes a conceptual model for assessing information source similarity (similarity_score) that integrates four components: cosine similarity of content embeddings, thematic clusterization, timed semantic influence for detecting temporal-semantic dependencies between sources, and network proximity in the information flow graph. The model is extensible and can incorporate additional metrics such as sentiment analysis, named entity recognition, and toxicity assessment. An information system architecture for practical implementation on Telegram data is also proposed.
- 30 June 2026
- Published by Yuriy Fedkovych Chernivtsi National University in Security of Infocommunication Systems and Internet of Things
- Vol. 4 (1) , 01007
Open AccessEvaluating Information Sources on Social Media: A Systematic Review and Conceptual Model
This article presents a systematic review and conceptual framework for analyzing information sources in social networks, with a focus on Telegram as the dominant news platform in Ukraine. The relevance of the study is determined by the dramatic shift in media consumption patterns following Russia's full-scale invasion in February 2022: Telegram's audience tripled from 20% to 60% within months, and by 2025 it remains the primary news source for 52% of the adult population, surpassing television (25%), YouTube (32%), and other platforms. Despite this dominance, only 29% of users trust Telegram channels, creating a paradox where mass consumption coexists with low trust and minimal verification ‒ only 6% of Ukrainians consult fact-checking organizations. The article examines the mechanisms of information manipulation, including the five-level cascading narratives model (from "Russian World" to alternative information) and the five-level amplification pyramid that describes the transformation of deliberate disinformation into unwitting misinformation through state media, expert legitimization, portal localization, social media dissemination, and organic user spread. The 4D tactics framework (Dismiss, Distort, Distract, Dismay) systematizes the core techniques of these operations. A comparative analysis of ten methods for information source analysis spanning 2008-2025 reveals a clear evolution from isolated statistical approaches through graph-based methods to modern hybrid systems integrating natural language processing, graph neural networks, and attention mechanisms. The highest performance was demonstrated by the Lightning system (AUC 86.83%) and OnlineAgglomerative (93.2% accuracy). However, none of the reviewed methods simultaneously provides comprehensive semantic, temporal, and network analysis for source evaluation. To address this gap, the article proposes a conceptual model for assessing information source similarity (similarity_score) that integrates four components: cosine similarity of content embeddings, thematic clusterization, timed semantic influence for detecting temporal-semantic dependencies between sources, and network proximity in the information flow graph. The model is extensible and can incorporate additional metrics such as sentiment analysis, named entity recognition, and toxicity assessment. An information system architecture for practical implementation on Telegram data is also proposed.
- Open Access
- 30 June 2026
- Published by Yuriy Fedkovych Chernivtsi National University in Security of Infocommunication Systems and Internet of Things
- Vol. 4 (1) , 01003
Research on Computer Vision Methods for Human Identification in Low Visibility Conditions
The article discusses the pressing scientific and practical problem of dependable human detection and identification in situations of low visibility, including dense fog, thick smoke, or extremely inadequate lighting. The research offers a systematic analysis of hybrid approaches based on the complementary integration of data from both the visible and infrared spectra. It is suggested that by synergizing these modalities, it may be possible to overcome the fundamental limitations of mono-spectral sensors, supporting robust performance in environments where traditional optical or thermal cameras individually fail. The article synthesizes a conceptual architectural model for a two-stream neural network specifically for unmanned aerial vehicles based on a comprehensive analysis and systematization of current sensor fusion methods. Particular focus is being placed on the theoretical development of a pre-assessment module logic. This module can dynamically switch between processing modes based on real-time image entropy analysis to make the best use of computing power. The research identifies a suitable technological stack, which includes end-to-end restoration architectures and lightweight integration blocks, such as temperature-guided lightweight fusion, adapted for the hardware constraints of embedded systems. The analysis indicates potential benefits of this hybrid approach in reducing spectral pollution and supporting adherence to data privacy regulations, particularly the General Data Protection Regulation, compared with mono-spectral alternatives. The article concludes by outlining strategic opportunities for the advancement of autonomous search and rescue systems, with a particular emphasis on the potential of federated learning principles for secure and decentralized model updates in practical settings.
- 30 June 2026
- Published by Yuriy Fedkovych Chernivtsi National University in Security of Infocommunication Systems and Internet of Things
- Vol. 4 (1) , 01003
Open AccessResearch on Computer Vision Methods for Human Identification in Low Visibility Conditions
The article discusses the pressing scientific and practical problem of dependable human detection and identification in situations of low visibility, including dense fog, thick smoke, or extremely inadequate lighting. The research offers a systematic analysis of hybrid approaches based on the complementary integration of data from both the visible and infrared spectra. It is suggested that by synergizing these modalities, it may be possible to overcome the fundamental limitations of mono-spectral sensors, supporting robust performance in environments where traditional optical or thermal cameras individually fail. The article synthesizes a conceptual architectural model for a two-stream neural network specifically for unmanned aerial vehicles based on a comprehensive analysis and systematization of current sensor fusion methods. Particular focus is being placed on the theoretical development of a pre-assessment module logic. This module can dynamically switch between processing modes based on real-time image entropy analysis to make the best use of computing power. The research identifies a suitable technological stack, which includes end-to-end restoration architectures and lightweight integration blocks, such as temperature-guided lightweight fusion, adapted for the hardware constraints of embedded systems. The analysis indicates potential benefits of this hybrid approach in reducing spectral pollution and supporting adherence to data privacy regulations, particularly the General Data Protection Regulation, compared with mono-spectral alternatives. The article concludes by outlining strategic opportunities for the advancement of autonomous search and rescue systems, with a particular emphasis on the potential of federated learning principles for secure and decentralized model updates in practical settings.
- 8 April 2026
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 18 (2) , 151-207
Information Technology for VR Training Evaluation with First Aid Skills Improvement to Detecting Human Resource Damage in Emergencies based on Behavioural Methods
Traditional first aid preparation methods often fail to reproduce realistic stress levels and to simulate visual difficulty in identifying lesions in critical situations. In emergencies, delays in recognising injuries or errors in protocols result in critical losses of human resources. The use of computer graphics and virtual reality technologies enables you to create a safe yet highly realistic environment for rescuers to test and improve their skills. The article presents an integrated methodological framework for assessing the effectiveness of VR first-aid training in conditions of damage to civilian infrastructure. The main focus is on developing mathematical models and algorithms to identify and evaluate the quality of rescuers' actions by analysing digital interaction signals in a virtual environment. A composite efficiency indicator is proposed that combines normalised parameters for reaction time, manipulation accuracy, stress level, and immersion. The work aims to formalise a mathematical model to assess the effectiveness of VR training in developing skills for lesion identification and first aid provision, using quantitative metrics. The study aims to identify statistically significant differences in learning speed and skill retention between groups using VR simulations and traditional methods. The project aims to validate innovative content creation methods, including mobile photogrammetry, to visualise damaged infrastructure and victim models. The study used a comprehensive approach that includes mobile photogrammetry and generative neural networks to create a library of 3D assets with varying degrees of detail. Performance score is based on composite indicator that integrates normalised data on reaction time, manipulation accuracy, error count, immersion rate. Linear mixed models, exponential approximations, and bootstrap estimation of effect stability were used to analyse hierarchical data and individual learning trajectories. The experimental part includes the use of mobile photogrammetry and generative neural networks to create realistic 3D models of affected environments and identify types of injuries (bleeding, burns, unconsciousness). To analyse the dynamics of learning and maintaining skills, models with mixed effects and exponential forgetting curves are used. The results confirm that the use of VR technologies provides a statistically significant acceleration in the development of automated skills for lesion identification and assistance compared to traditional methods. The proposed approach is a scalable tool for preparing civil and rescue services to act in critical situations. Experimental data showed that the integral performance score in the VR group increased from 0.42 0.10 to 0.76 0.08, while in the control group it increased from only 0.40 0.09 to 0.55 0.10 (p < 0.001). The largest effect was observed in the bleeding arrest scenario, where the effect size (Cohen's d) reached 2.3. The analysis of forgetting curves confirmed the superiority of VR: the skill loss rate in the VR group was 0.25, providing knowledge retention 1.8 times longer than in the control group (0.45). The study confirmed that VR simulations significantly accelerate the formation of automated behaviour patterns and reduce reaction time in extreme conditions. The proposed mathematical assessment model provides objective feedback and standardisation of the rescue training process. The results indicate the high practical value of introducing such tools into training programs for civilian and military structures to minimise losses in real emergencies.
- 8 April 2026
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 18 (2) , 151-207
Information Technology for VR Training Evaluation with First Aid Skills Improvement to Detecting Human Resource Damage in Emergencies based on Behavioural Methods
Traditional first aid preparation methods often fail to reproduce realistic stress levels and to simulate visual difficulty in identifying lesions in critical situations. In emergencies, delays in recognising injuries or errors in protocols result in critical losses of human resources. The use of computer graphics and virtual reality technologies enables you to create a safe yet highly realistic environment for rescuers to test and improve their skills. The article presents an integrated methodological framework for assessing the effectiveness of VR first-aid training in conditions of damage to civilian infrastructure. The main focus is on developing mathematical models and algorithms to identify and evaluate the quality of rescuers' actions by analysing digital interaction signals in a virtual environment. A composite efficiency indicator is proposed that combines normalised parameters for reaction time, manipulation accuracy, stress level, and immersion. The work aims to formalise a mathematical model to assess the effectiveness of VR training in developing skills for lesion identification and first aid provision, using quantitative metrics. The study aims to identify statistically significant differences in learning speed and skill retention between groups using VR simulations and traditional methods. The project aims to validate innovative content creation methods, including mobile photogrammetry, to visualise damaged infrastructure and victim models. The study used a comprehensive approach that includes mobile photogrammetry and generative neural networks to create a library of 3D assets with varying degrees of detail. Performance score is based on composite indicator that integrates normalised data on reaction time, manipulation accuracy, error count, immersion rate. Linear mixed models, exponential approximations, and bootstrap estimation of effect stability were used to analyse hierarchical data and individual learning trajectories. The experimental part includes the use of mobile photogrammetry and generative neural networks to create realistic 3D models of affected environments and identify types of injuries (bleeding, burns, unconsciousness). To analyse the dynamics of learning and maintaining skills, models with mixed effects and exponential forgetting curves are used. The results confirm that the use of VR technologies provides a statistically significant acceleration in the development of automated skills for lesion identification and assistance compared to traditional methods. The proposed approach is a scalable tool for preparing civil and rescue services to act in critical situations. Experimental data showed that the integral performance score in the VR group increased from 0.42 0.10 to 0.76 0.08, while in the control group it increased from only 0.40 0.09 to 0.55 0.10 (p < 0.001). The largest effect was observed in the bleeding arrest scenario, where the effect size (Cohen's d) reached 2.3. The analysis of forgetting curves confirmed the superiority of VR: the skill loss rate in the VR group was 0.25, providing knowledge retention 1.8 times longer than in the control group (0.45). The study confirmed that VR simulations significantly accelerate the formation of automated behaviour patterns and reduce reaction time in extreme conditions. The proposed mathematical assessment model provides objective feedback and standardisation of the rescue training process. The results indicate the high practical value of introducing such tools into training programs for civilian and military structures to minimise losses in real emergencies.
- 8 April 2026
- Published by MECS Publisher in International Journal of Computer Network and Information Security
- Vol. 18 (2) , 149-180
An Innovative Method for Detecting Fake News Distribution Sources based on Machine Learning Technology and Graph Theory
An innovative approach to identifying rapidly spreading false information is to create a targeted graph and its subsequent clustering. A method for detecting rapidly spreading fake messages in social networks has been developed. K-means, Louvain, and Leiden algorithms were applied to identify large communities in graphs, enabling the rapid detection of fake news. A modified fake news detection algorithm based on k-means and Leiden can group fake news clusters, enabling rapid identification of widely spreading news. The combination of Leiden for structural analysis of communities and SVM for classification provides an optimal balance between accuracy (F1-score = 0.87) and completeness of fake detection (Recall = 97%), allowing the system to be used both for analysing large datasets and for monitoring new publications. The Lei-den algorithm demonstrated the highest modularity (Q = 0.7212), which is 4.8% better than Louvain (Q = 0.6884), and detected 40 structural communities. The modified method has a lower modularity (Q = 0.5584), since modularity is not calculated for K-means.
- 8 April 2026
- Published by MECS Publisher in International Journal of Computer Network and Information Security
- Vol. 18 (2) , 149-180
An Innovative Method for Detecting Fake News Distribution Sources based on Machine Learning Technology and Graph Theory
An innovative approach to identifying rapidly spreading false information is to create a targeted graph and its subsequent clustering. A method for detecting rapidly spreading fake messages in social networks has been developed. K-means, Louvain, and Leiden algorithms were applied to identify large communities in graphs, enabling the rapid detection of fake news. A modified fake news detection algorithm based on k-means and Leiden can group fake news clusters, enabling rapid identification of widely spreading news. The combination of Leiden for structural analysis of communities and SVM for classification provides an optimal balance between accuracy (F1-score = 0.87) and completeness of fake detection (Recall = 97%), allowing the system to be used both for analysing large datasets and for monitoring new publications. The Lei-den algorithm demonstrated the highest modularity (Q = 0.7212), which is 4.8% better than Louvain (Q = 0.6884), and detected 40 structural communities. The modified method has a lower modularity (Q = 0.5584), since modularity is not calculated for K-means.
- Open Access
- 27 February 2026
- Published by National Technical University Kharkiv Polytechnic Institute in Bulletin of the National Technical University "KhPI" A series of "Information and Modeling"
- Vol. 1 (1(15)) , 155-167
Formation and analysis of information cascades based on timed semantic influence
This study is devoted to the analysis of information flows and the search for hidden connections between Telegram news channels. The main goal is to develop a method to help identify sources with the most significant influence and model how information spreads on the network. The work proposes the Timed Semantic Influence (TSI) algorithm - it allows you to assess the connections between messages, taking into account not only the content but also the time of publication. Thanks to this method, the structure of news distribution is built in the form of tree-like cascades from the initial message to the channels that picked it up. As a result, the channel that most often triggers news waves was identified. Among the topics that caused the most excellent resonance after eliminating general air alerts, news about significant geopolitical events and international support for Ukraine dominated. The proposed approach opens up the opportunity to identify key players in the media space and determine the topics that most influence public opinion. Tabl.: 1. Bibliogr.: 11 items.
- 27 February 2026
- Published by National Technical University Kharkiv Polytechnic Institute in Bulletin of the National Technical University "KhPI" A series of "Information and Modeling"
- Vol. 1 (1(15)) , 155-167
Open AccessFormation and analysis of information cascades based on timed semantic influence
This study is devoted to the analysis of information flows and the search for hidden connections between Telegram news channels. The main goal is to develop a method to help identify sources with the most significant influence and model how information spreads on the network. The work proposes the Timed Semantic Influence (TSI) algorithm - it allows you to assess the connections between messages, taking into account not only the content but also the time of publication. Thanks to this method, the structure of news distribution is built in the form of tree-like cascades from the initial message to the channels that picked it up. As a result, the channel that most often triggers news waves was identified. Among the topics that caused the most excellent resonance after eliminating general air alerts, news about significant geopolitical events and international support for Ukraine dominated. The proposed approach opens up the opportunity to identify key players in the media space and determine the topics that most influence public opinion. Tabl.: 1. Bibliogr.: 11 items.
- Open Access
- 5 December 2025
- Published by Lutsk National Technical University in COMPUTER-INTEGRATED TECHNOLOGIES: EDUCATION, SCIENCE, PRODUCTION
- No. 61,p. 82-90
Моделювання та підвищення надійності ідентифікації суб’єктів доступу в СКУД на основі гібридних PCA-алгоритмів
Аналіз сучасних викликів, які супроводжують ефективну роботу сучасних систем контролю та управління доступом (СКУД) дозволяє виокремити їх ключові проблеми: значне зростання об’ємів ідентифікаційних ознак, які перевантажують традиційні методи, та високу чутливість до нестандартних умов зчитування, що призводить до формування помилок. Метод головних компонент дозволяє ефективно вирішити проблему зниження розмірності даних та прискорити пошук. Проте цей метод залишається вразливим до неідеальних умов, що обмежує його застосування та надійність у динамічному середовищі СКУД. Для систематизації та детального розуміння процесу ідентифікування було застосовано структурний підхід із моделюванням за методологією IDEF0, який вказав на важливість етапу порівняння перетвореного образу з еталонними. Проаналізовано інтеграцію методу головних компонент із алгоритмами згорткових нейронних мереж, машинного та глибокого навчання на основі виявлення прихованих нелінійних закономірностей та стійких ознак. Для оцінки ефективності гібридних методів використано галузеві стандарти: FAR, FRR та EER. Результати дослідження підтверджують актуальність гібридних підходів для забезпечення максимальної безпеки та ефективності функціонування сучасних СКУД
- 5 December 2025
- Published by Lutsk National Technical University in COMPUTER-INTEGRATED TECHNOLOGIES: EDUCATION, SCIENCE, PRODUCTION
- No. 61,p. 82-90
Open AccessМоделювання та підвищення надійності ідентифікації суб’єктів доступу в СКУД на основі гібридних PCA-алгоритмів
Аналіз сучасних викликів, які супроводжують ефективну роботу сучасних систем контролю та управління доступом (СКУД) дозволяє виокремити їх ключові проблеми: значне зростання об’ємів ідентифікаційних ознак, які перевантажують традиційні методи, та високу чутливість до нестандартних умов зчитування, що призводить до формування помилок. Метод головних компонент дозволяє ефективно вирішити проблему зниження розмірності даних та прискорити пошук. Проте цей метод залишається вразливим до неідеальних умов, що обмежує його застосування та надійність у динамічному середовищі СКУД. Для систематизації та детального розуміння процесу ідентифікування було застосовано структурний підхід із моделюванням за методологією IDEF0, який вказав на важливість етапу порівняння перетвореного образу з еталонними. Проаналізовано інтеграцію методу головних компонент із алгоритмами згорткових нейронних мереж, машинного та глибокого навчання на основі виявлення прихованих нелінійних закономірностей та стійких ознак. Для оцінки ефективності гібридних методів використано галузеві стандарти: FAR, FRR та EER. Результати дослідження підтверджують актуальність гібридних підходів для забезпечення максимальної безпеки та ефективності функціонування сучасних СКУД
- 4 December 2025
- Published by Central Ukrainian National Technical University in Central Ukrainian Scientific Bulletin. Technical Sciences
- Vol. 2 (12(43))
Experimental Analysis of Deep Neural Networks for Automated Object Classification Using MRI Images
Dmytro Uhryn, Professor, Doctor of Technical Sciences, Professor of the Department of Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0000-0003-4858-4511, e-mail: d.ugryn@chnu.edu.ua Oleksandr Dorenskyi, Associate Professor, PhD in Information Technology (Candidate of Technical Sciences), Associate Professor of Cybersecurity and Software Department, Central Ukrainian National Technical University, Kropyvnytskyi, Ukraine, ORCID: https://orcid.org/0000-0002-7625-9022, e-mail: dorenskyiop@kntu.kr.ua Oleksii Iliuk, Senior System & Businness Analyst, Team Lead, Temabit LLC, Kyiv, Ukraine, ORCID: https://orcid.org/0000-0002-0904-3045, e-mail: olexiyilyukm@gmail.com Yuriy Ushenko, Professor, Doctor of Physical and Mathematical Sciences, Head of the Department of Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0000-0003-1767-1882, e-mail: y.ushenko@chnu.edu.ua Kateryna Shkidina, Master’s degree Student in Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0009-0001-8536-8095, e-mail: shkidina.kateryna@chnu.edu.ua The article presents a system analysis and comparative study of the efficiency of modern deep neural networks for the task of automated brain tumor classification based on MRI images. Three architectures were used in the study ResNet50, DenseNet121, and EfficientNet-B0, which belong to the most widely adopted models in computer vision. The experimental part is based on a medical dataset that was preprocessed using standard augmentation and normalization methods. Quantitative results showed that ResNet50 achieved an accuracy of 92%, DenseNet121 reached 74%, and EfficientNet-B0 demonstrated the highest performance with an accuracy of 97%. Additional qualitative metrics supported these findings: the F1-score for EfficientNet-B0 reached 0.96, indicating a strong balance between precision and recall, while ResNet50 achieved an F1-score of 0.91, and DenseNet121 scored 0.73. Visualization of classification results showed that all models correctly identified the tumor class, but their confidence levels varied from 0.74 in DenseNet121 to 0.97 in EfficientNet-B0. The qualitative analysis confirmed the suitability of EfficientNet-B0 for cases where fast and accurate inference is prioritized under limited computational resources. Despite higher computational costs, ResNet50 can be effective in tasks that require robustness and maximum precision. The obtained results highlight the significant potential of deep learning models in medical diagnostics and the development of intelligent decision-support systems in neuroradiology.
- 4 December 2025
- Published by Central Ukrainian National Technical University in Central Ukrainian Scientific Bulletin. Technical Sciences
- Vol. 2 (12(43))
Experimental Analysis of Deep Neural Networks for Automated Object Classification Using MRI Images
Dmytro Uhryn, Professor, Doctor of Technical Sciences, Professor of the Department of Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0000-0003-4858-4511, e-mail: d.ugryn@chnu.edu.ua Oleksandr Dorenskyi, Associate Professor, PhD in Information Technology (Candidate of Technical Sciences), Associate Professor of Cybersecurity and Software Department, Central Ukrainian National Technical University, Kropyvnytskyi, Ukraine, ORCID: https://orcid.org/0000-0002-7625-9022, e-mail: dorenskyiop@kntu.kr.ua Oleksii Iliuk, Senior System & Businness Analyst, Team Lead, Temabit LLC, Kyiv, Ukraine, ORCID: https://orcid.org/0000-0002-0904-3045, e-mail: olexiyilyukm@gmail.com Yuriy Ushenko, Professor, Doctor of Physical and Mathematical Sciences, Head of the Department of Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0000-0003-1767-1882, e-mail: y.ushenko@chnu.edu.ua Kateryna Shkidina, Master’s degree Student in Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0009-0001-8536-8095, e-mail: shkidina.kateryna@chnu.edu.ua The article presents a system analysis and comparative study of the efficiency of modern deep neural networks for the task of automated brain tumor classification based on MRI images. Three architectures were used in the study ResNet50, DenseNet121, and EfficientNet-B0, which belong to the most widely adopted models in computer vision. The experimental part is based on a medical dataset that was preprocessed using standard augmentation and normalization methods. Quantitative results showed that ResNet50 achieved an accuracy of 92%, DenseNet121 reached 74%, and EfficientNet-B0 demonstrated the highest performance with an accuracy of 97%. Additional qualitative metrics supported these findings: the F1-score for EfficientNet-B0 reached 0.96, indicating a strong balance between precision and recall, while ResNet50 achieved an F1-score of 0.91, and DenseNet121 scored 0.73. Visualization of classification results showed that all models correctly identified the tumor class, but their confidence levels varied from 0.74 in DenseNet121 to 0.97 in EfficientNet-B0. The qualitative analysis confirmed the suitability of EfficientNet-B0 for cases where fast and accurate inference is prioritized under limited computational resources. Despite higher computational costs, ResNet50 can be effective in tasks that require robustness and maximum precision. The obtained results highlight the significant potential of deep learning models in medical diagnostics and the development of intelligent decision-support systems in neuroradiology.
- 4 December 2025
- Published by Central Ukrainian National Technical University in Central Ukrainian Scientific Bulletin. Technical Sciences
- Vol. 2 (12(43))
Method of Integral Analysis of Relationships between Information Sources based on Temporal-Semantic Metrics
Artem Kalancha, PhD Student, Department of Computer System Software, Yuriy Fedkovych Chernivtsi National University, Сernivtsi, Ukraine, ORCID: https://orcid.org/0009-0004-1451-7470, e-mail: kalancha.artem@chnu.edu.ua Dmytro Uhryn, Professor, Doctor of Technical Sciences, Professor of the Department of Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0000-0003-4858-4511, e-mail: d.ugryn@chnu.edu.ua Oleksandr Dorenskyi, Associate Professor, PhD in Information Technology (Candidate of Technical Sciences), Associate Professor of Cybersecurity and Software Department, Central Ukrainian National Technical University, Kropyvnytskyi, Ukraine, ORCID: https://orcid.org/0000-0002-7625-9022, e-mail: dorenskyiop@kntu.kr.ua Vasyl Lytvyn, Doctor of Technical Sciences, Professor of the Department of Information Systems and Networks, Lviv Polytechnic National University, Lviv, Ukraine, ORCID: https://orcid.org/0000-0002- 9676-0180, e-mail: vasyl.v.lytvyn@lpnu.ua This study is devoted to the analysis of the relationships between information sources in the Ukrainian information space, in particular between news and political Telegram channels. The paper identifies the main shortcomings of existing analysis methods, which are mainly focused on the English-language context and do not take into account the dynamic nature of information sources. As an alternative to these methods, a new method of integral analysis is proposed, which combines several independent metrics at the same time - lexical similarity and temporal-semantic influence. The proposed and implemented approach allows us to assess not only the semantic similarity of sources, but also the nature of their interactions over time, forming a multidimensional matrix of relationships. The results obtained demonstrate the effectiveness of the integral approach for identifying hidden structures of information influence, which can be used for the further development of monitoring systems.
- 4 December 2025
- Published by Central Ukrainian National Technical University in Central Ukrainian Scientific Bulletin. Technical Sciences
- Vol. 2 (12(43))
Method of Integral Analysis of Relationships between Information Sources based on Temporal-Semantic Metrics
Artem Kalancha, PhD Student, Department of Computer System Software, Yuriy Fedkovych Chernivtsi National University, Сernivtsi, Ukraine, ORCID: https://orcid.org/0009-0004-1451-7470, e-mail: kalancha.artem@chnu.edu.ua Dmytro Uhryn, Professor, Doctor of Technical Sciences, Professor of the Department of Computer Science, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine, ORCID: https://orcid.org/0000-0003-4858-4511, e-mail: d.ugryn@chnu.edu.ua Oleksandr Dorenskyi, Associate Professor, PhD in Information Technology (Candidate of Technical Sciences), Associate Professor of Cybersecurity and Software Department, Central Ukrainian National Technical University, Kropyvnytskyi, Ukraine, ORCID: https://orcid.org/0000-0002-7625-9022, e-mail: dorenskyiop@kntu.kr.ua Vasyl Lytvyn, Doctor of Technical Sciences, Professor of the Department of Information Systems and Networks, Lviv Polytechnic National University, Lviv, Ukraine, ORCID: https://orcid.org/0000-0002- 9676-0180, e-mail: vasyl.v.lytvyn@lpnu.ua This study is devoted to the analysis of the relationships between information sources in the Ukrainian information space, in particular between news and political Telegram channels. The paper identifies the main shortcomings of existing analysis methods, which are mainly focused on the English-language context and do not take into account the dynamic nature of information sources. As an alternative to these methods, a new method of integral analysis is proposed, which combines several independent metrics at the same time - lexical similarity and temporal-semantic influence. The proposed and implemented approach allows us to assess not only the semantic similarity of sources, but also the nature of their interactions over time, forming a multidimensional matrix of relationships. The results obtained demonstrate the effectiveness of the integral approach for identifying hidden structures of information influence, which can be used for the further development of monitoring systems.