Перегляд за Автор "Sobko, O."
Зараз показуємо 1 - 15 з 15
Результатів на сторінці
Налаштування сортування
Документ Analysis of Precision of Finding the Destroyed Remains Buildings on Photos using MobileNetV3 and ViT Neural Networks(2025) Dydo, R.; Sobko, O.; Molchanova, M.; Mazurets, O.; Мазурець, Олександр ВікторовичThis study presents a comparative analysis of the precision and recall of MobileNetV3 and Vision Transformer (ViT) neural networks in detecting destroyed building remains from photographic data. Using a curated dataset of disaster-zone images, both models were trained and evaluated on key performance metrics. Results show that while both architectures performed well, ViT consistently achieved higher accuracy and generalization, particularly in complex material classes. The findings support the use of ViT in high-precision post-disaster assessment systems and highlight its potential for integration into automated, real-time damage detection platforms.Документ Approach to Identification of Artificial Intelligence-Generated People Images by Means of Machine Learning(2024) Zharnovskyi, O.; Mazurets, O.; Sobko, O.; Мазурець, Олександр ВікторовичThe result of the work is the development of a method of identification of images of people generated by artificial intelligence by means of machine learning. The developed method allows for efficient image identification and can be integrated into mass media and social networks for automatic verification of image authenticity. In addition, the network can be constantly improved and adapted to new methods of image generation to prevent the spread of false information.Документ Convolutional Neural Network Architecture for Image-Based Architectural Style Recognition(2025) Mushtyn, O.; Sobko, O.; Molchanova, M.; Mazurets, O.; Мазурець, Олександр ВікторовичThis paper presents a convolutional neural network–based approach to architectural style classification using the MobileNetV2 architecture. The model leverages transfer learning and fine-tuning techniques to adapt a pre-trained network for multi-class classification across 25 architectural styles. A modular intelligent system is proposed, encompassing data preprocessing, training, evaluation, classification, and user interaction. The approach includes advanced methods such as image augmentation, dropout regularisation, and learning rate scheduling to improve generalisation and reduce overfitting. Experimental results demonstrate high classification accuracy and robustness to variations in image quality, perspective, and lighting. The model also provides interpretable outputs through attention mechanisms, supporting transparency and deeper analysis of stylistic features. The proposed system holds significant potential for applications in heritage documentation, urban analysis, and educational tools, while addressing challenges related to data imbalance, hybrid styles, and the need for scalable, adaptive solutions. Future directions include integrating multimodal data, continuous learning, and deployment in real-world heritage and planning contexts.Документ Datalogic Model for Determining the Clients Compatibility Based on Questionnaires Data Analysis by Artificial Intelligence Means(2024) Zhuk, D.; Mazurets, O.; Sobko, O.; Klimenko, V.; Мазурець, Олександр ВікторовичTherefore, the Datalogic Model for Determining the Clients Compatibility Based on Questionnaires Data Analysis by Artificial Intelligence Means was designed. To determine compatibility between users, the system uses a test in which points will be added for each matching criterion. After comparing the questionnaires according to these criteria, the system displays the percentage of compatibility of the questionnaires, which is calculated based on the matches of the given criteria. The more criteria match, the higher the match percentage. The need to create a database for the information system, which is intended for many users, was also described in detail. Using a database will allow efficient and convenient storage and processing of data about users and their interestsДокумент Datalogic structure for intelligent system for areas localization in photos with the remains of buildings using neural network(2025) Dydo, R.; Sobko, O.; Klimenko, V.; Mazurets, O.; Мазурець, Олександр ВікторовичThis paper presents the development of a datalogic structure for an intelligent system designed to detect and localize areas in photographic images that contain remains of destroyed buildings using neural networks. The system integrates pre-processing, object detection via the YOLO model, and multiclass classification of building materials. A relational database was designed to store and manage information about images, segments, detected materials, experiments, and classification metrics. This structured approach ensures efficient data handling, supports analytical reporting, and enables retraining of models based on historical data, contributing to more accurate and scalable damage assessment solutions in post-conflict or disaster-struck areas.Документ Designing CNN Neural Network Model for Detecting Fractures of Lower Extremities by X-ray Images(2024) Мазурець, Олександр Вікторович; Kharysh, I.; Sobko, O.; Mazurets, O.The problem of using CNN neural network model for detecting fractures of lower extremities by X-ray images was investigated. In particular, the Dense Convolutional Network architecture is used, which is one of the advanced convolutional neural networks, which is well suited for the task of identifying bone fractures in X-ray images due to its efficiency and ability to store information at all levels of the network. It differs from traditional CNNs in that each layer is connected to all previous layers, which allows storing information and facilitates the transfer of gradients during training. Regarding the advantages of DenseNet over other networks, it has fewer parameters compared to others, which reduces the need for computing resources. Also improved feature retention, where the model better remembers and uses features from earlier layers, improving its ability to recognize small details important for fracture detection.Документ Intelligent System for Neural Network Detection of Fake Document Images for Automated Personality Identification(2024) Zharnovskyi, O.; Sobko, O.; Klimenko, V.The outcome of the work is the development of a method for the neural network detection of fake document images and an information system for personal identification. The system utilizes the developed method and allows for the automated assessment of the authenticity level of a photographed identity document image using the developed approach. The research has demonstrated that the method achieves an accuracy of 95.16%. The developed information system for detecting fake document images can effectively prevent fraud in personal identification systems. It can aid in identifying forged document images that may be used for identity theft, credit card fraud, or other criminal activities.Документ Object-Oriented Approach for Ethnic Enmity Detection in Text Messages by NLP(2024) Molchanova, M.; Mazurets, O.; Sobko, O.; Boiarchuk, I.; Мазурець, Олександр ВікторовичThe effectiveness of the method was studied using the developed software by comparing the obtained answers with the validation set, and the trained FastForest machine learning model was evaluated using the metrics MicroAccuracy, MacroAccuracy, LogLoss, ConfusionMatrix, f1-measure, and Recall. Without changing the working training set, the metrics values were as follows: MicroAccuracy 0.9890, MacroAccuracy 0.9889, and LogLoss 0.0463. It was developed a software implementation of the method for detecting manifestations of ethnic hatred in text messages of social Internet networks by NLP tools, which uses natural language processing techniques and converts input data in the form of a trained FastForest classifier and an input text message into output data in the form of a percentage of ethnic hatred in a test message of social Internet networks.Документ Practical Approach for Detection by Deep Learning of Target Objects of Subject Area Based on Semantic Connectivity Indicators in Audio Database(2024) Mazurets, O.; Sobko, O.; Vit, R.; Pasternak, V.; Мазурець, Олександр ВікторовичA practical approach for detection was performed by deep learning of target objects in the subject area based on semantic connectivity indicators in an audio database. A database was also created for the software that detects actors in Ukrainian-language audio data for the media sphere. This database includes the necessary tables that provide convenient storage and organization of information about actors, audio data and transcription. It is a key tool for analysing and processing audio data in the media field, which allows you to efficiently identify, classify and analyze the information contained in audio files, taking into account their context and content.Документ Practice Implementation of the Method for Analysis and Formation of Representative Text Datasets(2024) Sobko, O.Proposes the practice implementation of the method for analysis and formation of representative text datasets, designed for the analysis and formation of representative text samples of data according to the FATE principle of fairness for subject areas. The studied efficiency proves that developed method allows performing the analysis of the representativeness of text datasets and bringing them to representative form according to various aspects of the FATE fairness principleДокумент Real Time Detection the Person Emotion State Using Neural Network(2024) Hladun, O.; Mazurets, O.; Molchanova, M.; Sobko, O.; Мазурець, Олександр ВікторовичThe method of determining the emotional state of a person in real time by neural networks tools is proposed, which uses a convolutional neural network and allows detecting 7 basic emotional states of a person with an accuracy of more than 80% for each of the emotions.Документ Relation Datalogic Model for Determining the Diagnosis Based on Intellectual NLP-analysis of Symptom Description(2024) Mazurets, O.; Sobko, O.; Klimenko, V.; Kozenko, Y.; Мазурець, Олександр ВікторовичA relational data model for a database was designed for the automated determination of a diagnosis based on textual descriptions of symptoms using NLP tools, along with corresponding software in the form of a desktop application. By utilizing an SQLite database, reliable storage and organization of medical information can be ensured. This enables high-speed data access and efficient processing of user queries. All of this together forms a robust foundation for creating a useful application that can significantly streamline the diagnosis determination process in medical practice and enhance the quality of healthcare services. Additionally, a software application was implemented to determine a diagnosis based on textual descriptions of symptoms using NLP tools, utilizing the developed relational data model for the database. The created software product includes functions for determining the probable diagnosis and text description, entering new information, and working with the dataset. The functional structure of the information system for diagnosing diseases based on textual descriptions was described, consisting of three subsystems: working with the database, diagnosis, preprocessing of text data, and the corresponding database.Документ Виявлення та класифікація кіберзалякувань у цифрових текстах засобами штучного інтелекту(Хмельницький національний університет, 2024) Собко, О.; Sobko, O.У статті запропоновано комплексний підхід до виявлення та класифікації кіберзалякувань у цифрових текстах за допомогою штучного інтелекту. Підхід складається з трьох етапів: оцінювання та коригування репрезентативності датасету з урахуванням етичних критеріїв, нейромережевого виявлення та класифікації кіберзалякувань за різними типами (віковими, релігійними, етнічними, гендерними тощо), а також візуальної інтерпретації результатів моделі. Підхід дозволяє забезпечити неупередженість та відповідність етичним вимогам, а також надає пояснення рішень моделі щодо кожного виявленого типу кіберзалякування, що підвищує прозорість і довіру до систем штучного інтелекту. Результати дослідження підтверджують ефективність підходу, зокрема точність не нижче 94% для моделей BiLSTM і BERT для виявлення та класифікації кіберзалякувань, а також успішну адаптацію текстових датасетів до репрезентативних розподілів.Документ Метод нейромережевого виявлення кібербулінгу з використанням хмарних сервісів та об'єктно-орієнтованої моделі(Хмельницький національний університет, 2024) Молчанова, М.; Мазурець, О.; Собко, О.; Кліменко, В.; Андрощук, В.; Molchanova, M.; Mazurets, O.; Sobko, O.; Klimenko, V.; Androschuk, V.У роботі пропонується практичний підхід до виявлення кібербулінгу із використанням нейронної мережі BiLSTM, навченої за допомогою хмарних сервісів та застосунку, що реалізовує запропонований метод виявлення кібербулінгу. Для дослідження ефективності запропонованого методу було створено об’єктно-орієнтовану програмну реалізацію середовища програмування PyCharm, а також ноутбук для виконання в хмарному сервісі «Google Colab» для навчання нейромережі.Документ Підходи до практичного аналізу обчислювальних алгоритмів(Хмельницький національний університет, 2021) Бармак, О.В.; Радюк, П.М.; Молчанова, М.О.; Собко, О.В.; Barmak, O.; Radiuk, P.; Molchanova, M.; Sobko, O.У роботі пропонується практичний підхід до визначення основних типів алгоритмів залежно від їх ефективності за зовнішнім виглядом програмного коду. Наведено приклади аналізу ефективності програмного коду для обчислювальної складності за зменшенням ефективності, що подається як (в асимптотичних позначеннях) О(1), О(logN), O(N), O(NlogN), O(N2), O(N3). Завдання дослідження полягає в аналізі програмного коду та визначенні умов, за яких алгоритм належить до того або іншого типу обчислювальної складності. Встановлено, що основними чинниками, за якими можна оцінити обчислювальну складність алгоритму за візуальним аналізом програмного коду є наявність у коді циклів, особливо вкладених, рекурсивність алгоритму тощо.