Перегляд за Автор "Yurchenko, D."

Зараз показуємо 1 - 2 з 2
  • Ескіз
    Документ
    Analysis of Dataset of Textile Materials Macro Images to Detect Data Leaks
    (2026) Merezhko, Ye.; Yurchenko, D.; Molchanova, M.; Mazurets, O.; Мазурець, Олександр Вікторович
    The paper analyzes a dataset of textile material macro images to detect redundancy-related data leaks that may negatively affect the reliability of neural network classification. The study emphasizes that textile datasets often contain visually similar or near-duplicate images due to repetitive fabric textures and repeated image acquisition conditions, which can lead to inflated evaluation results and reduced model generalization. The proposed approach combines perceptual hashing (pHash) and deep embeddings extracted from a pretrained ResNet18 model to identify and remove redundant images from the dataset. Experimental results demonstrate that dataset cleaning improves classification stability and increases the trustworthiness of CNN-based textile recognition models. The practical significance of the work lies in improving the quality of textile datasets for computer vision tasks, automated fabric classification, and intelligent textile recycling systems.
  • Ескіз
    Документ
    Approach to Using Cloud Services for Visual Analytics of Neural Network Analysis of Texts Emotional Tonality
    (2024) Yurchenko, D.; Mazurets, O.; Didur, V.; Molchanova, M.; Мазурець, Олександр Вікторович
    For the neural network analysis of the emotional tonality of messages, it is proposed to use a hybrid architecture neural network that combines the simultaneous advantages of the CNN and BiLSTM architectures. A neural network architecture was developed for determining emotional tonality, and the LIME model of interpreted model-agnostic explanations was used to visually explain the results of the neural network analysis of emotional tonality. This approach will make it possible to use all the advantages of neural network solutions, but to have an understanding for the user of what influenced these solutions.