ELARKHNU browse.title.page

Зараз показуємо 1 - 1 з 1

Convolutional Neural Network Architecture for Image-Based Architectural Style Recognition
(2025) Mushtyn, O.; Sobko, O.; Molchanova, M.; Mazurets, O.; Мазурець, Олександр Вікторович
This 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.