CLOUD INFORMATION TECHNOLOGY FOR NEURAL NETWORK ANALYSIS OF DESTROYED STRUCTURES USING VISUAL DATA FROM UAVS
DOI:
https://doi.org/10.35546/kntu2078-4481.2025.3.2.44Keywords:
cloud technologies, neural networks, UAVs, destroyed buildingsAbstract
The paper proposes an information cloud technology for neural network analysis of destroyed structures based on visual data obtained from unmanned aerial vehicles. The problem of fast and reliable identification of building remains is extremely relevant in modern conditions, since the results of such analysis are of critical importance for monitoring the consequences of hostilities, eliminating emergencies and planning restoration work. Traditional computer vision methods do not provide the necessary efficiency due to the scale of data, the variety of objects and the complexity of scenes. The use of deep neural networks integrated into the cloud environment allows overcoming these limitations, ensuring scalability and high accuracy of analysis. The developed technology involves the receipt of high-resolution photographs into the cloud environment, their pre-processing and the formation of a single training and validation dataset. In the cloud, YOLOv12 models are trained for detection and segmentation of potential building remains, as well as an ensemble of ten ResNet50 models for multi-class classification of selected fragments. The client part of the system provides preprocessing of new images and execution of neural network analysis based on previously trained models. As a result, the positions of residues are automatically determined and their assignment to specific categories of materials, including brick, concrete, wood, foam plastic, tiles, plastic and others.Experimental results confirm the effectiveness of the developed solution. The overall classification Accuracy of 0.965 was achieved, with the Precision value being 0.889, Recall – 0.965, and F1-measure – 0.925. The obtained results exceed the indicators of similar studies by at least 0.042, which demonstrates the competitiveness and feasibility of using the proposed approach. The use of cloud technologies allows for scalability of the solution, reduction of infrastructure costs and independence from the computing capabilities of local devices.
References
Kohns J., et al. Building damage assessment in natural disasters: A trans- and interdisciplinary approach combining domain knowledge, 3D machine learning, and crowdsourcing // Progress in Disaster Science. 2025. Vol. 26. P. 100427. https://doi.org/10.1016/j.pdisas.2025.100427
Hladun O., Mazurets O., Molchanova M., Sobko O. Real Time Detection the Person Emotion State Using Neural Network // Scientific Research: Modern Innovations and Future Perspectives : proceedings of the 2nd International scientific and practical conference (November 25–27, 2024, Montreal, Canada). Montreal, 2024. P. 119–123.
Mostofi S., Altunişik A. C., Başağa H. B., Okur F. Y., Yilmaz Z., Hadinata P., … Sezdirmez T. A Big Data-Enabled Decision Support Model for Post-Earthquake Damage Classification of RC Buildings: A Case Study on February 6, Kahramanmaraş Doublet Earthquakes // Journal of Earthquake Engineering. 2025. P. 1–23.
Hladun O., Zalutska O., Klimenko V., Mazurets O. Research on the effectiveness of classifying the remains of destroyed buildings using MobileNetV3 neural network architecture // Innovations in Science: From Theoretical Foundations to Practical Impact : proceedings of the 1st International Scientific and Practical Conference (May 12–14, 2025, Antwerp, Belgium). Antwerp, 2025. P. 158–162.
Didur V., Molchanova M., Mazurets O. Research on the effectiveness of neural network detection of plots with the destroyed buildings remains // Modern technologies and science: problems, new and relevant developments : proceedings of the XXI International Scientific and Practical Conference (May 26, 2025, Zaragoza, Spain). Zaragoza, 2025. P. 245–251.
Li T., Fei L. Exploring obstacles to the use of unmanned aerial vehicles in emergency rescue: A BWM-DEMATEL approach // Technology in Society. 2025. Vol. 81. Article 102863. https://doi.org/10.1016/j.techsoc.2025.102863
Alqudsi Y., Makaraci M. UAV swarms: research, challenges, and future directions // Journal of Engineering and Applied Science. 2025. Vol. 72, No. 1. P. 12. URL: https://doi.org/10.1186/s44147-025-00582-3
Dydo R., Sobko O., Klimenko V., Mazurets O. Datalogic structure for intelligent system for areas localization in photos with the remains of buildings using neural network // Modern Scientific Research: Theoretical and Practical Aspects : proceedings of the II International Scientific and Practical Conference (May 26–28, 2025, Riga, Latvia). Riga, 2025. P. 123–127.
Didur V. O., Molchanova M. O., Tyschenko O. O., Mazurets O. V. Approach for comparative analysis of effectiveness of using MobileNetV3 and ViT neural network models for graphical localization of destroyed buildings remains areas // Formation of Innovative Potential of World Science : proceedings of the IX International Scientific and Practical Conference (May 16, 2025, Waterford, Republic of Ireland). Waterford, 2025. P. 94–97.
Мазурець О. В., Петровський С. С., Дидо Р. А. Нейромережева модель для ідентифікації особистості за зображенням обличчя у реальному часі // Інформаційні технології і автоматизація : матеріали XVII міжнародної науково-практичної конференції (31 жовтня – 1 листопада 2024, Одеса, ОНТУ). Одеса, 2024. С. 655–658.
Бас І. С., Мазурець О. В., Молчанова М. О., Собко О. В. Дослідження ефективності методу автоматизованого визначення типу літального апарату за фотографічним зображенням // Збірник наукових праць за мате- ріалами XVI Всеукраїнської науково-практичної конференції «Актуальні проблеми комп’ютерних наук АПКН- 2024» (15–16 листопада 2024, Хмельницький). Хмельницький, 2024. С. 29–34.
Dydo R., Sobko O., Molchanova M., Mazurets O. Analysis of precision of finding the destroyed remains buildings on photos using MobileNetV3 and ViT neural networks // Science and Technology: New Horizons of Development : proceedings of the I International Scientific and Practical Conference (May 14–16, 2025, Prague, Czech Republic). Prague, 2025. P. 208–214.
Malaydakh V., Molchanova M., Shevchuk P., Mazurets O. Deep learning neural network architecture for determining sunflower growth stage from visual data // Modern Scientific Research: Theoretical and Practical Aspects : proceedings of the II International Scientific and Practical Conference (May 26–28, 2025, Riga, Latvia). Riga, 2025. P. 143–148.
Bas I. S., Kadynska V. D., Klimenko V. I., Mazurets O. V. Convolutional Neural Network Transfer Learning Method for Aircraft Image Classification // Scientific method: reality and future trends of researching : proceedings of the VI International Scientific and Theoretical Conference (June 6, 2025, Montreal, Canada). Montreal, 2025. P. 147–155.
Молчанова М. О., Мазурець О. В., Собко О. В., Кліменко В. І., Андрощук В. І. Метод нейромережевого виявлення кібербулінгу з використанням хмарних сервісів та об’єктно-орієнтованої моделі // Науковий журнал «Вісник Хмельницького національного університету», серія: Технічні науки. Хмельницький, 2024. № 2 (333). С. 200–206. https://doi.org/10.31891/2307-5732-2024-333-2
Tymofiiev I., Mazurets O., Hardysh D., Molchanova M. Neural Network Dual Architecture for Depression Detection Using Cloud Services // Scientific Research in the Era of Digital Technologies: Challenges and Opportunities : proceedings of the XLVІ International Scientific and Practical Conference (November 6–8, 2024, Barcelona, Spain). 2024. P. 84–88.
Yurchenko D., Mazurets O., Didur V., Molchanova M. Approach to Using Cloud Services for Visual Analytics of Neural Network Analysis of Texts Emotional Tonality // The Future of Scientific Discoveries: New Trends and Technologies : proceedings of the XLVІІ International Scientific and Practical Conference (November 13–15, 2024, Marseille, France). 2024. P. 108–113.
Ahmed M. I. B., et al. Deep Learning Approach to Recyclable Products Classification: Towards Sustainable Waste Management // Sustainability. 2023. Vol. 15, No. 14. P. 11138. https://doi.org/10.3390/su151411138.
Lin K., et al. Applying machine learning to fine classify construction and demolition waste based on deep residual network and knowledge transfer // Environment, Development and Sustainability. 2022. https://doi.org/10.1007/s10668-022-02740-6
Nežerka V., Zbíral T., Trejbal J. Machine-learning-assisted classification of construction and demolition waste fragments using computer vision: Convolution versus extraction of selected features // Expert Systems with Applications. 2024. Vol. 238. Article 121568. https://doi.org/10.1016/j.eswa.2023.121568.
YOLOv12: State-of-the-art object detection model [Електронний ресурс] // YOLOv12. 2025. Режим доступу: https://yolov12.com/
ResNet50: Convolutional Neural Network Model [Електронний ресурс] // PyTorch. 2025. Режим доступу: https://pytorch.org/hub/pytorch_vision_resnet/
Hladun O. V., Molchanova M. O., Zalutska O. O., Mazurets O. V. Effectiveness research of using ViT neural network architecture for classifying the destroyed buildings remains // Achievements of Science and Applied Research : proceedings of the 2nd International Scientific and Theoretical Conference (May 19–21, 2025, Dublin, Ireland). Dublin, 2025. P. 96–100.
