Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Analysis of remote sensing data using artificial intelligence techniques in order to assess the structural stability of buildings

Mavroudis Alkis

Full record

---

Summary

The preservation of buildings and infrastructure, increasingly vulnerable to damage from climate change-related natural disasters, remains a critical concern for both urban and rural areas, underscoring the need for advanced evaluation methods and mitigation strategies. Concurrently, advancements in Artificial Intelligence enable the development of tools that can automate significant aspects of structural stability assessment, enhancing efficiency and accuracy in response efforts. Regarding remote sensing data analysis in particular, Convolutional Neural Networks over the past decade, and more recently Vision Transformers, have shown promising results. However, existing publications often appear overfitted to address specific use cases, focusing on higher performance metrics rather than optimizing for in-field applications. This thesis proposes a unified pipeline for building damage assessment on the two primary sources of remote sensing data, satellite and aerial imagery, leveraging contemporary methods incorporating a Siamese U-Net approach for satellite images and a custom-trained YOLO model for drone footage. The core purpose of this work is to provide a toolkit for obtaining an overview using satellite imagery and enabling further investigation of areas of interest through the deployment of unmanned aerial vehicles. Experimental validations demonstrate the superior output in accuracy and inference speed of the proposal compared to baseline models, while extended testing in real-world scenarios in Greece and internationally highlights the generalizability of the process across a wide range of cases. The findings showcase the potential for real-time, deployable solutions in resource-constrained environments, bridging the gap between research and practical implementations. Ultimately, as automated disaster assessment continues to improve, the aggregation of analyzed data from previous events will become an invaluable resource for responding to crises moving forward.

Available Files

Services