A Joint Communication and Sensing Framework for Metaverse using Unmanned Aerial Vehicles
Jangirova, Sabina
Jangirova, Sabina
Author
Supervisor
Department
Machine Learning
Embargo End Date
2025-05-30
Type
Thesis
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
This thesis addresses the critical challenge of wildfire detection and response to disaster situations by integrating advanced unmanned aerial vehicle (UAV) networks with metaverse technologies and state-of-the-art deep learning approaches. We offer a two-part solution - a communication framework and a compact model for fire detection. By developing the metaverse-based communication framework for UAV network management, we optimize resource offloading, transmission latency, and energy consumption. A cooperative multiagent deep reinforcement learning algorithm based on a Double Deep Q-Network (DDQN) approach is developed to optimize resource allocation and association decisions across the network. We provide simulation results to confirm that the developed method achieves significantly higher cumulative rewards and lower operational costs compared to conventional techniques. In the second part of the thesis, a robust and efficient aerial fire detection model is developed for deployment on UAVs. The proposed solution employs the Mobile ViT-S architecture, which is effective in extracting local features via its convolutional neural network modules and global context modeling via transformer modules. To overcome the limitations of resource-constrained devices, knowledge distillation techniques are used to transfer critical insights from a larger teacher model (ViT/32) to a more compact student model. Extensive experiments on benchmark datasets such as Bo WFire, ADSF, and DFAN reveal that the distilled model not only achieves competitive classification accuracy and enhanced explainability but also significantly reduces inference latency and computational load. The contributions of this research bridge important gaps in wildfire monitoring and UAV-enabled network management. Our developed metaverse-enabled communication framework allows realtime resource optimization, and the lightweight fire detection model ensures rapid and accurate identification of fire events. These two components ensure rapid response to fire disasters even when the monitoring devices have resource constraints. These advancements promise significant benefits in safeguarding lives and the environment while advancing the capabilities of UAV-based surveillance systems.
Citation
Sabina Jangirova, “A Joint Communication and Sensing Framework for Metaverse using Unmanned Aerial Vehicles,” Master of Science thesis, Machine Learning, MBZUAI, 2025.
Source
Conference
Keywords
Metaverse, Convex Optimization, Disaster Management, Knowledge Distillation, Aerial Images