AI-EMPOWERED UAVS FOR RAPID DISASTER RESPONSE AND MANAGEMENT

Abstract

Efficient disaster response and survivor detection are essential in minimizing casualties and mitigating the impact of disastrous events. This thesis presents an innovative approach to addressing these challenges through a combination of Transfer Learning and Deep Reinforcement Learning (DRL).We introduce a Transfer Learning framework aimed at enhancing the speed and accuracy of survivor detection in aerial imagery by fine-tuning pre-trained convolutional neural networks (CNNs). Leveraging Transfer Learning significantly reduces computational costs while improving detection performance. Building upon this foundation, a novel DRL-based multi-drone targets visitation system designed for efficient disaster response missions in dynamic environments. Optimal drone deployment strategies are learned through extensive experimentation, and the system's adaptability to varying mission scales, target numbers, and area sizes is evaluated. The proposed framework offers a promising solution to enhance disaster response capabilities, outperforming traditional baseline policies in terms of energy consumption and mission efficiency. This thesis highlights the potential of Artificial Intelligence (AI) empowered Unmanned Aerial Vehicles (UAVs) technology in improving disaster management and guides future investigations in this critical field.