GUARDIAN: Goal-Oriented Unmanned Aerial Vehicles with Adaptive Real-Time Dynamic Intelligence for Advanced Neural Monitoring

Wildfires represent an escalating global threat requiring precise thermal monitoring, yet conventional methods lack the spatiotemporal resolution and adaptability for effective fire management and real-time response mechanisms. We present GUARDIAN, a goal-oriented IoT framework for wildfire monitoring that fuses a reconfigurable UAV swarm with advanced 3D neural reconstruction and language-guided intelligence. A dynamic graph models the wildfire environment, steering adaptive swarm formations to optimize multi-modal data collection. Statistical movement with dynamic resetting increases efficiency, feeding a Gaussian-splatting Neural Radiance Field (NeRF) pipeline that delivers high-fidelity 3D thermal reconstructions in real time. Integrated Large Language Model (LLM)-based Reinforcement Learning with Human Feedback (RLHF) refines swarm behavior, aligning it with reconstruction quality and operator-defined priorities. Experimental results show superior coverage efficiency and thermal accuracy over conventional methods, validating GUARDIAN's ability to provide actionable insights under diverse fire conditions. This scalable solution advances real-time wildfire monitoring, empowering ecological preservation and infrastructure protection with precision and adaptability.