A2D2: AI-Driven Autonomous Drone-Based Detection Network for Wildfires

Unmanned aerial vehicles (UAVs) or drones have promising advantages in wildfire detection, outperforming conventional systems in mobility and versatility. However, current drone operations such as path planning, maneuver control, and data analysis require significant manual work, restricting drone systems' efficiency and effectiveness. To reduce human efforts while improving performance, this paper proposes A2D2, an AI-driven Autonomous Drone-based Detection network, aiming to fully automate the drone operational process for wildfire detection. The proposed system is driven by AI models and automated solutions, including a deep computer vision (DCV) model for wildfire detection, a coverage path planning (CPP) method managing patrol navigation, and a deep reinforcement learning (DRL) model controlling drone's path towards the detected wildfires for inspection. This system programmatically manages the model functions in different stages to ensure autonomous operations with minimal human intervention. This paper describes the design, implementation, and experimental tests of the proposed system, including mathematical analysis and algorithms. The approaches and solutions introduced in this research contribute to the advancements of integrating AI technologies into autonomous drone systems and applications.