C047

AI-DRIVEN DRONE-BASED WILDFIRE DETECTION USING RGB-THERMAL IMAGING AND DEEP LEARNING MODELS

DR. PRABHA KUMARESAN, SHEIKH ADAM ZACHARY BIN SHEIKH NAZIRUDDIN

AFFILIATION
Faculty of Computing & Informatics, Multimedia University

Description of Invention

Wildfires pose a growing threat to ecosystems, safety, and economies, making early detection vital. This project introduces a drone-based detection system using RGB and thermal imaging with deep learning models—YOLOv8, EfficientNet-B0, and Fire Fusion-Net—trained on the FLAME-3 dataset. Designed for real-time, lightweight drone deployment, the system enhances mobility and cost-efficiency. The SVM classifier on Fire Fusion-Net features achieved 76% accuracy and 94% recall for smoke, though precision for non-smoke images was lower at 50%. Despite this, the system effectively identifies wildfire signs, offering a scalable, AI-driven solution for fast response, improved situational awareness, and reduced environmental and economic damage.