This survey provides a comprehensive overview of the latest advances in artificial intelligence (AI)-enabled drone systems for wildfire detection, with a particular focus on lightweight, edge-deployable architectures suitable for resource-constrained UAV platforms. We critically examine the landscape of computer vision approaches across both single- and multi-modal sensor pipelines. Our analysis reveals that YOLOv8 consistently offers one of the strongest trade-offs between detection accuracy, inference speed, and energy efficiency on edge devices. Emphasis is placed on adaptive camera management strategies dynamically switching between RGB and thermal imaging based on environmental conditions and the practical integration of gas sensors for early warning or fallback verification, while deliberately avoiding the high computational and power overhead associated with advanced multi-sensor data fusion. Analysis of real-world deployments highlights persistent challenges, including limited battery endurance, reliable detection of small or obscured fires, and robustness under environmental variability such as fog or low light. Beyond conventional directions, this survey points to promising research avenues including ultra-lightweight model optimization via quantization, pruning, and knowledge distillation, advanced adaptive sensing logic, predictive modeling of time-dependant fire spread dynamics using sequential UAV imagery, and the creation of more diverse, large-scale annotated datasets. This synthesis offers practical guidance for researchers and practitioners developing next-generation, truly field-ready wildfire detection systems.
Marrivada et al. (Mon,) studied this question.