| description abstract | Wildfires have become a persistent and growing global risk, causing increasing financial, human, and environmental damage. By all accounts and predictions, they will continue to rise in frequency and intensity throughout the 21st century. This paper begins by analyzing the physics of fire and outlines why detecting wildfires in their incipient stages is the most effective way to manage them. We review the various architectures and approaches adopted for wildfire detection, including spaceborne, airborne, fixed cameras, and sensor networks. The paper further analyzes the pros and cons of each approach and reviews recent deployments and published research. In particular, it focuses on the growing and significant role that Artificial Intelligence (AI) and Deep Learning (DL) play in improving the effectiveness of the aforementioned architectures. It examines recent algorithms and models published by various wildfire detection platforms and compares their effectiveness. The study suggests that the most effective solutions combine elements of the mentioned architectures, integrating different sensors to look for different fire signatures, and coupling them with sophisticated DL algorithms to maximize sensitivity while minimizing false alarms. An important trend is the advancement of low-power high-performance hardware architectures, enabling real-time operation of DL algorithms on an edge device with limited memory and processing resources. As seconds and minutes can significantly impact our ability to effectively suppress a wildfire, the ability to process data, in real-time at the network edge, even in remote, unpredictable, and fragile environment is crucial. | |
