Grant: Discovery Grants Program - Individual

Wildland and forest fires are a natural risk that poses a significant threat to ecosystems, wildlife, and human lives. With climate change, experts expect an increase in wildland fires in the coming years. This shows the necessity for effective approaches in the prevention and fighting against forest fires. This proposal presents a strategy for the use of artificial intelligence (AI) in the study of wildland and forest fires. The proposed research program aims to develop AI solutions for the improved management of wildland and forest fires. The specific objectives include developing deep learning algorithms for fire segmentation, generating synthetic fire images using generative AI and simulation platforms, predicting fire spread using AI algorithms, detecting hot spots by machine learning and computer vision, optimizing algorithms for embedded platforms, adapting algorithms to different environments, and exploring multiple spectrums for improved detection. For early detection of forest fires, AI can play a pivotal role by analyzing various types of imagery, including ground, aerial, and satellite images. By developing state-of-the-art deep learning and transformers algorithms, we aim to develop intelligent systems capable of identifying smoke plumes, flame signatures, and other fire-related indicators in real-time. The research will focus on optimizing the performance of these algorithms, enabling them to handle diverse environmental conditions and various fire types. In addition to the visible spectrum, the research will evaluate the use of multiple spectrums, such as infrared, thermal, and hyperspectral imaging. Our research also emphasizes domain adaptation techniques to facilitate the robust transfer of developed algorithms between different scenarios. By exploring methods that leverage existing knowledge from one fire-prone region to another, we can significantly reduce the time and effort required for algorithm deployment. Generative AI algorithms and photorealistic simulations will be developed to facilitate the creation of a comprehensive fire image dataset. These synthesized images, combined with real-world data, will serve as valuable inputs for training and testing the developed algorithms. Another important topic of this research is fire spread prediction. Utilizing multimodal data sources, advanced AI algorithms will be developed to predict the spread of fires over time. By integrating machine learning with fire behavior modeling, these algorithms will provide valuable insights into fire progression, thus helping in firefighting containment strategies. Underground fires often pose a long-term risk, can be challenging to detect and address effectively. By utilizing drones, thermal imaging and gas sensing technologies, we will develop a system capable of detecting hot spots associated with underground fires. This solution will enhance post-fire monitoring and support targeted mitigation efforts.