Pixel-Wise Machine Learning and Deep Learning Methods Implementation on Multi-Class Wildfire Mapping
B.S. (Bachelor of Science)
Department of Geographic and Atmospheric Sciences
Wildfires are destructive natural hazards. Artificial Intelligence (AI) has been a trendy topic in recent years due to its powerful applicability. This study focuses on the use of artificial intelligence (AI) in hazard management, specifically in the field of wildfire mapping. Machine learning and Deep learning are two subsets of AI. This study applied pixel-wise machine learning and deep learning methods to do multi-class mapping on two wildfire events in California, USA. The purpose of this research is to demonstrate the usefulness and advantages of using AI in the field of hazard management. The machine learning methods selected are Random Forest, eXtreme Gradient Boosting and Support Vector Machine. The deep learning method used is U-Net. The results indicate that U-Net did the best job at classifying wildfire events, while SVM had the best performance among machine learning algorithms. U-Net is, however, the most time-consuming model due to the nature of deep learning. There are some aspects of this study that can be improved. The models may be tuned to have better performances. And it will be better to use hand-labeled masks to make the deep learning model more useful in more complex conditions. This research emphasizes that the use of AI in hazard management can improve the accuracy and efficiency of wildfire mapping. It also highlights the potential for AI to be applied in other fields of hazard management. Overall, the study demonstrates the usefulness and advantages of using AI in wildfire mapping and provides insights into how this technology can be further optimized for hazard management.
Wu, Mingda, "Pixel-Wise Machine Learning and Deep Learning Methods Implementation on Multi-Class Wildfire Mapping" (2023). Honors Capstones. 1464.