Integrating remote sensing and U-Nets for wildfire detection

Abstract

Wildfires pose a significant threat to ecosystems and communities worldwide. Early and accurate detection is crucial for effective response and mitigation strategies, making monitoring systems essential for tracking and managing these disasters. Our proposed system utilizes satellites as a remote sensing source to monitor the Earth for real-time wildfire detection. We explore the effectiveness of the U-Net deep learning architecture using three differ- ent fire masks Intersection Masks, Voting Masks, and Murphy Masks on Landsat-8 satellite data. This allows alerting the relevant emergency services and facilitating a rapid response. Additionally, we conducted a comparative study evaluating the performance of U-Net against other commonly used techniques: image classification (InceptionV3), object detection (YOLOv3- tiny), and image segmentation (Fire-Net). The results demonstrate that U-Net is highly effective in wildfire detection, achieving significant perfor- mance metrics. Keywords: Wildfires, Image Segmentation, U-Net, Early Detection, Re- mote Sensing, Satellite Imagery.