Original title: Scale-based Approach for Active Wildfire Segmentation on Satellite Imagery
Matheus F. Kovaleski, Cristiano Premebida, João Ruivo Paulo · September 2026
Spotting active wildfires from space is tough. Fire pixels are rare and unevenly spread in satellite pictures, especially when fires are small, new, or low in density. Researchers studied how to outline fires using multispectral Landsat-8 images across different fire sizes. They built a method that groups nearby fire pixels and uses a statistical measure called an interquartile range to sort fires by size. This let them test how well three computer vision models handled different fire densities. They found that a model named U-Net proved the most robust across all conditions. SegFormer was also competitive, while DeepLabV3+ made cautious guesses that missed fires. Finally, using a shortwave infrared band called SWIR2 consistently yielded the best or near-best fire detection results.
Imagine you are looking for a specific coin dropped in a large, messy drawer. The coin is small and blends in with similar-looking items, so glancing quickly might miss it. In this paper, finding wildfires in satellite images is like that search—the fires are tiny spots among vast landscapes. The researchers found that using the right 'view' (the SWIR2 band) and the best magnifying tool (certain computer models) helps spot the fires reliably, even when they are just starting.
Early wildfire detection from satellites helps firefighters respond faster, potentially saving homes and lives. This work shows which specific data and methods make detection more reliable, especially for small, hard-to-see fires.
The abstract does not specify the exact number of images or fires used, nor does it compare performance on actual fire response outcomes. It focuses on model evaluation under controlled conditions, not real-world deployment or other detection methods.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Scale-based Approach for Active Wildfire Segmentation on Satellite Imagery — Matheus F. Kovaleski, Cristiano Premebida, João Ruivo Paulo
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.