Original title: Multimodal RGB-Infrared Combination for UAV-Based Wildfire Segmentation: A Comparative Study on FLAME3
Matheus F. Kovaleski, Luís Garrote, Cristiano Premebida, Jérôme Mendes +1 · September 2026
Spotting the exact edges of a wildfire from the air is tough, but drones can reach dangerous areas quickly and safely. Researchers wanted to know the best way to detect fires by combining normal color camera pictures with heat-sensing infraredImaging that captures heat radiation, making it possible to see fire through smoke or darkness. pictures. They tested three different computer vision models on drone data, trying out three ways to blend the color and heat data together. They also compared these fused systems against models using only color or only heat. They found that heat information plays the dominant role in helping the systems outline fires accurately. Blending the color and heat data at the feature level, while using modern transformer-based models, offered the most promising path forward for this drone-based fire tracking.
Imagine you're looking for a small leak under a kitchen sink in a dark cabinet. With just a flashlight, you might miss it because water blends with shadows. But if you also had a thermometer that could feel temperature changes from the leak, you'd spot it much quicker. In this paper, researchers are combining regular drone photos with heat images to better identify wildfires, similar to using both light and temperature to find a hidden problem.
Firefighters use drones to locate wildfires in hard-to-reach areas. Better detection from combined imagery can help crews respond faster and more safely, protecting communities and natural spaces.
This study is based solely on the FLAME3 dataset, so its findings may not generalize to other wildfire types or environments. It does not assess real-time performance, integration with UAV hardware, or operational costs. The work is comparative and identifies promising directions but does not deploy a full system for real-world use.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Multimodal RGB-Infrared Combination for UAV-Based Wildfire Segmentation: A Comparative Study on FLAME3 — Matheus F. Kovaleski, Luís Garrote, Cristiano Premebida et al.
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.