"Fire Segmentation in Multiple Scenarios with fragmentation feature"

"Increasing frequency of fire outbreaks across diverse scenarios has posed profound threats to both human society and ecological environments.Consequently, achieving high-precision semantic segmentation of Unmanned Aerial Vehicle (UAV) imagery has emerged as a critical research priority for disaster monitoring. However, accurate segmentation remains a formidable challenge, primarily due to the scarcity of multi-scene fire datasets, the inherent morphologcal complexity of fire, and the pronounced class imbalance characteristic of UAV-based fire imagery. To address these limitations, we constructed the Urban-System-Fire Dataset, encompassing four distinct fire scenarios, and introduced the concept of \u201dfire fragmentation feature\u201d to provide a systematic framework for classifying and interpreting complex fire imagery. "