Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study explores active wildfire segmentation using multispectral Landsat-8 imagery, addressing the challenges posed by sparse fire pixels, particularly in early-stage fires. By employing a data-driven protocol that analyzes fire-region size distributions, the researchers evaluated three segmentation architectures—U-Net, DeepLabV3+, and SegFormer—across varying local fire-region densities. The results indicate that U-Net consistently outperforms the others in robustness, while SWIR2 spectral band selection is crucial for enhancing segmentation accuracy in low-density fire scenarios.
U-Net outshines its competitors in wildfire segmentation, proving that the right spectral band can make all the difference in low-density fire detection.
Active wildfire mapping from satellite imagery is challenging due to the sparse and highly imbalanced nature of fire pixels, especially in early-stage or low-density fire observations. This work investigates the use of multispectral Landsat-8 imagery for active-fire segmentation under multi-scale wildfire size conditions. We propose a data-driven protocol to characterize fire-region size distributions through connected-component analysis and an interquartile range criterion, enabling the evaluation of model robustness across different local fire-region densities. Three segmentation architectures, U-Net, DeepLabV3+, and SegFormer, are evaluated under different SWIR-based spectral configurations. Results show that U-Net achieves the strongest robustness across the evaluated conditions, SegFormer provides competitive performance, and DeepLabV3+ tends to produce conservative predictions with reduced recall. Across architectures, SWIR2 consistently achieves the strongest or near-best results, highlighting its importance for active-fire segmentation in Landsat-8 imagery. These findings suggest that both spectral band selection and architectural design are critical for robust satellite-based active wildfire mapping trained on low active fire-pixel density images.