PHANTOM: Physics-Informed Hyperspectral Adversarial Network for Transformer-Optimized Methane Detection
IEEE Transactions on Geoscience and Remote Sensing · 2026 · 64 · 5532810-5532810
DOI: 10.1109/tgrs.2026.3741902 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Geoscience and Remote Sensing, 2026. Verfasst von Sachin Giri und 4 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Gefördert von: Center for Methane Emission Research and Innovation through the University of California, Merced Climate Action Seed Funds.
Abstract (Originalsprache)
Accurate detection of methane plumes in hyperspectral imagery is critical for environmental monitoring and greenhouse gas emission assessment. Traditional detection methods often struggle with complex atmospheric conditions, background variability, and subtle spectral signatures. We introduce the physics-informed hyperspectral adversarial network for transformer-optimized methane (PHANTOM) detection, a novel hybrid deep learning architecture that integrates physics-based constraints with advanced neural networks for robust methane plume segmentation. Our model’s core is a U-Net generator with an integrated Transformer bottleneck, designed to capture both local and global spatial dependencies in the data to produce high-fidelity methane abundance maps. The framework is uniquely constrained by two physical principles: 1) a Beer–Lambert reconstruction loss, which models the physical attenuation of surface radiance through the gas plume using a dedicated background estimation network and 2) a reaction–diffusion partial differential equation (PDE) loss, which regularizes the solution to encourage the spatial structure of the predicted plume to be physically plausible. An adversarial discriminator further refines the output, ensuring the generated maps are structurally indistinguishable from ground truth labels. By deeply integrating these physical priors into the training objective, our method produces results that are more physically consistent, robust, and interpretable than traditional data-driven approaches. Benchmarked against the state-of-the-art HyperSTARCOP model, PHANTOM demonstrates superior performance across all evaluation metrics with seven times fewer parameters. It also transfers to the independent spaceborne EMIT dataset without architectural changes, roughly doubling the $F1$ score of matched filter (MF) baselines.
Förderung
- Center for Methane Emission Research and Innovation through the University of California, Merced Climate Action Seed Funds
Zitieren
APA
Giri, S., Hollenbeck, D., Krzysiak, R., Chen, Y., & Du, W. (2026). PHANTOM: Physics-Informed Hyperspectral Adversarial Network for Transformer-Optimized Methane Detection. IEEE Transactions on Geoscience and Remote Sensing, 64, 5532810-5532810. https://doi.org/10.1109/tgrs.2026.3741902
Vancouver
Giri S, Hollenbeck D, Krzysiak R, Chen Y, Du W. PHANTOM: Physics-Informed Hyperspectral Adversarial Network for Transformer-Optimized Methane Detection. IEEE Transactions on Geoscience and Remote Sensing. 2026;64:5532810-5532810. doi:10.1109/tgrs.2026.3741902
BibTeX
@article{Giri2026,
author = {Giri, Sachin and Hollenbeck, Derek and Krzysiak, Rafal and Chen, YangQuan and Du, Wan},
title = {PHANTOM: Physics-Informed Hyperspectral Adversarial Network for Transformer-Optimized Methane Detection},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
year = {2026},
volume = {64},
pages = {5532810--5532810},
doi = {10.1109/tgrs.2026.3741902}
}
EUVIMED