Physics-Informed Spectral Modeling for Hyperspectral Imaging

This work addresses the challenge of analyzing hyperspectral imaging data for applications like remote sensing or medical imaging, offering improved performance and interpretability, though it appears incremental as it builds on existing physics-informed and deep learning approaches.

The paper tackles the problem of modeling hyperspectral imaging data by introducing PhISM, a physics-informed deep learning architecture that learns without supervision to disentangle observations and model them with continuous basis functions, resulting in outperforming prior methods on classification and regression benchmarks with limited labeled data and providing interpretable latent representations.