Deep Spectral Unmixing for Hyperspectral Imaging via Nonlinear Neural Operators
American Journal of Machine Learning · 2025
DOI: 10.71465/ajml3461 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal of Machine Learning, 2025. Verfasst von Joseph Davis. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Hyperspectral imaging has emerged as a pivotal technology in remote sensing, enabling precise material identification through high-dimensional spectral signatures. However, the low spatial resolution inherent to hyperspectral sensors frequently results in mixed pixels, where a single pixel contains spectral contributions from multiple distinct materials. Consequently, spectral unmixing—the decomposition of mixed pixels into pure endmembers and their fractional abundances—remains a fundamental challenge. Traditional approaches largely rely on the Linear Mixing Model, which fails to account for complex photon interactions, such as intimate mixtures and multi-path scattering. While recent deep learning techniques using Convolutional Neural Networks and Autoencoders have improved nonlinear unmixing, they often suffer from resolution dependence and limited generalization capabilities across varying spectral domains. This paper introduces a novel framework: Deep Spectral Unmixing via Nonlinear Neural Operators. By shifting the learning paradigm from mapping finite-dimensional vectors to learning operators between infinite-dimensional function spaces, we propose a discretization-invariant architecture capable of capturing global spectral dependencies and complex nonlinear interactions. Extensive experiments on both synthetic and real-world datasets demonstrate that our approach outperforms state-of-the-art geometric, statistical, and deep learning baselines in terms of reconstruction accuracy and abundance estimation.
Zitieren
APA
Davis, J. (2025). Deep Spectral Unmixing for Hyperspectral Imaging via Nonlinear Neural Operators. American Journal of Machine Learning. https://doi.org/10.71465/ajml3461
Vancouver
Davis J. Deep Spectral Unmixing for Hyperspectral Imaging via Nonlinear Neural Operators. American Journal of Machine Learning. 2025. doi:10.71465/ajml3461
BibTeX
@article{Davis2025,
author = {Davis, Joseph},
title = {Deep Spectral Unmixing for Hyperspectral Imaging via Nonlinear Neural Operators},
journal = {American Journal of Machine Learning},
year = {2025},
doi = {10.71465/ajml3461}
}
EUVIMED