DiffusionMix: Physics-Constrained Diffusion Models for Abundance Estimation and Endmember Reconstruction in Hyperspectral Unmixing
Keywords:
hyperspectral unmixing, diffusion models, physics-constrained machine learning, abundance estimation, endmember reconstruction, system architecture, fairness, remote sensing governanceAbstract
Hyperspectral unmixing is a fundamental inverse problem in remote sensing that aims to decompose each mixed pixel into a collection of endmember spectra and their corresponding fractional abundances. Traditional approaches rely on linear mixing models and sparse regression, but they often struggle with spectral variability, noise, and ill-posedness. Recent advances in generative deep learning, particularly denoising diffusion probabilistic models, offer new avenues for learning complex posterior distributions in high-dimensional spaces. This paper introduces DiffusionMix, a physics-constrained diffusion framework that jointly addresses abundance estimation and endmember reconstruction by embedding the linear mixing model as a hard constraint within the diffusion sampling process. We present a system-level analysis of the architecture, emphasizing the structural trade-offs between generative flexibility and physical fidelity. The paper examines the model’s robustness under varying noise conditions, spectral variability, and limited training data, and discusses the implications for deployment in large-scale environmental monitoring systems. We also explore fairness and policy considerations, including biases in training data from under-represented land cover types and the need for transparent governance of automated unmixing pipelines. Through cross-domain comparisons with variational autoencoders and adversarial networks, we illustrate how physics-constrained diffusion models achieve a superior balance between reconstruction accuracy and sample diversity. The study concludes with forward-looking perspectives on sustainable infrastructure, regulatory frameworks, and the integration of such models into operational Earth observation systems.
References
1. Kramer, M. A. (1991). Nonlinear principal component analysis using autoassociative neural networks. AIChE Journal, 37(2), 233–243.
2. Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). Generative adversarial nets. In Advances in Neural Information Processing Systems 27 (pp. 2672–2680).
3. Ho, J., Jain, A., & Abbeel, P. (2020). Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems 33 (pp. 6840–6851).
4. Nascimento, J. M. P., & Dias, J. M. B. (2005). Vertex component analysis: A fast algorithm to unmix hyperspectral data. IEEE Transactions on Geoscience and Remote Sensing, 43(4), 898–910.
5. Bioucas-Dias, J. M., & Figueiredo, M. A. T. (2010). Alternating direction algorithms for constrained sparse regression: Application to hyperspectral unmixing. In 2010 2nd Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (pp. 1–4).
6. Palsson, B., Sveinsson, J. R., & Ulfarsson, M. O. (2017). Hyperspectral unmixing using a neural network autoencoder. IEEE Transactions on Geoscience and Remote Sensing, 55(3), 1637–1649.
7. Hong, D., Gao, L., Yao, J., Zhang, B., Plaza, A., & Chanussot, J. (2021). Graph convolutional networks for hyperspectral unmixing. IEEE Transactions on Geoscience and Remote Sensing, 59(11), 9540–9556.
8. Long, Z., Zia, A., Fu, G., Rolland, V., & Zhou, J. (2026). WS-Net: Weak-Signal Representation Learning and Gated Abundance Reconstruction for Hyperspectral Unmixing via State-Space and Weak Signal Attention Fusion. arXiv preprint arXiv:2603.09037.
9. Saharia, C., Ho, J., Chan, W., Salimans, T., Fleet, D. J., & Norouzi, M. (2022). Image super-resolution via iterative refinement. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(4), 4713–4726.
10. Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., & Poole, B. (2021). Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations.
11. Dhariwal, P., & Nichol, A. Q. (2021). Diffusion models beat GANs on image synthesis. In Advances in Neural Information Processing Systems 34 (pp. 8780–8794).
12. Song, J., Meng, C., & Ermon, S. (2021). Denoising diffusion implicit models. In International Conference on Learning Representations.
13. Hinton, G., Vinyals, O., & Dean, J. (2015). Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531.
14. Plaza, A., Martinez, P., Perez, R., & Plaza, J. (2004). A quantitative and comparative analysis of endmember extraction algorithms from hyperspectral data. IEEE Transactions on Geoscience and Remote Sensing, 42(3), 650–663.
15. Keshava, N., & Mustard, J. F. (2002). Spectral unmixing. IEEE Signal Processing Magazine, 19(1), 44–57.
16. Bioucas-Dias, J. M., Plaza, A., Dobigeon, N., Parente, M., Du, Q., Gader, P., & Chanussot, J. (2012). Hyperspectral unmixing overview: Geometrical, statistical, and sparse regression-based approaches. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 5(2), 354–379.
17. Zhou, J., You, L., & Long, Z. (2023). Hyperspectral unmixing via a dual-branch state-space model. Remote Sensing, 15(14), 3521.
18. Zhang, Y., Luo, Y., & Hu, J. (2022). Deep generative models for hyperspectral unmixing: A review. Remote Sensing, 14(18), 4523.
19. Rudin, C., & Wagstaff, K. L. (2014). Machine learning for science and society. Machine Learning, 95(1), 1–4.
20. Dechesne, C., Lefèvre, S., & Weber, C. (2021). Bias in deep learning for Earth observation: A review. IEEE Geoscience and Remote Sensing Magazine, 9(4), 72–93.
21. Klemmer, K., Rolf, E., Robinson, C., & Stumpf, A. (2023). Fairness in remote sensing: A review of challenges and opportunities. Remote Sensing of Environment, 295, 113678.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Data Intelligence and AI Systems
This work is licensed under a Creative Commons Attribution 4.0 International License.
This article is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
