Interpretable deep network improves hyperspectral image noise removal

AXS-Net: Interpretable Deep Unfolding for Hyperspectral Image Denoising via Spectral Basis Unmixing and Structured Noise Refinement

Computer Vision and Pattern Recognition

Summary

Hyperspectral images, which capture data across many color bands, often include various types of noise like stripes and random disturbances that make analysis difficult. The authors propose a new method called AXS-Net that separates the clean image from different noise types by modeling the image as a combination of low-rank spectral components plus structured and random noise. Their approach unfolds an optimization process into a deep network with clear intermediate outputs that explain the underlying clean signals and noise patterns. Tests on multiple datasets show AXS-Net delivers strong denoising performance while clearly revealing the spectral elements and noise.

What this means in practice

  • For remote sensing engineers: Improve remote sensing image quality by separating clean spectral signals from various noise types using an interpretable deep network.
  • For medical imaging specialists: Enhance hyperspectral medical images by removing complex noise patterns while preserving meaningful spectral information for better diagnostics.

Authors

Ziyi Guan, Jianping Zhang, Zheng Yang

Abstract

Hyperspectral images (HSIs) are often degraded by mixed noise, including band-dependent Gaussian perturbations and structured artifacts such as stripes, dead-lines, and impulse noise. Most deep denoisers regress the clean image directly, entangling signal and structured noise. We instead model HSI denoising as $\Y=\A\X+\Snoise+\Nnoise$, where $\A\X$ is a low-rank spectral-subspace (unmixing) reconstruction, $\Snoise$ is structured sparse noise and $\Nnoise$ is residual Gaussian noise. The resulting regularized optimization problem is unrolled into AXS-Net, a $K$-stage alternating proximal-point framework. Each stage combines an analytic spectral-basis gradient step, an SSX-Block proximal operator for abundance coefficients, and an SBlock proximal operator for the structured residual with column-consistent and sparse priors. This optimization correspondence exposes interpretable endmembers, abundance maps, and structured-noise estimates. Across ICVL, CAVE, and Harvard datasets and five noise configurations, the proposed AXS-Net achieves strong in-domain accuracy and competitive zero-shot transfer, with consistent gains across all five noise regimes on ICVL and Harvard. The recovered structured-noise closely follows the synthetic reference, and the recovered spectral basis is smooth and band-ordered rather than an arbitrary set of latent channels.
View PDFOpen arXiv