Degradation-Aware Metric Prompting for Hyperspectral Image Restoration

Unified hyperspectral image (HSI) restoration aims to recover diverse degradations within a single model. However, current methods often rely on impractical explicit priors or opaque black-box representations that overfit to training distributions, hampering generalization to unseen scenarios. To bridge this gap, we propose Degradation-Aware Metric Prompting (DAMP), a novel framework that characterizes multi-dimensional degradations through interpretable spatial-spectral metrics. These metrics serve as Degradation Prompts (DP), enabling the model to capture shared characteristics across tasks and adapt to unknown corruptions. Central to our framework is the Degradation-Adaptive Mixture-of-Experts (DAMoE), where Spatial-Spectral Adaptive Modules (SSAMs) serve as experts that utilize learnable fusion coefficients to specialize in distinct degradation degrees. By using DP as a gating router, DAMoE dynamically activates specialized experts tailored to the specific degradation profile. Extensive experiments on natural and remote sensing HSI datasets demonstrate that DAMP achieves state-of-the-art performance and exhibits exceptional zero-shot generalization on unseen restoration tasks. Code is publicly available at \href{DAMP}{https://github.com/MiliLab/DAMP}.

Hyperspectral images capture hundreds of invisible light wavelengths beyond human vision, making them indispensable for environmental monitoring, agriculture, and satellite remote sensing. However, these images frequently suffer from various types of degradation, such as noise, low resolution, and blurriness. Existing restoration methods either require explicit prior knowledge of the exact type of damage affecting the image or operate as opaque "black boxes". We developed a novel framework called DAMP. It first accurately analyzes the degradation characteristics of an image and generates clear, interpretable "damage descriptions". It then uses a set of specialized "expert modules" that automatically activate the most appropriate experts based on the specific damage profile of each image. Our method achieves state-of-the-art performance on both natural and satellite hyperspectral datasets. Most importantly, it demonstrates excellent zero-shot generalization across multiple unseen restoration tasks, making hyperspectral image restoration far more practical and reliable for real-world complex scenarios.