Chromatic aberration-driven computational imaging for hyperspectral depth-of-field extension

Hyperspectral imaging systems provide abundant spectral information. However, owing to axial chromatic aberration and the limited depth of field of conventional optical systems, the best-focus planes at different wavelengths are axially displaced, causing wavelength-dependent variations in image sharpness that hinder subsequent spectral analysis. Existing extended-depth-of-field imaging methods typically rely on hardware-assisted schemes, such as reducing the aperture size, acquiring images at multiple focal settings, multi-focus image fusion, wavefront coding, or the design of specialized optical elements. These approaches require complex system calibration and substantially increase hardware cost and system complexity. To address this issue, we regard the inherent axial chromatic aberration of the lens as a physics-based geometric prior and exploit the wavelength-dependent differences in defocus among spectral bands. Accordingly, we propose a physics-driven, unsupervised, end-to-end network for all-in-focus hyperspectral image restoration,thereby enabling extended-depth-of-field hyperspectral imaging from a single acquisition of a spectral image sequence. Taking the captured defocused spectral sequence as input, the proposed method first integrates complementary information across spectral bands to construct an all-in-focus geometric guidance map. This map then serves as a shared structural constraint for progressively restoring each spectral band and accurately reconstructing spectral details across the entire wavelength range. The proposed method requires neither additional depth sensing, accurate blur-kernel estimation, nor complex optical modifications. Experimental results obtained in indoor (1–3 m) and outdoor (1.2–45 m) scenes demonstrate that the proposed method substantially enhances spatial details while effectively preserving spectral fidelity, providing an efficient and broadly applicable computational imaging framework for extended-depth-of-field hyperspectral imaging.