Cheng Dai

Cheng Dai

Researcher in thermal infrared hyperspectral imaging, physical vision, and efficient thermal perception.

HADAR-Based Thermal Infrared Hyperspectral Image Restoration

Published in arXiv, 2026

Thermal-infrared hyperspectral imagery provides rich scene information, but practical data are often degraded by sensor noise, spectral miscalibration, missing or corrupted bands, and structured artifacts. This work introduces HAIR (HADAR-based Image Restoration), a physics-driven restoration framework for ground-based TIR-HSI.

The method models thermal hyperspectral radiance using physically meaningful temperature, emissivity, and texture components. By combining the HADAR rendering equation with atmospheric downwelling radiative transfer, HAIR performs restoration through a TeX decompose-synthesize strategy that supports denoising, inpainting, spectral calibration, and spectral super-resolution.

The paper evaluates HAIR on outdoor DARPA Invisible Headlights data and in-lab FTIR measurements, showing improved objective accuracy, visual quality, and physical consistency compared with prior restoration methods.

Links: arXiv abstractHTML versionPDFGitHub

Recommended citation: Cheng Dai, Jiale Lin, Bingxuan Song, Yifei Chen, Jiashuo Chen, Xin Yuan, and Fanglin Bao. (2026). "HADAR-Based Thermal Infrared Hyperspectral Image Restoration." arXiv preprint arXiv:2605.13664.
Download Paper