ARTFEED — Contemporary Art Intelligence

FSCM: A GAN Framework for Infrared Hyperspectral Image Colorization

other · 2026-05-18

A new paper on arXiv introduces FSCM, a generative adversarial network (GAN) framework designed to colorize infrared hyperspectral images. Thermal infrared imaging is robust to illumination changes and smoke, aiding all-weather perception, but lacks natural color and fine texture, hindering target recognition and visual interpretation. Existing colorization methods rely on single-band images, causing structural distortion and semantic confusion due to insufficient spectral cues. Infrared hyperspectral images offer rich spectral data, but current single-band frameworks poorly model spatial-spectral coupling and weak textures. FSCM addresses these issues with a spectral-information-guided GAN, featuring a frequency-enhanced spatial-spectral state-space generator built from cascaded FSB units. Each FSB integrates state-space modeling and frequency enhancement to improve colorization quality. The paper is published on arXiv under ID 2605.15880.

Key facts

  • FSCM is a GAN framework for infrared hyperspectral image colorization.
  • Thermal infrared imaging is robust to illumination variations and smoke interference.
  • Existing infrared colorization methods use single-band images, leading to structural distortion.
  • Infrared hyperspectral images provide rich spectral responses and material information.
  • FSCM uses a spectral-information-guided GAN with a frequency-enhanced generator.
  • The generator consists of cascaded FSB units integrating state-space modeling.
  • The paper is available on arXiv with ID 2605.15880.
  • The approach targets all-weather perception and target recognition.

Entities

Institutions

  • arXiv

Sources