CTIS-GAN: computed tomography imaging spectrometry based on a generative adversarial network.

Area of Science:

• Optics and Photonics
• Computer Science
• Image Processing

Background:

• Computed tomography imaging spectrometry (CTIS) is a snapshot hyperspectral imaging technique.
• CTIS captures 3D data (2D spatial + 1D spectral) in a single exposure.
• The CTIS inversion problem is ill-posed, often requiring slow iterative algorithms.

Purpose of the Study:

• To leverage deep learning to significantly reduce CTIS computational costs.
• To develop a novel generative adversarial network (GAN) for rapid hyperspectral data reconstruction.
• To improve the quality and efficiency of CTIS data cube reconstruction.

Main Methods:

• A generative adversarial network integrated with self-attention was developed.
• The network exploits zero-order diffraction features specific to CTIS.
• The method was validated using simulation studies with real image datasets and noise analysis.

Main Results:

• The proposed network reconstructs a 31-band CTIS data cube in milliseconds (average ~16ms).
• Achieved higher reconstruction quality compared to traditional methods and state-of-the-art (SOTA).
• Demonstrated robustness against various levels of Gaussian noise.

Conclusions:

• The CTIS GAN framework offers a highly efficient and accurate solution for hyperspectral image reconstruction.
• The method significantly reduces computational time while enhancing data quality.
• The framework is adaptable for larger spectral/spatial dimensions and other compressed spectral imaging modalities.