Deep Learning Empowered Parallelized Metasurface Computed Tomography Snapshot Spectral Imaging

Area of Science:

• Optics and Photonics
• Computational Imaging
• Medical Technology

Background:

• Snapshot spectral imaging captures spatial-spectral data rapidly but is limited by bulky optics.
• Current systems are unsuitable for space-constrained applications like endoscopy and cellular imaging.

Purpose of the Study:

• To develop an ultracompact snapshot spectral imaging system.
• To overcome the size limitations of existing spectral imaging technologies.

Main Methods:

• Proposed a parallelized metasurface computed tomography strategy.
• Integrated seven multifunctional sub-metasurfaces for multi-angle spectral data acquisition.
• Employed generative adversarial deep neural networks for image reconstruction.

Main Results:

• Reduced optics volume from cm³ to sub-mm³ scale.
• Achieved single snapshot imaging in 38 ms.
• Obtained a spectral resolution of 10 nm (450-650 nm range).

Conclusions:

• The developed technique significantly miniaturizes snapshot spectral imaging systems.
• Enables integration into microscopy and endoscopic devices for advanced medical diagnosis.