Deep learning-enhanced spectral ghost imaging with accelerated and high-fidelity reconstruction

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning Applications

Background:

• Ghost imaging reconstructs images using light correlations.
• Spectral ghost imaging extends this to the spectral domain by modulating light components.
• Traditional methods require extensive measurement data.

Purpose of the Study:

• To develop a computational spectral ghost imaging method.
• To improve image reconstruction fidelity and reduce measurement time.
• To demonstrate the efficacy of a deep learning approach for spectral ghost imaging.

Main Methods:

• Implementation of computational spectral ghost imaging.
• Development of a deep learning framework: Spectral Ghost Imaging using Convolutional Neural Network (SGICNN).
• Training SGICNN exclusively on simulated data for image reconstruction and denoising.

Main Results:

• SGICNN achieved high-fidelity spectral image reconstruction from only 8000 realizations.
• This surpasses the accuracy of images reconstructed with 100,000 measurements.
• Demonstrated over a 10x reduction in acquisition time without compromising image quality.

Conclusions:

• The proposed SGICNN method offers robust and straightforward spectral ghost imaging.
• Achieves significant reduction in measurement acquisition time.
• Shows strong potential for remote spectral sensing and high-resolution integrated spectrometers.