Transformer Enabled Dual-Comb Ghost Imaging for Optical Fiber Sensing

David Dang, Myoung-Gyun Suh, Maodong Gao, Dong-Chel Shin, Aggraj Gupta, ByoungJun Park, Can Uzundal, Beyonce Hu, Yucheng Jin, Wilton J. M. Kort-Kamp, Ho Wai Howard Lee

Keywords: Sparse Sensing, Ghost Imaging, Compressed sensing, Deep learning reconstruction

TL;DR: We propose a dual-comb ghost imaging system with a transformer-based Optical Ghost-GPT model that enables fast, robust, and high-quality image reconstruction at ultra-low sampling ratios, with potential for minimally invasive fiber-based endoscopy.

Abstract: Ghost imaging (GI) reconstructs images from single-pixel measurements but remains hindered by slow pattern projection and noise-sensitive reconstruction. We present a dual-comb ghost imaging framework that addresses these limitations. Dual optical frequency combs generate hundreds of uncorrelated speckle patterns in parallel, enabling snapshot bucket-sum detection through a single-core fiber without spatial or spectral scanning. For recovery, we introduce Optical Ghost-GPT, a transformer-based model that achieves real-time, high-fidelity reconstruction at ultra-low sampling ratios. This combination of dual-comb hardware and deep learning significantly outperforms classical GI in speed, robustness, and image quality. As a proof of concept, we highlight fiber-based endoscopy as a key application, where our approach could deliver minimally invasive, high-resolution imaging with sub-millimeter hardware.

Submission Number: 4