https://doi.org/10.1140/epjp/s13360-025-06636-y
Regular Article
Self-supervised deep learning ghost imaging through scattering media based on dual physical models
Changchun University of Science and Technology, Changchun, China
Received:
29
March
2025
Accepted:
7
July
2025
Published online:
27
July
2025
This study proposes a self-supervised deep learning method for GI through scattering media, termed DMSSGI, which integrates dual physical models (GI and light scattering) with neural networks to achieve high-quality image reconstruction without relying on labeled training data. The method first combines the GI physical model with a deep network (GI-Net) to extract features from bucket detector signals, where the scattering medium is placed between the light source and the object, affecting the illumination pathway. A scattering medium model (S-Net) is then designed to restore images by jointly optimizing transmission, atmospheric light, and clean image estimation through self-supervised training guided by physical constraints. By eliminating dependency on external datasets and leveraging physics-driven supervision, DMSSGI significantly enhances generalization across diverse scattering scenarios. Experimental results demonstrate that DMSSGI outperforms existing GI and scattering media imaging methods (e.g., DGI, DCP, GIDC) even at ultralow sampling rates (as low as 12.5), with superior PSNR (>25 dB) and SSIM (>0.8) metrics under strong scattering (transmission = 0.4). This work provides a robust and efficient solution for GI through complex scattering environments, bridging the gap between physical models and data-free deep learning.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to Società Italiana di Fisica and Springer-Verlag GmbH Germany, part of Springer Nature 2025
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.