Deep learning-based position-aware imaging through scattering media

Imaging through scattering media is a fundamental challenge in optics, as scattering scrambles spatial information and severely degrades image visibility. Recent deep learning-based scattering imaging methods have demonstrated high reconstruction accuracy under trained conditions; however, their performance often deteriorates when the scattering configuration deviates from the training condition. In this study, we propose a position-aware scattering imaging framework that estimates the lateral position and axial depth of a diffuser from speckle images and selects an image reconstruction model trained for the corresponding scattering condition. By introducing the diffuser position estimation into image reconstruction, the proposed approach avoids large-scale multi-condition training and enables robust image reconstruction under spatially shifted diffuser conditions. Experimental results demonstrate that the proposed method improves reconstruction robustness against lateral and axial displacement of the diffuser compared with conventional single-position training models. This framework provides a computationally efficient solution for deep learning-based imaging through scattering media under diffuser displacement.