Reconstruction of moving small targets through scattering media: A hierarchical network approach integrating event information

Abstract

Optical scattering presents substantial challenges for imaging systems across various domains, significantly complicating the acquisition of target information. Existing techniques for imaging through scattering media primarily address static targets. However, continuously moving targets will introduce motion blur into the speckle image, thus severely affecting the reconstruction quality. To address this problem, we innovatively introduce an event camera and propose a two-stage speckle reconstruction network (TSR-Net), which effectively integrates speckle and event information. TSR-Net first deblurs speckle images in its first stage, followed by reconstructing moving targets from the refined speckle images in the second stage. Event data is leveraged throughout the reconstruction process, being extracted and fused at multiple levels to enhance the backbone network's performance in deblurring and reconstruction, thereby guiding training more effectively. The dedicated datasets of speckle images were collected and processed to evaluate our approach. Experimental results highlight the superior reconstruction performance of the proposed method, especially for small pixel-level objects in continuous motion.