Imaging reconstruction of targets in highly turbid and inhomogeneous liquids via TSIDF

With the aim of meeting critical technological demands in fields such as underwater resource exploration, marine rescue, and oceanographic research, this paper presents an innovative exploration of target detection and imaging technology within highly turbid water bodies. In such environments, the profuse and random distribution of suspended particles induces a strong multiple light scattering effect, leading to significant distortion of the target light field information and severe degradation of the imaging quality. Notably, even after ballistic photons are virtually extinguished, the randomized photons still carry target characteristics. To extract the target information embedded within these randomized photons, a transmittance scattering intensity distribution function (TSIDF) based on a Mixed Gaussian model is introduced to characterize the optical scattering properties of highly turbid media. Furthermore, by leveraging this statistical characterization, we introduce a novel deep neural network architecture, URBF-Net, which achieves effective decoupling of target light field information from scattering effects. The experiments utilized milk solutions at concentrations of 0.15%, 0.2%, 0.25%, and 0.4% to simulate highly turbid environments. Under conditions where the target light energy suffers scattering attenuation up to 126.4 dB, successful three-dimensional reconstruction of targets immersed in turbid media was accomplished using a time-of-flight (TOF) 3D imaging system. Significant progress was also achieved in dynamic imaging tests of live fish, with the structural similarity index (SSIM) of the reconstructed images exceeding 0.90. The methodology proposed in this study demonstrates that randomized scattering photons indeed retain target optical information, thereby providing substantial experimental evidence for underwater imaging technology in extremely turbid conditions.