Underwater reflective single-pixel imaging based on parallel networks through strong scattering media under low sampling rates

Abstract

Single-pixel imaging (SPI) boasts higher detection sensitivity and a broader detection bandwidth, and it exhibits significant advantages in extremely low-light conditions and underwater imaging involving scattering media. However, with the increase in the intensity of scattering media in underwater environments, the image reconstruction quality of SPI severely deteriorates. In this paper, an underwater reflective SPI system based on parallel networks is designed and built to achieve high-quality imaging in turbid waters at low sampling rates. The proposed network consists of an upper branch and a lower branch, and the upper branch consists of a multi-scale initial feature extraction network and a multi-scale feature transformation network to enhance the network's ability to learn high-frequency information, and the lower branch is mainly responsible for generating images. Additionally, a multi-scale structural similarity index measure and normalized mean square error are also designed as loss functions to better learn images features with varying sizes. Simulations and experiments demonstrate that the network is capable of reconstructing underwater objects at a 3.125% sampling rate and 90 NTU turbidity, and it indicates that the network has strong generalization abilities.