Polarization-resolved semantic segmentation for classification and detection of materials in a turbid underwater environment

With the development of computer vision technology, deep learning-based segmentation methods have achieved impressive results in conventional environments using intensity images. However, in turbid scattering environments, many network models struggle to accurately segment targets using intensity images. Here, a polarization-resolved material segmentation network based on the Mueller transform matrix is proposed to mitigate the scattering effects, realizing polarization-driven accurate segmentation for target material classification and detection in turbid underwater environments. The ablation experiments validate the effectiveness of the proposed network architecture, and the experiment results show that the approach outperforms existing segmentation methods, especially in turbid scattering environments. This work provides deeper insights into polarization-driven imaging and material segmentation tasks in challenging environments.