Underwater polarization image de-scattering utilizing a physics-driven deep learning method.

Area of Science:

• Optics and Photonics
• Computer Vision
• Machine Learning

Background:

• Polarization imaging effectively suppresses underwater backscattered light, making it valuable for subsea applications.
• Current learning-based polarization methods often lack interpretability and generalizability due to their data-driven nature.

Purpose of the Study:

• To develop an interpretable and generalizable polarization de-scattering method for underwater imaging.
• To integrate active polarization imaging principles with deep learning for enhanced performance.

Main Methods:

• A novel deep learning network incorporating a polarization feature-refined block to focus on specific polarization information.
• Direct prediction of active polarization imaging model parameters by the network, removing the need for manual estimation.
• Development of specialized loss functions to restore polarization and intensity information in degraded underwater images.

Main Results:

• The proposed method successfully generates clear de-scattered underwater images.
• Experimental results show superior performance compared to existing advanced methods across diverse target materials and turbidity levels.
• The method effectively restores polarization information and improves intensity recovery.

Conclusions:

• The combined active polarization imaging model and deep learning approach offers an interpretable and generalizable solution for underwater de-scattering.
• The technique significantly enhances the quality of underwater images, outperforming current state-of-the-art methods.
• This work advances the application of polarization imaging in challenging underwater environments.