Underwater Target Detection Utilizing Polarization Image Fusion Algorithm Based on Unsupervised Learning and...

Area of Science:

• Computer Vision
• Optical Engineering
• Marine Technology

Background:

• Underwater images suffer from low brightness, blurriness, and lost details due to light absorption and scattering.
• Conventional intensity cameras provide limited information in aquatic environments.

Purpose of the Study:

• To develop a deep learning-based method for fusing underwater polarization and intensity images.
• To improve the quality and detail of underwater imagery for enhanced visual analysis.

Main Methods:

• A deep fusion network was designed to merge polarization and intensity images.
• An unsupervised learning framework guided by an attention mechanism was employed for image fusion.
• A custom dataset of underwater polarization images was created and augmented for training.

Main Results:

• The fused underwater images exhibited significantly improved detail, with a 24.48% increase in information entropy and a 139% increase in standard deviation compared to intensity images.
• The method outperformed other fusion-based approaches in image processing quality.
• An improved U-net network facilitated feasible target segmentation even in turbid water conditions.

Conclusions:

• The proposed deep fusion method enhances underwater image quality effectively.
• The technique offers robustness, self-adaptability, and faster operation speeds, suitable for ocean detection and underwater target recognition.