Underwater ghost imaging based on generative adversarial networks with high imaging quality.

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Ghost imaging is valuable for underwater active optical imaging due to its non-local and long-distance capabilities.
• Underwater environments degrade ghost imaging quality, limiting its practical applications.

Purpose of the Study:

• To propose an improved underwater ghost imaging method using generative adversarial networks (GANs).
• To enhance the reconstruction performance and image quality of ghost imaging in challenging underwater conditions.

Main Methods:

• A GAN-based method incorporating a U-Net generator with double skip connections and an attention module.
• Training the network using a combined loss function: weighted adversarial loss, perceptual loss, and pixel loss.

Main Results:

• The proposed method demonstrates effective improvement in target reconstruction for underwater ghost imaging.
• Experimental and simulation results validate the enhanced performance compared to traditional methods.

Conclusions:

• The developed GAN-based ghost imaging technique significantly enhances underwater imaging quality.
• This advancement supports the further development of active optical imaging for underwater targets.