A W-Shaped Self-Supervised Computational Ghost Imaging Restoration Method for Occluded Targets.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Computational Imaging

Background:

• Ghost imaging is a powerful technique for object reconstruction, but it faces challenges with occluded targets.
• Traditional methods struggle to recover high-quality images when objects are partially hidden.
• Computational ghost imaging offers flexibility but requires efficient reconstruction algorithms.

Purpose of the Study:

• To introduce a novel self-supervised learning method for enhancing ghost imaging of occluded objects.
• To improve the quality and efficiency of image reconstruction in ghost imaging.
• To demonstrate the efficacy of a W-shaped neural network for ghost imaging preprocessing.

Main Methods:

• Development of a W-shaped neural network architecture.
• Application of self-supervised learning for image preprocessing.
• Implementation of the W-shaped self-supervised computational ghost imaging (WSCGI) method.
• Validation through numerical simulations and experimental setups.

Main Results:

• The WSCGI method significantly improved the quality of reconstructed images for occluded objects.
• Enhanced efficiency in the ghost imaging reconstruction process was achieved.
• Numerical simulations and experimental results confirmed the method's superiority over existing techniques.
• Demonstrated successful ghost imaging of previously challenging occluded targets.

Conclusions:

• Self-supervised learning is a promising approach for advancing ghost imaging techniques.
• The W-shaped neural network effectively preprocesses images, boosting reconstruction performance.
• The WSCGI method offers a robust solution for ghost imaging of occluded objects.