An MEM-DMD-Enabled Ghost Imaging System Enhanced by a Hybrid CNN-GAN for High-Resolution Imaging Under Scattering...

Area of Science:

• Optics and Photonics
• Computational Imaging
• Micro-systems Engineering

Background:

• Conventional ghost imaging (GI) struggles with scattering conditions due to fixed illumination patterns.
• High-resolution imaging in challenging environments like fog requires advanced techniques.
• Micro-Electro-Mechanical Systems digital micromirror devices (MEMS-DMDs) offer high-speed pattern programmability.

Purpose of the Study:

• To develop a MEMS-DMD-enabled ghost imaging framework for high-resolution imaging under scattering conditions.
• To implement adaptive illumination strategies for improved sampling efficiency and reconstruction stability.
• To leverage learning-based reconstruction for enhanced image quality from compressed measurements.

Main Methods:

• Utilized a MEMS-DMD for high-speed, adaptive illumination pattern selection during ghost imaging.
• Employed a hybrid convolutional neural network-generative adversarial network (CNN-GAN) for image reconstruction.
• Evaluated performance under modeled fog conditions, comparing against traditional GI methods.

Main Results:

• Achieved 23-40% gains in Peak Signal-to-Noise Ratio (PSNR) and 18-26% in Structural Similarity Index Measure (SSIM).
• Reduced the number of required measurements by up to 60% compared to traditional methods.
• Demonstrated superior reconstruction quality and stability through adaptive pattern selection.

Conclusions:

• Integrating programmable MEMS hardware with learning-based reconstruction is effective for imaging in scattering media.
• The proposed framework offers significant improvements for applications in remote sensing, environmental monitoring, and surveillance.
• Adaptive illumination via MEMS-DMDs enhances ghost imaging performance in challenging conditions.