Denoising ghost imaging under a small sampling rate via deep learning for tracking and imaging moving objects.

Area of Science:

• Computational imaging
• Machine learning applications
• Optical sensing

Background:

• Traditional ghost imaging (GI) demands extensive sampling, hindering applications with dynamic targets.
• High sampling rates are a bottleneck for real-time imaging and processing of moving objects.

Purpose of the Study:

• To develop and evaluate a deep learning approach for enhancing ghost imaging quality.
• To significantly reduce the number of required samples in ghost imaging, particularly for dynamic scenes.

Main Methods:

• Utilized a convolutional denoising auto-encoder network trained on numerical data.
• Employed the denoised outputs for trajectory reconstruction and image clarity enhancement.
• Applied cross-correlation based ghost imaging for object reconstruction.

Main Results:

• Achieved enhanced image quality at sampling rates as low as 3.7%.
• Successfully denoised blurry images acquired with minimal samplings.
• Reduced the number of required samples by two-thirds for reconstructing moving objects.

Conclusions:

• Deep learning effectively denoises images and reconstructs moving objects in ghost imaging.
• This method drastically lowers sampling requirements, improving efficiency and applicability.
• The approach offers a viable solution for ghost imaging challenges with dynamic targets.