Reconstruction of partially obscured objects with a physics-driven self-training neural network

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Imaging Science

Background:

• Terahertz (THz) imaging offers unique capabilities for non-ionizing 3D scene reconstruction.
• Traditional holographic imaging requires phase information, which is often unavailable.
• Reconstructing 3D scenes from intensity-only THz data presents significant challenges, especially with occlusions.

Purpose of the Study:

• To develop an artificial-intelligence-supported method for 3D holographic imaging using coherent terahertz radiation.
• To reconstruct 3D scenes from intensity-only measurements, overcoming the lack of phase information.
• To address the challenge of reconstructing partially occluded objects in THz imaging.

Main Methods:

• Utilized a physics-informed deep learning (DL) algorithm incorporating angular spectrum theory.
• Generated a synthetic dataset of diffraction patterns for initial training.
• Employed a self-training strategy with a physics-informed neural network (NN) using both synthetic and unlabeled experimental THz data.
• Implemented a recursive training process to incorporate unknown ground truth experimental images.

Main Results:

• Successfully reconstructed 3D scenes from intensity-only THz images.
• Demonstrated effective object reconstruction even with partial occlusions.
• Showcased the method's robustness in data-scarce scenarios.
• Validated the approach using both simulated and experimental datasets.

Conclusions:

• The proposed physics-informed DL approach significantly advances 3D THz imaging capabilities.
• The method effectively reconstructs 3D scenes from phase-absent holographic data.
• This technique holds substantial potential for applications requiring non-invasive 3D imaging in the THz spectrum.