Deep learning based coherent diffraction imaging of dynamic scattering media.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Artificial Intelligence

Background:

• Lensless coherent diffraction imaging using ptychographic iterative engine (PIE) is effective but struggles with dynamic scattering media.
• Applications in atmospheric pollution, seawater detection, and medical imaging are hindered by interference from scattering.

Purpose of the Study:

• To develop a novel deep learning approach for image reconstruction in lens-free coherent diffraction imaging through dynamic scattering media.
• To enable PIE imaging in complex environments previously unsuitable for the technique.

Main Methods:

• A computational deep learning model was developed for dynamic scattering medium image reconstruction.
• The neural network's effectiveness was evaluated under various scattering concentrations.
• Hybrid training with different scattering concentrations assessed the model's generalization ability.

Main Results:

• The proposed deep learning method successfully achieved PIE lens-free imaging despite non-static scattering media interference.
• The neural network demonstrated effectiveness in PIE image recovery across different scattering conditions.
• Experimental results confirmed the model's ability to generalize to varying scattering levels.

Conclusions:

• The novel deep learning approach overcomes limitations of PIE in dynamic scattering media.
• This method expands the practical applications of PIE in complex environments.
• The technique offers significant potential for atmospheric, oceanic, and biomedical imaging.