Patching-based deep-learning model for the inpainting of Bragg coherent diffraction patterns affected by detector...

Area of Science:

• Physics
• Materials Science
• Computational Science

Background:

• Bragg coherent diffraction imaging (BCDI) is crucial for nanoscale material characterization.
• Detector gaps in BCDI data introduce significant artifacts, compromising reconstruction accuracy.
• Restoring missing intensity data is essential for reliable BCDI analysis.

Purpose of the Study:

• To develop a deep-learning algorithm for inpainting missing intensity data in BCDI patterns.
• To mitigate artifacts caused by detector gaps in BCDI reconstructions.
• To enhance the reliability and accuracy of BCDI experiments, especially at high resolutions.

Main Methods:

• A deep-learning algorithm was trained using cropped sections of diffraction data.
• The neural network predicts intensity values for regions affected by detector gaps.
• A patching strategy was employed to combine predictions and complete diffraction peaks.

Main Results:

• The deep-learning method successfully inpainted detector gaps in BCDI patterns.
• Gap-induced artifacts were effectively removed from reconstructed objects.
• The approach demonstrated robustness on both simulated and experimental datasets.
• The method is scalable to experimental data arrays of any size.

Conclusions:

• The proposed deep-learning algorithm provides a robust solution for BCDI data inpainting.
• This technique significantly improves the quality of reconstructed objects from BCDI data.
• The method is vital for advancing high-resolution BCDI applications by enabling the use of more complete datasets.