Deep learning for improving non-destructive grain mapping in 3D.

Area of Science:

• Materials Science
• Crystallography
• Imaging Techniques

Background:

• Laboratory X-ray diffraction contrast tomography (LabDCT) is crucial for non-destructive 3D grain structure characterization.
• Accurate grain reconstruction depends on precise segmentation of diffraction spots in LabDCT images.
• Conventional filtering methods often result in over/under-segmentation, especially for low signal-to-noise or small spots, and require parameter tuning.

Purpose of the Study:

• To develop an efficient and accurate deep learning method for background noise cleaning in LabDCT images.
• To improve the segmentation of diffraction spots for enhanced grain reconstruction.

Main Methods:

• A deep learning neural network was developed for background noise reduction.
• The network was trained using synthesized LabDCT images combined with experimental background features.
• The trained network was applied to remove noise from experimental LabDCT images of various samples and conditions.

Main Results:

• The deep learning method effectively removed background noise from LabDCT images.
• Processed images showed improved diffraction spot segmentation compared to conventional filtering.
• Grain reconstructions derived from deep learning processed images demonstrated significantly better grain mapping.

Conclusions:

• Deep learning offers a superior approach to background noise cleaning in LabDCT compared to standard filtering techniques.
• This advancement leads to more accurate 3D grain structure characterization and mapping.
• The developed deep learning network provides a robust solution for improving LabDCT image analysis.