Multi-scale digital holographic reconstruction with deep learning.

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional hologram reconstruction methods face challenges with multi-scale and large objects, including complex operations and zero-order interference.
• Accurate and efficient reconstruction is crucial for digital holography applications in optical imaging and detection.

Purpose of the Study:

• To propose a novel deep learning-based hologram reconstruction method addressing the limitations of existing techniques.
• To enhance the speed, resolution, and accuracy of reconstructing holograms of multi-scale and large objects.

Main Methods:

• Developed a UHC-Net model incorporating ConvNeXt and hierarchical polymerization HANC modules for hologram reconstruction.
• Utilized a mixed dataset of analog and digital holograms for training and validation to improve network generalization.

Main Results:

• The proposed deep learning method significantly improves reconstruction speed compared to traditional approaches.
• Maintained high resolution and accuracy in reconstructed images, demonstrating effectiveness for large-size holograms.
• The use of analog holograms in the training dataset enhanced the network's generalization capabilities.

Conclusions:

• The UHC-Net model offers an efficient and accurate solution for hologram reconstruction, particularly for large-scale objects.
• The study provides a valuable reference for integrating deep learning with digital holography for advanced optical imaging.
• Enhanced network generalization through mixed data sets is key for practical applications in optical detection.