Single-exposure height-recovery structured illumination microscopy based on deep learning.

Area of Science:

• Optical Metrology
• Surface Metrology
• Computational Imaging

Background:

• Structured Illumination Microscopy (SIM) is crucial for 3D surface topography reconstruction.
• Traditional methods like phase-shift (PS) and frequency analysis have limitations including low efficiency and loss of high-frequency information.
• Existing techniques require complex calculations for modulation-to-height mapping.

Purpose of the Study:

• To develop a deep learning-enabled single-exposure method for direct 3D surface measurement.
• To improve the efficiency and accuracy of 3D surface reconstruction compared to conventional SIM techniques.
• To integrate denoising, modulation calculation, and height mapping within a single neural network framework.

Main Methods:

• A deep learning model was trained to directly restore height information from a single fringe image.
• The neural network encompasses denoising, modulation calculation, and height mapping functionalities.
• Modulation-based structured illumination microscopy (SIM) principles were adapted for single-exposure data acquisition.

Main Results:

• The proposed deep learning method achieves direct height information restoration from one fringe image.
• Experimental results show higher accuracy and efficiency compared to traditional Fourier methods.
• The method successfully provides accurate and rapid surface measurements for diverse structures.

Conclusions:

• Deep learning enables a highly efficient single-exposure approach for 3D surface measurement using SIM.
• This method overcomes the drawbacks of traditional multi-image and single-image decoding techniques.
• The proposed technique offers a promising solution for fast and accurate 3D surface topography reconstruction.