Deep-learning-enhanced ice thickness measurement using Raman scattering.

Area of Science:

• Physics
• Materials Science
• Artificial Intelligence

Background:

• Accurate ice thickness measurement (ICM) is crucial for various applications.
• Detecting the ice-water interface using Raman scattering presents challenges due to subtle spectral differences.
• Existing methods like Gaussian fitting (GFM) can be complex and less efficient.

Purpose of the Study:

• To develop a novel deep residual network (DRN) for improved ice-water interface detection in ICM.
• To transform interface detection into a Raman spectra identification problem.
• To offer a more accurate, faster, and cost-effective ICM method.

Main Methods:

• A deep residual network (DRN) was designed and trained using pre-collected Raman spectral data.
• The DRN was trained to predict Raman spectra of ice and water, thereby identifying the interface.
• The proposed DRN method was compared against the Gaussian fitting method (GFM).

Main Results:

• The developed DRN successfully identified ice-water interfaces with high accuracy and speed.
• The DRN approach demonstrated superior performance compared to the state-of-the-art GFM.
• Experimental results validated the feasibility and effectiveness of the proposed DRN for ICM.

Conclusions:

• The novel DRN provides an effective and efficient solution for ice-water interface detection in ICM.
• This AI-driven method offers a simpler, lower-cost, and more accurate alternative to conventional techniques.
• The study highlights the potential of deep learning in advancing Raman spectroscopy applications for material characterization.