Fast Nano-IR Hyperspectral Imaging Empowered by Large-Dataset-Free Miniaturized Spatial-Spectral Network

Area of Science:

• Spectroscopy and Imaging
• Materials Science
• Nanotechnology

Background:

• Nanoscale infrared (nano-IR) imaging provides label-free molecular contrast.
• Current nano-IR imaging is slow due to point-by-point acquisition of 3D data cubes.
• Accelerating nano-IR imaging is crucial for broader applications.

Purpose of the Study:

• To develop a deep learning model (SS-Net) combined with compressive sampling for rapid nano-IR imaging.
• To improve the imaging speed of various nano-IR techniques without compromising data quality.
• To establish a new paradigm for high-speed nano-IR microscopy.

Main Methods:

• Developed a miniaturized deep-learning model, spatial-spectral network (SS-Net).
• Implemented compressive sampling in both spatial and spectral domains.
• Trained SS-Net using subsampled nano-IR data with an efficient loss function incorporating image and spectral similarity.

Main Results:

• Achieved up to 10-fold speed improvement on stimulated Raman-scattering datasets with state-of-the-art performance.
• Demonstrated 7-fold speed improvement on atomic force microscopy infrared (AFM-IR) microscopy.
• Showcased up to 261.6-fold faster imaging speed on nanoscale Fourier transform infrared (nano-FTIR) microscopy.
• Validated generalization on AFM-force volume-based multiparametric nanoimaging.

Conclusions:

• The SS-Net approach significantly accelerates nano-IR imaging speed.
• This method offers a versatile and efficient solution for various nano-IR techniques.
• Enables new possibilities for advanced research in materials, photonics, and beyond.