Deep Learning-Enabled Multimodal AFM Image Enhancement: Correlation Analysis between Surface Topography and...

Area of Science:

• Materials Science
• Nanotechnology
• Biophysics

Background:

• Atomic Force Microscopy (AFM) is crucial for nanoscale material characterization.
• Multiphysics field noise suppression and feature enhancement in AFM data present significant challenges.
• Deep learning shows promise in multiscale feature extraction and data visualization.

Purpose of the Study:

• To propose a multimodal data-fusion-based image enhancement model for AFM data.
• To capture surface topographical characteristics and facilitate correlative analysis of physical properties.
• To discover latent correlations between topographical characterization and physical property analysis.

Main Methods:

• A deep learning enhancer framework utilizing a convolutional neural network (CNN) was developed.
• The model extracts and enhances features from multiscale AFM data.
• An in-house three-probe AFM system synchronously acquired topographical and physical property data.

Main Results:

• The model precisely identified chromosome-associated regions in super-resolution (SR) topographical images.
• Accurate surface morphological features of chromosomes were acquired.
• Correlation analysis revealed spatial positions of chromosomal domains (short arms, centromeres, long arms) and their structural correspondence with physical properties.

Conclusions:

• The proposed deep learning model enhances AFM image analysis for materials science.
• It enables precise identification of nanoscale topographical features and their correlation with physical properties.
• This opens new avenues for exploring structure-property relationships in materials.