Image reconstruction for sub-sampled atomic force microscopy images using deep neural networks.

Area of Science:

• Materials Science
• Nanotechnology
• Microscopy

Background:

• Atomic Force Microscopy (AFM) enables high-resolution surface imaging.
• Increasing AFM imaging rates is crucial for dynamic processes.
• Undersampling in AFM acquisition presents significant image reconstruction challenges.

Purpose of the Study:

• To develop an advanced method for reconstructing AFM images acquired with undersampling.
• To enhance the accuracy and quality of reconstructed AFM images from limited data.
• To evaluate the performance of a novel deep neural network (DNN) approach for AFM image reconstruction.

Main Methods:

• A deep neural network (DNN) comprising sequential RED-net and U-net sub-networks was designed.
• The DNN was trained end-to-end using simulated and experimental AFM images with μ-path sub-sampling patterns.
• Performance was compared against established optimization-based reconstruction methods: basis pursuit, total variation minimization, and inpainting.

Main Results:

• The proposed DNN approach demonstrated superior image quality in reconstructing μ-path sub-sampled AFM images.
• Simulations and experimental results confirmed the DNN's effectiveness over existing methods.
• The DNN successfully addressed the challenge of accurate image reconstruction from limited AFM measurements.

Conclusions:

• Deep neural networks offer a powerful solution for accurate AFM image reconstruction from undersampled data.
• The presented DNN architecture significantly improves image quality in high-rate AFM imaging.
• This work advances the capabilities of AFM for studying dynamic nanoscale phenomena.