Accelerating Scanning Probe Microscopy Using Sparse Reconstruction Algorithms

Overview

Venkatakrishnan, Oak Ridge National Laboratory (ORNL) Scanning Probe Microscopy uses a sharp probe to scan a material's surface in a zig -zag raster pattern. Though highly accurate, raster scans can be slow and inefficient, limiting the study of surface changes over time. While previous research has used metho ds such as Gaussian Process Optimization and Compressed Sensing, few studies have addressed the combination of baseline sparse reconstruction methods and deep learning for scanning acceleration. Our study addresses this gap by utilizing biharmonic inpainti ng and Convolutional Neural Networks on varying sparsity levels that model different scanning speeds, with our experiments showing significant results. Our biharmonic inpainting approach, the baseline method, had a max average testing error of ~7%, while our deep learning approach showcased a max average error of ~4%. These performance metrics, along with further out of distribution studies conducted in our research, highlight that our deep learning approach can be highly effective compared to previous baseline methods of image reconstruction. By showing a decrease in error compared to baseline methods, our study offers a newer approach to accelerating scanning probe microscopy through deep learning techniques.

Competition history

• JSHS 2025 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.