About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Autonomous Platforms for Designing and Understanding Materials
|
| Presentation Title |
Sparse Sampling and Inpainting for High-Throughput Scanning Transmission Electron Microscopy |
| Author(s) |
Alex William Robinson, Jack Wells, Daniel Nicholls, James Hainsworth, Romanas Sonkinas, Nigel D. Browning |
| On-Site Speaker (Planned) |
Alex William Robinson |
| Abstract Scope |
With recent advances in aberration correctors and bright electron sources, the STEM is now the state-of-the-art tool for acquiring images at the highest spatial resolution. However, due to the nature of the data acquisition modality (a scanning probe with detectors collecting various signals arising from scattering), temporal resolution is limited to around one full frame per second with reasonable acquisition conditions. This problem gets worse when we consider more advanced methods such as EDS, EELS, and 4D-STEM.
One method which has demonstrated increased STEM frame rates is sparse sampling and inpainting, a form of compressive sensing whereby only a (pseudo-) random subset of probe locations is acquired over the desired field-of-view. This subsampled data is recovered in real-time using a GPU accelerated inpainting algorithm provided by SenseAI Vision.
We shall present the latest results of subsampling and inpainting to increase the throughput of STEM experiments without loss of fidelity. |