|About this Abstract
||2018 TMS Annual Meeting & Exhibition
||Algorithm Development in Materials Science and Engineering
||Deep Learning and Dynamic Sampling for Smart Data Acquisition in Scanning Electron Microscopy
||Yan Zhang, G. M. Dilshan Godaliyadda, Nicola Ferrier, Emine Gulsoy, Charles Bouman, Charudatta Phatak
|On-Site Speaker (Planned)
In conventional point-based scanning modalities for imaging or spectroscopy, each pixel measurement can take up to a few seconds, which translates into several hours of data acquisition time for large image sizes. This is often true for energy-dispersive X-ray spectroscopy (EDX) in a scanning electron microscope (SEM) which is widely utilized in materials science for determining elemental compositions. In this work, we will present a dynamic sampling method based on supervised learning algorithm and convolutional neural networks for data acquisition in a SEM. We will demonstrate the results for two modalities: (1) secondary electron imaging, and (2) EDX mapping. Our method is capable of achieving high quality images and elemental maps with as low as 30% sampling from all available pixels. We will discuss the impact of various algorithms and the experimental implementation of these algorithms for smarter data acquisition resulting in reduced time and radiation exposure of the sample.
||Planned: Supplemental Proceedings volume