Subsampling Methods for Fast Electron Backscattered Diffraction Analysis

TL;DR

This paper introduces a fast EBSD data acquisition method using subsampling and inpainting, enabling high-quality reconstruction of material maps from only 10% of the data, thus significantly speeding up the process.

Contribution

It proposes a novel subsampling and inpainting approach for 4D EBSD data, utilizing Bayesian dictionary learning for efficient reconstruction.

Findings

High-quality image reconstruction from 10% subsampled data

Effective use of Bayesian dictionary learning for EBSD data

Significant acceleration in EBSD data acquisition

Abstract

Despite advancements in electron backscatter diffraction (EBSD) detector speeds, the acquisition rates of 4-Dimensional (4D) EBSD data, i.e., a collection of 2-dimensional (2D) diffraction maps for every position of a convergent electron probe on the sample, is limited by the capacity of the detector. Such 4D data enables computation of, e.g., band contrast and Inverse Pole Figure (IPF) maps, used for material characterisation. In this work we propose a fast acquisition method of EBSD data through subsampling 2-D probe positions and inpainting. We investigate reconstruction of both band contrast and IPF maps using an unsupervised Bayesian dictionary learning approach, i.e., Beta process factor analysis. Numerical simulations achieve high quality reconstructed images from 10% subsampled data.