Indexing electron backscatter diffraction patterns with a refined template matching approach

Abstract

Graphical AbstractAbstract Electron backscatter diffraction (EBSD) is a well-established method of characterisation for crystalline materials. Using this technique, we can rapidly acquire and index diffraction patterns to provide phase and orientation information about the crystals on the material surface. The conventional analysis method uses signal processing based on a Hough/Radon transform to index each diffraction pattern. This method is limited to the analysis of simple geometric features and ignores subtle characteristics of diffraction patterns, such as variations in relative band intensities. A second method, developed to address the shortcomings of the Hough/Radon transform, is based on template matching of a test experimental pattern with a large library of potential patterns. In the present work, the template matching approach has been refined with a new cross correlation function that allows for a smaller library and enables a dramatic speed up in pattern indexing. Refinement of the indexed orientation is performed with a follow-up step to allow for small alterations to the best match from the library search. The refined template matching approach is shown to be comparable in accuracy, precision and sensitivity to the Hough based method, even exceeding it in some cases, via the use of simulations and experimental data collected from a silicon single crystal and a deformed α-iron sample. The speed up and pattern refinement approaches should increase the widespread utility of pattern matching approaches.

Advancing characterisation with statistics from correlative electron diffraction and X-ray spectroscopy, in the scanning electron microscope

“…EBSD patterns for each point were processed in MATLAB, with background correction, radial cropping, and hot-pixel/split-chip fixes performed using the AstroEBSD package developed and presented by Britton et al previously [23]. EBSD patterns were originally captured with an aspect ratio of 4:3, but these were cropped to squares to simplify the refined template matching indexing (as discussed by Foden et al [12]) prior to creation of the data matrix. EDS spectra were processed in MATLAB.…”

“…We label our data first using EBSD pattern analysis. In this work we apply the Refined Template Indexing approach developed by Foden et al [12] to assign phase and orientation to each of the point labels. This method involves cross-correlating of test EBSPs (in this case the RC-EBSPs) with a database of library patterns, sampled with the fundamental zone of SO(3) space for each crystal structure with a specified angular frequency.…”

“…Considering the full AOI of nine tiles, Figure 8 presents the dataset processed with w = 1 (EBSD weighting) and w = 0.1 (EDS weighting, identified as the local minimum in standard deviation of χcomb in section 3.2). RC-EBSPs were indexed using the Refined Template Matching procedure [12], and RC-spectra were quantified with Bruker Esprit 2.1.…”

“…a small phase). Signal classification can be improved through the use of template matching [12][13][14] against simulated patterns using dynamical diffraction theory [11,15].…”

“…A ~40,000 point map can be reduced to a few hundred RC-EBSPs and RC-spectra each with a superior signal-to-noise ratio than a single EBSD pattern or EDS spectrum. In this work we analyse the EBSD patterns using the refined template matching approach developed by Foden et al [12] with a selection of possible candidate structures. Adopting a template matching approach allows us to utilise the fine detail in RC-EBSPs (weakly reflecting bands, band widths, etc) that PCA is able to extract.…”