(shown in red), M2 matrix phase (shown in green), and
a strain relieving interface between the two shown in
blue. Figure 11c,d,e shows each cluster individually.
While the example is relatively simple, it serves to bridge
machine learning methods with high-quality experimental
data that, due to its intrinsic complexity, is typically
only qualitatively analyzed. While it may be possible to
manually distinguish phases and assign them to the
atomic species, performing this task done quantitatively
for a large number of frames quickly becomes a monumental
feat.
Imaging in k-space
The concept of image frame analysis can be further
extended to image sequences, which are perfectly suited
for analysis using the same multivariate statistical
methods. As an example, we turn to an image sequence of
reflection high-energy electron diffraction (RHEED) data
acquired during deposition of SrRuO3 on (001) SrTiO3.
The (00) or specular spot is closely monitored for signs of
oscillations, which would indicate a layer-by-layer growth
of the film on the substrate. As a first approximation,
these oscillations arise due to a filling of incomplete layers
(which reduces step density and therefore increases the
intensity of the diffracted beam), until the layer is
complete followed by more roughening as more material
is deposited, with a corresponding decrease in intensity
of the specular spot [81]. The process continues as