Autonomous detection of whistler-mode chorus elements in the Van Allen
radiation belts using morphological signal processing and pattern
recognition techniques
Abstract
Autonomous large-scale detection of whistler-ode chorus elements in the
Van Allen radiation belts has been an open computational challenge. This
is primarily due to: (i) Variability of the spectral morphology of
chorus elements, and (ii) Structured background interference from
hiss-like chorus that can make elements difficult to detect using
traditional signal processing and pattern recognition techniques. We
will present computational techniques drawing on pixel connectivity,
signal-to-noise (SNR) considerations as well as supervised and
unsupervised pattern recognition techniques. Specifically, we will
explore the efficacy of popular machine learning techniques trained on
unfiltered spectral images versus those trained on culled features
generated by unsupervised feature extraction techniques. Representative
results will be presented based on magnetic field measurements taken by
the EMFISIS instrument suite in the Van Allen probes mission.