A Framework for Automated Detection of Ground-Motion Signals on Seismic
and Infrasound Arrays using Computer Vision, Machine Learning, and
Sensor Fusion
Abstract
The recent deployment of nodal seismic instruments and DAS sensor
networks has opened up a new chapter in observational seismology.
Besides earthquake-generated signals, these new types of seismic array
record a larger variety of low-frequency acoustic (infrasound) signals
and noise than traditional instruments. Due to the near-surface
deployments of these seismic sensing systems, the composition of the
signals includes ground-motion sources that are associated with the
atmosphere and manmade events. Meanwhile, with rapidly-growing data
volumes recorded by nodal and DAS sensors, designing an automated system
that detects time periods that are contaminated by these ground motions
is essential. This presentation presents a comprehensive framework for
the automated detection of ground-motion signals on seismic or
infrasound instruments using computer vision, machine learning, and
sensor fusion. Extensive experiments were conducted on Sage Brush Flats
nodal array in Southern California, and our framework provides a
benchmark performance for detecting aircraft events (one of the major
ground-motion sources) and demonstrates the advantage of using an array
of sensors. This framework is flexible and can also be extended to
detecting other common types of ground motions, and to fusing other
types of sensors. By combining models/systems that detect other types of
ground motion, a complete characterization of the ground motions
recorded by nodal and DAS sensor networks could be obtained. In
addition, the data from other types of sensors such as All-Sky cameras
and ADS-B or Mode S receivers could be integrated into this framework to
further refine the detection performance.