Accession Number:

AD1104459

Title:

Neural Network Models for Nuclear Treaty Monitoring: Enhancing the Seismic Signal Pipeline with Deep Temporal Convolution

Descriptive Note:

Technical Report,01 Sep 2017,01 Jul 2020

Corporate Author:

AIR FORCE INSTITUTE OF TECHNOLOGY WRIGHT-PATTERSON AFB OH WRIGHT-PATTERSON AFB United States

Personal Author(s):

Report Date:

2020-07-01

Pagination or Media Count:

149.0

Abstract:

Seismic signal processing at the IDC is critical to global security, facilitating the detection and identification of covert nuclear tests in near-real time. This dissertation details three research studies providing substantial enhancements to this pipeline. Study 1 focuses on signal detection, employing a TCN architecture directly against raw real-time data streams and effecting a 4 dB increase in detector sensitivity over the latest operational methods. Study 2 focuses on both event association and source discrimination, utilizing a TCN-based triplet network to extract source-specific features from three-component seismograms, and providing both a complimentary validation measure for event association and a one-shot classifier for template-based source discrimination. Finally, Study 3 focuses on event localization, and employs a TCN architecture against three-component seismograms in order to confidently predict back azimuth angle and provide a three-fold increase in usable picks over traditional polarization analysis.

Subject Categories:

  • Seismic Detection and Detectors
  • Acoustic Countermeasures
  • Acoustic Detection and Detectors

Distribution Statement:

APPROVED FOR PUBLIC RELEASE