22–26 Jun 2015
Europe/Vienna timezone

T3.3-P20. Melcepstral Coefficients Used as Input to a Neural Network for Identification of an Expanded Set of Atmospheric Nuclear Explosions and Bolides

Not scheduled
Poster 3. Advances in sensors, networks and processing

Speaker

Robert Kemerait (U.S. Air Force Technical Applications Center)

Description

In our previous research, Melcepstral coefficients were extracted from a number of infrasonic hand-digitized atmospheric explosion waveforms from the 1962 Operation Dominic series of atmospheric nuclear tests. These explosions were shown to have a distinctive pattern of melcepstral coefficients which can be modeled with synthetically-generated waveforms. In this follow-on research, the authors have accomplished two major additions: the first being the expansion of the database to include additional atmospheric nuclear explosions, and a significant number of additional Bolides, as well as a number of surface chemical explosions. More importantly, we have designed a Neural Network that takes these coefficients and outputs the class that best fits (explosions or bolides). Finally, we have done a preliminary investigation on how the melcepstrum/neural network identifies underground nuclear and chemical explosions which are detected by one, or more, infrasound arrays, as compared with earthquakes.

Primary author

Robert Kemerait (U.S. Air Force Technical Applications Center)

Presentation materials

There are no materials yet.