Strombolian Eruption Detection Using an Onsite-Ready Convolutional Neural Network on Mount Erebus, Antarctica
Abstract
A small and efficient convolutional neural network was developed to enable onsite detection and recording of strombolian eruptions in the Ray lava lake atop Mount Erebus, Antarctica. Detection can be achieved by training the neural network with infrared images of the lava lake that were shot periodically over a time span of several years (Peters et al., 2013).
This research expands upon previous work (Dye, Morra 2019) which used a similar method with the Inception v3 Convolutional Neural Network. In that research a correct detection rate of 84% was achieved. We aim to achieve a more accurate result by building a smaller neural network designed specifically for the task of identifying and categorizing types of strombolian eruptions. A smaller neural network also requires less calculations per classification and therefore could run faster and be used on a small, low power device onsite. This allows for storage of more events with greater frame rates by discarding time periods determined by the neural network to be irrelevant. These advancements should improve the detection rate and supply us with more data useful in our attempt to predict large strombolian eruptions. Peters, N., Oppenheimer, C., and Kyle, P., 2013, Autonomous thermal camera system for monitoring the active lava lake at Erebus volcano, Antarctica: Geoscientific Instrumentation, Methods and Data Systems Discussions, v. 3, no. 2, p. 627-647, doi: 10.5194/gid-3-627-2013.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.V11C..16D
- Keywords:
-
- 1099 General or miscellaneous;
- GEOCHEMISTRY;
- 1916 Data and information discovery;
- INFORMATICS;
- 1942 Machine learning;
- INFORMATICS;
- 3699 General or miscellaneous;
- MINERALOGY AND PETROLOGY