Event and Noise Discrimination Using Deep Learning

Date: 4/24/2019

Time: 03:00 PM

Room: Elliott Bay

We are surrounded by noises originating from a myriad of sources. Short period geophone based seismometers like Raspberry Shake are sensitive to vibrations between .5 to 25 Hz which coincides with urban and cultural noise. Discriminating small to medium events from noise in urban settings by an automatic AI system is the next important step to having usable data from short period geophones.

Detecting local and regional seismic events with conventional amplitude ratio techniques is challenging due to high numbers of triggers generated by short period urban seismic stations.

In this paper, we discuss a methodology based on deep learning to create an AI model that can discriminate between noises and events. We curated a dataset from the Raspberry Shake network that contains a labeled list of events and noises from different sources in urban settings. We created a Deep Neural Network based on AlexNet architecture and trained it in our curated dataset. Our AI model achieves high accuracy in distinguishing events from noise.

Presenting Author: Vikraman Karunanidhi

Authors