Implementación de una Red Neuronal Convolucional para la clasificación de ruido sísmico y señales sísmicas

Abstract

This article develops a detection methodology of seismic events based on Convolutional Neural Network in order to improve the processing of huge rawdata of seismograms. This is a trained model by using 45 thousand signals between seismic event and noise, databases of the Geophysical Institute of Peru (IGP) and STEAD (Stanford Earthquake Dataset), for testing we use 15 thousand instances, I obtain a promising results, the confusion matrix, 99.18% of success and an accuracy of 99.69%. This method shows better performance that STA/LTA classical algorithm with less false-positives in 20% of 1000, it could improve the real time monitoring.