Multi-channel fusion for seismic event detection and classification

Abstract

Automatic detection and identification of seismic events is an important task that is carried out constantly for seismic monitoring. This monitoring process results in a seismic event bulletin that contains information about the detected events, their locations and, magnitudes and type (natural or man made event). Current automatic seismic bulletins comprise a large number of false alarms, which have to be manually corrected by and analysts. The progress in machine learning methods and the availability of a big historic seismic archives emerge the template based seismic detection methods. We propose a two stage processes for detection and classification of seismic events. First an energy detector is applied to every channel. Then, we fuse data from multiple channels by applying a multiview kernel based construction. The framework produces a reduced mapping in which every seismic waveform is classified as related to seismic noise, explosion or earthquake.