Naslov
Earthquake—explosion discrimination using genetic algorithm-based boosting approach

Autori
Orlić, Nikša ; Lončarić, Sven

Izvornik
Computers & geosciences (0098-3004) 36 (2010), 2; 179-185

Vrsta, podvrsta i kategorija rada
Radovi u časopisima, članak, znanstveni

Ključne riječi
seismogram classification; digital signal processing; genetic algorithm; pattern recognition
(Seismogram classification; Digital signal processing; Genetic algorithm; Pattern recognitionseismogram classification; digital signal processing; genetic algorithm; pattern recognition)

Sažetak
An important and challenging problem in seismic data processing is to discriminate between natural seismic events such as earthquakes and artificial seismic events such as explosions. Many automatic techniques for seismogram classification have been proposed in the literature. Most of these methods have a similar approach to seismogram classification: a predefined set of features based on ad-hoc feature selection criteria is extracted from the seismogram waveform or spectral data and these features are used for signal classification. In this paper we propose a novel approach for seismogram classification. A specially formulated genetic algorithm has been employed to automatically search for a near-optimal seismogram feature set, instead of using ad-hoc feature selection criteria. A boosting method is added to the genetic algorithm when searching for multiple features in order to improve classification performance. A learning set of seismogram data is used by the genetic algorithm to discover a near-optimal feature set. The feature set identified by the genetic algorithm is then used for seismogram classification. The described method is developed to classify seismograms in two groups, whereas a brief overview of method extension for multiple group classification is given. For method verification, a learning set consisting of 40 local earthquake seismograms and 40 explosion seismograms was used. The method was validated on seismogram set consisting of 60 local earthquake seismograms and 60 explosion seismograms, with correct classification of 85%.

Izvorni jezik
Engleski

Znanstvena područja
Geologija, Elektrotehnika, Računarstvo

POVEZANOST RADA