A neural network approach to seismic event identification using reference seismic images

A data compression neural network is known to have both data compression capability and generalization capability. In this study, a data compression neural network is trained using reference seismic images to rapidly identify natural earthquakes and underground explosions. The method developed is based on the generalization properties of the trained neural network and a quantitative measure of the degradation of the reconstructed image over the population of similar events and dissimilar events (i.e. explosions vs earthquakes). As examples, this approach is applied to a dataset of 11 natural earthquakes and 11 mining (chemical) explosions recorded by the NORESS array in Norway. Preliminary results using this neural network method show very promising performance