Speaker
Description
Recent advances in convolutional neural networks (CNNs) have brought impressive detection capabilities to one- and three-component seismic stations. Still, the highest sensitivity to repeating events is obtained by beamforming signals over a seismic array. We propose a new neural network architecture that combines the two, by introducing a two-dimensional convolutional layer that encodes the propagation time delays between array stations. This results in a purely empirical model, which does not rely on the plane-wave approximation of traditional beamforming. We demonstrate the model by detecting and classifying repeating blasts from multiple mining sites in northern Fennoscandia. Results are compared to those obtained by empirical matched field processing, a highly sensitive method which, similarly, does not require signals to be coherent under the plane-wave model.
Promotional text
We present a machine learning method for seismic arrays, which aims to improve verification capabilities through higher detection sensitivity and better identification of repeating sources.
