f Analysis of split-shear waves from near-offset VSP data using a neural network

Neural networks provide a way of recognizing particular patterns and extracting their characteristic parameters. Here they are applied to the different patterns of shear-wave particle motion recorded by orthogonal horizontal geophones. It is postulated that the major component of this particle motion is from shear-wave splitting, and that it can be parameterized as the polarization direction of the leading shear-wave and the time-delay between split shear-waves. The network is trained initially to recognize patterns of shear-wave splitting with known parameters from synthetic seismograms. Once training has been accomplished, the network can now be used to recognize patterns which have unknown characteristics, but which still lie within or not far outside the experience of the network. It is hoped to use the network to rapidly quantity any patterns of shear-wave splitting and to analyse large sets of seismic data.