Processing of seismic signals for pattern recognition

The study describes the preparation of seismic signals in order to identify the best technique for input to a pattern recognition scheme. The signal is first filtered with zero phase shift high and low pass Butterworth filters. It is then subjected to adaptive filtering and finally moving average filtering. Spectral decomposition in terms of circular functions is done via conventional Fourier and log P maximum entropy analysis. Spectral decomposition in terms of sequency functions, Walsh, and Chebyshev, is also performed. The Walsh decomposition is done with the conventional fast operator. The Chebyshev decomposition is done with an optimization procedure. Results, based on these various operations, are presented for an underground nuclear explosion and several earthquakes.