Fractal analysis with applications to seismic pattern recognition of nuclear explosion

Abstract

Based on the processing and analysis of seismic signals originating from underground nuclear explosions and natural earthquakes, it is illustrated that the seismic signals in the time domain possess the characteristics of statistical self-affine fractals, whilst the fractal dimension D yielded from logarithmic power spectrum does not serve as an effective feature for seismic pattern recognition. Moreover, it is found that the signal "energy" at each scale of the wavelet decomposition relates closely to the scale, and that an apex appeared on the "energy spectrum" of the detail signal, hence, the two kinds of features advocated are very likely to be utilized in seismic pattern recognition applications. The provided recognition results show the improvement and performance achieved by the proposed feature extraction and selection methods.