Hierarchical syntactic pattern recognition and hough transformation for the automatic reconstruction of seismic patterns

Hierarchical syntactic pattern recognition and Hough transformation are proposed for automatic reconstruction of seismic patterns on the seismogram. In a one-shot seismogram, the direct wave pattern and refracted wave pattern are straight lines and the reflected wave partem is a hyperbola. In a stacked seismogram, the horizontal reflection pattern is a straight line and the diffraction wave patterns are hyperbola. In the first step, the patterns are hierarchically decomposed/recognized into single pattern as straight line pattern or hyperbolic pattern by using syntactic pattern recognition. In the second step, the Hough transformation technique will be used in the reconstruction, one pattern by one pattern. The system of syntactic pattern recognition includes envelope generation, linking process in the seismogram, primitive recognition, grammatical inference, and syntax analysis. In order to make recognition processing more efficient, peaks on each trace are first detected from the envelope. Linking pattern is established by using a linking algorithm based on the pattern growing technique and a function approximation algorithm. Amplitude-dependent encoding is used in the assignment of primitives. The K-tail finite-state inference is used in the grammatical inference procedure. Error-correcting finite-state automata are used in the syntax analysis. The system of Hough transformation includes Hough transformation, parameter determination and reconstruction of the patterns. Two parameter determination methods are used: accumulator array method and clustering algorithm. The clustering algorithm uses K-means algorithm with bottom-up hierarchical method and the PFS-value method. Experimental results in a simulated one-shot seismogram and a simulated stacked seismogram are presented. The seismic patterns can automatically be recognized and reconstructed by computer.