Self-organizing neural network for picking seismic horizons

Applications of self-organizing feature maps to various cognitive tasks have been demonstrated by Kohonen. In this paper, we design an algorithm based on self-organization model with a linear array feature maps to pick the seismic horizons in the seismogram. This algorithm creats a vector quantizer by adjusting weighting vectors from the input-vectors. Weights are initially set to snail random values. From the property of seismic horizon, a set of neurons that joined together in a one dimensional linear array is used. In the self-organizing process, the distribution of weights is tending into an approximate form which best imitates the structure of the input density. We have applied the algorithm on simulated and real seismograms. This self-organizing neural network technique can be used as a pre-processing of the seismic pattern recognition and improve the seismic interpretations.