Correntropy-based SOM for Waveform classification

Self-organizing mapping (SOM) is one of the most famous classification method in seismic facies analysis. Traditional self-organizing map network and its variation methods usually use the Euclidean distance to measure the similarity between input data and weights of neuron node. A Euclidean distance is mainly used to measure the global similarity between the random variables, while a correntropy is the local similarity measure of random variables and very useful for non-Gaussian signal. In general, the geological sedimentation has its own regular pattern. The data of seismic imaging is a response of geological sedimentation. If the geological sedimentation has regularity, the seismic data basically presents characteristics of non-Gaussian distribution. Therefore, we proposed a classification method, which introduce the maximum correntropy as a new distance measure criterion to the SOM. The field data example demonstrates that the proposed method can effectively delineate the distribution and boundary of channels in seismic data. Therefore, it is of great significance to improve the accuracy of reservoir interpretation.