Artificial Neural Networks for Seismic Images Enhancement

A processing sequence was developed based on both the Hilbert transform and the wavelet transform. The sequence’s objective is to enhance certain characteristics that may be of interest to evaluate oil occurrence through seismic data. Once processing is applied to a set of data, which belongs to one or more seismic sections, the outputs are used as training sets of an artificial neural network designed with the objective of learning to apply any processing sequence to any seismic data. This neural network is grounded in the feedforward structure, so the data flows in one direction. This neural network is built around an input layer, an output layer, and a hidden layer. The hidden layer is resultant from fifteen neurons. Not only the training but also the execution of the network considers vector arrays as inputs and outputs. This fact means that the network is executed to each trace. The neural network has demonstrated to be able to correctly apply the processing sequence. Furthermore, it performed better in terms of computational cost. In consequence less time is required to apply the network than the processing sequence.