SEISMIC SIGNAL CLASSIFICATION USING PERCEPTRON WITH DIFFERENT LEARNING RULES

Abstract

In a seismogram, there exist many kinds of wavelets. It can be classified into two classes. One class is normal; the other is abnormal. The abnormal may be the bright spot pattern caused from the gas sand zone. It has the properties of high amplitude and low frequency content in the wavelets. Perceptron is adopted to classify these 10 classes. Four different learning rules are used in the training of perceptron. Those are the fixed increment, normalized perceptron, fractional correction, and absolute correction rules. The experiments are in the simulated and real seismograms. For the comparison of four learning rules in the experiment of simulated seismogram, the absolute correction rule can get the fastest convergence. Then, it is applied to the real seismic data. The bright spot pattern can be detected. The result can improve the seismic interpretation.