Noisy Seismic Traces Classification Using Principal Component Analysis

Abstract

Seismic Data, an exploration method of sending energy or sound waves into the earth and recording the wave reflections to reveal essential subsurface rock information including type, size, shape and depth. Seismic data acquisition typically produces a significant data size. Seismic files within a survey may include useless noisy traces that increase the file size. Noisy traces have some noticeable features which can be exploited to aid the denoising process. In this work, the features were formulated based on the Principal Component Analysis (PCA) to automatically distinguish excellent traces from noisy traces. PCA projects the seismic trace features to a lower dimension with only two features. To classify and detect noisy traces, we first select the dataset and generate Gaussian noise, then add the noise to the selected dataset and then normalize the traces before extracting the features: threshold algorithm, histogram algorithm, and zero-crossing algorithm and finally apply the PCA to obtain the projected data. In this work, two types of artificial noises were generated. It is shown that PCA is able to separate two types of noisy seismic traces. PCA projections show that at high noise contamination, the method is unable to separate the noisy and clean seismic traces.