Seismic Structures Classification Using Novel Features from Seismic Images

Abstract

Seismic facies can be used as novel features to classify different classes of seismic structures. Classification of seismic structure is beneficial for mineralogy, grain size approximation, the permeability of deposition units, and the identification of areas of interest. To extract features of seismic images, the following extraction methods were used: Discrete Wavelet Transform Features, Discrete Cosine Transform Features, Discrete Fourier Transform Features, and Gabor Features. The classification methods being considered are Support Vector Machine (SVM), Random Forest (RF), Fast Decision Trees (FDT), and Naïve Bayes (NB). The proposed study uses the LANDMASS database, composed of two datasets, LANDMASS-1, with 17,667 images, and LANDMASS-2, with 4,000 images. The datasets contain seismic images of four different classes of seismic structures; Chaotic, Fault, Horizon, and Salt Dome. The outcome of this study proves that the combination of Forest Tree classification method and the Discrete Cosine Transform Features extraction method achieved the highest accuracy, which was around 94.17% - higher than that achieved considering similar methods reported in the extant literature.