Robust prestack seismic facies analysis using shearlet transform-based deep learning

One of the primary purposes of seismic stratigraphy is to evaluate the components of seismic layer relationships within a depositional chronology. Prestack seismic images contain a wealth of information, such as variations in the offset and azimuth of a seismic event, and naturally produce higher-resolution seismic facies analysis results than poststack data. However, prestack data usually suffer from potential unreliability issues due to low signal-to-noise ratios. As this is often overlooked, the present facies analysis methods sometimes fail to extract accurate features from prestack images, which inevitably influences the facies analysis results. To address this issue, this article provides a robust data-driven technique for extracting offset-temporal features via shearlet transform-based deep convolution autoencoders (STCAEs). Unlike the present facies analysis in the time domain, STCAE can optimally represent prestack images at multiple scales and directions through the two-dimensional shearlet transform, which preserves fine edges while suppressing noise in prestack images. Subsequently, robust features are extracted from prestack images in a data-driven manner through a contractive convolutional autoencoder network. We compare our method with other advanced methods and demonstrate the advantages of the proposed approach in classifying seismic layers in prestack data.