An Outlook on Seismic Diffraction Imaging Using Pattern Recognition

A seismic image is formed by interactions of the seismic wavefield with geological interfaces, in the form of reflections, diffractions, and other coherent noise. While in conventional processing workflows reflections are favoured over diffractions, this is only beneficial in areas with uniform stratigraphy. Diffractions form as interactions of the wavefield with discontinuities and therefore can be used to image them. However, to image diffractions, they must first be separated from the seismic wavefield. Here we propose a pattern recognition approach for separation, employing image segmentation. We then compare this to two existing diffraction imaging methods, plane-wave destruction and f-k filtering. Image segmentation can be used to divide the image into pixels which share certain criteria. Here, we have separated the image first by amplitude using a histogram-based segmentation method, followed by edge detection with a Sobel operator to locate the hyperbola. The image segmentation method successfully locates diffraction hyperbola which can then be separated and migrated for diffraction imaging. When compared with plane-wave destruction and f-k filtering, the image segmentation method proves beneficial as it allows for identification of the hyperbolae without noise. However, the method can fail to identify hyperbolae in noisier environments and when hyperbolae overlap.