Application of AI-Based Super-Resolution in Seismic Image Processing

Seismic data acquisition is an essential stage in the petroleum exploration and production. Seismic acquisition is done using a large 2D array of sensors that are laid out onshore or offshore. The raw data undergoes several processing steps to produce a 3D seismic image of the subsurface. Seismic images can be obtained from such a 3D volume by taking inline or crossline slices of the volume. Due to the limited number of sensors used in a given survey, as well as limitations on the physical sensor, most legacy seismic data have a Low Resolution (LR), which hinders seismic interpretation. Methods that aim in producing High Resolution (HR) seismic images in the last two decades either focus on high-density seismic acquisition, or on capturing and recording a wideband of frequencies. The first type would add more traces to the seismic image, and thus increasing the horizontal resolution. The second type would contribute to increasing the vertical resolution. The disadvantage of these approaches is either having to deal with and process enormous amount of seismic data, or by requiring advanced and more sophisticated sensors to capture the wide range of frequencies. An alternative approach is the use of image super-resolution. Image super-resolution is the problem of producing HR image from an LR one. Building on this, we propose the application of state-of- the-art Convolutional Neural Network (CNN) models that are built for the super-resolution task for seismic images. Since seismic images interpretation benefits from image analysis algorithms that utilize Human Visual System (HVS) characteristics, objective as well as subjective image quality assessment will be carried out to evaluate the super-resolution model used for seismic images. Furthermore, applications of the use of super-resolution for a seismic interpretation task, such as facies classification, will be investigated.