Filtering techniques and numerical solutions related to image enhancement and automatized geological interpretation of seismic data

Seismic reflection data have been acquired since the late 1920?s and have played an important part of oil and gas exploration since the early 1950?s. Seismic reflections represent changes in acoustic impedance in the subsurface, and the geologist interprets these reflections to develop maps of geological surfaces and obtain a conceptual understanding of the geological evolution, such as sedimentation history and tectonic events. Today, the geologist interprets large fully migrated 3D seismic volumes which are created from acquisition and processing of seismic reflection data. Ideally, the process of interpreting seismic images is an integrated process based on geophysical knowledge and an intuitive geological understanding. The state-of-the-art seismic interpretation workflow aims to extract qualitative and quantitative information from the seismic data. Geological features of interest may include seismic horizons, sequence boundaries such as unconformities, and faults. While the interpretation of seismic data is essential in order to accumulate knowledge and build an understanding of the subsurface, some elements of the interpretation workflow can be tedious, subjective and in some cases even trivial. With increasing computational power, data science continues to evolve and provide new digital tools applicable to various disciplines, including geoscience and thereby seismic interpretation. This thesis investigates aspects of automated seismic interpretation. The overall objective has been to implement digital tools with geological and geophysical knowledge and workflows in order to automatically extract information from seismic data. To reach this objective, we have developed new data-driven methods that take advantage of existing methods and algorithms from the fields of image processing, signal processing and machine learning. In order to address a broad range of key elements of the seismic interpretation workflow, the thesis work has included: 1) Semi-automatic extraction of individual faults and unconformities from 3D seismic data using image processing tools (Paper I) 2)Automatic correlation and extraction of seismic multi-horizons from 3D seismic data using non-local pattern recognition for trace matching (Paper II) 3)Data-driven identification of stratigraphic units in 3D seismic data using unsupervised machine learning (Paper III) 4)Image to image transformation in order to improve seismic images and to extract attribute information using supervised machine learning (cGANs) (Paper IV)

1)Semi-automatic extraction of individual faults and unconformities from 3D seismic data using image processing tools (Paper I) 2)Automatic correlation and extraction of seismic multi-horizons from 3D seismic data using non-local pattern recognition for trace matching (Paper II) 3)Data-driven identification of stratigraphic units in 3D seismic data using unsupervised machine learning (Paper III) 4)Image to image transformation in order to improve seismic images and to extract attribute information using supervised machine learning (cGANs) (Paper IV)

The proposed doctoral work will focus on imaging techniques in geoscience, and how images of the subsurface can be enhanced through digital filtering and image processing. With this, the aim is to improve and automatize how the earth's interior is interpreted. The candidate will conduct research on use of numerical methods for optimization of the filter algorithms. The motivation and overall goal is to develop automatic/semi-automatic interpretive techniques for seismic data. Attention will be especially directed to the classification and interpretation of structural trends and stratigraphic sequences. Case studies with data from Lundin Norway AS, acquired in the southwest Barents Sea, will be implemented with the thesis work. Imaging techniques that are similar to the geophysical reflection seismology (such as X-ray surveys, MR- and CAT-scans and echo sounding) are being used in different industries and image processing techniques used in other disciplinary fields will be exploited in this work. Such techniques are e.g. the non-local means algorithm for image de-noising, used in the medical industry today. Other examples such as face recognition techniques can also be mentioned. Thus, the candidate will benefit from techniques and knowledge related to other research fields, and will conduct new research that is valuable in other industries where image processing is relevant.