Unsupervised Machine Learning Approaches for Salt Dome Identification and Seismic Image Enhancement via Deep Learning

Abstract This thesis embarks on a comprehensive exploration of seismic data analysis, focusing on two critical facets: salt dome detection and seismic image enhancement. The study integrates unsupervised machine learning, particularly K-means clustering, with advanced deep learning techniques, specifically employing the Super-Resolution Generative Adversarial Network (SRGAN) model. In the pursuit of salt dome detection, we employ K-means clustering in tandem with the OpenCV library. This powerful combination enables the precise demarcation of salt dome boundaries within seismic images, thereby advancing the accuracy of geological interpretations and subsurface structure assessments. Concurrently, our application of the SRGAN model has brought about a significant enhancement in the resolution of seismic images. The resultant improvement not only augments the visual quality of these images but also simplifies the interpretation of seismic 3D volumes. This technological innovation empowers seismic data interpreters with unparalleled levels of detail, offering novel perspectives for the analysis of subsurface structures. The ramifications of our research extend to diverse practical applications in the fields of geophysics and energy exploration. The refined salt dome detection methodology provides a more profound understanding of subsurface features, while the enhancement of seismic image clarity enriches the overall interpretative process. Collectively, these advancements represent a pivotal step forward in seismic data analysis, enriching geological and geophysical research and advancing the capabilities of subsurface imaging.