A comprehensive review of deep learning techniques for salt dome segmentation in seismic images

Abstract

Salt dome detection in seismic images is a critical aspect of hydrocarbon exploration and production. Salt domes are subsurface structures formed from the accumulation of salt deposits and can trap oil and gas reservoirs. Seismic imaging techniques are used to visualize the subsurface structures and identify the presence of salt domes. Historically, the process of detecting salt domes in seismic images was done manually, which was time-consuming and required the input of domain experts. However, in recent years, automated methods using seismic attributes and machine learning algorithms have been developed to improve the efficiency of salt dome detection. Deep learning-based methods have shown promising results in salt body segmentation, and several techniques have been proposed in recent years. This review examines recent deep-learning architectures for salt body segmentation in seismic images, offering a concise overview of the various models proposed in the literature. It delves into established benchmark datasets, highlighting potential limitations and emphasizing the importance of data quality for robust models. It explores performance evaluation metrics used in the literature to capture a more comprehensive picture of segmentation performance. This paper identifies several promising areas for further research and development opportunities to refine and enhance the current state-of-the-art salt body segmentation in seismic images. This comprehensive analysis provides a valuable roadmap for researchers and practitioners interested in understanding how deep learning can be utilized for salt body classification in seismic exploration.