Anis MOHAMMEDI2024-12-092024-12-092024https://dspace.estin.dz/handle/123456789/19The exploration and extraction of hydrocarbon resources, such as oil and natural gas, have long been crucial endeavors for meeting the world’s ever-increasing energy demands. One of the key challenges in this domain lies in accurately identifying and mapping geological structures that can serve as potential reservoirs for these valuable resources. Among these structures, salt domes have garnered significant attention due to their propensity to trap hydrocarbons within their intricate formations. Salt domes are massive underground deposits of salt that have been pushed upward through overlying sedimentary rock layers over millions of years, creating dome-like structures. These formations can act as traps for hydrocarbons, making their precise delineation a critical step in the exploration and extraction processes. Traditionally, the interpretation of seismic data has relied on manual analysis by skilled experts, a time-consuming and laborintensive task that can take weeks or even months to complete. With the advent of deep learning techniques, particularly in the field of computer vision, new opportunities have emerged to automate and expedite the interpretation of seismic data. Semantic segmentation, a foundational task in computer vision, aims to partition an image into semantically meaningful regions, offering a promising solution for the accurate delineation of salt domes in seismic images. By leveraging state-of-the-art deep learning architectures, this research endeavors to develop robust and accurate models for automating the detection and segmentation of salt domes in seismic data. The overarching goal of this research is to explore and evaluate various deep learning techniques for semantic segmentation, with a specific focus on their application in the detection and delineation of salt domes in seismic images. Through a comprehensive literature review, dataset acquisition and preprocessing, model implementation and fine-tuning, and rigorous evaluation, this study aims to contribute to the advancement of automated seismic data interpretation. The work plan for this research will involve several key steps, including: 1. Conducting a thorough literature review to understand the current state-of-the-art techniques and methodologies employed in semantic segmentation, with a particular emphasis on their applications in seismic data analysis. 2. Evaluate the performance of the developed models using appropriate metrics, such as intersection over union (IoU) and compare their results against established baselines and state-of-the-art methods. 3. Investigate techniques for semi-supervised learning and domain adaptation to address the challenges of limited labeled data and domain gaps across different seismic datasets, analyze the strengths and limitations of the employed deep learning techniques, identifying areas for further improvement and potential future research directions. By addressing these objectives, this research aims to contribute to the advancement of automated seismic data interpretation, ultimately facilitating more efficient and cost-effective hydrocarbon exploration while promoting sustainable energy practices.Salt Domes are subsurface geological formations critical for hydrocarbon exploration and storage. They often serve as traps for hydrocarbons, significantly contributing to the defining characteristics of reservoirs in the oil and gas industry. Accurately identifying and delineating these structures is essential for efficient resource management. Traditional methods for detecting salt domes rely heavily on manual interpretation, which is timeconsuming and prone to errors. Recent advancements in Deep Learning, particularly in Semantic Segmentation, offer promising solutions to automate and enhance this process. This study explores the application of state-of-the-art Deep Learning models, including Convolutional Neural Networks (CNNs), U-Net architectures, and Transformer-based approaches, for the Semantic Segmentation of Salt Domes in Seismic Images. Generative AI techniques are also examined for data augmentation and enhancing model robustness. The integration of these advanced models aims to improve the precision and reliability of Salt Dome identification, potentially transforming the field of geophysical exploration. Keywords— Salt Domes, Semantic Segmentation, Deep Learning, Generative AI, CNN, UNet, Transformers, Seismic Imaging, Hydrocarbon ExplorationenThesis