International Journal of Academic Multidisciplinary Research (IJAMR) ISSN: 2643-9670 Vol. 6 Issue 2, February - 2022, Pages:27-33 www.ijeais.org/ijamr 27 Evaluating Performance of Semi-Supervised Clustering with Limited Features for Segmenting Salt Bodies in Seismic Images Shadi Abudalfa Software Development Department University College of Applied Sciences Gaza, Palestine sabudalfa@ucas.edu.ps Abstract: In recent years, major hydrocarbon discoveries have been made by exploring subsalt hydrocarbon plays. Finding a proper model for identifying the salt deposits is great importance for identifying salt-related drilling hazards. To simplify the process of salt body extraction from seismic data and better assess the quality of the extracted salt bodies, an automated system is needed. In this work, we deal with detecting salt bodies in seismic data by using very limited number of features that used for texture-based segmentation along with presenting a hybrid semi-supervised clustering technique. Experiment results have shown that the presented technique provides remarkable accuracy. Keywords—Semi-Supervised; Clustering; Texture-based Segmentation; Seismic Images; Salt Bodies 1. INTRODUCTION Seismic imaging [1] is an active technique used to illustrate the nature of earth's layers located below the surface of the ground. The resulted seismic data gives the explorationist a lot of details about the geology of the subsurface. Seismic data is collected by using a computerized system that registers different seismic reflection events based on various forms of rocks and fluids located in each layer. Mainly, there are three types of dimensional seismic data: 2D, 3D, and 4D. The 2D seismic image shows a single slice of the earth's layers while the 3D seismic image shows a volume shape of earth core. Whereas, the 4D seismic data is an extension of 3D seismic image by showing 3D volumes at different times. Fig. 1 illustrates example of different 2D seismic images. Fig. 1. 2D seismic images Seismic data is widely used with the field of extracting oil and natural gas. Seismic data help oil and gas companies to improve performance of drilling the earth's layers by selecting optimum locations. Thereby, the extracted amount of oil and gas will be maximized with decreasing the cost. Currently, recent artificial intelligence systems have been employed to explore place of oil and gas. One of the most important tasks achieved by these systems is based on detecting a specific geologic structure in seismic data. This geologic structure is referred to as “salt dome”. Salt dome [2] is a mushroom-shaped diaper formed by the deposition of salt. It is an important geologic structure, because it is common for a salt dome to trap petroleum reservoirs. Detecting salt boundaries in seismic data is a challenging problem because of several reasons. The extraction should allow for the possibility that there are sediments within the salt body. The algorithm of detection must handle complex geometries, e.g., top and base salt should not be limited to surfaces that are single-valued in depth. In addition, the quality of the extracted salt body should be assessed. Thus, that the manual quality control and editing to produce the final result is reduced to a minimum. Fig. 2 demonstrates the shape of salt dome within a seismic image. The current state of the art in salt body extraction is mainly based on using complex operations. Most of these operations mimic deep learning algorithms for interpreting salt bodies. To simplify process of extracting salt bodies from seismic data with reasonable performance, we present a semi-supervised clustering technique. The presented technique provides remarkable results with a very limited number of features (attributes).