Unsupervised Clustering on Paleochannel Detection with Curvature-Based Attribute

Abstract

Abstract This research focuses on the application of advanced seismic data processing techniques and unsupervised clustering algorithms to accurately identify and map paleochannels in the high-resolution Maui seismic dataset. Pre-processing steps were applied to enhance data quality, followed by the calculation of a positive curvature attribute used as input for multiple clustering algorithms. The study evaluates four state-of-the-art unsupervised clustering methods and employs Principal Component Analysis (PCA) for result refinement and differentiation. By analysing the effectiveness of the algorithms, we demonstrate that Fuzzy C-means and Birch provide optimal segmentation of paleochannels. This automated approach not only overcomes the challenges of noisy data but also opens avenues for similar applications in seismic interpretation. The research underscores the significance of the unsupervised method in facilitating efficient resource exploration, emphasizing the integration of PCA to enhance segmentation accuracy and reduce human intervention. This study contributes to the sustainable management and extraction of natural resources, underscoring the potential of advanced techniques in remote sensing analysis.