Remote sensing data processing using convolutional neural networks for mapping alteration zones

Geological mapping faces challenges with traditional methods, prompting the exploration of streamlined approaches. This study employs convolutional neural networks (CNNs) on Landsat 8, Landsat 9, and ASTER data for mapping alteration zones in Broken Hill, a mineral-rich region in the west of New South Wales, Australia. CNNs, adept at extracting features, outperform support vector machine and multilayer perceptron methods. Landsat 9 excels in mapping iron oxide areas, while ASTER, paired with CNNs, provides the most accurate map for argillic and propylitic alteration zones. The findings underscore the effectiveness of CNN in enhancing geological mapping precision, particularly in identifying subtle alterations indicative of mineralisation processes.