Machine learning-based lithological mapping and mineral prospecting using hyperspectral and multispectral remote sensing in Wadi Halfa, north Sudan
Year: 2025
Type: article
Abstract: At present, the global demand for mineral resources is critical, leading nations to focus on exploration. Remote sensing is a cost-effective tool, especially in harsh terrains. This study conducted lithological mapping in Wadi Halfa, North Sudan, using algorithm-based remote sensing, field observations, and petrographical analysis to detect iron ore and barite deposits. Multisensor optical datasets (L9, L8, and S2) were integrated to effectively delineate the lithological units. In addition, PRISMA hyperspectral data, with its detailed spectral signatures, improved spatial distribution patterns of barite and iron oxides across the study area. Image processing techniques (false colour composites, principal component analysis, minimum noise friction, band ratios) detected hydroxyl-bearing minerals, ferric, and ferrous oxides. Support Vector Machine (SVM), Artificial Neural Network (ANN), and Mahalanobis Distance Classifier (MDC) achieved overall accuracies of 95.51 %, 94.59 %, and 98.99 %, respectively. The study helped interpret the spatial relationship between barite and iron oxides. Four types of iron ore with more than three distinct layers were identified, including (a) oolitic ironstone, (b) ferruginous sandstone, (c) ferruginous ironstone, and (d) Banded Iron Formation (BIF) during field investigations, petrographic examinations, and chemical analysis validated remote sensing findings, revealing iron ore (62.7 % Fe) and barite (63.9 % Ba) concentrations. An economic assessment confirmed the presence of economic reserves suitable for exploitation. This research is recommended for broader application, particularly in machine learning for delineating iron ore and barite deposits in complex sedimentary sequences. The realization of machine learning algorithms emphasizes their potential to enhance lithological mapping in sedimentary sequences, suggesting a promising direction for future research. • Remote sensing and MLA mapped iron ore and barite (95.5% accuracy); field and laboratory confirmed rich mineable reserves. • Significant concentrations of iron ore (62.7 % Fe) and barite (63.9 % Ba).
Authors: Abazar M.A. Daoud, Ali Shebl, Mutwakil Nafi, Abdelmajeed A. Elrasheed, Árpád Csámer, Péter Rózsa
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,370
OpenAlex ID: W7077065941