2025/10/04 by Ghannadpour, Seyyed Saeed, Esmaelzade Kalkhoran, Samane, Beiranvand Pour, Amin
#ASTER #Advanced image processing #Alteration mapping #FCM clustering #Fusion #Hydrothermal alteration zones #Iran #K-Means-SVM #Machine Learning algorithms #Porphyry Cu exploration #SAM
paper · doi:10.57647/j.ijes.2025.17565
Machine learning (ML) algorithms can be applied to remote sensing satellite imagery to improve the accuracy of mineral and lithology mapping in the initial stages of a mineral exploration program. In this study, both unsupervised and supervised ML algorithms, namely K-Means clustering, Support Vector Machine (SVM), Fuzzy C-Means (FCM) clustering, and Spectral Angle Mapper (SAM), are evaluated and fused for the accurate mapping of alteration zones in the Zafarghand porphyry Cu region in central Ir an using ASTER satellite imagery. The alteration zones of potassic, phyllic, argillic, propylitic, and silicification were documented in the Zafarghand porphyry deposit, which could be identified and mapped with ASTER satellite images and ML algorithms. The ML algorithms analysed the digital pixel numbers (DNs) of the satellite images from the ASTER sensor to determine hydrothermal alteration zones. K-Means clustering was used for grouping the data, while SVM was used for categorization and prediction. K-Means clustering achieved an accuracy of 78.83%, while the SVM method achieved an accuracy of 98.25% in identifying hydrothermal alteration zones, especially propylitic and phyllic alteration zones. Moreover, stacking the bands and applying FCM clustering with SAM improved the precision of hydrothermal alteration detection, and the accuracy of K-Means-SVM after stacking reached 96.2%. Using the fusion of these ML algorithms, a thorough investigation of alteration zones was developed, improving the overall accuracy and effectiveness of the alteration mapping procedure. The developed ML-based approach was then used to create a map of hydrothermal alteration in the Zafarghand porphyry Cu region, which was verified in field investigations and petrographic analyses. The Xie-Beni index for FCM clustering was 0.25833, indicating high accuracy and reliability in the clustering process. The results show that the fusion of supervised and unsupervised ML algorithms, such as K-Means-SVM, combined with methods like FCM clustering and SAM, has the potential to identify complex geological features and significantly improve the porphyry Cu exploration program at low cost.