Date of Award


Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy in Geological Engineering (PhD)

Administrative Home Department

Department of Geological and Mining Engineering and Sciences

Advisor 1

Snehamoy Chatterjee

Advisor 2

Thomas Oommen

Committee Member 1

Chad Deering

Committee Member 2

William H. Farrand


The sustainable exploration of mineral resources plays a significant role in the economic development of any nation. The lithological maps and surface mineral distribution can be vital baseline data to narrow down the geochemical and geophysical analysis potential areas. This study developed innovative spectral and Machine Learning (ML) methods for mineral and lithological classification. Multi-sensor datasets such as Airborne Visible/Infrared Imaging Spectrometer-Next Generation (AVIRIS-NG), Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), Advanced Land Observing (ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR), Sentinel-1, and Digital Elevation Model (DEM) were utilized. The study mapped the hydrothermal alteration minerals derived from Spectral Mapping Methods (SMMs), including Spectral Angle Mapper (SAM), Spectral Information Divergence (SID), and SIDSAMtan using high-resolution AVIRIS-NG hyperspectral data in the Hutti-Maski area (India). The SIDSAMtan outperforms SID and SAM in mineral mapping. A spectral similarity matrix of target and non-target classes based optimum threshold selection was developed to implement the SMMs successfully. Three new effective SMMs such as Dice Spectral Similarity Coefficient (DSSC), Kumar-Johnson Spectral Similarity Coefficient (KJSSC), and their hybrid, i.e., KJDSSCtan has been proposed, which outperforms the existing SMMs (i.e., SAM, SID, and SIDSAMtan) in spectral discrimination of spectrally similar minerals. The developed optimum threshold selection and proposed SMMs are recommended for accurate mineral mapping using hyperspectral data. An integrated spectral enhancement and ML methods have been developed to perform automated lithological classification using AVIRIS-NG hyperspectral data. The Support Vector Machine (SVM) outperforms the Random Forest (RF) and Linear Discriminant Analysis (LDA) in lithological classification. The performance of SVM also shows the least sensitivity to the number and uncertainty of training datasets. This study proposed a multi-sensor datasets-based optimal integration of spectral, morphological, and textural characteristics of rocks for accurate lithological classification using ML models. Different input features, such as (a) spectral, (b) spectral and transformed spectral, (c) spectral and morphological, (d) spectral and textural, and (e) optimum hybrid, were evaluated for lithological classification. The developed approach has been assessed in the Chattarpur area (India) consists of similar spectral characteristics and poorly exposed rocks, weathered, and partially vegetated terrain. The optimal hybrid input features outperform other input features to accurately classify different rock types using the SVM and RF models, which is ~15% higher than as obtained using spectral input features alone. The developed integrated approach of spectral enhancement and ML algorithms, and a multi-sensor datasets-based optimal integration of spectral, morphological, and textural characteristics of rocks, are recommended for accurate lithological classification. The developed methods can be effectively utilized in other remote sensing applications, such as vegetation/forest mapping and soil classification.