Mapping ilmenite deposit in Pulmudai, Sri Lanka using a hyperspectral imaging-based surface mineral mapping method

Abstract

Mineral detection using remote sensing techniques is important since it saves the time and effort of carrying out manual land surveys. In this paper a novel algorithm, which can be used to detect ilmenite using hyperspectral image analysis is discussed. To investigate this task, a hyperspectral image obtained from the Earth Observing-1 (EO-1) satellite's Hyperion sensor was used. In the proposed algorithm, first, principal component analysis (PCA) was used for dimensionality reduction and an Euclidean distancebased method was used to extract the pixels containing soil. Thereafter, lab spectral data of typical ilmenite deposits were considered as the reference and a correlation factor analysis was carried out to determine the soil pixels, which are most likely to contain ilmenite and most unlikely to contain ilmenite. Using these two sets of pixels, a training set was constructed to apply Fisher's discriminant analysis (FDA) in order to separate the dataset into two distinct classes-ilmenite and non-ilmenite. Based on the spectral similarity, each pixel of the image was classified under one of these classes. This paper also introduces a probability-based approach to obtain results that are more accurate. A probability density function was designed considering the spatial distribution of the mineral. Thereafter, classification was done considering the probability measure as well. Lab tests performed on the soil samples collected from the locations, which were detected by the algorithm validate that the algorithm is accurate.

Single-target mineral detection with site-specific endmember extraction for survey points identification: A case study of Jaffna, Sri Lanka

“…Besides, owing to greater spatial resolution, HSIs available via satellites or airborne imaging systems, have been used to analyze minerals considering soil variability at different RS scales and to measure rock microstructures (Van Ruitenbeek et al, 2019) as well. Further, the use of generic laboratory-generated spectral libraries has been the prominent method used in most digital lithological studies (Black et al, 2016;Ekanayake et al, 2019;Feng et al, 2003;Grebby et al, 2011;Tziolas et al, 2020) whereas generation of a site-specific signature has not been a focus. Moreover, the minerals of which the generic spectral signatures were used for lithological mapping were either assumed to be present in the site (Ninomiya and Fu, 2019;Yu et al, 2012) or known from an already conducted survey of the site (Pour et al, 2018;Xiong et al, 2011).…”

“…To reduce misclassification of other minerals as 'limestone' and for computational efficiency, the HSI was pre-classified into subcomponents. For this, sub-component analysis (Ekanayake et al, 2019) was used in combination with vertex component analysis (VCA) (Nascimento and Dias, 2005). As suggested in Ranasinghe et al (2019), pixels were categorized into four classes: soil, water, vegetation, and sand.…”

“…Mineral identification was performed on the pixels classified under the soil class. However, the algorithm proposed in Ekanayake et al (2019) requires a priori knowledge regarding the composition of the soil to construct a representative spectral signature and the laboratory references for every major mineral compound that is assumed to exist. Nonetheless, it is not necessary to have information about the mineral composition of the soil as such reference collection is immaterial for an algorithm with an application of identifying probable survey locations for an earmarked mineral.…”

A Robust Stochastic Approach to Mineral Hyperspectral Analysis for Geometallurgy

“…Another very interesting work that combines dimensionality reduction and statistical analysis for mineral classification problem was presented by Ekanayake et al [109]. This work attempts to obtain ilmenite content mapping for a real case by using principal component analysis (PCA) as a dimensionality reduction strategy.…”

Convolutional Autoencoder for Blind Hyperspectral Image Unmixing

“…Hyperspectral image (HSI) data has been extensively used in remote sensing as it provides a glut of spectral information to analyze and classify materials [1]- [4] in an area. Though, the respective sensory methods often offer higher spectral resolution, the same cannot be told regarding their spatial resolution.…”