Multi-sensor analysis of Aguas Teñidas, Magdalena, Sotiel and Majada deposits, Iberian Pyrite Belt, Spain: Analytical drill core-scanning for optimized exploration of ore, ULiège
Simoes Teixeira Mendes, Pedro
Promotor(s) : Pirard, Eric
Date of defense : 24-Aug-2018 • Permalink :
|Multi-sensor analysis of Aguas Teñidas, Magdalena, Sotiel and Majada deposits, Iberian Pyrite Belt, Spain: Analytical drill core-scanning for optimized exploration of ore, ULiège
|Simoes Teixeira Mendes, Pedro
|Date of defense :
|Committee's member(s) :
Pons Perez, Juan Manuel
|Number of pages :
|Engineering, computing & technology > Geological, petroleum & mining engineering
|Université de Liège, Liège, Belgique
|Master en ingénieur civil des mines et géologue, à finalité spécialisée en "geometallurgy (EMERALD)"
|Master thesis of the Faculté des Sciences appliquées
[fr] This master thesis is a result of investment considering an optimistic economic scenario for the
use of raw materials in Europe in the recent years. The work focuses mainly on Hyperspectral
Images in the VNIR-SWIR range but also on X-Rays Transmission analysis to assist the first
steps of a broad project known as ANCORELOG - Analytical Core Logging. A multi-sensor
scanning technique is applied to improve prospection of cupriferous and polymetallic (Cu-Pb-Zn)
ores in Aguas Teñidas, Magdalena, Sotiel and Majada deposits in a fast and consistent way.
Initially, the methodology includes a practical procedure in MATSA - Minas de Aguas Teñidas,
Iberian Pyrite Belt, Spain.
Information regarding exploration and ore treatment activities was essential for the selection of
40 segments of drill cores as samples used for geological characterization, density measurement
and chemical analysis of valuable metals, followed by interpretation of image analysis techniques.
Acquisition with sensors was carried out at the University of Liege, Belgium. Spectral analysis
indicates contrasting responses ruled by lithologies from two types of geological deposits:
Volcanogenic Massive Sulphides and exhalative hosted by shales. Supervised classification of
Hyperspectral Images using Machine Learning for core box lithological mapping was applied with
a robust strategy to identify optimal parameters in practical applications.
Considering time as a critical variable along with computational efforts, classification clearly
shows to be dependent on the level of investment in cameras and hardware. Fisher Discriminant
and Linear Support Vector Classifier (LSVC) were considered as the most promising algorithms
(perClass library) to discriminate barren rocks, alteration degrees and mineralization types.
Classification performance reflects the variation of absorption features in the spectra profile, being
more consistent for volcanic material than fine clastic rocks. Both algorithms show outstanding
results with an impressive capability to deal with a large number of classes. Average accuracy is
close to 100% when LSVC is combined to a large number of features. When either VNIR or
SWIR cameras is chosen, the classification error increases but satisfactory results (approximately
80% accuracy) are obtained when Fisher is used. In this case accuracy can be improved by
increasing the training set size or applying post-post processing techniques without compromising
X-Rays are efficient to differentiate sulphides from barren rocks. Even though a slight
discrimination between ore types is possible by analysing the Transmitted-Attenuated profiles,
an objective segregation at mineral level is still a challenge due to sensor limitations such as poor
contrast and sample thickness. The results of this work successfully meet the first steps of
ANCORELOG with promising business opportunities. A real classification performance can only
be evaluated when applying algorithm decisions to a complete drill core on site. It should consider
class changes to adapt the training set to a real geological section where transitional contacts and
high degree of textural variation are impractical to be sampled in a representative manner. Future
works can improve pixel-wise segmentation and ore characterization when combined to XRF and
Cite this master thesis
The University of Liège does not guarantee the scientific quality of these students' works or the accuracy of all the information they contain.