XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 1157
This application is designed to semi-automatize mineral
grain analysis from microscope pictures and uses the power
of cutting-edge image processing and machine learning
algorithms to automatically detect, classify, and analyze
mineral grains within geological samples. This tool aims
at accelerating the analytical process and at reducing the
manual labor.
MATERIAL USED
The material used for this study comes from a heavy min-
erals sands mining operation that primarily extracts heavy
mineral sands, including ilmenite, rutile, and zircon, which
are valuable components used in the production of various
industrial and consumer products. The mineral separation
process involves multiple stages of density separation fol-
lowed by magnetic and electrostatic separations. In order to
optimize the separation, a close monitoring of the perfor-
mance of each step or equipment is necessary.
For this study over 100 samples were sampled from dif-
ferent stream locations in the processing plant. They were
all analyzed for elemental composition using XRF and ana-
lyzed using Scanning Electron Microscopy (SEM) based
automated mineralogy to assess the variability in mineral
composition in each stream. The main minerals are :zir-
con, rutile, ilmenite, pseudorutile, leucoxene, anatase and
quartz. In lesser amounts the following minerals can be
found: calcium aluminum iron silicates (epidote and gre-
nat) and aluminosilicates (staurotide, tourmaline, kyanite).
Most of these minerals have very clear and defined color,
which make these samples good candidate for the machine
learning approach.
The photographs used in this study were taken using
a Tomlov DM602 numerical microscope. On this type of
basic entry level microscope the scale is not provided and
since the size of the material is not used in the present dem-
onstration it is not indicated on the following figures how-
ever, the average size of the material is 250 microns, which
can be used as a rough indication of grain size.
Figure 1 is a sample of rutile product and will serve as
base image for the algorithm demonstration.
ALGORITHMS STRUCTURE
This section describes the main methods proposed for the
grain sorting algorithm.
The approach consists of Three main stages: (1) seg-
mentation and partition of input images, (2) feature extrac-
tion, (3) classification of sub-images.
Segmentation
In digital image processing and computer vision, image seg-
mentation is the process of partitioning a digital image into
multiple image segments, also known as image regions or
image objects (sets of pixels). The goal of segmentation is to
simplify and/or change the representation of an image into
something that is more meaningful and easier to analyze.
Image segmentation is typically used to locate objects and
boundaries (lines, curves, etc.) in images. More precisely,
image segmentation is the process of assigning a label to
every pixel in an image such that pixels with the same label
share certain characteristics.
The result of image segmentation is a set of segments
that collectively cover the entire image, or a set of contours
extracted from the image. Each of the pixels in a region
are similar with respect to some characteristic or computed
property, such as color, intensity, or texture. Adjacent
regions are significantly different color respect to the same
characteristic(s). In this case, the goal of the segmentation
step is to locate each individual grain.
Segment Anything Model (SAM)
The Segment Anything Model (SAM) was released by Meta
in 2023 (Kirillov et al., 2023). It is built on Transformer
Vision models (Kirillov et al., 2023) trained on the large
visual corpus (SA-1B) containing more than 11 million
images and 1.1 billion masks. SAM has strong zero-shot
performance on a variety of segmentation tasks. However,
SAM is trained on general world case scenarios with popu-
lar structures. SAM has learned a general notion of what
Figure 1. Sample of rutile product
This application is designed to semi-automatize mineral
grain analysis from microscope pictures and uses the power
of cutting-edge image processing and machine learning
algorithms to automatically detect, classify, and analyze
mineral grains within geological samples. This tool aims
at accelerating the analytical process and at reducing the
manual labor.
MATERIAL USED
The material used for this study comes from a heavy min-
erals sands mining operation that primarily extracts heavy
mineral sands, including ilmenite, rutile, and zircon, which
are valuable components used in the production of various
industrial and consumer products. The mineral separation
process involves multiple stages of density separation fol-
lowed by magnetic and electrostatic separations. In order to
optimize the separation, a close monitoring of the perfor-
mance of each step or equipment is necessary.
For this study over 100 samples were sampled from dif-
ferent stream locations in the processing plant. They were
all analyzed for elemental composition using XRF and ana-
lyzed using Scanning Electron Microscopy (SEM) based
automated mineralogy to assess the variability in mineral
composition in each stream. The main minerals are :zir-
con, rutile, ilmenite, pseudorutile, leucoxene, anatase and
quartz. In lesser amounts the following minerals can be
found: calcium aluminum iron silicates (epidote and gre-
nat) and aluminosilicates (staurotide, tourmaline, kyanite).
Most of these minerals have very clear and defined color,
which make these samples good candidate for the machine
learning approach.
The photographs used in this study were taken using
a Tomlov DM602 numerical microscope. On this type of
basic entry level microscope the scale is not provided and
since the size of the material is not used in the present dem-
onstration it is not indicated on the following figures how-
ever, the average size of the material is 250 microns, which
can be used as a rough indication of grain size.
Figure 1 is a sample of rutile product and will serve as
base image for the algorithm demonstration.
ALGORITHMS STRUCTURE
This section describes the main methods proposed for the
grain sorting algorithm.
The approach consists of Three main stages: (1) seg-
mentation and partition of input images, (2) feature extrac-
tion, (3) classification of sub-images.
Segmentation
In digital image processing and computer vision, image seg-
mentation is the process of partitioning a digital image into
multiple image segments, also known as image regions or
image objects (sets of pixels). The goal of segmentation is to
simplify and/or change the representation of an image into
something that is more meaningful and easier to analyze.
Image segmentation is typically used to locate objects and
boundaries (lines, curves, etc.) in images. More precisely,
image segmentation is the process of assigning a label to
every pixel in an image such that pixels with the same label
share certain characteristics.
The result of image segmentation is a set of segments
that collectively cover the entire image, or a set of contours
extracted from the image. Each of the pixels in a region
are similar with respect to some characteristic or computed
property, such as color, intensity, or texture. Adjacent
regions are significantly different color respect to the same
characteristic(s). In this case, the goal of the segmentation
step is to locate each individual grain.
Segment Anything Model (SAM)
The Segment Anything Model (SAM) was released by Meta
in 2023 (Kirillov et al., 2023). It is built on Transformer
Vision models (Kirillov et al., 2023) trained on the large
visual corpus (SA-1B) containing more than 11 million
images and 1.1 billion masks. SAM has strong zero-shot
performance on a variety of segmentation tasks. However,
SAM is trained on general world case scenarios with popu-
lar structures. SAM has learned a general notion of what
Figure 1. Sample of rutile product