1162 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
same color can be in a different cluster if they show a higher
color gradient. This shows that these criteria are not suf-
ficient to separate the grains efficiently for this application.
CONCLUSIONS
This preliminary research extended the SAM framework
and applied it to the segmentation of grains in geological
samples of sand. The viability and performance of SAM for
geological samples segmentation was validated especially
in the case of sand where particles are liberated it showed
high accuracy. The following specific parameters within
the SAM’s code can be adjusted for better results in the
context of sand samples: box_nms_thresh, crop_nms_thresh,
pred_iou_thresh and point_per_side. The last one in particu-
lar plays a role in the performance of SAM and is directly
related to the running time of the algorithm. Further tests
could be done on finding the balance between time effi-
ciency and performance of segmentation to be accurate and
representative in a day to day operations context.
The classification using k-means algorithm based
on color is accurate for samples with distinct mineralogy
(black ilmenite and white or transparent zircon for exam-
ple) but shows its limitation as soon as a mineral can have
multiple colors (such as the rutile presented) or two miner-
als have the same color (quartz and zircon are both white or
transparent). This means that either new parameters need
to be added to the algorithm (such as size, shape, etc.) or
a more solid algorithm for classification such as DBScan
needs to be implemented, specifically for this deposit. In
digital image processing the quality of the image is nec-
essary and a bank of images can only be used efficiently
if representative of the future use conditions (light, back-
ground, reflection, sharpness, etc). Therefore, the images
necessary for the image bank will be collected at the plant
during operations so they can be as representative taken in
the condition of daily operations are being collected and
will be implemented in the near future.
Figure 9. Classification using k-mean algorithm results for rutile product
(a) (b)
REFERENCES
Berry, R., Walters, S. G., &McMahon, C. 2008.
Automated mineral identification by optical micros-
copy. Australasian Institute of Mining and Metallurgy
Publication Series, 91–94.
De Castro, B., Benzaazoua, M., Chopard, A., &Plante, B.
2022. Automated mineralogical characterization using
optical microscopy: Review and recommendations.
Minerals Engineering, 189, 107896.
Gomes, O. D. F. M., Iglesias, J. C. A., Paciornik, S., &
Vieira, M. B. (2013). Classification of hematite types
in iron ores through circularly polarized light micros-
copy and image analysis. Minerals Engineering, 52,
191–197.
Hołyńska, B., Lankosz, M., Ostachowicz, J., &Wolski, K.
1985. On-stream XRF measuring system for ore slurry
analysis and particle-size control. The International
Journal of Applied Radiation and Isotopes, 36(5),
369–373.
Iglesias, J. C. Á., Santos, R. B. M., &Paciornik, S. 2019.
Deep learning discrimination of quartz and resin
in optical microscopy images of minerals. Minerals
Engineering, 138, 79–85.
Khajehzadeh, N., Haavisto, O., &Koresaar, L. 2016.
On-stream and quantitative mineral identification
of tailing slurries using LIBS technique. Minerals
Engineering, 98, 101–109.
Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C.,
Gustafson, L., ...&Girshick, R. 2023. Segment any-
thing. arXiv preprint arXiv:2304.02643. doi: 10.48550
/arXiv.2304.02643.
Koh, E. J., Amini, E., Spier, C. A., McLachlan, G. J., Xie,
W., &Beaton, N. 2024. A mineralogy characterisa-
tion technique for copper ore in flotation pulp using
deep learning machine vision with optical microscopy.
Minerals Engineering, 205, 108481.
same color can be in a different cluster if they show a higher
color gradient. This shows that these criteria are not suf-
ficient to separate the grains efficiently for this application.
CONCLUSIONS
This preliminary research extended the SAM framework
and applied it to the segmentation of grains in geological
samples of sand. The viability and performance of SAM for
geological samples segmentation was validated especially
in the case of sand where particles are liberated it showed
high accuracy. The following specific parameters within
the SAM’s code can be adjusted for better results in the
context of sand samples: box_nms_thresh, crop_nms_thresh,
pred_iou_thresh and point_per_side. The last one in particu-
lar plays a role in the performance of SAM and is directly
related to the running time of the algorithm. Further tests
could be done on finding the balance between time effi-
ciency and performance of segmentation to be accurate and
representative in a day to day operations context.
The classification using k-means algorithm based
on color is accurate for samples with distinct mineralogy
(black ilmenite and white or transparent zircon for exam-
ple) but shows its limitation as soon as a mineral can have
multiple colors (such as the rutile presented) or two miner-
als have the same color (quartz and zircon are both white or
transparent). This means that either new parameters need
to be added to the algorithm (such as size, shape, etc.) or
a more solid algorithm for classification such as DBScan
needs to be implemented, specifically for this deposit. In
digital image processing the quality of the image is nec-
essary and a bank of images can only be used efficiently
if representative of the future use conditions (light, back-
ground, reflection, sharpness, etc). Therefore, the images
necessary for the image bank will be collected at the plant
during operations so they can be as representative taken in
the condition of daily operations are being collected and
will be implemented in the near future.
Figure 9. Classification using k-mean algorithm results for rutile product
(a) (b)
REFERENCES
Berry, R., Walters, S. G., &McMahon, C. 2008.
Automated mineral identification by optical micros-
copy. Australasian Institute of Mining and Metallurgy
Publication Series, 91–94.
De Castro, B., Benzaazoua, M., Chopard, A., &Plante, B.
2022. Automated mineralogical characterization using
optical microscopy: Review and recommendations.
Minerals Engineering, 189, 107896.
Gomes, O. D. F. M., Iglesias, J. C. A., Paciornik, S., &
Vieira, M. B. (2013). Classification of hematite types
in iron ores through circularly polarized light micros-
copy and image analysis. Minerals Engineering, 52,
191–197.
Hołyńska, B., Lankosz, M., Ostachowicz, J., &Wolski, K.
1985. On-stream XRF measuring system for ore slurry
analysis and particle-size control. The International
Journal of Applied Radiation and Isotopes, 36(5),
369–373.
Iglesias, J. C. Á., Santos, R. B. M., &Paciornik, S. 2019.
Deep learning discrimination of quartz and resin
in optical microscopy images of minerals. Minerals
Engineering, 138, 79–85.
Khajehzadeh, N., Haavisto, O., &Koresaar, L. 2016.
On-stream and quantitative mineral identification
of tailing slurries using LIBS technique. Minerals
Engineering, 98, 101–109.
Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C.,
Gustafson, L., ...&Girshick, R. 2023. Segment any-
thing. arXiv preprint arXiv:2304.02643. doi: 10.48550
/arXiv.2304.02643.
Koh, E. J., Amini, E., Spier, C. A., McLachlan, G. J., Xie,
W., &Beaton, N. 2024. A mineralogy characterisa-
tion technique for copper ore in flotation pulp using
deep learning machine vision with optical microscopy.
Minerals Engineering, 205, 108481.