1160 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
increases the number of objects detected (41, 114 and 136
respectively), but the running time also increases (50s, 115,
and 246s respectively).
Figure 7 shows the original image (a) and the segmented
results for points_per_side=15 (b) points_per_side=33 (c)
and points_per_side=50. In the highlighted sections in red
we can observe that some grains are not segmented by
SAM. It is important to note that for example grain labeled
1 is detected in the first iteration (points_per_side=15) and
in the third iteration (points_per_side=50) but not in the
second iteration (points_per_side=33). These results mean
that optimization needs to be found between performance
of the segmentation algorithm and the analysis time. This
especially plays a role in the context of mineral processing
operations.
Classification
Figure 8 shows the result of the classification algorithm on a
mixed sample of pure products Ilmenite (black grainvs) and
Zircon (white or transparent grains). The original image
can be found in Figure 7(a). The classification algorithm
seems to perform well, finding 51% of zircon and 40% of
ilmenite in proportions of pixels and 59% (80 grains) of
zircon and 41% (56 grains) of ilmenite. Visually the separa-
tion is clear and the identification is correct.
Figure 9 shows the result of the classification algo-
rithm on the original picture (a) with four clusters (b). The
expected result would be four categories as follow:
• dark brown grains corresponding to dark rutile
• maroon grains corresponding to staurotide
• orange and yellow grains corresponding to rutile
• white yellow grains corresponding to anatase
The classification criteria are defined as the average of
median pixel value and interquartile range for each color
Figure 5. Example of clustering calculated by the k-means algorithm
0
50
100
150
200
250
300
0 20 40 60
points_per_side
Figure 6. Comparison between the points_per_side
parameter and the running time of the algorithm. Data
analysis performed on 16cores at 2.30GHz
Table 1. Performance of SAM algorithm in number of
objects detected compared to the points_per_side setting
Points_per_sides 15 33 50
Running time (s) 50 115 246
Number of objects detected 41 114 136
run
g415me
(s)
increases the number of objects detected (41, 114 and 136
respectively), but the running time also increases (50s, 115,
and 246s respectively).
Figure 7 shows the original image (a) and the segmented
results for points_per_side=15 (b) points_per_side=33 (c)
and points_per_side=50. In the highlighted sections in red
we can observe that some grains are not segmented by
SAM. It is important to note that for example grain labeled
1 is detected in the first iteration (points_per_side=15) and
in the third iteration (points_per_side=50) but not in the
second iteration (points_per_side=33). These results mean
that optimization needs to be found between performance
of the segmentation algorithm and the analysis time. This
especially plays a role in the context of mineral processing
operations.
Classification
Figure 8 shows the result of the classification algorithm on a
mixed sample of pure products Ilmenite (black grainvs) and
Zircon (white or transparent grains). The original image
can be found in Figure 7(a). The classification algorithm
seems to perform well, finding 51% of zircon and 40% of
ilmenite in proportions of pixels and 59% (80 grains) of
zircon and 41% (56 grains) of ilmenite. Visually the separa-
tion is clear and the identification is correct.
Figure 9 shows the result of the classification algo-
rithm on the original picture (a) with four clusters (b). The
expected result would be four categories as follow:
• dark brown grains corresponding to dark rutile
• maroon grains corresponding to staurotide
• orange and yellow grains corresponding to rutile
• white yellow grains corresponding to anatase
The classification criteria are defined as the average of
median pixel value and interquartile range for each color
Figure 5. Example of clustering calculated by the k-means algorithm
0
50
100
150
200
250
300
0 20 40 60
points_per_side
Figure 6. Comparison between the points_per_side
parameter and the running time of the algorithm. Data
analysis performed on 16cores at 2.30GHz
Table 1. Performance of SAM algorithm in number of
objects detected compared to the points_per_side setting
Points_per_sides 15 33 50
Running time (s) 50 115 246
Number of objects detected 41 114 136
run
g415me
(s)