Environmental Engineering Reference
In-Depth Information
Tabl e 3. 3
Validation results for the 10 composite images. From Tessier
et al.
[26]
True surface composition (%)
Estimated surface composition (%)
Image
Ore
Soft
Medium
Hard
Soft
Medium
Hard
1
Dry
20
50
30
19.0
50.5
30.5
2
Dry
20
60
20
25.6
56.9
17.5
3
Dry
20
30
50
12.1
24.9
63.0
4
Dry
20
40
40
21.3
34.7
44.0
5
Dry
20
30
50
22.0
28.9
49.1
6
Wet
20
50
30
23.6
61.9
14.8
7
Wet
20
60
20
29.2
67.4
3.4
8
Wet
20
30
50
22.8
34.6
42.6
9
Wet
30
40
30
39.3
47.2
13.5
10
Wet
20
30
50
29.2
39.4
31.4
Mean
Overall
21
42
37
24.4
44.6
31.0
Dry
20
42
38
20.0
39.2
40.8
Wet
22
42
36
28.8
50.0
21.1
Correlation between true and estimated
Dry
0.73
0.90
0.84
Wet
0.65
0.47
0.63
Weight %
Number
Experiment #
0%
0%
100%
100%
Soft
Medium
Hard
of images
12
12
1
33
66
0
26
2
66
33
0
18
3
0
66
33
19
4
0
33
66
27
1
1
3
3
5
66
0
33
31
9
9
6
33
0
66
30
7
60
20
20
23
8
20
20
60
26
9
20
60
20
23
10
10
10
33
33
33
30
4
4
2
2
11 (Pure MS)
100
0
0
24
7
7
8
8
12
0
100
0
17
13
0
0
100
42
14 (Pure DS)
0
100
0
75
15 (Pure NT)
0
100
0
52
100%
100%
5
5
6
6
0%
0%
11
11
13
13
16 (Pure G)
0
0
100
50
17 (Pure P)
0
0
100
46
0%
0%
100%
100%
Har d
Har d
(a)
(b)
Figure 3.26
Design of experiment for conveyor belt application. From Tessier
et al.
[26]
was preserved in this final application. In the end, about 560 images were collected
including both dry and wet rocks. The machine vision approach was applied to
each image and the rock classification results are provided in Figure 3.27, which
presents a comparison between the true and estimated weight compositions (solid
line and dots, respectively). To convert surface composition obtained by imaging
into weight composition, the estimated surface composition were weighted accord-
ing to the density of each mineral (see Table 3.2).
Very promising results were obtained, particularly for dry rock mixtures. The
trends are followed quite well, although some confusion between medium and hard
Search WWH ::
Custom Search