Multivariate Image Analysis in Mineral Processing - Advanced Control and Supervision of Mineral Processing Plants - page 128

Tabl e 3. 2 Main characteristics and class assignment for the five mineral types. From Tessier et al.
[26]

Mineral type Impact work

index

Ni grade Density

( kg

Hardness Class

m 3 ) /

Massive sulphide (MS) Low High 4000 Soft 1

Disseminated sulphide (DS) Medium Medium 2300 Medium 2

Net textured (NT) Medium Medium 2300 Medium 2

Gabbro (G) High Waste 3000 Hard 3

Peridotite (P) High Waste 2700 Hard 3

era provides 8 bit images of 1024 1376 pixels spatial resolution. The camera was

installed 90 cm above the conveyor leading to a 32

×

24 cm field of view. The con-
veyor was located at COREM's pilot plant (Quebec City, Canada). More details on
the hardware used, including the lighting system, are available in Tessier et al. [26].

×

3.5.3.2 The Mineral Recognition Problem

Rock type identification for this mineral system is challenging for three reasons:

1. The visual characteristics of individual rock fragments are themselves very het-
erogeneous. Their color and surface texture patterns vary from one face to the
other. This can be clearly observed in the massive sulphide and gabbro images
available in Figure 3.22. Hence, the intra-class variability in the rock fragments
visual features is important, at least for some of them.
2. A strong overlap in some color and surface textural features between rock frag-
ments belonging to different classes is observed in Figure 3.23 A-B. Image A is
a composite image made from larger images of each rock type. The color signa-
ture of the pixels segmented by the blue region drawn on the MS rock image was
quantified using MIA. The pixels belonging to other rock types but having sim-
ilar color features as MS are identified by the blue color in image B. The strong
color overlap between MS, DS and NT rocks is clearly observed. Hence, inter-
class variability may be small between some of the classes ( i.e. some classes may
overlap in the classification problem).
3. The rock surfaces can be completely dry or wet depending on weather conditions.
Surface moisture modifies the visual appearance of the fragments, their color
mainly. Wet rocks are systematically darker when wet as shown in Figure 3.23 C
(dry) and D (wet).

The machine vision approach will address each of these issues in order to achieve
accurate rock type classification.