Multivariate Image Analysis in Mineral Processing - Advanced Control and Supervision of Mineral Processing Plants - page 120

160

211
373
800

140

120

100

80

60

40

20

0

1 3 7 15 31 63

(Scaled) Bubble Diameter

211 373 800

Figure 3.17 Bubble size histograms computed for three froth images. From Liu et al. [37]

3.5.2.2 Froth Color Features

Color features consists of black holes (BH) and clear windows (CL). Both of these
are shown in Figure 3.18(a). When the froth is poorly loaded with mineral particles,
bubbles size is large and regions not covered by particles starts appearing on top
of bubbles. The black spots located on top of the bubbles of the froth image shown
in Figure 3.18(a)(left) correspond to the so-called CL. Black holes appear in the
opposite situation: when the froth is overloaded with mineral particles and bubbles
are very small. The BH shown in Figure 3.18(a) (right) indicate the onset of froth
collapse. The color features used in classification of froth state correspond to the
number of pixels belonging to either BH or CL. The method used for computing
these for each incoming image is illustrated in Figure 3.18.
A composite image made of two froth images, a poorly loaded (left) and a heavily
loaded froth (right) is shown in Figure 3.18(a). Multivariate image analysis applied
to this composite image ( i.e. , PCA applied to the unfolded composite image) yields
the t 1 - t 2 score density histogram shown in Figure 3.18(b). The white mask appear-
ing on that histogram, found by trial and error, encompasses most pixels having a
black (or nearly black) color as shown in Figure 3.18(c). The pixels falling under-