Self-Organizing Maps and Unsupervised Classification - Neural Networks: Methodology and Applications - page 438

Fig. 7.31. Representation of the patterns, of the observation subsets and of the
geographical zones that are allocated to neurons 33, 82 and 93 (PRSOM map 10 × 10
using App cod2 as learning set). Figures ( a ), ( b ), ( c ) represent the neuron spectra,
Figures ( d ) and ( e ) represent the associated learning set spectra for neurons 33 and
82. The zones that are associated to neurons 33 and 93 have been colored in black
in figure ( f ) and ( g )

Table 7.3. Confusion matrix allowing the comparison between the expert-based
classification (GAC image) and the PRSOM-based classification. On this map, the
hundred neurons have been labeled using majority vote from expert data

PRSOM + majority vote

Class 1 Class 2 Class 3 Class 4

Marine aerosols 0.8 0.04 0 0.16

Expert Clouds 0.03 0.91 0.01 0.05

Case 2 0.03 0.22 0.71 0.03

Desert aerosols 0.1 0.04 0 0.86