Image Analysis, Classification and Change Detection in Remote Sensing, 3rd Edition

The Kohonen self-organizing map 353

8.6 The Kohonen self-organizing map

The Kohonen self-organizing map (SOM), a simple example of which is sketched

in Figure 8.4 , belongs to a class of neural ne tworks which are trained by com-

petitive learning (Hertz et al., 1991; Ko honen, 198 9). It is very useful as a

visualization tool for exploring the class structure of multispectral imagery.

The layer of neurons shown in the figure can have any geometry, but usually

a one-, two-, or three-dimensional array is chosen. The input signal is the

observation vector g = (g

1

, g

2

. . . g

N

)

⊤

, where, in the figure, N = 2. Each

input to a neuron is associated with a synaptic ...