Satellite Image Classification Using Cascaded Architecture of Neural Fuzzy Network - Advanced Techniques in Knowledge Discovery and Data Mining - page 212

Kohonen's SOMs. In the next section, the acquisition, implementation, and modi-
fication of the three groups of inputs are given. In Section 8.3, the integration of
CNFM by cascading Kohonen's SOM and SONFIN is modeled. Experimental re-
sults are shown in Section 8.4. Finally, some conclusions are reached in Section
8.5.

Supervised
neural
fuzzy network

Unsupervised
neural network

Unsupervised
neural network

Unsupervised
neural network

Statistical
features

Spectral
features

Spatial
features

Fig. 8.1. System architecture of CNFM.

A region among a
center-pixel 3 X 3

gray-
values

Kohonen's
SOM

2D
coordinates

Desired
cluster i

A region among a
center-pixel 6 X 6

Co-occurrence
matrix

ASM,
IDM,
CON,
ENT.

Neural
fuzzy
network
(SONFIN)

2D
coordinates

C luster j

Kohonen's
SOM

A region among a
center-pixel 8 X 8

Wavelet
decomposition

2D
coordinates

Kohonen's

SOM

Energies,
entropies.

Fig. 8.2. A detailed view of CNFM.