Affiliations: Departamento de Electrónica e Ingeniería
Electromecánica, Universidad de Extremadura, Avda. de Elvas s/n, 06071
Badajoz, Spain. E-mail: {miguel, carlos, horacio,
ramon}@nernet.unex.es | Departamento de Física, Universidad de
Extremadura, Avda. de Elvas s/n, 06071 Badajoz, Spain. E-mail:
[email protected]
Note: [] Corresponding author: M. Macías-Macías, Centro
Universitario de Mérida, Sta. Teresa de Jornet, 38, 06800 Mérida, Spain.
Tel.: +34 387068, ext: 2538
Abstract: In this work a back propagation neural network is used for the
segmentation of Meteosat images covering the Iberian Peninsula. The images are
segmented in the classes land (L), sea (S), fog (F), low clouds
(C_L), middle clouds (C_M),
high clouds (C_H) and clouds with vertical growth
(C_V). The classification is performed from an initial
set of several statistical features based on the gray level co-occurrence
matrix (GLCM) proposed by Welch [1]. This initial set of features is made up of
144 parameters and to reduce its dimensionality three methods for feature
selection have been studied and compared. The first one includes genetic
algorithms (GA), the second is based on principal component analysis (PCA) and
the third uses independent component analysis (ICA). These methods are
conceptually very different. While GA interacts with the neural network in the
selection process, PCA and ICA only depend on the values of the initial set of
features.
