S. Delgado, C. Gonzalo, E. Martínez, Á. Arquero

 

Abstract

 

Remote sensing can be defined as the technique that facilitates the acquisition of land surface
data without contact with the material object of observation. The development of tools for
analyzing and processing multispectral images captured by sensors aboard satellites has
provided the automation of tasks that could not be possible otherwise. The main problem related
with this discipline is the large volume of data of multidimensional nature that must be handled.
The concept of spectral index emerged as an idea to reduce the number of dimensions to one,
and thus facilitate the study of different features associated to the types of land cover categories
that exhibits a multispectral image. Formally, a spectral index is defined as a combination of
spectral bands whose function is to enhance the contribution of one type of land cover
mitigating the rest of covers. In this work a no-supervised methodology to analyze and discover
spectral indexes based on growing self-organizing neural network (GCS-Growing Cell
Structures) is presented.