Self-organizing map for hyperspectral image analysis

Abstract

In this paper we present a neural network methodology used for classifying an hyperspectral image referencied as Indian Pines. The network parameters (learning and neighborhood function) are adjusted using a test battery generated from the image, selecting the values that give the best robutness and discrimination capacity. The availity of ground truth allows us to intoduce a new stadistical measure to quantify the resulting classification accuracy. The results of this methodology show an accuracy of 80% in the classification. © Springer-Verlag Berlin Heidelberg 2001.