An improved neural network technique for data dimensionality reduction in satellite imagery

摘要

This paper presents an application of back-propagation neural network based mapping scheme of multispectrale data images.udThe approach exploits the ability of neural networks for non-linear projection of multidimensional data, and their advantagesudover traditional methods. An updating rule for this network, based on the Conjugate Gradient Algorithm is used. The mainudadvantage of this algorithm is the speedup of the convergence rate. Performance evaluation using a Landsat image of Kénitraudregion (Morocco) is carried out. Classification results of the proposed algorithm outperform those obtained using conventionaludmethods.