A two phases neural processing techniques for automatic remote sensing sub-image classification

A neural network approach for remote sensing image data classification is proposed. The basic characteristic of neural networks to allow the extraction of informations from a great amounts of data has been used to simultaneously handle image portions from the available bands representing each image portion. To guarantee good generalization performance and rapid convergence, in the present paper a two phase neural procedure is proposed. In the former a feedforward neural network is used to obtain an image data compression. In the latter a neural classifier is trained on the compressed data. The proposed procedure allows the virtual elimination of convergence problems, while conside-rably reducing the overall computation cost; moreover it sensibly betters the classification performance of samples not contained in the training set. The results of several tests are reported.