Machine Extraction of Landforms from Multispectral Images Using Texture and Neural Methods

This study makes an attempt towards machine generation of landform maps from optical remote sensing data. The automation in our approach is to the extent that the training of multilayer perceptrons (MLP) used as a classifier is carried out offline, and subsequently the trained MLP is used to identify the landform classes in a given unknown satellite image. Emphasis of this paper is on exploring potential of gray level co-occurrence (GLC) texture statistics computed from a reasonably extensive database created using multispectral images for landform discrimination. GLC texture statistics form the feature of the pattern vector used for training the MLP. Generalization results are assessed using the cross validation mechanism. Our results for aeolian landforms suggest the textural method to be promising in machine extraction of landforms