Neural nets classification techniques for the tutelage of cultural and natural resources

Themultispectrality of satellite images and the continuing improvement in the resolving power of the last generation of satellites makes satellite images useful not only for land cover classification but also for identification and classification tasks at a detail useful for archaeology, urban classification and relative risk assessment. The availability of multi-scalar and multi-spectral images asks for power tools for processing this kind of data. In this paper we investigate the use of unsupervised artificial neural networks like Self-Organizing Maps and supervised neural nets like Multi-LayerPerceptron for feature extraction and classification of satellite images. The study area is a coastal plain in the southern part of Italy, which is of high interest for the Cultural and Natural resources there preserved. Firstlywe employ theKohonen’s SelfOrganizingMap (SOM) as a strategy for an unsupervised analysis of an IKONOS scene, employing in our analysis both the spectral information from its four MultiSpectral channels together with the spatial information extracted from the Panchromatic band. Secondly, we implement a supervised strategy on the same dataset, based on two steps: (1) features extraction step, which exploits the previous unsupervised results, (2) the discrimination stepwhich uses a Multilayer Percepetron (MLP) neural network. The training of the Multlayer Perceptron is carried out using a training dataset which has been built mainly on the basis of information extracted by visual inspection of the Panchromatic image, but also using the results of the previous segmentation step over Ikonos data and other ancillary information available to us. 1 STUDYAREA, DATASET, PROCESSING TOOLS The study area is a coastal plain in the southern part of Italy, which is of high interest for theCultural andNatural resources there preserved. The area is located in the alluvial plain of the Salerno Gulf, more precisely on the left bank side of the Sele River. Within its boundaries is also situated an archaeological site, the ancient city of Poseidonia-Paestum, that preserves ruins dated 600 BC. Land use is primarily agricultural, but during the last sixty years an urbanization phenomenon has arisen, associated not only to farm practice, but also to the growing 643 GeoInformation in Europe, M.A. Gomarsca (ed.) 2007 Millpress, Netherlands, ISBN 9789059660618