F. Roli and S. Vitulano (Eds.): ICIAP 2005, LNCS 3617, pp. 1190 – 1197, 2005. © Springer-Verlag Berlin Heidelberg 2005 An Application of Neural and Probabilistic Unsupervised Methods to Environmental Factor Analysis of Multi-spectral Images Luca Pugliese 1 , Silvia Scarpetta 3 , Anna Esposito 1,2 , and Maria Marinaro 1,3 1 IIASS, Istituto Internazionale per gli Alti Studi Scientifici "E.R.Caianiello", Via G.Pellegrino, 19 – Vietri sul Mare - Salerno {iiass.luca@tiscali.it, iiass.annaesp@tin.it 2 Dipartimento di Psicologia, Seconda Università di Napoli, Via Vivaldi 43, Caserta 3 Dipartimento di Fisica "E.R.Caianiello", Università degli Studi di Salerno, Via S.Allende, Salerno, Italy and INFM and INFN Sezione di Salerno, Italy silvia@sa.infn.it, marinaro@sa.infn.it Abstract. In this paper we test the performance of two unsupervised clustering strategies for the analysis of LANDSAT multispectral images of the Temples of Paestum Area in Italy. The classification goal is to identify environmental fac- tors (soils, vegetation types, water) on the images, exploiting the features of the seven LANDSAT spectral bands. The first strategy is a fast migrating means technique based on a Maximum Likelihood Principle (ISOCLUST algorithm), and the second is the Kohonen Self Organizing Map (SOM) neural network. The advantage of using the SOM algorithm is that both the information on classes and the similarity between the classes are obtained (since proximity cor- responds to similarity among neurons). By exploiting the information on class similarity it was possible to automatically colour each cluster identified by the net (assigning a specific colour to each of them) thus facilitating a successive photo-interpretation. 1 Introduction The analysis of remotely sensed multispectral data is of great interest for improving the knowledge of the Earth surface and remarkably contributes to the development of policies for planning and monitoring environmental resources [12]. The standard approach to the analysis of such images involves the grouping of image data into a finite number of discrete clusters or classes that identify the distribution, over the land, of environmental factors such as soils, vegetation, urban areas, and rivers. For several decades, such clustering has been implemented using classical statistical ap- proaches, mostly based on the Maximum Likelihood Principle (MLP) [1-2], assuming that clusters can be modelled as a multivariate normal distribution. However, geo- graphical phenomena are not randomly distributed in nature and are not always dis- played in the image with a normal distribution. Therefore, other methods have been suggested to overcome their limitations, among which Artificial Neural Networks (ANN). Recent developments in the field have shown that supervised NN algorithms