Unsupervised spectral pattern recognition by Self-Organizing Maps

This work consists of three main parts. In chapter 1 the Self-Organizing Maps (SOMs), proposed by T. Kohonen (1982), are analysed in their particular features. In order to apply the proposed SOM method to the analysis of geophysical data, the architecture and processes involved in the training of a SOM map are examined in detail. A useful and commonly used method to visualize a SOM map (the U-matrix, Uni ed distance matrix) is shown and a procedure is proposed for the automatic detection of clusters on the map. The second part of this work (chapters 2 and 3) aims to show the results of some applications of the proposed SOM method to analyse geophysical data. In particular, in chapter 2, the SOM process is used to study the dynamical regimes of volcanic systems, starting from the acquired tremor. In the third chapter the SOM method has been applied to the HVSR technique (or H/V spectral ratio or Nakamura's method) with the intent to improve this method that allows the identi cation of the fundamental frequency that characterizes the sedimentary deposits of a site in a cheap and relatively easy way. The third part of this work consists in an appendix that shows the scripts for the implementation of the proposed SOM method. The SOM process has been entirely implemented using the free software environment R (http://www.r-project.org/ ). Some packages were already available in the repositories of R software (http://cran.r-project.org/web/packages/ ) and were adapted to implement some parts of the process, some other parts have been originally developed from scratch.