One-Dimensional Kohonen Networks and Their Application to Automatic Classification of Images

This paper analyses the results obtained when different topologies of 1-Dimensional Kohonen Networks where used to classify color pictures taken to Popocatépetl Volcano (located in the State of Puebla, México, active and monitored since 1997). Due to the fact that volcano images needed to be correlated with other experiments in the research center where this was carried out, it was required to find out if classification was based on the intensity of the images or the topology of the pixels, that is, their connectivity. To do so, we addressed two approaches: one analyzing the results obtained when pixels in the images were permuted; the other using in the clusters generated by the network a novel metric previously tested to be invariant to topology of pixels, and comparing the results to the obtained when using the Euclidean distance. Index Terms – 1-Dimensional Kohonen Networks, non-supervised Image Classification, Metrics on Euclidean Spaces, Non-Supervised Classification.