Abstract:
A hierarchical Self-Organizing Map has been developed for solving classification problems, where several measurements have been taken from one object. The algorithm will be explained with the use of the practical example where it is the aim to classify eggs according to their shell state. Broken eggs will be separated in this way from intact eggs. The classification architecture actually consists of two different SOMs. The first SOM clusters the data in an unsupervised way. Afterwards, the ordered activations of each object are collected and fed to the second SOM which actually clusters the ordered activations and assosiates them with a class, which is presented as a binary vector. This class-vector is assigned to every node in the second map as an output weight and it is learned with Kohonen's learning rule.