A self-organization neuro-fuzzy network applied to a seismic signal classification problem

The marriage between neural networks and fuzzy logic systems can lead to improved systems in which the best of both worlds are combined, viz. learning capability as well as the ability to handle real-life ambiguities and uncertainties gracefully. In this contribution a heterogeneous neuro-fuzzy network is proposed as a solution for a problem relevant in the mining industry: the classification of seismic signals according to their generating sources. Different stages in the traditional fuzzy system are implemented in consecutive layers in the network, resulting in an architecture reminiscent of radial basis function networks. In contrast to some other neuro-fuzzy networks in which a rule-base is derived by the rule-elimination-algorithm, fuzzy rules are generated with the aid of a fuzzy adaptive resonance theory network. Besides leading to reduced training times, the proposed rule generation algorithm can result in a better understanding of the signals being classified