A clustering algorithm of the underwater acoustic signal with the dynamic adjustable number of clusters

This paper proposes a modifed K-MEANS training algorithm with the dynamic adjustable number of clusters.The algorithm trains the clustering centres and the number of clusters under the supervision of priori labels of samples.Therefore,it can effectively jump out of a local minimum for the squared error cost function and decrease the overlap between patterns.The clustering efficiency of the new algorithm has been demonstrated by clustering experiment of the real underwater sound signal.